You are on page 1of 83

Page 1

Practical Rf Engineering
Davood Mirzahosseini
o vaD od Miro haz sseini

Page 2

A Practical Guide to the Principles of Radio Design.
There are many books on the subject of radio design, but the majority of them are either too basic and do not really give enough information, or they immediately launch into advanced theory that, unless you are a mathematical genius, is just confusing and unhelpful. Of course some basic maths is mandatory in engineering and, where this is appropriate, I shall quote the relevant formulae and expressions - but only in their final practical form as engineering tools. Let the academics concern themselves with theory. CAD is now considered to be an essential design tool in the electronics industry, but it is no substitute for basic understanding and knowledge. It may tell you if your design will work, but not if it was an appropriate choice in the first place, nor how to properly test and debug your finished design.

reference book : 1.Rf Circuit Design,Chris Bowick,1997 2.Radio Frequency Circuit Design,W. Alan Davis, Krishna K. Agarwal, 2001 3.Circuit Design for Rf Transceivers,D.Leenaers,j.V.D.Tang,C.S.Vaucher,2001 4.Rf and Microwave communication Circuits,Devendra K. Misra,2001 5.Rf and Microwave Handbook,Mike Golio,2001 6.Advanced Techniques in Rf Power Amplifier Design,Steve C.Cripps,2002 7.Practical Rf Handbook,Ian Hickman,2002 8.Circuit and Systems for Wireless Communications,M.Helfenstein,G.S.Moschytz,2002 9.Practical Rf System Design,William Egan,2003 10.Modern Receiver front-ends,j.Laskar,B.Matinpour,S.Chakraborty,2004 11.Short-range Wireless Communication,Alen Bensky,2004 12.The Design of Modern Microwave Oscillators for Wireless ,U.L.Rohde,A.K.Podder,G.Bock,2005

Page 3 The book is divided into three parts, as follows: Part 1 The Receiver Front End Page
4 5 6 7 12 23 26

Contents Receiver Noise Antennas Front-end Filters The LNA Bias circuits, decoupling and layout Measurement techniques

Contents Mixers The Local Oscillator The Phase Locked Loop IF Filters IF Amplifiers Direct Conversion Measurement techniques

Part 2 Mixers, Local Oscillators and the IF

29 30 33 39 46 49 53 54

Part 3 Transmitters & Modulation Techniques

Contents Propagation Modulation VCOs Frequency Multipliers Up Converters Driver Stages The Power Amplifier Linearity Techniques and Spectral Purity PIN Switches Isolators and Circulators Directional Coupler Power Supply management Measurement techniques Appendices

56 57 58 63 64 64 65 66 70 74 75 76 78 79 83

About the author: Davood Mirzahosseini I have had more than 10 years experience as an RF design engineer, the last 5 years as an RF Design Specialist working on new designs. I have worked for some of the leading companies in the field of communications. My expertise has included the design of low-noise amplifiers,Mixer,VCO, phase-locked loops, RF amplifier and power amplifier,Transmitter and Receiver for wireless communication.

KNTU Edition . Davood Mirzahosseini - Copyright 2007.

Page 4

Practical Rf Engineering Part I

By Davood Mirzahosseini, KNTU.
Copyright 2007.

Receiver Noise Antennas Front-end Filters The PIN switch The Dielectric Resonator Filter The Helical Resonator Filter L-C Filters Microstripline Filters The Duplexer/Diplexer Filter Coaxial Cavity Filters The LNA LNA Design Optimum Noise match. Stability Factor Third-order Intercept Point. (IP3). Practical LNA Balanced LNA. Bias circuits, decoupling and layout Simple bias Active bias Decoupling and layout Measurement techniques 5 6 7 7 8 9 9 10 10 11 12 12 15 16 18 20 21 23 23 23 24 26

Page 5

The Receiver Front End

1.1. Receiver Noise At frequencies below 30MHz, noise is predominantly atmospheric and environmental. The noise contribution from the first amplifier stage is therefore relatively unimportant. As the frequency reaches the VHF and UHF bands however, the first amplifier and any preceding filters become the dominant noise source in the receiver chain. If we consider an ideal noise-free receiver, then the absolute sensitivity is limited by the theoretical noise-floor. This is determined by the expression Pn = kTB (where k is Boltzmanns constant, T is temperature in degrees Kelvin and B is the system bandwidth) and, at a normal ambient temperature of 17C, approximates to -174dBm in a 1Hz bandwidth. Absolute noise floor at 17C = -174dBm/Hz From the above expression, it may be seen that the absolute noise floor of any receiver system is defined by the final signal bandwidth (i.e. system bandwidth at the demodulator). Also, in any practical receiver, the receiver itself will add to the noise and its Noise Figure must be added directly to the above figure. Noise Figure is defined as: Noise Figure (NF) = 10 x log [S/N ratio at the input / S/N ratio at the output]. For a complete receiver system, this is more usefully defined as: Noise Figure (NF) = (actual S/N dB at the output - S/N dB for a noise-free receiver). [Equ. 1] If we consider the fact that a 3dB noise figure equates to a 30% loss in maximum line-of-sight range, it may readily be seen that it is very important to keep the receiver noise figure to an absolute minimum. The overall performance of a receiver system is usually defined by its minimum detectable signal, or Mds and is this is given by the following expression: Mds = -174 + NF + 10 (log BW), where BW is system bandwidth in Hz. Hence, to calculate the Mds for a given receiver, see the following example: A receiver has a system bandwidth of 200kHz and a noise figure of 5dB. From equation 2 above: Mds = -174 + 5 + 10 (log of 200,000) = -174 + 5 + 53 = -116dBm. Note that this is not the same as the receiver sensitivity. In any practical system, the demodulator requires a minimum signal/noise ratio. In an analogue system, this signal/noise is generally expressed as SINAD (Signal/Noise and Distortion) and is typically about 12dB, but for digital systems it is more usually defined as carrier/noise ratio (C/N) and is typically about 9dB. In the above example, the sensitivity of a practical (digital) receiver is therefore given by: Receiver sensitivity = -116 + 9 = -107dBm. [Equ. 2]

Page 6 1.2. Antennas. Antenna design is a highly specialised field that is of limited interest, except to those who work in it. However, a section on receiver front-ends cannot be entirely complete without some mention of antennas. I shall therefore restrict this discussion to VHF/UHF antennas, where the dimensions are comparable to a half or quarter wavelength at the frequency of operation and the matching impedance is (usually) 50. For ideal (line-of-sight) propagation, the range versus transmitter power follows an inverse square law. This means that in order to double the effective range, the transmitter power must increase by a factor of four. Another factor to be considered is that the free-space path loss increases with the operating frequency and hence, for a given transmitter power, the maximum range is limited by the absolute receiver sensitivity. For fixed installations, the range can be significantly improved by the use of a high-gain directional antenna. The natural impedance of a half-wave dipole is approximately 75. Historically, the impedance used for domestic TV receivers and VHF broadcast radio has always been 75, which conveniently matches to a dipole. This means that a quarter-wave monopole would have an impedance of 37.5. In fact, because the preferred working impedance is 50, a monopole antenna is deliberately made electrically shorter than /4 to increase its impedance. Antenna gain is generally measured in terms of dBi, which means the relative gain (in dB) over an isotropic radiator (i.e. a radiator having a uniform field in both the horizontal and vertical planes). For example, a quarter-wave monopole antenna with a ground plane has a gain of about 2.5dBi, whereas a multiple element UHF Yagi type antenna can have a forward gain of more than 18dBi. This extra gain is free - that is it consumes no power and contributes no additional noise, the only penalty being in the size and complexity of the antenna. Fixed installations frequently use co-linear antennas (effectively, several resonant elements connected in series to increase gain), or phased arrays. The latter, as the name implies, comprises a number of separate antennas, each driven in phase by using carefully adjusted lengths of transmission line. The great benefit of the phased array is that its polar diagram can be steered by adjusting the relative phase of the driven elements, changing its properties between omni-directional and directional. For higher frequencies, the entire phased array can be constructed on a single printed-circuit board using patch antennas and with microstripline elements providing the required phase shifts. Using these techniques, it is possible to achieve antenna gains of more than 23dBi. For mobile systems, where omni-directional properties are required, it is rare to achieve an antenna gain of more than 5dBi and, for small hand-held equipment such as mobile phones, the gain can be less than unity. Fig. 1.1. A double Yagi antenna for 1.27GHz. This antenna has a forward gain of 21dBi.

Page 7 1.3. Front-end Filters. A hand-held device will almost invariably be a transceiver; that is to say that it must be capable of transmitting and receiving with the same (integral) antenna. This also applies to most vehiclemounted (PMR) equipment. For fixed installations it may be possible to have separate antennas, but if these are co-sited (i.e. in close physical proximity to one another), it will still be necessary to prevent the transmitter signal from swamping the receiver. In all of these cases, some form of switching and/or filtering will be necessary between the antenna and the transmit/receive paths. This switch/filter will have an insertion loss and, because it must come before any amplifier stage, it will directly add to the noise figure of the receiver. The mode of operation may be either: Simplex the same frequency is used for both transmit and receive functions. These cannot take place simultaneously, but are sequential by means of a press-to-talk or a voice-activated switch. Connection to the antenna will usually be via a PIN switch (or relay) and a band-pass filter. Diplex transmit and receive functions are simultaneous, using different frequencies and with filters to separate transmitter and receiver paths. * There is some confusion about the distinction between duplexers and diplexers. I prefer to define them in the way given here, but others may disagree. Duplex this is a combination of simplex and diplex, specifically suited to digital modulation systems such as TDMA, where transmit and receive functions are confined to sequential time slots, but are also at different frequencies and are switched continuously at a rapid rate. The modulation is time-compressed during transmission and time-expanded on reception to give effective continuous duplex operation. In making a decision on what type of filter is best for your application, you will need to consider the following factors: a). What is the operating frequency of your receiver? This will limit the choices available to you. b). Is it a frequency band that is already widely used? There are many inexpensive off-the-shelf dielectric resonator filters and diplexers for the popular frequency bands. c). Is physical size an important consideration? Dependent on your operating frequency, this will limit your choices. d). Will the receiver be required to operate in an environment where there are high level out-of-band signals? In a duplex system, the duplexer/diplexer acts as a filter and any additional filter is best placed after the first amplifier stage (LNA). Remember that any insertion loss at the input will directly add to your noise figure. The various types of input filters/duplexers are described below: 1.3.1. The PIN switch. In HF man-pack radios, the antenna will be electrically much shorter than /4 and therefore a high impedance. In order to produce the same transmitter power, an impedance transformer is used to match the antenna, producing a much higher RF voltage. At frequencies below about 5MHz, the RF voltage can be several kilovolts and here the relay reigns supreme. For VHF/UHF equipment, the PIN switch has generally replaced the transmit/receive relay, due to its much smaller size. PIN switches are most commonly found in simplex systems (i.e. where transmit and receive functions do not occur simultaneously). A good PIN switch should have an insertion loss of typically less than 0.5dB. PIN switches will be discussed in further detail in Part Three of this book.

Page 8 1.3.2. The Dielectric Resonator Filter. At frequencies in the 800MHz to 3GHz region the most common type of filter is the ceramic dielectric resonator type, due to its compact size and relatively high Q. This type of filter usually comprises three or four edge-coupled resonators to give a band-pass response with low ripple in the pass-band. As with any filter, the number of elements (poles) determines the steepness of the skirts and the degree of out-of-band rejection, but more poles inevitably means greater insertion loss. For a front-end filter, the insertion loss will typically be in the range 2 -3dB, with the aim to keep this as low as possible. Below 800MHz the physical length of the resonators becomes rather large and they therefore become less attractive. Conversely, at frequencies beyond 3GHz, the very short physical length makes their construction impractical. The response of typical dielectric resonator filter is shown in Fig. 1.2. below, with an actual filter shown in Fig. 1.3.

Fig. 1.2. Response of a typical 3-pole Dielectric Resonator Filter.

Fig. 1.3. A 4-pole Dielectric Resonator Filter for 2.4GHz, mounted on a pcb.

Page 9 1.3.3. The Helical Resonator Filter. Another type of filter that can be useful in the 200MHz to 1GHz range is the helical resonator type. This comprises an open-ended coil (or helix) inside an enclosed air cavity and is edge-coupled to the next cavity through a slot in the case wall. Usually this type offers a limited range of adjustment, by means of a metal screw core. At the lower frequencies, this can be smaller and cheaper than a comparable dielectric resonator type but, due to the decreasing physical dimensions, is impractical above about 1.5GHz. Insertion loss is higher than a comparable dielectric resonator type. A typical 3-pole surface-mount filter is shown in Fig. 1.4. below.

Fig. 1.4. A 3-pole surface-mount Helical Filter in the 5CHT range by Toko. Available range 395 - 510MHz .

1.3.4. L-C Filters. Below 400MHz it is more likely that a conventional L-C filter will be used at the input. As the frequency is reduced below 200MHz, the noise figure becomes less significant by comparison with interference and environmental noise and so higher losses at the input may be tolerated. For fixed frequency applications, a simple band-pass filter as shown in Fig. 1.5. may be used. For a repeatable design, 5% inductors and 2% capacitors are required. If the receiver is to operate over a wide range of frequencies, all or part of the tuning capacitors could be replaced by varicap tuning diodes.

Fig. 1.5. Schematic of a typical 2-pole Band-Pass Filter for 440MHz.

More complex filters with up to 4-poles may be required for certain applications such as PMR sets. This type of equipment needs to operate in a high RF noise environment and here sensitivity is traded for selectivity.

Page 10 1.3.5. Microstripline Filters. It is also possible to design filters using microstrip lines, but below about 3GHz these will occupy a large board area and, on FR4 substrate material, the Q will be low with a high insertion loss. Beyond 3GHz, it becomes mandatory to use a superior substrate material such as Teflon or Alumina. This is fortunate because, as previously described, most other types of filter cease to be viable at these frequencies. As it becomes more difficult to meet the gain and noise figure requirements, the filter is usually placed after the LNA, leaving the input broadband. For fixed installations, such as a satellite TV receiver, the antenna itself is frequency selective and limits the out-of-band response. 1.3.6. The Duplexer/Diplexer Filter. For a duplexer (or diplexer), it is common to combine the filters for receive and transmit paths in a single integrated block. This can be a low-pass and a high-pass filter, with a crossover point midway between the transmit and receive frequencies, or it may be a band-pass filter on the receiver side with a band-stop (notch) at receiver frequency on the transmit side. Two dielectric resonator type duplexers are shown in Figures 1.6. and 1.7. below.

Fig. 1.6. An extremely compact Duplexer from Toko for the 2GHz band , shown on a golf tee for size comparison.

Fig. 1.7. A 900MHz high power (10W) Duplexer from TDK. This has a total of 11 resonator elements, 5 in the transmit path and 6 in the receiver path.

Page 11 1.3.7. Coaxial Cavity Filters. For base-stations and other applications where a very high performance is required, the coaxial cavity filter can handle high RF power, gives a much higher Q with lower insertion loss and has good outof-band rejection, but is physically large and very expensive. A duplexer filter will probably have four or more cavities (poles) on each side, with adjustment screws for fine tuning (see Fig. 1.8). A typical filter of this type is milled from a solid brass block and then silver-plated for low RF loss. Adjustment screws allow precise adjustment for an optimally flat response and minimum loss in the pass-band. The filter response will be carefully trimmed by a skilled engineer, using a network analyser and, once set, the adjustment screws are locked in place.

Fig. 1.8. A cavity duplexer for the 900MHz GSM band. This filter has an insertion loss of less than 0.7dB and better than 40dB isolation.

1.3.8. Waveguide Filters. The waveguide filter is really an extension of the cavity filter, but specifically applied to frequencies above 5GHz where the cavities become very small. As with the cavity filter, these are very expensive and would only be used for highly specialised applications. It has been included here only to complete the list of filter types.

Fig. 1.9. A typical waveguide filter for rectangular waveguide.

1.3.9. SAW (Surface Acoustic Wave) Filters. SAW filters are now being used in the front-end of GSM and WCDMA systems and will be covered in Part Two, under IF filters.

Page 12 1.4. The LNA (Low Noise Amplifier). The LNA is the first amplifier device in the receiver chain and, as the name implies, its function is to amplify the incoming signal to a level that is well above the noise threshold of subsequent stages, without significantly adding to that noise itself. As described in Section 1.3, the passive circuits at the input directly add to the overall noise figure and it is very important to keep these losses to an absolute minimum. 1.4.1. LNA Design. A good LNA design should meet the following requirements: High gain (>15dB). Low noise figure (<2dB). Unconditionally stable. Good input return loss (<10dB). High third-order intercept point *(see section 1.4.4.) In addition, for use in a hand set, the following are also important: Low supply voltage (<3V). Low current consumption (<10mA). Small footprint. Low cost. Unfortunately, these are not always mutually compatible and a good design must balance the tradeoffs against the requirements for the particular application. When considering the design of an LNA, it is first necessary to choose the active device. For a handset, this will probably be a bipolar transistor, although other types are available. Suitable devices are clearly labelled low-noise in the manufacturers data sheets . However, it is necessary to look beyond this - the device must also meet other important criteria, as listed above. Suitable devices fall into 3 main categories: 1. Bipolar transistor. Examples: HBFP-0420, BFP420, BFG425W 2. FET/PHEMT (pseudomorphic high-electron-mobility transistor). Examples: ATF34143, CFH800 3. MMIC (monolithic microwave integrated circuit). Examples: MGA-53543, CF750 Generally, the PHEMT (there are variations on this, such as e-HEMT) devices have a lower noise figure and higher gain than bipolar types, but at the expense of higher current. Most, but not all, also need a negative bias supply. MMICs can offer a quick and convenient solution, requiring little or no design effort for a predictable result, but they are invariably more expensive than discrete types. The manufacturers data sheet will indicate the optimum operating voltage and current to give the specified noise figure (see Fig.1.10 on the following page), but it will be necessary to make certain trade-offs. The MGA-53543 MMIC offered by Agilent is suggested for WCDMA applications, but at 5V and 53mA, it is unlikely to be seriously considered as the LNA in battery powered equipment. True, this device has a really excellent IP3, but the trade-off is too great. For mains-powered or vehicle-mounted equipment, current is less likely to be an issue, but cost may be important. FAQ 1. What topology should I use for my LNA bipolar, FET/PHEMT or MMIC? This will depend on cost and power supply considerations, as well as the application. FET/PHEMT devices usually have a lower noise figure and higher IP3, but are more expensive and may require a negative supply for the bias circuitry. MMICs require very little design effort, but are usually the most expensive solution. Some MMICs offer additional functionality in the same package (e.g. LNA + mixer or two LNA stages), which may be an attractive option.

Page 13

Fig. 1.10. Typical noise figure plot for a low-noise bipolar transistor (Infineon BFP420).

In a good design, the LNA and input circuits should be the dominant factor in the overall noise figure of the receiver. In order to achieve this, the gain of the LNA device needs to be as high as possible The gain budget diagrams in Figs 1.11 and 1.12 below, illustrate this point. The input duplexer filter has a fixed loss of 3dB before the LNA, which itself has a noise figure of 1.5dB. The overall noise figure for the LNA is therefore 4.5dB. In a typical receiver architecture, this might be followed by a band-pass filter, a mixer and an IF filter with a combined insertion loss of 12dB, before reaching the first IF gain stage, in this case with 26dB gain and a 2dB noise figure. Fig. 1.11. Gain budget diagram with 20dB LNA gain, showing overall noise figure of 5.18dB

Fig. 1.12. Gain budget diagram with 12dB LNA gain, showing overall noise figure of 7.68dB

Page 14 Note that in Fig. 1.11 the LNA has a gain of 20dB and the overall degradation in noise figure is only 0.68dB (from 4.5dB to 5.18dB). In Fig. 1.12 nothing has changed except the gain of the LNA, but note that with only 12dB gain, the noise figure is degraded by more than 3dB (from 4.5dB to 7.68dB). The above examples have been computed using a simple, but very useful CAD program called SYSCALC (Arden Technologies Inc), but if you want to work it out the hard way it is necessary to use the expression for cascaded noise factor: F = F1 + (F2-1)/G1 + (F3-1)/G1.G2 + (F4-1)/G1.G2.G3 +.etc, [Equ. 3]

where F is the total noise factor, F1, F2..etc are the noise contributions of successive elements, and G1, G2..etc are the gain values of successive elements. But Take Care - these are all ratios, not dB and each passive element must be treated as if it were an amplifier with a fractional gain. See example. Example: Using the line-up in Fig.8, we have: F1 = 2 the ratio of 3dB F2 = 1.4 the ratio of 1.5dB F3 = 15.8 the ratio of 12dB (the 3 passive elements in series) F4 = 1.6 the ratio of 2dB G1 = 0.5 for 3dB loss G2 = 15.8 for 12dB gain G3 = 0.063 for 12dB loss Entering these figures in the expression for cascaded noise factor: F = 2 + (1.4 -1)/0.5 +(15.8 -1)/0.5x15.8 + (1.6 - 1)/0.5x15.8x0.063 Hence, F = 2 + 0.8 + 1.87 + 1.2 = 5.87 To convert this back to a dB figure: NF = 10 x log 5.87 = 7.68dB * which agrees with the result in Fig.1.11.

At frequencies above about 3GHz, it is often not possible to obtain sufficient gain with a single LNA stage and two or more stages may be needed. To limit broadband response and to ease the blocking requirement, it is common practice to include a band-pass filter between the first and second stages, but it is important to keep the insertion loss of this filter as low as possible so that the first stage is still the dominant noise source (see Fig. 1.13 below).

Fig. 1.13. Gain budget diagram showing overall noise figure for 2-stage LNA of 5. 09dB. In the above example, the first LNA has a gain of 12dB, followed by a filter with 3dB insertion loss and then a second LNA stage. The overall gain is therefore (12 - 3) + 12 = 21dB. Note that the noise figure is still dominated by the first stage alone, subsequent stages adding only 0.59dB. This configuration would be used where the gain of a single device is inadequate.

Page 15 FAQ 2. How much gain do I need in my LNA? Remember, in order to minimise the effect of noise in later stages, the overall gain (including the insertion loss at the input) should exceed about 15dB, although this can be less if the device is followed by an active mixer (i.e. having positive gain rather than a 6dB insertion loss). If it is not possible to get sufficient gain from a single stage, you may need a second LNA stage. 1.3.2. Optimum Noise match. For minimum noise figure, the LNA device should be matched for its optimum noise impedance rather than the system input impedance (usually 50 Ohms). These are NOT the same, although for the best devices they are fairly close and, through the careful use of feedback it is possible to make them even closer. In reality, the difference in the overall noise figure is likely to be quite small and, for many applications, it is probably better to settle for a slightly higher noise figure in order to present a good return loss at the input. Moreover, the condition for minimum noise is frequently not compatible with a good stability factor, especially at frequencies outside the pass-band and the designer needs to ensure that the LNA does not become an oscillator (see section 1.4.3). In the noise figure plot shown in Fig. 1.10, it may be seen that for the Infineon BFP420 device, the difference between the noise figure for a 50 match and the noise match (Zsopt) is less than 0.2dB. Where a duplexer with a typical insertion loss of 3dB precedes the LNA, there is very little benefit to be gained from saving just 0.2dB in a 4.5dB overall noise figure. If, on the other hand, the specification demands a noise figure of better than 1.5dB say, as might be required for a base-station with a low-loss, high Q input filter, then even 0.2dB is significant. For such a demanding application, however, a balanced LNA can give real benefits (see Fig. 1.22), as will be discussed in section 1.4.6. FAQ 3. How can I decide what is an acceptable noise figure for the active device in my LNA? The system specification will usually require a minimum sensitivity for a fixed bit-error rate (BER), or a Sinad figure for an analogue system. Using the relationships given in section 1.1, determine the maximum system noise figure that will meet this requirement. Now take the worst-case figure for insertion loss of any circuit elements before the LNA (duplexers, filters, PIN switches, etc) and subtract this from your maximum system noise figure. It is assumed that your LNA will have sufficient gain to make it the dominant noise source and, to allow for some contribution from later stages, subtract a further 1dB. The resultant figure is the maximum noise figure that is acceptable for the active device. Example: A receiver requires a minimum sensitivity of -105dBm for a 0.02% BER and a C/N of 9dB at the demodulator to achieve this result. From the figures in section 1.1, we see that the maximum system noise must be less than 7dB. If the worst-case insertion loss for the duplexer is 3dB and we allow a further 1dB for noise contribution in later stages, the maximum LNA device noise figure must be less than 3dB. To allow for tolerances and some degradation at elevated temperatures, we would choose a device with a noise figure better than 2dB. FAQ 4. Do I need to match my LNA for optimum noise? If the insertion loss of any input filters or switching circuits exceeds 3dB, then it is probably not worthwhile to match for optimum noise. It is assumed that the system impedance is 50 and, unless the optimum noise match is already close to 50, a mismatch will result in poor return loss and can lead to instability. For a high-quality system where the input losses are less than 1dB, a noise match can be worthwhile. For the best possible performance and where cost and size are unimportant, the balanced LNA configuration described in section 1.4.6. should be considered.

Page 16 1.4.3. Stability Factor. Ideally, your LNA should be unconditionally stable - that is, it will not become unstable with any combination of impedances at the input and output. This condition should be checked over a wide frequency range and well beyond the frequency of operation of the LNA. It is quite common for an LNA design to appear stable at the operating frequency, but to become unstable (and oscillate) at a frequency in the high GHz range. There are several methods of measurement for stability, but one of most common is given by the Rollett Stability Factor, designated K, and this must always be greater than unity for an unconditionally stable design. Its value is given by: K = (1 + [S11.S12 - S21.S12]2 - S112 - S122)/2.S11.S12 * [Equ. 4]

But dont worry, we shall not even attempt to work this out numerically. Here is one case where CAD is more or less essential. Provided that you have modelled your LNA correctly, you can plot the K factor across a wide range of frequencies and check that it remains greater than unity. As a rough guide, 0.8 - 1 is not ideal but you can probably live with it, anything less than 0.8 is likely to give an unstable design with reactive mismatch, and a negative value means guaranteed oscillation! Whilst on the CAD, this will be a good opportunity to observe the effect of a small value resistor in series with the output and of a small emitter/source inductor/microstripline (see Figs 1.18 and 1.22). The effect of the series resistor (typically 10 -22) is to improve the stability factor with only a small loss of gain in band. A similar effect can be achieved by a resistor across the output shunt inductor; in this case typically 220 - 680. Fig. 1.14. The two plots shown are for an actual LNA using the Agilent ATF34143 PHEMT device, with Vd-s = 3V and Id = 40mA K factor and gain versus series resistor for values of 0, 10 and 22 ,with 2 x 2mm microstriplines in the source leads. At least 15 is needed to bring the K factor to greater than unity in the passband.




Return loss versus series resistor for values of 0, 10 and 22 , with 2mm microstriplines. Note that 22 is necessary to achieve the desired 10dB input return loss. The microstriplines in the source (two of them in parallel) are essential (see next page).



Page 17 Similarly, the source/emitter inductor not only improves K, but can also rotate S11 closer to Zopt. The effect of the latter is to improve both the noise figure and the input return loss, with only a small loss in gain. For the example using the ATF34143 on the previous page, the inductance is formed by two microstriplines just 1mm wide and 2mm long representing less than 0.5nH of inductance, and is more or less mandatory in this design. But be careful, too much inductance here will have the reverse effect and can actually cause instability (see Figs 1.15 and 1.16). Note that the dimensions given for the microstriplines assume a board thickness of 0.51mm. For thicker board, the ground via inductance will be significant and the lines should be modified or omitted. Fig. 1.15. K factor and gain with a 22 series resistor and 3mm microstriplines in the source. Note that the stability factor seems to be further improved, with only a small loss of gain. But - see Fig. 1.16.

Fig. 1.16. Effect on K at high out-of-band frequencies when line lengths are increased from 2 to 3mm. Note that the K factor changes abruptly from unconditionally stable to a large negative value (and certain oscillation) at 7GHz.

Another factor affecting the stability factor is the proper decoupling of the bias supply and the output load. It is not sufficient to decouple these points at the LNA operating frequency; they must also be decoupled for much lower frequencies to limit the LF response. It can be shown that the presence of large signals can cause inter-modulation products at low frequencies that modulate the bias supply, causing degradation in both the noise figure and the IP3 (more on this in section 1.4.4). Useful Tip 1: If the measured gain of your completed LNA is much lower than predicted, it is probably oscillating. Instability can often occur at very high frequencies and may not be seen on a spectrum analyser unless it is extended to its maximum range (you need at least a 7GHz spectrum analyser). Another clue is if the bias conditions are all wrong. Carefully probe the decoupling nodes, whilst watching the output, to discover if any of them is live. Review your layout and decide if additional decoupling or grounding is required. If you have used CAD for your design, check the stability factor over a wide band, well beyond your frequency of operation. But first make sure your imported S-parameters extend to the maximum frequency of your scan, as no CAD program can predict what will happen at frequencies for which it has no data.

Page 18 1.4.4. Third-order Intercept Point. (IP3) In section 1.1, it was shown that receiver sensitivity is determined by the noise floor and the overall noise figure. Another factor that can affect receiver sensitivity is intermodulation distortion (IMD). IMD is produced when there are two or more strong interfering signals at the input of a non-linear device. The action is similar to that of a mixer and produces outputs that have the relationship (2f1- f2), (2f2- f1), etc. For HF receivers, IP3 is much more important than noise figure. These are third-order products; that is for every 1dB increase in the input level the IMD output level increases by 3dB, and the greater the number of frequencies, the greater the number of third-order products. If there is a large number of interferers, the integrated effect will raise the noise floor of the receiver with a consequent loss of sensitivity. This is especially important in CDMA applications, where there can be a large number of signals present simultaneously in the input pass-band. The Third-order Intercept Point, IP3, is used as a convenient measure of the linearity of an amplifier stage (or stages) and hence the occurrence of intermodulation products. Note that the actual IP3 value is a theoretical concept only, since the amplifier response goes into compression well before it can reach IP3 (see Fig. 1.17). The example below shows how important it can be to have a high IP3 in any environment where large interfering signals may be encountered.
Input versus Third-Order Intercept Point (IP3)

Amplifier response, showing compression

Fig. 1.17. Plot of input power versus output power for an amplifier with 10dB gain, showing the third-order product (yellow trace) and its intercept with the extended first-order plot (violet trace).

First-order response


Third-order response

Output dBm





-10 -10





Input signal level dBm

Example: A receiver has a wanted input signal level of -50dBm at a frequency of 450MHz. The IP3 of the amplifier is +20dBm and its gain is 10dB (see Fig. 1.16). There are two interferers present, f1 = 446MHz and f2 = 448MHz. (2f2- f1) gives (896 - 446) = 450MHz (the same as the wanted frequency). In this example, the level of the wanted signal will be -40dBm at the output of the amplifier and it may be seen that the same level will be produced (and the receiver blocked) if the two interferers each has a level of -10dBm. By further inspection, it is apparent that interferers even at -30dBm level will block a wanted signal at -110dBm, still within the expected range of sensitivity for many receivers.

Page 19 The point at which the output is compressed by 1dB (i.e. amplifier gain falls by 1dB) is known as P1dB and, as a rule-of-thumb, IP3 is P1dB+10dB. IP3 is normally given as a figure for the output of an amplifier (OIP3). The input IP3 (IIP3) will be OIP3 minus the gain of the amplifier. The IP3 can often be improved by a deliberate mismatch at the device output (load-pull), but this can only be determined experimentally and must be offset against poor output return loss. It is important to maintain a good IP3 up to, and including the mixer; large offset signals beyond that point will be removed by the IF filter. With the advent of CDMA systems, it has become a challenge to the designer to design a front-end with the highest possible IP3 without consuming a prohibitively large amount of dc power. Useful Tip 2: As a rough approximation, the IP3 of an amplifier stage may be predicted by multiplying the dc input power by 5 and converting to an equivalent dBm figure. Example: An LNA has Vc-e = 4V and Ic = 10mA, so dc input power is 40mW. Multiply by 5 to get 200mW and convert this to dBm. IP3 = 10 x log 200 = 23dBm. (This is the output IP3. For input IP3, subtract the stage gain). 1.4.5. Band-Pass Filters. Although the LNA will have some selectivity, it will still have significant gain at frequencies well outside the pass-band. It is therefore usual to follow the LNA by a good band-pass filter to prevent these out of band signals from reaching the mixer, where their effect could seriously degrade the IMD performance of the receiver. The band-pass filter is normally either a dielectric resonator type or a helical resonator type, as already described in sections 1.3.2 and 1.3.3. Ideally, it should be just wide enough to encompass the pass-band, with a steep roll-off on either side to give maximum out-of-band rejection. As long as the LNA has adequate gain, an insertion loss of 3 - 4dB is not a problem. Occasionally, an L-C filter or a microstripline filter will be used instead, but both of these have their limitations. Due to the low associated Q of both the inductors and the capacitors, an L-C filter would not normally be considered at frequencies above 500MHz. Even at lower frequencies, the tolerances on the inductors and capacitors makes the response unpredictable unless adjustable inductors are used, and these are expensive and time-consuming to set up. Microstripline filters are a possibility at higher frequencies, but these too are unpredictable unless constructed on a stable, low loss substrate, such as alumina. Fig. 1.18 below shows the construction of a typical edge-coupled filter with a 200MHz wide pass-band, centred on 4.4GHz.

Fig. 1.18. A microstripline filter for 4.4GHz, constructed on an alumina substrate. The insertion loss in the pass-band is less than 2dB and the rejection is typically 20dB at 300MHz offset

Page 20 1.4.6. A Practical LNA. Although it is possible to design an LNA without the use of CAD, there is no doubt that the task is made much easier with a good simulation program such as Microwave Office, Serenade, ADS or Libra. Even some of the older programs, such as Super Compact or Touchstone can be used with good results. The free Appcad program is better than nothing, but is not recommended. Fig. 1.19. Schematic and gain plot for a 900MHz LNA.

Fig. 1.20. Return loss and Noise Figure plots for the above LNA.

Fig. 1.21. Stability Factor (K) for the above LNA.. Figs 1.19 and 1.20 above show a practical design for a 900MHz LNA, using the Infineon BFP420, and Fig.1.21 shows its Stability Factor. The active device is biased at 3V and 5mA, which is the point corresponding to its minimum noise figure. Note that this device requires an output shunt resistor and emitter microstriplines (there are two in parallel because this device has two emitter tabs) to give a stability factor greater than unity in the pass-band. A shunt resistor was found to give better results than a series resistor for this design. The gain at mid-band is 19.3dB. It may be seen that the input return loss is offset and only just meets the 10dB in-band target . This compromise is necessary to achieve the best noise figure of about 1.1dB.

Page 21 1.4.7. Balanced LNA. The balanced LNA, as the name implies, has two parallel paths for amplification and uses Wilkinson splitters/combiners or hybrid couplers to split and then re-combine the RF signal. A typical circuit is shown in Fig. 1.22 on the following page. The addition of the two quarter-wave transmission lines gives the balanced configuration special properties, as will be described. The operation of the balanced amplifier is frequently misunderstood. The input splitter, by definition, must give a 3dB insertion loss at the input of each amplifier. As with the single LNA, insertion losses at the input directly add to the noise figure and cannot be recovered. It might therefore be assumed that the balanced LNA will have a noise figure that is at least 3dB worse than the single LNA but, in fact, this is not so. The secret lies in the balanced configuration. By re-combining the outputs of the two amplifiers, the wanted RF signals are in phase and therefore add coherently to recover the 3dB loss at the input. But, the noise component is essentially noncoherent and therefore does not add at the output. We therefore have the same overall gain and noise figure as might be achieved with a single LNA. What then, is the point of it all? First, the IP3 of the balanced amplifier is improved by a factor of 3dB this is often important in the design of a base-station. Second, and perhaps more importantly, it may be shown that the inclusion of the /4 lines in alternate branches of the balanced pair, effectively cancels the return loss at both input and output ports. This means that the LNA devices can be matched for optimum noise (Zsopt) to obtain the best possible noise figure without the need to worry about poor return loss. In the example shown in Fig. 1.22, a return loss of better than 24dB (1.01:1 VSWR) was measured at both ports - almost a perfect match. The explanation for this is as follows: Consider an RF signal present at the input port. This is split equally to feed the two LNA stages but, due to the mismatch at the LNA device, a reflected signal (return loss) is returned to the input with a 180 phase-shift. But, for the path with the /4 line, the input signal is phase-shifted by 90 when it reaches the LNA and the reflected signal is phase-shifted by a further 90 on its return, making a full 360. Thus, the reflected signals from the two paths are in anti-phase and cancel one another. A similar effect occurs at the output port. Note that it is important that the difference in path length is 90 for the two LNA stages, if the optimum performance is to be realised. This means that the finite path length on the non phaseshifted side must also be added to the 90 phase-shifted side to maintain the 90 relationship. Advantages of the balanced LNA are as follows: 3dB improvement in IP3. May be matched for optimum noise to give improved noise figure. /4 lines give excellent VSWR at both input and output ports. Very stable design. Continues working even if one amplifier fails, although with 6dB loss in gain and 3dB increase in noise figure. This can be important where extremely reliable operation is required. Disadvantages of the balanced LNA are: Twice as many components, therefore higher cost. Requires much greater board area, especially for Wilkinson couplers and /4 lines. Twice the power consumption .

Page 22 Fig. 1.22. Schematic of a typical Balanced LNA design for 900MHz

The circuit in Fig.1.22 shows a design using Agilent ATF34143 PHEMT devices. For optimum noise and good IP3, both devices are operated at a current of 40mA (80mA total from the supply). Active biasing is used here because it is important to ensure that the current in the two halves is the same. The gate voltages need to be negative with respect to source and the bias networks are therefore returned to a 5V supply. The purpose of the two microstriplines in each of the source connections has already been described in section 1.4.3. Note that the dimensions of the lines are determined by the board thickness, in this case 0.51mm (one layer of a multi-layer board). For a thicker board, the inductance of the ground vias will increase the effective inductance and the lines should be shortened, or possibly omitted entirely. The series 12 resistors are also necessary with this device, in order to achieve a good Stability Factor. The plots shown in Figs 1.14, 1.15 and 1.16, section 1.4.3 are for the same device (ATF34143) used as a single LNA. It will be seen that the input return loss was only just meeting the 10dB requirement. With the balanced LNA design, figures of 26dB and 31dB were measured for input and output return loss, respectively. At 900MHz, the losses associated with the microstripline Wilkinson splitters/combiners on FR4 board are quite small. However, at 1,800MHz the use of a discrete hybrid coupler will give lower loss than the on-board Wilkinson and hence a benefit in noise figure.

Page 23 1.5. Bias circuits, decoupling and layout. 1.5.1. Simple bias circuit. Above about 500MHz it really is not a good idea to use emitter bias stabilisation, as is conventional at lower frequencies. It has been shown in section 1.4.3 that some emitter inductance is desirable for LNA applications, but it is difficult to control the inductance of discrete components and it is far better to ground the emitter, either directly or via a microstripline of defined length. This means that other methods of bias stabilisation must be found. The simple bias arrangement shown in Fig. 1.23 works quite well, provided that there is an adequate differential between the supply voltage and the required collector voltage on the transistor. If the value of resistor R1 becomes too low, then the dc negative feedback will no longer be able to control the working point of the transistor. With the values shown, this circuit gives acceptable performance over temperature and has the merit of extreme simplicity and virtually zero cost. Fig. 1.23. Simple bias circuit for Vc = 3V and Ic =5mA from a 5V supply. Transistor hfe is assumed to be approximately 100.

1.5.2. Active bias circuits The active bias circuit shown in Fig. 1.24 gives excellent control of the transistor working point over extreme variations in temperature and transistor hfe and, with the addition of a diode (Fig. 1.25) will work effectively with as little as 0.5V differential between supply voltage and the desired collector voltage. In this circuit, the PNP control transistor Q1 behaves as a voltage reference source, where its emitter voltage is determined by the ratio of R2 and R3 and is given by: Ve = Vs[R3/(R2 + R3)] + 0.7 where Vs is the supply voltage. The 0.7V must be added on to take account of Vb-e. It may be seen that the entire circuit acts as a constant current regulator, with the RF transistor current being set by the value of resistor R1 and given by: Ic = (Vs -Ve)/R1 Fig. 1.24. Active bias circuit for Vc = 3V and Ic =5mA from a 5V supply. In this example Ve is 2.95V, hence Ic = (5 -2.95)/0.39 = 5.25mA. Note that resistor R4 is included only to allow proper decoupling of the bias supply and, provided that its value is low, it will not affect the bias conditions. [Equ. 5]


Page 24 Figures 1.25 and 1.26 below show variations on the basic circuit for active bias. In order to operate the RF transistor at a higher voltage, the voltage drop across R1 becomes quite small. The Vb-e of transistor Q1 changes at a rate of approximately 2mV/C over temperature and, for 50C change in temperature this amounts to 100mV. If the voltage drop across R1 is only 0.5V, this represents a 20% change in voltage and hence in transistor current (bad). In Fig. 1.24 a diode is added in series with R2. As the diode voltage drop is also temperature dependent, this acts in the same direction as the Vb-e drop and compensates for it, greatly reducing the effects of temperature. The circuit of Fig. 1.26. shows a further improvement. Here, a dual current mirror transistor is used (Philips BCV62) and, because the two halves are perfectly matched, the temperature tracking is very good. Note that in both cases, the value of R2 must be adjusted to allow for the additional 0.7V drop.

Fig. 1.25. Active bias with temperature compensating diode.

Fig. 1.26. Active bias with current mirror transistor.

Where it is important to precisely define the operating conditions of a transistor (or FET), active biasing is essential. For a PHEMT device, an additional resistor would be included from the junction of R4 and the bias inductor to a negative supply (as shown in Fig. 1.22), thereby allowing the device gate to operate with a negative voltage. The main disadvantages of active biasing are the additional board space required and increased cost. 1.5.3. Decoupling and layout. All discrete components have an associated series inductance and a shunt capacitance. At frequencies below about 200MHz, these are relatively insignificant and can be ignored, but at higher frequencies they must be accounted for. For optimum RF decoupling, a capacitor value should be chosen such that it is self-resonant at the frequency of interest (i.e. its series inductance is exactly cancelled by its capacitance value to give a virtual short-circuit). Generally speaking, wire-ended capacitors are not suitable for frequencies above 200MHz, because the associated lead inductance is too high. A typical 0603 SMD chip capacitor has a series inductance value of approximately 0.7nH. Using the expression for resonance, C = 1/w2L , with L = 0.7nH we find that: 220pF is the nearest standard value for self-resonance at 440MHz. 47pF is the nearest standard value for self-resonance at 900MHz. 10pF is the nearest standard value for self-resonance at 1,800MHz.

Page 25 Also, it is not generally understood that the effective value of a capacitor increases with frequency. This is because the inductive reactance (vectorj+) appears in series with the capacitive reactance (vectorj-) to give a modified value. For example, a 10pF capacitor has a nominal reactance of 16 at 1GHz. If its lead inductance is 1nH, this represents a nominal reactance of 6. By vector addition, the actual reactance is 6 + (-16) = -j10 ....equivalent to a 16pF capacitor! Thus, the 10pF capacitor is effectively changed to 16pF at 1GHz. Note that at 2GHz, this same capacitor will cease to be a capacitor at all. 12 + (-8) = +j4 ....which is entirely inductive. At higher frequencies, it becomes increasingly important to ensure good grounding of and around the active device. Ground vias need to be as close as possible to the components they are grounding and it should always be remembered that even a via has a series inductance value of (typically) 0.1nH. This can be significant at frequencies above 1GHz and multiple vias are often needed to reduce the impedance. Useful Tip 3: Avoid the use of unnecessary ground plane on the top (active) layer of the pcb. Unless this is specifically required as part of a coplanar transmission line, it only increases the probability of unwanted signals being coupled from other parts of the circuit. All ground connections should take the shortest route to the internal (or bottom layer) ground plane using vias, or multiple vias. Above 800MHz, the use of R-C emitter (or source) biasing should be strictly avoided. The associated parasitic inductance and capacitance of these components is virtually guaranteed to cause instability. Any un-decoupled emitter resistors will seriously degrade the noise figure. Remember that all of the connecting tracks on your printed circuit board will act, to some degree, as microstrip transmission lines. Even a short length of track can significantly affect the impedance at the point to which it is connected. If in doubt, include all track lengths in your CAD model and only discard them if you are sure that their effect is negligible. All signal tracks that are not specifically part of a matching network or filter should be made 50 impedance. Always try to keep your layout as compact as possible. A design that is unnecessarily spread out not only wastes board space, but is more likely to be unstable. At very high frequencies, even a solid ground plane cannot be relied upon to give zero impedance. If two ground vias are separated by a distance that is a significant fraction of a wavelength at signal frequency, then they can no longer be considered as a common point. If these are decoupling points for the same amplifier stage, then instability can occur. When decoupling the power supply and bias feed points, it is not uncommon to see double or triple decoupling. This is necessary to limit the out-of-band gain of the stage, which may otherwise lead to instability and degrade the noise figure. For example, if the collector supply has a 220 resistor that is decoupled by a 47pF capacitor, this will be effective only at frequencies above 900MHz. At much lower frequencies, the 47pF will be high impedance and the collector load will now become the 220 resistor, possibly giving a substantial gain. A second, much larger value of capacitor (say 10nF) in parallel with the 47pF ensures that the decoupling remains effective. Frequently a third capacitor is added, which may be as high as 10F and this, as well as limiting out of band gain down to very low frequencies, also removes common-mode noise conducted by the power supply feed. The use of metal screening cans and covers should be used with care. Where these are considered necessary, they must be effectively grounded along the entire periphery. Failure to do so will not only render the can ineffective as a screen, but can actually cause it to act as an antenna and to radiate spurious signals to other equipment. At very high frequencies the dimensions may approach a quarter or half wavelength and the screening can become a resonant cavity. In such cases, the screening can may cause instability in, say, an amplifier stage by introducing feedback between output and input. One solution to this problem would be the inclusion of additional partitions inside the screening can.

Page 26 1.6. Measurement Techniques. Note that all of the following measurement techniques assume a working impedance of 50. If this is not the case, it will be necessary to convert your impedance and make allowance for insertion losses. 1.6.1. LNA gain. Method A: The best way to measure LNA gain is with a Network Analyser, especially since this instrument can also measure the input and output return losses at the same time. The following procedure assumes the use of one of the HP/Agilent range of instruments. It is first necessary to calibrate the instrument, including the connecting cables up to the point of connection to the LNA under test, and for the specific frequency range of interest. Select the 2-port mode of operation and carry out the full calibration procedure - open, short, load for both ports and then connected together for transmission. Isolation can be omitted. It is also important to set the source power level; the default for most instruments is 0dBm and this is too high. If your LNA has a gain of 20dB, your amplifier may be driven into compression. Also, some older instruments will themselves go into compression at this level. A source power of -20dBm is recommended. Connect your LNA to the network analyser and, of course, to a suitable power supply. On the instrument display select your measurement to magnitude. The default vertical scale is 10dB/div, but I recommend that you change this to 5dB/div. Select S21 to display your gain plot - you can use the markers to give an exact reading at any particular frequency. With the same set-up, simply select S11 and S22 to display your input and output return losses. Method B: You can measure gain using a Spectrum Analyser and a Signal Generator. Set the signal generator frequency to the centre of your frequency band and set the level to -20dBm. To calibrate out the connecting cables, first connect them together between the signal generator and the spectrum analyser and adjust the analyser display so that this level is about half-way up the screen. Now set this point as your reference marker and switch to delta marker. When you insert your LNA, this will give you a precise reading of your LNA gain at that frequency. Repeat as necessary for other frequencies or, if preferred, select max.hold on your display and simply sweep your signal generator across the frequency band of interest. 1.6.2. Return Loss. If you have used Method A above to measure LNA gain, then return loss can be displayed on the same set-up by selecting S11 and S22. If you do not have a Network Analyser, it is rather more difficult and requires the use of a Directional Coupler, a Signal Generator and a Spectrum Analyser. Set the signal generator output level to about -5dBm (or the highest level that you are sure will not drive your amplifier into compression) and connect its output to the directional coupler (in the forward direction). Connect the output of the directional coupler to the spectrum analyser and set this level as your reference level. Select delta marker. Now connect the reverse output from the directional coupler to the spectrum analyser and the forward output to your LNA. You must load the output of your LNA with a 50 dummy load. The spectrum analyser will display a reading for reflected power, but this is not the figure for input return loss; to obtain this you must add on the loss factor for the directional coupler. E.g. If the analyser reading is -30dB and you are using a 20dB coupler, the return loss figure is (-30 + 20) = -10dB. Similarly, to measure the output return loss, connect the directional coupler between the output of the LNA and the spectrum analyser to set a reference level, then load the output of the directional coupler with a 50 dummy load and connect the reverse power output to the spectrum analyser. The return loss will again be given by the difference in the two readings plus the loss factor of the coupler.

Page 27 1.6.3. Noise Figure. Method A: The obvious and most efficient way to measure the noise figure of your LNA is with a Noise Meter. Again, HP/Agilent make a range of instruments, their latest offering being a highly sophisticated piece of equipment with a matching price-tag. When using a noise meter, always ensure that the noise source head is connected directly to the input of your LNA. Any intervening cable can give rise to inaccurate readings. Method B: If you do not have access to a Noise Meter, you will at the very least need a noise source. This should be a properly calibrated device with a known ENR figure (Equivalent Noise Ratio). A typical ENR figure would be in the range 10 - 15dB. You will also need a broadband amplifier module (such as those manufactured by Mini-Circuits) with a known gain and noise figure, and a Spectrum Analyser. Fig. 1.27. Noise measurement set-up.
Power supply

Noise Source
ENR = 14.2dB

LNA Amplifier

Spectrum Analyser

Using the set-up in Fig.1.27, set the spectrum analyser to the centre frequency of the LNA under test, with minimum bandwidth setting and the input attenuator at zero. Make sure that the displayed level is well above the noise floor of the analyser itself by temporarily removing the input. Use trace averaging to give a steady display. With the noise source switched off, measure the noise level. Now switch on the noise source and measure it again, giving it time to settle at the new average. The two readings taken will be in dBm and it is necessary to convert them to actual power levels, using the relationship NT = 10(dBm/10) . Example: Noise off reading = -122dBm, so NT = 10(-122/10) = 10 -12.2 mW = 6.31 x 10 - 13 mW Noise on reading = - 109dBm, so NT1 = 10(-109/10) = 10 -10.9 mW = 1.259 x 10 -11 mW Ratio NT1/NT = 19.95 This ratio is known as the Y factor and is the basis of all noise measurements. We can now obtain the noise figure, using the following expression: NF = ENR - 10 x log(Y - 1) where Y = NT1/NT [Equ. 6]

From the example then, we get NF = 14.2 - 10 x log(19.95 -1) = 14.2 -12.78 = 1.42dB. Note that we have taken no account of any noise contribution from the broadband amplifier. Provided that our LNA gain exceeds 15dB, we can ignore this. Otherwise it will be necessary to use the equation for cascaded noise figure (Equ. 3 on page 14). This is the principle upon which noise figure meters are based, but I am sure you will agree that it is much easier to use a Noise Meter. For system noise figure (i.e. the noise figure of a complete receiver chain), we use the much simpler method given by Equ.1 on page 5, but this will be fully discussed in Part 2, section 2.7.5 of this book.

Page 28 1.6.4. IP3. The IP3 of an amplifier is measured using the two-tone method. This requires a Spectrum Analyser, two Signal Generators and a combiner (see Fig. 1.28 for set-up). Two frequencies are chosen with a convenient spacing, such that they are both within the amplifier pass-band and that the expected products are also within the pass-band. For example, if the pass-band is 880 - 900MHz, two suitable frequencies would be 889MHz and 891MHz, because their products (plus and minus the difference) at 887MHz and 893MHz are still within the pass-band. Sig. Gen. 1 LNA or other amplifier under test Spectrum Analyser

6dB Combiner

Sig. Gen. 2

Fig. 1.28. Set-up for 2-tone measurement of IP3. Using the above set-up, first set the two signal generators to the required frequencies and set their output levels to about - 20dBm and both the same. Connect the output from the combiner directly to the spectrum analyser and adjust the display so that both signals are visible near the centre of the screen and with an appropriate span, such that their products will also be on screen. Check that both signals are equal to within less than 0.5dB and, if not, adjust one of the signal generator outputs until they are equal. Now connect in the amplifier and slowly increase the outputs from both signal generators together in exactly equal steps until the distortion products appear on either side of the two outputs. Adjust the vertical scale on the analyser as necessary to avoid the display going off the top of the screen. Check the relative amplitudes of the main outputs and the distortion products and keep increasing the signal generator levels until there is a difference of as near as possible to 40dB. At this point, measure the precise amplitude of the main outputs; if they are not exactly the same check that you have increased both signal generators by the same amount. Add 20dB to your figure to give the actual IP3 for your amplifier. The reason for this is that, if the output level increases by 20dB, the third-order product would increase by 60dB (i.e. 3 x 20). As previously explained in section 1.3.4., this is a theoretical limit only and is used as a convenient figure of merit. In reality, the amplifier would go into hard saturation about 8dB lower than the IP3 figure (see P1dB). Example: measured output level for 40dB products is +12dBm, IP3 is therefore 12 + 20 = 32dBm. Note that the input IP3 (IIP3) will be given by subtracting the amplifier gain from the above figure. If you try to correlate this figure with your measurements, remember that there is a 6dB insertion loss in the combiner. 1.6.5. P1dB. This is the easiest of all to measure, requiring only a Signal Generator and a Spectrum Analyser. Connect the signal generator to the input of your amplifier and its output to the spectrum analyser. Using an appropriate frequency in the pass-band, set the signal generator level to give an amplifier output of about -5dBm (it is assumed that your P1dB will be significantly higher than this). Use peak search to set your reference marker on this output and then select delta marker. Increase the output of the signal generator in 1dB steps, checking that each step of input gives a corresponding 1dB step in output. When this no longer happens, find the level at which the actual output is 1dB less than it should have been if it had continued to rise in 1dB steps - this is your P1dB figure.

Page 29

Practical Rf Engineering Part II

By Davood Mirzahosseini, KNTU.
Copyright 2007.

Passive Mixers Active Mixers The Local Oscillator Phase Noise in Oscillators Oscillator Design Practical VCO Design LO Buffers The Phase Locked Loop Synthesiser Programming IF Filters Crystal Filters Ceramic Filters SAW Filters L-C Filters Group Delay IF Amplifiers Double Conversion The Second Mixer and LO Demodulation Noise Blanking and Data Recovery Spurious Responses and Blocking Direct Conversion Measurement techniques 30 31 32 33 33 35 37 39 39 42 46 46 46 47 48 48 49 50 50 51 51 52 53 54

Page 30

Mixers, Local Oscillators and the IF

2.1. Mixers. The function of the mixer is to convert the receiver RF signal to a fixed frequency IF, by mixing it with a locally-generated oscillator signal (local oscillator, or LO). This means that selective filtering, most of the system gain, and demodulation can all be carried out at a convenient fixed frequency. Generally, when two signals are combined in a non-linear element, other frequencies are generated , the principal of these being the sum and the difference between the two input frequencies. For example, if frequencies of 100MHz and 145MHz are mixed, we get 245MHz and 45MHz at the output. Either of these could be selected as our IF but, for receiver applications, it would be usual to choose 45MHz. The higher frequency would selected for a transmitter (up-converter), but this will be covered in Part Three of this book.. Usually the IF is at a lower frequency than the input RF, but this is not always the case. Sometimes, where a very broad tuning range is required (such as in a multi-band HF communications receiver), it is more convenient to use a first IF of 45MHz, say, and then to convert down again in a second mixer. Using a high first IF and a second conversion has other benefits, as will be discussed in section 2.5. In mixer applications, the local oscillator signal should be very much larger than the RF input signal. Normally, a received RF signal would, in any case, be quite small. However, where strong interfering signals are present, especially if these are in the receiver pass-band, they can easily be of comparable amplitude to the LO. In section 1.4.4 , it was shown that inter-modulation products can occur in the presence of two or more large signals and that these are related to the IP3 of the device. In passive mixers (see section 2.1.1), the IP3 is mainly determined by the LO drive power. For receive-only applications such as broadcast receivers, a poor IP3 may not be a problem, but for communications systems it is probable that there will be other users transmitting on nearby channels that can cause blocking in your receiver. In base-stations for the mobile phone network, where all channels are expected to meet the minimum sensitivity requirement, it is common to see LO levels as high as +17dBm at the receive mixer. This is illustrated in Fig. 2.1 below. Fig. 2.1. Effect of LO drive power on receiver IP3. In the set-up shown, a level 3 mixer (IP3 = +7dBm) is compared with a level 17 mixer (IP3 = +30dBm). Input IP3 Gain (dB) -3 (IIP3) is almost 10dB better with the OIP3 1 (dBm) 100 IIP3 1 (dBm) higher level mixer.
OIP3 2 (dBm) IIP3 2 (dBm) 100

20 25 25

-3 100 100

-6 7 30

8.00 6.49 -1.51 15.83 7.83

It was stated above that the outputs from the mixer are the sum and difference frequencies, but there are also many other unwanted (spurious) responses. In any new design, we need to be aware of these as they can often cause problems. A mixer is also an excellent harmonic generator and, no matter how well the LO is filtered, the mixer itself will generate multiples of the LO frequency up to a high order. These harmonics can then mix with other signals, or their harmonics, to produce an output in the IF band. Particular attention needs to be given to systems with double conversion (and therefore having a second local oscillator) and synthesised LOs, having a reference clock frequency.

Page 31 A number of simple computer programs are available that can calculate the spurious responses, but dont forget to also include the reference clock frequency. I have personally experienced a problem where the 15th harmonic of the reference clock was mixing with the 5th harmonic of the second LO to produce an in-band spur on the first IF. Identifying the cause of the problem can often be the most difficult task and the one described above was eventually found only by working backwards and eliminating everything else. Good layout is essential for a high performance mixer and its associated local oscillator. The LO must be isolated and carefully screened from other parts of the RF circuit and similarly, where a synthesiser is used, the reference clock must be kept well away from the RF section. A particular problem, although not actually associated with the mixer, is the occurrence of deaf channels. This was a problem with early designs for mobile phone handsets using a 13MHz clock, where the 72nd and 73rd harmonics of 13MHz fell on receiver channels 5 and 70 in the 947MHz receiver band. One might expect the 72nd and 73rd harmonics to be insignificant, but with a receiver sensitivity of typically -105dBm, any signal that is equal to or greater than -105dBm will cause a loss of receiver sensitivity (deaf channel). It is therefore important to keep any clock signals well away from the receiver front end and to round off the edges of the clock to minimise its harmonic content. 2.1.1. Passive mixers. Mixers are either passive (i.e. do not use active devices), or active (use active devices). In its simplest form, the mixer function can be performed by a single diode, but this has low input impedance and gives no inter-port isolation. Diode mixers (using special low-loss diodes) are often found in very high microwave applications, where the received signal needs to be converted to IF right at the input (usually a waveguide assembly). Sometimes a balanced configuration is used (2 diodes), but by far the most common type is the double balanced mixer. This has a four diodes in a ring bridge formation and bifilar-wound balanced RF transformers at the RF and LO ports (see Fig. 2.4). There are many different types of double balanced mixers offered by manufacturers (see Figs 2.2 and 2.3), dependent on the frequency and the particular application. Advantages: Does not require a power supply. Simple design (when bought in as a manufactured component). Good IP3 (dependent on LO level). Good inter-port isolation (40dB) - although this can still be a problem with very high LO levels. Disadvantages: Requires high level of LO drive (+3dBm to + 27dBm). Termination sensitive (needs good 50 match). Physically larger size than most active mixers. High insertion loss (6dB minimum) * *Typical conversion loss is 6dB and can be up to 10dB. Half the power goes into each of the IF components, plus the insertion loss due to the coupling transformers, etc.

Fig. 2.2. Three different types of double balanced mixers (surface-mount).

Fig. 2.3. A mixer with SMA connectors for test and measurement applications.

Page 32

Fig. 2.4. Schematic for a double balanced mixer. Note that the L and R ports are generally interchangeable. 2.1.2. Active mixers. These can take many different forms: Bipolar transistor. This is a very low-cost solution that is generally used in cheap commercial equipment. The baseemitter junction of the transistor acts as a diode mixer, with the advantage of built-in amplification by the transistor itself. The required LO signal level is relatively low and can conveniently be injected at the emitter of the transistor. Cheap AM radios also use the same transistor as the local oscillator (selfoscillating mixer). Advantages: Very low cost and simple design. Low LO power requirement. Low noise. Some conversion gain. Disadvantages: Very poor IMD (low IP3). Little or no isolation. FET. Similar to the transistor mixer but works on the transconductance principle. Generally produces a lower level of harmonics and other spurs. A variant on this, the dual-gate FET mixer gives improved isolation by injecting the LO on gate 2. A further variant is the quad-FET commutative mixer, which is similar to the diode ring and has some perceived advantages. Advantages (not quad-FET type): Low cost. Low noise, some conversion gain. Disadvantages (not quad-FET type): Poor IMD (low IP3). Poor isolation (better with dual-gate). Advantages (quad-FET type only): Very good IMD and IP3 possible. Low noise. Disadvantages (quad-FET type only): Expensive. Conversion loss.

Integrated Circuit. These generally use a balanced, or double balanced mixer configuration and include an LO amplifier to allow a much lower level of LO drive (typically -6dBm). Some types, based on the transistor tree (see Fig. 2.5), have excellent IMD performance and good LO isolation. IC mixers usually have an overall conversion gain. There are many ICs available that do not stop at the mixer, but include the entire IF subsystem and demodulator. Indeed, it is now unusual to see a double-conversion receiver design that does not use one of these ICs as the second mixer and IF subsystem. There are even complete integrated radio chips, such as the Motorola MC13135, that include two mixers (for double conversion systems). This type of IC makes radio design extremely simple but, inevitably, there are some compromises to be made.

Page 33 Advantages: Require much lower levels of LO drive power. Physically smaller than most passive types (the Agilent IAM91563 has a tiny SOT-363 package). Good IMD and some types have excellent IP3. Good inter-port isolation. Generally insensitive to termination impedance. Usually give a conversion gain. Often include additional functionality (e.g complete IF subsystem). Disadvantages: IP3 not as good as diode ring. Often (but not always) more expensive than diode ring. 2.2. The Local Oscillator. In any receiver system, the LO is required to produce a stable signal at a frequency separated from the input frequency by the IF. Depending upon the receiver requirement, this LO may be tuneable over a continuous range (broadcast receiver), stepped by fixed increments (channelised receiver), or operated at a single fixed frequency (single channel receiver or as the second LO in a dual conversion receiver). The precise architecture to achieve this will depend upon the application. A tuneable LO will almost certainly use one or more varicap diodes as the tuning element, controlled by a variable dc voltage (the mechanically operated variable capacitor is now a museum piece). A stepped LO will also use varicap diodes, but in this application the control voltage will be from a synthesiser loop, (see section 2.3). For a fixed frequency LO, the oscillator will generally use a quartz crystal, although it is also quite common to use a synthesiser in this application too. Where the oscillator is controlled by a synthesiser loop, it is more usually referred to as a VCO, or voltage-controlled-oscillator. A wide range of manufactured modular oscillators is available to cover most of the VHF and UHF bands and it is often more cost-effective to use one of these, rather than to design one yourself (see Fig. 2.6). Fig. 2.6. Some typical VCO modules. Advantages of a bought-in VCO are: Predictable and repeatable performance. Guaranteed phase-noise, output power and control range. Compact construction. 2.2.1. Phase Noise in Oscillators. Because the LO signal is mixed with the received signal, any noise generated by the oscillator itself will add to the receiver noise. The predominant LO noise is referred to as phase noise because it appears as phase modulation of the LO signal. In a poorly designed oscillator, the phase noise can seriously degrade the receiver performance. For a perfect oscillator, the output display on a perfect spectrum analyser would be a single vertical line. The effect of phase noise is to extend the output on either side of the carrier. The spectral density of this noise decreases with offset from the carrier, initially by a 1/f2 relationship (6dB/octave), but at larger offsets (the break point) it becomes 1/f3 until it ultimately reaches the system noise floor.

Fig. 2.5. The transistor tree type mixer.

Page 34 When the VCO is part of a phase locked loop, the phase detector will effectively cancel the VCO noise within the loop bandwidth. The dominant noise here is due to the synthesiser dividers and the reference frequency source. Outside the loop bandwidth, the VCO becomes the dominant noise source and the specification for a VCO will quote figures for SSB (single sideband) phase noise power at a specific offsets (1kHz, 10kHz and 100kHz) from the carrier or, alternatively, a figure may be given for the integrated phase noise (i.e. the integrated noise power over the output spectrum). Fig. 2.7 below demonstrates the effects of phase noise inside and outside the loop bandwidth of a phase locked loop, and of the choice of loop bandwidth. The phase noise at the centre of the trace is determined by the reference oscillator only (plus any effect due to the dividers and the PLL itself), but the phase noise outside the loop bandwidth is determined predominantly by the VCO. In the left-hand trace, the choice of loop bandwidth is entirely inappropriate and, unless there were a good reason for this, it would be regarded as a poor design. It may be readily seen that the total noise density actually increases at the limits of the loop bandwidth, where the loop gain becomes unity, and this clearly indicates that the loop bandwidth is 80kHz (20kHz per division on the display). The righthand trace shows exactly the same loop, but with a loop bandwidth of about 15kHz. Here the noise falls away gracefully as expected. If we examine the figures for phase noise at a 50kHz offset, in the first case we have -85dBc/Hz, and in the second case we have 105dBc/Hz (20dB improvement). Unfortunately, as always there is a compromise to be made. As the loop bandwidth is decreased, the lock time will increase and the loop will also become more susceptible to other influences, such as microphony.

200kHz scan width, 3kHz analyser B/W. Phase noise increases at limits of loop bandwidth. -35dB at 40kHz offset = -70dBc/Hz (-35dB in 3Khz bandwidth, as given by 10xlog3000).

200kHz scan width, 3kHz analyser B/W. Phase noise decreases smoothly beyond loop bandwidth. -65dB at 40kHz offset = -100dBc/Hz.

Fig. 2.7. Spectrum analyser display showing the effects of VCO phase noise and loop bandwidth. In a narrow-band receiver, phase noise effectively increases the channel bandwidth to cause what is known as reciprocal mixing. For multi-channel equipment, the channel filter is designed to ensure that signals in the adjacent channels do not interfere with the wanted channel. In fact, due imperfect filters, there will always be some level at which the adjacent channel will affect the wanted channel, but this will be defined by the system specification for adjacent channel blocking. The presence of any significant level of phase noise at an offset of half the channel width, will effectively extend the response of the receiver into the adjacent channel, totally negating the benefit of the channel filter and seriously degrading receiver performance. The phase noise will be further discussed in section 2.2.5.

Page 35 2.2.2. Oscillator Design. In a perfect resonant circuit with zero losses, oscillation would continue indefinitely without the need for external power. In reality there are always losses to overcome and, in order to sustain oscillation, some external amplifying device is required. This amplifying device should be self-limiting and must provide only enough gain to cancel the losses and restore the overall loop gain to unity. With too much gain, the oscillator will be driven into saturation and heavy distortion. There are two main categories of oscillator; the feedback type, which relies on positive feedback from the amplifier device, and the negative resistance type, which relies on a negative resistance element in the path of the resonant circuit. Most oscillators in the VHF/UHF frequency range are based upon either the Colpitts type, or the Clapp type, although other variants are possible. In fact, it is also possible to make an oscillator using a negative-resistance diode. The tunnel diode is now considered to be little more than a curiosity, but the Impatt diode is commonly used for oscillators at millimetre wavelengths (>30GHz), where it can produce appreciable amounts of power. Fig. 2.8 below shows the three possible configurations for the Colpitts type oscillator, all of which are electrically equivalent. The most common is probably the grounded collector version, although the grounded base version has a better phase-noise performance. The grounded emitter version is rarely seen. Fig. 2.8. Equivalent circuit representation of a Colpitts oscillator, showing the 3 main configurations.

At first sight, the Clapp oscillator (shown in Fig. 2.9 below) looks remarkably similar to the grounded collector Colpitts type. In fact, its operation is somewhat different. Instead of relying on a 180 phase shift across a parallel resonant circuit to give positive feedback, the transistor provides a negative resistance in the series resonant circuit (0 phase shift) comprising the inductor and 3 capacitors. The Clapp oscillator is especially useful for crystal oscillators, where the crystal can be operated in the (preferred) series-resonant mode. Also, when compared with the grounded collector Colpitts type, it has less loading effect on the resonant circuit (and hence higher Q).

Fig. 2.9. Equivalent circuit representation of a Clapp oscillator.

Page 36

Fig. 2.10. Practical circuits for the three versions of Colpitts oscillator.

Note that some of the circuit elements have been rearranged for practical reasons. Fig. 2.11. A practical circuit for a Clapp oscillator, used with an overtone crystal. The crystal is operated in its low impedance series resonant mode and the parallel resistor, Rd (usually about 330), is needed to ensure that the crystal does not operate in its high impedance parallel resonant mode. Inductor Lr, together with the 3 capacitors in series, is chosen to resonate at the wanted crystal frequency.

Fig. 2.12. Another practical circuit for use with overtone crystals, this time based on the grounded base type Colpitts. With careful choice of components, this configuration can be used for a very low phase noise design. The RF output coupling capacitor should be small enough to have negligible loading effect on the resonant circuit.

Although bipolar transistors are shown as the active device in all of the above circuits, J-FETs or even MMICs can be used. Similarly, the tuned circuit may be replaced by a quartz crystal, a dielectric resonator or a coaxial resonator.

Page 37 2.2.3. Practical VCO Design. A good VCO should meet the following requirements: Low phase noise. Stable and repeatable design. Adequate VCO gain - this is the measure of frequency versus control volts (MHz/V). Adequate RF output power (most bought-in designs give about 0dBm) The effects of phase noise on the LO have already been discussed, and the designer needs to ensure that the VCO phase noise is not excessive. It may be shown that the phase noise is proportional to the inverse square of the working Q of the tuned circuit. It therefore follows that the Q needs to be as high as possible. This becomes a problem at frequencies above 500MHz, where the maximum Q for even the best inductors and high-quality chip capacitors is quite low. One possible solution for higher frequencies is to run the VCO at half the frequency and then use a frequency-doubler stage. Although the doubler will itself increase the noise by a factor of n2 (where n = 2), this will generally be offset by a higher Q (noise decreases as 1/Q2). In fact, the frequency doubler approach has an advantage (especially when used in a transmitter) that VCO isolation is improved, and hence frequency-pulling is reduced. Another, and more usual, solution to the problem of low Q, is to use dielectric resonators, coaxial lines or even, where space permits, coaxial cavity resonators. On a high permittivity substrate, such as alumina, it is also possible to use a microstripline as the resonant element and this is how many bought-in VCO modules are constructed. However, this technique is not normally available to the experimenter, and the performance of microstripline on FR4 type substrates is rather poor. Other factors affecting phase noise are the noise figure of the amplifier in the oscillator, the voltage across the tuned circuit, and the RF output power. The first of these can be addressed by choosing a good low-noise RF transistor for the active device, and the last by ensuring that the RF coupling to the output is very light (a VCO buffer/amplifier is in any case mandatory). The voltage across the tuned circuit is largely pre-determined by the available supply voltage and the architecture of the oscillator design. For multi-channel equipment it is often necessary for the oscillator (VCO) to cover a relatively large range of frequencies. At first sight, this would not seem to be too much of a problem, but with the limited output voltage range available from modern synthesisers, the choice of tuning components is critical. Many of the older types of varicap diode have a good capacitance ratio, but only with a large range of control voltage (often as much as 26V). With modern battery-powered equipment, the control voltage range is typically less than 4 volts, and this dictates the use of a hyper-abrupt varicap diode (that is, one that has a very large capacitance ratio for a small change in applied voltage). Unfortunately, the hyper-abrupt varicap generally has an inferior Q to the standard types, although the more recently introduced types are much improved. For a wide range VCO, the VCO gain (frequency shift in MHz/V) needs to be high, but not too high. The value must be high enough to cover the desired frequency range with the available voltage swing, but an excessive gain makes the VCO susceptible to any induced voltage on the control line and can degrade phase noise performance. Due allowance must be made for component tolerances and temperature dependency, and this will considerably increase the figure for the required gain to give an adequate safety margin. This makes it very difficult to produce an optimal design without the need for some manual adjustment (unacceptable for mass production) to centralise the tuning range. For this reason, it becomes more attractive to use a bought-in VCO that comes with a guaranteed specification for frequency range, gain and phase noise.

Page 38 Example: Consider the requirement for a GSM receiver: The VCO range to cover 124 x 200kHz channels is 24.8MHz. If the maximum control voltage range is 4V, the minimum VCO gain required will be 6.2MHz/V, with zero allowance for tolerances. But the capacitance value of the varicap has a tolerance of typically 20% at either end and is also temperature-dependent. Also, the frequency versus capacitance relationship is an inverse square law. A tolerance analysis will show that the practical VCO gain needs to be at least 12MHz/V and will typically be more like 25MHz/V. At 25MHz/V, an induced voltage of just 1mV on the control line will cause 25kHz of deviation on the VCO frequency. A practical VCO circuit for the 440MHz band is shown in Fig. 2.13. This was designed as a receive LO for a receiver covering 425 - 470MHz, with a 45MHz IF. The LO covers 380 - 425MHz with a 3V range of control voltage and was required to meet phase noise specification of -120dBc/Hz at 25kHz offset. It is not claimed that this is the ultimate in good design, but given the constraints of small physical size and the use of easily available components, it performed quite well. With the exception of the main tuning inductor, inductors are 0805 size chip types and all resistors and capacitors are 0603 chip types. In order to meet the required phase noise and VCO gain, it was necessary for the tank inductor to be an adjustable type and to have as high Q as possible. For this application a surface-mount Toko 13nH adjustable (with aluminium tuning core) was used.

Fig. 2.13. A practical VCO design for 380 425MHz

Your attention is drawn to the following points about the design: a). An RF choke is used in the transistor collector so that the tank circuit can be lightly coupled, with its bottom end at ground potential. b). The oscillator output is extremely lightly coupled by a 1pF capacitor. This is followed by a buffer amplifier to improve isolation and to amplify the output to +3dBm. c). In order to achieve the required control range, a hyper-abrupt varicap has been used. To minimize the effect on the Q of the tank circuit, the varicap is lightly coupled with a 5.6pF capacitor. This gives a rather non-linear control characteristic, but is essential for good phase-noise.

Page 39 2.2.4. LO Buffers. It has already been explained that, for good phase noise, the local oscillator (VCO) should be as lightly loaded as possible. For broadcast receivers, such considerations are generally unimportant, but for narrow band systems it is mandatory to follow the LO with a buffer/amplifier. A bought-in VCO will almost certainly include a buffer, but the output power level is unlikely to exceed about +3dBm. With a synthesiser loop, it is common to follow the VCO with a 6dB splitter pad, so the available LO power will be reduced to -3dBm. This is adequate for most active mixers, as these invariably include an amplifier for the LO, but further amplification will be required for a passive mixer. For a level 17 mixer (+17dBm), we need a further 20dB of gain. The amplifier should also be low noise, or the benefit of a low phase noise VCO will be partially lost. A disadvantage of passive mixers is that, for best performance, they need a good 50 termination on all ports. The LO amplifier should therefore be matched for 50 output and have low return loss. Although the mixer will itself generate harmonics, it is desirable to avoid additional harmonic power and a low-pass filter is often included at the amplifier output. In order to further ensure a good 50 termination, it is also common to precede the mixer by a 3dB resistive pad. The total gain required in the LO buffer amplifier could therefore be as much as 23dB. 2.2.5. Phase Noise. The calculation for maximum acceptable phase noise to achieve a given receiver performance is not at all straightforward. Fortunately, it is invariably specified at system level and the practical engineer is unlikely to be asked to do this. The designer therefore only needs to know what the requirement is and to ensure that his/her design can meet it. It is difficult to measure the phase noise of a VCO without special test equipment. A free-running oscillator will not be sufficiently stable for narrow-band measurements and it will need to be included in a phase locked loop to obtain a stable frequency. Some spectrum analysers are able to give a direct measurement in dBc/Hz, but this will only be valid for readings above the noise floor of the analyser itself. For large frequency offsets, the readings will be unreliable and a more complex test set-up is needed. See section 2.7 on Measurement Techniques for a further details. With the increasing use of digital modulation in modern communications equipment, it is now more common to specify a figure for error vector magnitude (EVM). This has a direct relationship with the integrated phase noise and is a more practical and convenient measurement. Top of the range test equipment, such as the Agilent Vector Signal Analyser, can measure EVM directly. 2.3. The Phase Locked Loop. There has already been much written on the design of phase locked loops, indeed there are several books that devote themselves entirely to this subject. The theoretical analysis necessarily involves complex mathematical calculations and, since it is the stated intention of this book to avoid such mathematics, I shall limit the discussion to purely practical issues. For those who are interested in a more detailed approach, I can recommend the website at where free information is available, together with a useful CAD program. The practical engineer will wish to design a functional phase locked loop with the minimum of effort. Unless you are designing a fixed-frequency loop for a critical application, it is in any case pointless to use a precise method of calculation, since the parameters themselves are not constant. The VCO gain will change over its frequency range, as will the synthesiser divider ratio. A mid-range value is generally assumed. Of course, the finished design must meet the requirements - that is, it should have an adequate phase margin, appropriate loop bandwidth and lock time, and low phase noise, but for all except the most demanding applications, the following procedure will apply.

Page 40 2.3.1. Design Procedure for Phase Locked Loops. 1. Choose your VCO. If you are buying in a ready-made VCO, then obviously it must cover the range of frequencies that you expect, and with a given range of control voltage. The operating voltage and temperature range must be compatible with your requirements and, for battery-powered equipment, the supply current is also important. Check that the phase noise is acceptable (usually a figure is given for 10kHz offset, but more recently the figure may be given for integrated phase noise). If you are designing your own VCO, then follow the guidelines given in section 2.2.3. 2. Choose your synthesiser chip. Practically all of the currently available synthesiser chips use a charge-pump type phase detector. These offer several benefits over the older voltage type phase detector, one of which is that it uses a passive loop filter instead of an active (op-amp based) filter. Unless your VCO is only required to operate at a single fixed frequency (as for a second LO in a dual conversion receiver), you will need a dual-modulus synthesiser with a built-in prescaler. Divider ratios available are typically 8/9, 32/33, 64/65 or 128/129. The choice will depend upon your VCO frequency (higher frequencies will need a larger division in the prescaler) and the maximum operating frequency of the main synthesiser block. National Semiconductor offer an excellent range to suit most needs. The LMX2316 is an inexpensive low power type for use up to 1.2GHz and is an excellent choice for battery-powered equipment. For a higher performance, the LMX2325 might be considered. Other synthesiser manufacturers include Philips, Motorola and AMD. 3. Choose your comparator frequency. For multi-channel equipment, this will be pre-determined by the channel spacing - e.g. for GSM applications, the channel spacing is 200kHz and hence the comparator frequency should also be 200kHz. For PMR applications, the channel spacing is 12.5kHz and the comparator frequency will therefore also be 12.5kHz. Note that the comparator frequency must also be an exact sub-multiple of the VCO frequency - for example, 144.1MHz will divide by 0.0125, but 144.13MHz will not. For this reason, 144.13MHz would not be a viable channel frequency. 4. Choose your reference frequency. The reference must be divisible by a whole number to give your comparator frequency. 13MHz will divide by 65 to give 200kHz or by 1,040 to give 12.5kHz, but 18.432MHz (a standard reference for DECT) cannot be divided to give 200kHz or 12.5kHz. In order to reduce the probability of in-band spurious products, it is a good idea to make the reference frequency as high as possible, although this will usually be limited by the maximum available divider ratio in the synthesiser chip. The reference oscillator is the main factor in determining the overall frequency stability of the system and also the phase noise performance inside the loop bandwidth. A simple crystal oscillator will have good phase noise, but relatively poor frequency stability with temperature variation. It is better to use a TCXO (Temperature Controlled crystal Oscillator) or VTCXO (Voltage and Temperature Controlled crystal oscillator). The latter type permits fine adjustment of the frequency by an external voltage and is occasionally used to allow direct frequency modulation (within the loop bandwidth). 5. Choose your loop bandwidth. As a general rule-of-thumb, the loop bandwidth should be a factor of ten times less than the phase comparator frequency. This allows adequate suppression of reference spurs, which would otherwise be present on the VCO output, giving the effect of greatly increased phase noise. Thus, for 200kHz channel spacing, the loop bandwidth should not exceed 20kHz, but for 12.5kHz channel spacing, the loop bandwidth must be only 1.25kHz. It follows that this also affects the loop settling time, which is a measure of how fast the loop can react to a change in frequency - the lower the loop bandwidth, the slower the settling time.

Page 41 6. Determine the values for your Loop Filter. For most applications, a third-order loop is recommended - this means that in addition to the standard lead/lag integrating network, there will be a further pole with a higher frequency break point. The purpose of this third pole is to attenuate the reference sidebands, which would otherwise appear on the VCO output to cause spurious responses in the adjacent channels. A typical loop filter is shown in the circuit at Figure 2.14. Note that, for good noise performance, capacitors C1 and C2 should be high quality types. Fig. 2.14. A loop filter for a third-order loop. The main filter comprises C1, C2 and R1. R2, C3 provide an additional break point for the suppression of reference sidebands. The standard calculation invariably yields a high value for R2, but it is not always understood that R2 will directly add to the VCO noise figure. In practice, this value may be reduced within reason by scaling R2/C3, without significant effect on the main loop. In order to calculate your loop filter, you will need the following information: a). The charge-pump current. This is given in mA and will usually be in the range 1 - 10mA. b). The VCO gain. This is given in MHz/V and will typically range from about 2MHz/V for a VCO in the 25 - 75MHz region to more than 50MHz/V for a VCO in the UHF bands. Because the VCO gain will not be constant across its control range, the mid-band value should be used. Some programs require this to be given in radians/sec, in which case the above figures should be multiplied by 2 x 106 . Eg. 2MHz/V is equivalent to 12.56x106rad/sec. c). The total divider ratio, N (i.e. from VCO to comparator). Again, the mid-band value should be used. d). The comparator frequency (in kHz). As already stated, this will usually be the same as the channel spacing. e). Desired loop bandwidth (in kHz). In the absence of other directives, this should be one-tenth of the comparator frequency. f). The desired Phase Margin. A good design should have a phase margin of not less than 40 degrees. Note that phase margin is also related to damping factor. For critical damping, d = 0.7, and this is equivalent to a phase margin of 37 degrees. g).The minimum acceptable attenuation of the reference sidebands (specifed by National as T1/T3 ratio). If sidebands are unimportant, this ratio can be 0% and the third pole (R2/C3) omitted. As was stated at the beginning of this section (page 39), it is not the purpose of this book to discuss transfer functions and the complex mathematics associated with calculation of loop filter values. Instead, it is expected that the practical engineer will use a computer program. I can recommend EasyPLL, which can be found on the Internet at: Select Webench and then Wireless Simulation. 7. Other Considerations. For good low-noise design, the power supply for the PLL must be stable and clean. It is therefore mandatory to use a separate, well decoupled voltage regulator for the PLL section. For some very critical applications, the VCO will often have its own separate voltage regulator. Layout is also very important, especially in the area of the loop filter and VCO control line. The designer must ensure that the loop components are kept well away from any potential source of interference

Page 42 Digital switching signals can be especially problematic. Care must be taken with the routing of the reference (clock) signal, since any interference on this line will appear as spurs on the PLL output. Ideally, the entire PLL section should be self-contained in its own screened compartment. Fig. 2.15. Block diagram for a typical Synthesiser chip. National Semiconductors type LMX2316.

2.3.2. Programming the Synthesiser. Most modern synthesiser chips use serial data input and data is loaded into the registers on receipt of a load enable pulse. A typical synthesiser chip will have three programmable dividers, generally referred to as the R, A and B counters (some manufacturers use the letters M and N, but their function is the same). The exact programming and the data sequence obviously vary from one device to another, but the counters will require three binary data words to set up the dividers. The table of binary values below may prove helpful in the determination of divider ratios.
20 1 21 2 22 4 23 8 24 16 25 32 26 64 27 128 28 256 29 512 210 1024 211 2048 212 4096 213 8192 214 16384 215 32768

The R counter. This is the reference divider and must be programmed to divide the reference frequency down to the comparator frequency. From section 2.3.1, items 3 and 4, it was seen that the comparator frequency will determine the step size and that for a channelised system, the step size will be the same as the channel spacing. Hence, with a 13MHz reference, the divider ratio required to give 12.5kHz channel steps must be 13/0.0125 = 1040. Converting 1040 to binary gives us 1040 = 1024 + 16 = 210 + 24 = 10000010000, which requires 11 data bits. For the LMX2316 synthesiser, the R counter has 14 bits, so this number must be preceded by three more zeros. This is further complicated by the fact that the data register is 21 bits, configured so that there are 5 bits before and 2 bits after the 14-bit word. In fact, the final 2 bits are control bits and these determine which of the three counters is programmed. For the R counter the control word is 00 and the first 5 bits are also generally set to zero. For serial data, the MSB (most significant bit) is input first, so the data in the register actually appears reversed. The complete data word would therefore look like:
C1 C2 R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11 R12 R13 R14 R15 R16 R17 R18 R19

0 LD


LSB ... 14 bit R divider MSB

Test bits

This may appear somewhat daunting for the novice, but if one understands binary arithmetic and the concept of a serial data stream, it is actually very simple.

Page 43 The P counter. An important feature of modern synthesiser chips is the built-in dual modulus prescaler (P counter). This initially divides the high frequency from the VCO to a frequency that can be handled by the main dividers. The dual modulus capability allows the prescaler to switch between P and (P+1). It has already been said that the VCO frequency must be exactly divisible by the comparator frequency, but this is modified by the prescaler and it is important to realise that the divider ratios can only be whole integers. Consider the example of two channel frequencies, 384MHz and 384.025MHz say. These would be directly divisible by 0.0125 to give the whole integers 30,720 and 30,722. But, when first divided by the prescaler (divide by 32 for the LMX2316), we have frequencies of 12MHz and 12.00781MHz at the input to the main dividers The second of these frequencies is not divisible by 0.0125. This problem is solved by the swallow counter function and is best explained as follows: In the example above, 384MHz will divide by 32 and then by 960 to give the 12.5kHz comparator frequency. For 384.025MHz, the main (B) counter is still set to 960, but the A counter is set to a value of 2. This means that for two counts of the clock, the prescaler is changed to 33, or (P + 1), and therefore effectively swallows two divider pulses. The VCO can then run two counts high, giving the desired frequency. This can be repeated for 31 frequency steps (P - 1) by setting the A counter with values 0 - 31, until the next whole integer is reached for the B counter (961 in this example). The A and B counters. In the example of the LMX2316 synthesiser, the main divider comprises a 5-bit swallow counter ( the A counter) and a 13-bit counter (the B counter), together making a total of 18-bits. This tells us that the maximum divider ratio is (218- 1) = 262,143. The divider ratio N is given by dividing the required VCO frequency by the comparator frequency (as set by the R counter). N must always be a whole integer and, in a multi-channel system, normally the lowest VCO frequency will be taken as the starting point. The values for the A and B counters are given by the expression N = (P x B) + A, where P is the prescaler divide ratio, B is the divisor and A is the remainder. This can be more conveniently expressed as B = N/P and A = N - (P x B). To give a simple example, suppose our VCO frequency is 320MHz. With a comparator frequency of 12.5kHz, N = 320/0.0125 = 25,600 and B = 25,600/32 = 800. As this is a whole integer (i.e. there is no remainder), the A counter will be set to 0 and B = 800. The value 800 is given by 512 + 256 + 32 = 29 + 28 + 25 = 1100100000, which requires 10 data bits. This number must be preceded by 3 more zeros to make up 13-bits and the complete data word will be:
C1 C2 N1 N2 N3 N4 N5 N6 N7 N8 N9 N10 N11 N12 N13 N14 N15 N16 N17 N18 N19


LSB. 5 bit A counter .MSB

LSB . .. 13 bit B counter .. MSB GO

Using the same example, suppose our required VCO frequency is now 320.05MHz. N = 320.05/0.0125 = 25,604 and B = 800.125. This is no longer a whole integer, so the B counter will still be set to 800, ignoring the remainder. Now A = 25,604 - (32 x 800) = 25,604 - 25,600 = 4. The A counter must therefore be set to 4 and this is given by 00100 in the 5-bit counter. In fact, we have no need to perform the second part of this calculation because we know that each increment of the A counter will advance the VCO frequency by one channel, through 31 steps until the next whole integer is reached (at 320.4MHz). The word to give a frequency of 320.05MHz is shown below:
C1 C2 N1 N2 N3 N4 N5 N6 N7 N8 N9 N10 N11 N12 N13 N14 N15 N16 N17 N18 N19


LSB. 5 bit A counter .MSB

LSB . .. 13 bit B counter .. MSB GO

Page 44 Function and Initialization registers. The format for function and initialization will depend upon the particular synthesiser chip and are fully explained in the data sheets for the relevant device. The LMX2316 has a 21-bit data register and an 18-bit function latch. The function latch is used to set charge-pump polarity, tri-state output, fastlock mode, etc. The last 2 bits are control bits. Data is loaded each time the Load Enable (LE) input is set high and the status of the two control bits will determine which of the registers is programmed. After initialization and loading of the R counter, only the values in the N counter will need to be changed. Software. The synthesiser will invariably be programmed using a micro-controller chip and an EPROM. Fixed data will be stored in the EPROM and, for PMR applications, a manual user interface (keypad) allows the user to select any of the available channels. For mobile phone applications, automatic selection of channels is controlled by the base station and the handset can change channels many times, even during a conversation. From this latter requirement, it follows that the synthesiser must have a fast settling time and hence a relatively high loop bandwidth. Example for PLL design: A receiver is required to cover the frequency band 432 - 440MHz in 25kHz steps (320 channels). The receiver first IF is 45MHz, so the required LO range will be 387 - 395MHz. The Synthesiser: The comparator frequency will determine the step size, so this must be 25kHz. The reference oscillator can be any convenient frequency that is exactly divisible by 25kHz. A VCXO (voltage controlled crystal oscillator) is available at 13MHz, so the reference divider (R counter) will be set to 520. It will be seen that 387MHz does not divide by 32 and then 0.025 to give a whole integer. This does not mean that the LMX2316 synthesiser is unsuitable, but the frequency range would be extended down to 386.4MHz, where it is divisible. These extra frequencies can easily be disabled in software. Alternatively, the LMX2306 has a 8/9 prescaler and 387MHz is divisible by 8 and 0.025. N = 387/0.025 = 15,480 and, for the LMX2306 synthesiser, B = N/P = 15480/8 = 1935. 1,935 is 11110001111 in binary. We must precede this with two more 0s to make up the 13 bits, hence the B counter must be loaded with 1111000111100. The A counter will add up to 7 steps (0 - 7) at intervals of 0.025MHz, until the next change in the B counter value. A = 1,936 will give 387.2MHz. For the highest frequency, N = 15,800 and B = 1,975. Although the A counter in the LMX2306 is 5-bits, only the first 3 are used. 0 - 7 in binary is 000 to 111. Two more zeros are added to make up 5-bits, hence 00111. To load the N counter with the values for 387MHz, the programming bit stream would be as follows. The first bit, the GO bit, (MSB) N[19] is used for FastLock operation (see data sheet). The next 13 bits, (N[18]N[6]) shifted in, are the B counter value, 1111000111100. Bits N[5]N[1] are the A counter and are 00000 in this example. The final two bits (the control bits) are 1,0 identifying the N counter. The VCO: The VCO must, of course, be designed to cover the required frequency range of 387 - 395MHz, but in order to allow for tolerances and temperature effects, the range will need to be greater than this. It must also be capable of covering the required range with the available control voltage from the synthesiser chip. For the LMX2306 in this example, the control voltage range is only 0 - 4 volts. Also, the synthesiser operation will be unreliable at the extremes and it is therefore advisable to restrict the usable voltage range to 0.5 - 3.5 volts. With such a limited range of voltage, a hyperabrubt varicap tuning diode is mandatory. It is possible to amplify the output voltage range, using an op-amp, but this necessarily requires a higher voltage rail and the use of a type A active loop filter (different method of calculation). It is usual for the VCO to have some kind of manual adjustment to allow the range to be centralised during the initial set up. With a bought-in VCO, the range will be factory preset.

Page 45 It may be seen that the required VCO range covers an 8MHz span. If the control voltage range is 3 volts, we may deduce that the VCO gain must be at least 3MHz/V. In practice, we should design our VCO with a gain of about 10MHz/V. Remember that the higher the VCO gain, the more susceptible it becomes to external sources of noise and interference. Good layout and careful screening is therefore very important. The Loop Filter: Values for mid-range should be used in the calculation. Figures are as follows: 1. Charge Pump current, , mA = 1 (for LMX2306) 2. VCO Gain, MHz/V = 10 3. Divider ratio, N = 15,640 4. Comparator frequency, kHz = 25 5. Phase Margin, , degrees = 45 .default figure ( should be >40 deg.). 6. Loop Bandwidth, kHz = 2.5 (nominally one-tenth of the comparator frequency). 7. T3/T1 ratio * 40% The schematic for the completed design is shown in Fig.2.16 below. Fig. 2.16. Schematic for a typical PLL.

Loop filter values were initially calculated using EasyPLL and then optimised to use standard component values. Predicted performance for this design is as follows: Phase Margin, = 46 degrees Damping Factor = 0.83 Loop Bandwidth = 2.42kHz T3/T1 ratio = 22% Note that R2/C3 can be scaled to reduce phase noise (say 10k and 680pF), without significant effect on the loop. FAQ 5. How do I choose a synthesiser chip for my design? Aside from availability and cost issues, this will depend on your VCO frequency and the step size. Some synthesiser chips have low phase noise, and you must decide if this is important in your design. Clearly, the prescaler must be capable of operation at the VCO frequency and must give an output frequency that is divisible by the comparator frequency. For example, at 1,800MHz, a 32/33 prescaler would give 56.25MHz. This is not directly divisible by 200kHz and would require additional steps in the A counter. The possible merits of using an alternative prescaler ratio should be examined.

Page 46 2.4. IF Filters. It was shown that the front-end filters define the overall receiver bandwidth by attenuating the out-ofband frequencies. The IF filters define the receiver channel bandwidth by attenuating all frequencies that are outside the desired channel frequency. It is the quality of the IF filter that will determine the adjacent channel rejection and the blocking rejection for all offset channels. The ideal filter response would have very low insertion loss within the channel bandwidth and an infinite insertion loss outside the channel bandwidth (sometimes referred to as a brick-wall filter). This ideal can be achieved with digital filters and DSP, but digital processing will necessarily take place at the second IF, and the first IF filter still plays an important role in limiting the response to other offset channels (blocking). There are a number of clearly defined types of IF filter, as follows: 2.4.1. Crystal Filters. As the name implies, the crystal filter is based on the use of quartz crystals as the resonant elements and, as such, their use is confined to narrow-band applications (from a few kHz up to about 25kHz). Due to physical size, they are also restricted to first IF applications where the frequency is 10.7MHz, or higher. Common frequencies are 10.7MHz, 21.4MHz, 45MHz and 70MHz. The filters may be supplied singly (2-pole) or in matched pairs (4-pole), according to the steepness of the skirt and hence the adjacent channel rejection. Common packages are metal can (Fig. 2.17 below), or surface-mount types. 6-pole and 8-pole versions are also available, but these will be in block form and are factory pre-aligned. Fig. 2.17. Typical crystal filters in a wire-ended, metal can package. Correct matching is important to maintain the specified bandwidth and a flat band-pass response. Appropriate values will be specified by the manufacturer and a typical schematic is shown in Fig.2.18 below. Designers will often attempt a lossless match by using inductors as impedance matching, but this may not be a good idea. Unless close-tolerance inductors are specified, it greatly increases the chance of a mismatch, leading to a poor performance. By the time the signal reaches the IF stages, noise figure is no longer an issue and the additional 3dB loss with resistive matching can easily be made up in the following amplifier stage. Fig. 2.18. Circuit for a 4-pole filter using 2 x 2-pole filters (matched pair). The values of C1 and C3 must be modified to include strays on the circuit board. 2.4.2. Ceramic Filters. The type of ceramic filters used in the IF are totally different from the ceramic resonator type used in the RF front-end. Here, their use is generally confined to frequencies of 10.7MHz and below, taking over from crystal filters, where the latter would become physically impractical. There are 2, 4 and 6pole types, dependent on application, and these come in a range of pre-defined bandwidths. 10.7MHz types are specifically designed for broadband (FM) systems, and a similar range covers 5 - 6MHz for

Page 47 TV sound systems. In a dual-conversion receiver, ceramic filters are used almost exclusively in the second IF at 455 - 470kHz (or single conversion systems where the IF is 455kHz). These will be 4 or 6-pole types and will define the overall system bandwidth. Matching is less critical than with crystal filters and, with integrated IF systems, the matching is generally provided by the IC itself. A further variant is the single pole resonator. This is used in place of an inductor in the quadrature detector for narrow band FM systems and has the merit of not requiring manual adjustment. It is also frequently used to make a simple oscillator, using a digital inverter or gate IC (see Fig. 2.20). Fig. 2.19. Some different types of ceramic filter. Clockwise from top left, these are: 4-pole 455kHz block filter. 6-pole 455kHz block filter. 2-pole 10.7MHz wideband filter. 2-pole 6MHz filter for TV sound. Single pole 455kHz resonator. Fig. 2.20. Simple 455kHz oscillator using a resonator and a digital inverter IC.

2.4.3. SAW Filters. The SAW (Surface Acoustic Wave) filter is an electro-mechanical device and, as the name implies, generates a transverse acoustic wave across the surface of a substrate material. The wave is initiated at one end of the device with a transducer and is received at the other end by a second transducer. The precise architecture of the device controls not only the centre frequency, but also the bandwidth and the shape of the filter response. Whereas the ceramic IF filter is used exclusively at low frequencies, the use of SAW filters is restricted to frequencies above 40MHz, where physical dimensions become practical. Their use is particularly attractive in wideband digital systems (GSM, DECT, WCDMA, etc), where they give a precisely controlled response in the given bandwidth. They are also widely used in analogue TV IF systems at 45MHz, giving the required asymmetrical response for vestigial sideband reception. SAW resonators are frequently used in simple oscillators (e.g Remote car key). Data for a typical filter is shown in Fig. 2.21 below. Good matching is essential to achieve the desired response and low group delay (see section 2.4.5). Fig. 2.21. Extract from the data sheet for a DECT IF filter at 110.592MHz (by Murata).

Page 48 2.4.4. L-C Filters. The old IF transformer is now virtually a museum piece, but the L-C filter can still be found in some specialist applications. For IF frequencies or bandwidths other those for which standard filters are available, the L-C filter may be the only option. The main problem will be repeatability. Even with close-tolerance inductors and capacitors, the spread may be unacceptable without manual adjustment (usually with variable inductors). This in turn requires difficult (and costly) set-up procedures and is to be avoided unless there is no alternative. A typical configuration for an L-C band-pass filter is shown below. Fig. 2.22. Band-pass filter centred at a frequency of 1MHz, with 150kHz bandwidth (3dB). The predicted response is shown below. Note that close tolerance components are essential here.

2.4.5. Group Delay. An important parameter in filter performance, especially in digital systems, is Group Delay. When a signal is passed through a filter, it will be subject to a timing delay, dependent on the bandwidth of the filter. This is not normally a problem with narrow-band filters, but with the wide-band filters used in digital systems, the phase-shift (or delay) will be affected by the flatness of the filter response in the pass-band. An ideal filter has a completely flat response in the pass-band and should therefore have a constant delay for all modulating frequencies. Most filters will quote a figure for ripple, and this is a measure of its flatness (expressed as a dB ratio), but this can be adversely affected by poor matching. The extent to which the filter delay changes with modulating frequency is known as Group Delay, and should ideally be very low. When the signal is modulated with a digital data stream, a poor group delay will cause variations in timing at the output of the filter. These timing errors can, in severe cases, result in missed data bits and increased BER (bit error rate).

Page 49 2.5. IF Amplifiers. Preferred IF frequencies are 455kHz, 10.7MHz, 21.4MHz, 45MHz and 70MHz. Higher frequencies are used for specialist applications such as DECT and GSM. The requirements for the IF amplifier will depend upon the type of system in which it is used. The IF amplifier will generally provide most of the system gain, but there are major differences between IF systems for FM/GMSK applications and for AM/SSB/QPSK applications. For an FM/GMSK system, the final amplifier stages will be driven into hard limiting on all but the weakest signals. This is a desirable feature because any incidental amplitude modulation, especially that due to interference or rapid fading, will be removed by the limiting action. For AM systems, it is obviously important to preserve the amplitude component and, for this reason, the IF amplifier must not be driven into limiting. It is therefore essential to employ some form of Automatic Gain Control (AGC). In its simplest form, this is done by taking the dc component of the rectified signal at the demodulator and using it to reduce the gain in the IF amplifier stages, having first removed the modulation component with a long time-constant R-C filter. A typical schematic is shown in Fig. 2.23 below. For SSB and QPSK systems, there is no proportional dc level at the output of the demodulator and a more complex arrangement is required, with a separate detector. This is generally referred to as RSSI (Received Signal Strength Indicator). Basic AM broadcast receivers will use a single IF at 455kHz. This is adequate for the low frequencies normally associated with AM broadcast, but for a good quality HF communications receiver, double conversion will be used (see section 2.5.2). FM broadcast receivers use a single IF at 10.7MHz. With a requirement for 180kHz bandwidth, the choice of 10.7MHz offers practical filter design and adequate sensitivity at the demodulator. Narrow band FM systems (PMR and digital data links) will use double conversion, with a first IF of 10.7MHz or 45MHz and a second IF of 455kHz. Fig. 2.23. Basic AGC system for an AM receiver. Note that the schematic assumes a positive voltage to reduce gain. For a negative voltage, the polarity of the diode will be reversed.

2.5.1. Discrete IF amplifiers. If a receiver employs a mixer and an IF filter with high insertion loss, most of the front-end gain could be lost. This means that, even with a good LNA, the overall noise figure will be affected by the first IF amplifier and care will be needed in its design. With the increasing trend toward integration, the choice of RF transistors is strictly limited. Many of the older types are no longer available, and it is likely that those that are available will have too much gain at frequencies below 100MHz, leading to instability. At first sight, a single-stage gain of 38dB may sound attractive, but in practice this will almost certainly oscillate. The solution to this problem is to reduce gain through negative feedback. One obvious method would be to include an un-decoupled emitter resistor, but this is definitely not recommended, as it will directly increase the noise figure. Another possibility is to use a low value inductor (or microstripline) in the emitter, but this must be chosen with great care and will need to be

Page 50 shunted by a low value resistor (say 22 ohms) to avoid certain oscillation. The best solution is to use resistive feedback between collector and base (see Fig. 2.24). The capacitor is required as a dc block only, and should have very low reactance at the IF frequency. This circuit may be optimised for a specific gain, using simulation software, but otherwise a value of about 680 ohms will generally be a reasonable compromise. The feedback capacitor may sometimes be shunted with a much lower value (10pF) to ensure feedback is maintained up to the highest frequencies. Fig. 2.24. Using resistive negative feedback on a transistor amplifier stage to reduce gain.

Useful Tip 4: Modern transistors generally have a very high Ft. This means that, even with a low IF, your IF amplifier can oscillate at a high frequency. Unless you are specifically looking for it, this may not be obvious, and will manifest itself by poor IMD and multiple spurious responses. It is therefore important to have good RF decoupling and proper grounding, just as with an RF amplifier stage. 2.5.2. Double Conversion. Why have double conversion? As the receiver frequency increases, so must the first IF increase to avoid problems with image response. For example, if a receiver at 100MHz were to have a first IF of 455kHz, the image would only be 910kHz away, at 100.91MHz. It would be impossible to reject a frequency so close to the wanted frequency and image rejection will be practically zero. With a first IF of 10.7MHz, the image will be 21.4MHz away and, aside from now being well outside the passband, this can be rejected much more easily. At higher frequencies, the first IF should be 45MHz, 70MHz or even higher. The first IF filter will define the channel width and, as previously discussed, this is fine for wideband modulation. But, for narrowband systems we now have a problem. Consider a receiver for narrowband FM (NBFM), with a first IF of 45MHz and having a modulation deviation of only 3kHz. At 45MHz, a deviation of 3kHz represents a tiny percentage of the centre frequency and would be impossible to detect. Also, it would be swamped by residual noise. However, if we now employ a second conversion to change 45MHz down to 455kHz, the problem is solved. A second IF filter at 455kHz will clean-up the signal and a demodulator can now easily resolve the 3kHz deviation. 2.5.3. The Second Mixer and LO. The first IF filter will usually be followed by at least one stage of amplification before feeding into the second mixer. This may be a discrete amplifier stage, as described in section 2.5.1, or part of an IF subsystem IC as described in section 2.5.4. At this point in the receiver chain, we are no longer concerned with issues such as IMD and spurious responses, and a low-level mixer is quite suitable. However, phase noise is still important, as this will be directly imposed on the wanted signal. It is common practice to use a crystal oscillator as the second LO, as this will have inherently low phase noise and good stability. For a first IF at 10.7MHz , the crystal will be 10.245MHz and fundamental mode. For higher first IFs, overtone crystals will be used and the configuration of the oscillator must be changed to suit (see Fig. 2.25 on the following page). It may also be necessary to provide a means of adjustment for

Page 51 fine tuning, so that the IF can be centred in the passband. This could be a manual adjustment by a variable inductor or trimmer capacitor, or with a varicap diode and AFC loop. It is sometimes convenient to use a high-side VCO for the first IF and this will invert the data. For an analog signal, inversion may be unimportant, but for digital data the polarity of the recovered signal is very important. Using a high-side second LO will restore the correct polarity. For example, at 45MHz, either 44.545MHz or 45.455MHz crystals will give a 455kHz second IF, but with opposite polarities for recovered data. Fig. 2.25. Typical second LO as part of an IF system IC, showing variants for fundamental and overtone crystals.

2.5.4. RF/IF System ICs. Due to their wide availability and ease of design, it is unusual to see a receiver design that does not use an IC for the IF subsystem. Exceptions to this are high-quality communications receivers and base-station equipment, where the system specification demands superior performance. IF subsystems comprise the second mixer and LO (as described above), a high gain IF amplifier and a demodulator (FM or AM, as appropriate). Many ICs also contain a noise gate, or squelch circuit and this will be described in section 2.5.6. Manufacturers include Motorola (e.g. MC3357, MC3371), Philips (e.g. SA615D) and Toko (e.g. TK10931V). It has already been stated that AM receivers must not be driven into limiting and must therefore employ some form of gain control (AGC). A similar requirement applies to SSB, QPSK, CDMA, etc, where the amplitude component must be preserved. For the ultimate simplicity and ease of design, there are also some ICs (e.g. Motorola MC13135) that contain all the functionality for a complete Dual Conversion, Narrowband FM receiver, although the maximum RF operating frequency is limited to below 300MHz. 2.5.5. Demodulation. AM detectors. For simple AM detection, a diode is used to rectify the carrier and a low-pass filter then removes the RF component to leave the recovered audio signal. IF subsystem ICs for AM will usually include a demodulation circuit on the chip. FM detectors. For FM systems, the quadrature detector is now used almost exclusively and this will invariably be included in the subsystem IC. It may be shown that when the incoming signal is added to a signal that is phase-shifted by 90 degrees, the mean dc level is proportional to the phase (frequency) difference (the modulation). By filtering the RF, only the modulation component remains. The 90 degree shift is provided by a resonant circuit and the Q of this circuit will determine the bandwidth and sensitivity of the detector. Frequently, a ceramic resonator will be used for this function (see section 2.4.2), as this will give a predictable performance without need of adjustment. Detectors using discrete diodes and inductors are now only found in old equipment, pre-dating the development of system ICs.

Page 52 SSB. The use of SSB (single-sideband) is restricted almost exclusively to the HF radio bands (3 - 30MHz), where its narrow bandwidth and greater power efficiency are useful in long-range communications. Simple SSB detection can be achieved by re-inserting the carrier at the receiver with an adjustable RF oscillator at the IF (BFO). Modern systems will perform this function automatically. FSK. This is simply digital FM for data applications, and standard FM detection applies. GMSK, CDMA, OFDM and other types of digital demodulation. There are many types of digital modulation and most of these are application specific. Demodulation is performed by the DSP (Digital Signal Processing) section of an ASIC (Application Specific IC). Intelligent software plays an ever-increasing role in digital communications. 2.5.6. Noise Blanking. As the received signal becomes weaker, the signal-to-noise ratio at the demodulated output becomes impaired. For voice links, this is measured as SINAD, and a figure of 12dB is generally considered to be the minimum acceptable. Below this, the user is subjected to a loud and objectionable hissing sound. Noise will be at a maximum when there is no signal, as will happen in stand-by mode. It is therefore desirable to remove this noise signal by a process known as Noise Blanking, or Squelch. A typical speech signal has relatively low energy at the highest frequencies, whilst noise has a high energy level across the entire bandwidth. If the received signal is passed through a high-pass filter and then rectified, the resultant dc level will be a measure of the noise content. By setting a variable threshold, this dc level may be used to operate a gate, which mutes the audio at a pre-determined noise level. Many of the IF subsystem ICs have a built-in squelch circuit (see Fig. 2.26 below). Fig. 2.26. A typical squelch circuit as part of an IF system IC. The Squelch control sets the threshold at which the gate switch mutes the audio output.

2.5.7. Data Recovery. With a digital FM/FSK system, the recovered signal is more tolerant of poor signal to noise ratio (or carrier to noise, C/N, in this context), but the effect of noise is to cause phase jitter on the rising and falling edges of the data pulses. As the noise increases, this phase jitter leads to unacceptable timing errors in the recovered data and, in order to minimise the effect, the data is initially fed through a

Page 53 data-slicer circuit. This comprises a high-gain comparator circuit, with a variable threshold set by the mean dc level of the data itself (see Fig. 2.27 below). By this means, only the centre portion of the signal is sampled (at the zero-crossing point), thereby minimising the effect of jitter. Fig. 2.27. A basic data-slicer circuit. The values for the integrator will be determined by the data rate. As the incoming data becomes more noisy, the comparator will clean up the signal to give an output that is relatively free from jitter. 2.5.8. Spurious Responses and Blocking. There are many computer programs and spreadsheets that can be used to predict spurious responses. The number of possible responses is increased by using double-conversion but, fortunately, most of these will only be apparent at very high levels of interferer and may be minimised by good design. The first, and usually most significant spurious response will be the first IF image, but there are many more and some of these are not immediately obvious, often resulting from mixes of high harmonics. It was mentioned in section 2.1 that the reference oscillator for the synthesiser can also give rise to spurious responses and blocking. Even very high order harmonics can cause de-sensitisation of the receiver, if they fall in the passband (The 72nd harmonic of 13MHz is at 936MHz in the GSM band). Example: Consider a receiver covering the band 442 - 448MHz and having a first IF at 45MHz. The second LO frequency will be 44.545MHz (for a 455kHz second IF). The 9th harmonic of the second LO will be 9 x 44.545 = 400.905MHz And the 11th harmonic will be 11 x 44.545 = 489.995MHz These will give rise to fixed in-band responses: 400.905 + 45 = 445.905MHz, and 489.995 - 45 = 444.905MHz. These are both in the receiver pass band. Note also that the 10th harmonic at 445.45MHz is itself in band and could cause de-sensitisation at that frequency. One might expect that these high-order harmonics will be very small, but without careful layout and decoupling, they can easily become a problem. 2.6. Direct Conversion. A technique that has found some favour, due to its simplicity, is Direct Conversion. As the name implies, direct conversion receivers do not use an IF, and convert the received signal directly down to base-band. An immediate advantage is that there is no image response and there is no need for any IF filters. All filtering and processing is done digitally at the base-band frequencies. The disadvantage is that the local oscillator will be at the same frequency as the received signal, and great care is needed to prevent the LO from swamping the front-end and de-sensitising the receiver. This, in turn, means that the LO and mixer must necessarily be low power (poor IMD). It is therefore essential that the front-end should have as much selectivity as possible, with good signal-handling capabilities. Direct conversion receivers will invariably use a specialist IC for all the major functionality and it is not my intention to include any detailed description here.

Page 54 2.7. Measurement Techniques. The following measurements are those most commonly performed on a complete receiver system. Note that a working impedance of 50 is assumed throughout. 2.7.1. Adjacent channel rejection. This is effectively a measure of the performance of the IF channel filter(s). Because the filters can never be perfect, there will be an area of overlap between signals in the wanted channel and signals in the adjacent channel (see Fig. 2.28 below). The presence of a strong signal in the adjacent channel will spill energy into the wanted channel and, if the wanted signal is weak, it will be swamped. Note that a similar effect will be caused by the presence of high phase noise on the LO (reciprocal mixing). Fig. 2.28. Typical IF filter response, showing the overlap of signals in an adjacent channel. There will be a corresponding adjacent channel on the other side of the wanted channel.

Adjacent channel performance is generally measured by injecting a low-level signal at the wanted frequency and adding a second (modulated) signal at the adjacent channel frequency. The output from the receiver is monitored on a Sinad meter (analog systems), or a BER test set (digital systems). The amplitude of the unwanted signal is slowly increased until a failed Sinad, or BER reading is seen. The ratio between the power levels on the two signal generators is a measure of the adjacent channel rejection (in dB). A typical test set-up is shown in Fig. 2.29 below. Fig. 2.29. Test set-up for measurement of Adjacent Channel rejection.
Sig.Gen 1. Wanted channel frequency. 3dB Combiner RECEIVER UNDER TEST SINAD or BER test set.

Sig.Gen 2. Adjacent channel frequency.

Measurement is usually made at a specified input level near to maximum receiver sensitivity. With modulated signals on both channels, the level of the adjacent channel signal is slowly increased until the receiver output fails. Note that the same set-up is used for Blocking and Co-channel suppression tests (below). 2.7.2. Blocking performance. The blocking performance of a receiver is defined by its ability to withstand large interfering signals at frequencies outside the normal receiver pass-band. The system specification will often specify which frequencies, or bands of frequencies are to be tested and the power levels at which they should be applied. The test set-up is identical to that shown above, except that the interfering signal will be un-modulated. When applying specific levels, remember to include the 3dB loss in the combiner. It is common to add a margin for errors due to mismatch at the receiver input, when the interfering signal frequency is well outside the normal pass-band. A 20dB pad may be used, provided that the required levels can still be met.

Page 55 2.7.3. Co-channel suppression. This is a measure of the ability of the receiver to operate in the presence of another signal on the same channel as the wanted signal. Clearly, the receiver cannot work if the interfering signal is larger than the wanted signal, but an FM system should theoretically operate with only 3dB difference. In reality, a figure of 4 - 6dB would be expected. The interfering signal must initially be set about 20dB below the wanted signal level and is un-modulated. It is then slowly increased (in 0.5dB steps) until the output fails. The difference between the two levels is the co-channel suppression. 2.7.4. Receiver sensitivity. This is a measure of the minimum resolvable signal level for the receiver and will be defined by the system specification. It is directly related to the system noise figure (see 2.7.5 below). For a simple analogue system, sensitivity can be measured using a signal generator and a modulation meter (SINAD). The signal generator should be modulated to the specified FM deviation, or AM depth, with a pure 1kHz sine-wave. The RF input level will then be reduced until a specified SINAD figure (usually 12dB) is indicated. Sensitivity is then given by the indicated RF level. For digital systems, a specialist Test Set will be required, with the ability to measure bit-error rate (BER). A suitable RF source should be modulated with a representative digital data stream and the demodulated output from the receiver will be fed to a BER measuring device, synchronised with the modulator. The RF input level will be reduced until a pre-defined BER figure is indicated (say 0.02% BER). This will give the required figure for sensitivity. 2.7.5. System noise. System noise is the overall noise performance of the receiver, from front-end to demodulator. It has already been shown that the most significant contribution to noise will be the RF input filters and the LNA, but subsequent stages will also contribute noise, especially in a poorly designed receiver. Synthesiser phase noise and poorly filtered power supplies can seriously degrade the system noise figure. Unlike the noise measurement for a front-end, system noise can be readily measured with a spectrum analyser. This is the true measure of receiver sensitivity and, although not precise, it will give a good indication of the overall noise figure. In order to make a measurement, we need to sample the IF signal at the end of the receiver chain, but before demodulation. The receiver input should be fed with an un-modulated signal, on channel, at a known level of about -100dBm. The IF output should then be displayed on the spectrum analyser. Set the controls for 10kHz bandwidth and centralise the display. The IF output should be clearly visible and well above the noise floor. Now carefully measure the peak amplitude of the wanted signal and of the noise floor adjacent to the peak (use signal averaging and delta to give a direct reading of the difference). This is actually the C/N value in 10kHz bandwidth. For wideband systems, increase the measurement bandwidth to suit (for GSM, use 100kHz) Example: When displayed on a spectrum analyser, the IF output of a receiver appears as a signal that is 29dB above the average noise floor in a 10kHz bandwidth. The RF input level is -100dBm. The noise in a 10kHz bandwidth for a noise-free receiver would be -174 + 40 = -134dBm and, for a -100dBm input the expected difference would be 34dB. Note that the corresponding figures would be -124dBm and 24dB in a 100kHz bandwidth. The measured difference is 29dB, so it follows that the system noise figure is 34 - 29dB = 5dB. Although this is not an accurate measurement, it is usually within 1dB.

Page 56

Practical Rf Engineering Part III

By Davood Mirzahosseini, KNTU.
Copyright 2007.

Propagation, the Link Budget Modulation, Amplitude Modulation Frequency Modulation Modulation methods, up-mixers, synthesisers, etc. Phase Modulation Digital Modulation, FSK, GMSK, QPSK, etc. VCOs Frequency Multipliers Up Converters Driver Stages The Power Amplifier PA Efficiency and Class of operation, A, B, C, etc. Class D, E and F Linearity Techniques and Spectral Purity Linearisation, Polar Modulation Transmitter Output Filters PIN Switches Isolators and Circulators Directional Coupler for Power Control Power Supply management Measurement techniques Appendices 57 58 60 61 62 62 63 64 64 65 66 67 69 70 72 73 74 75 76 78 79 83

Page 57

Transmitters & Modulation Techniques

3.1. Propagation. Below about 30MHz, refraction in the upper atmosphere causes bending of the radio waves to follow the contours of the Earth (rather like a series of mirrors), thus greatly increasing the distance over which communication is possible. As frequency increases, so the angle of refraction decreases until, at about 30MHz, a range of up to 5,000km is possible in a single hop, with just 1 watt transmitter power. Unfortunately, this phenomenon is unreliable, varying with the time of day (propagation improves during the hours of darkness) and atmospheric conditions. In the lower HF band, the effect of different path lengths (multipath) can mean that two or more components of the same signal arrive in anti-phase, causing serious fading. For reliable communication, the transmitter power needs to be high enough to allow an adequate fade margin. At frequencies above 30MHz, communication is restricted to line-of-sight (transmitter and receiver antennas are directly visible to one another above the curvature of the Earth). At ground level, the maximum range is further reduced by ground effect and by topographical features, buildings and trees. In an urban environment, the predicted range will be reduced by as much as 75%, this effect becoming worse at frequencies above 1GHz. As the height above ground level increases, the range will increase also, not only due to increased line of sight, but also because ground losses are reduced. This is why most transmitter antennas are sited at the top of tall buildings, or on mountain tops. Beyond the Earths atmosphere (in space), the range is limited only by the path loss (see below). An excellent demonstration of this is the SKY TV satellite. This is in geo-synchronous orbit at a distance of 35,700km from Earth and has line-of-sight coverage over almost half the globe. Output power at the transmitter is about 100 watts, but the signal can be received by millions of receivers throughout the Western world. Compare this with ordinary ground-based TV broadcast transmitters, where transmitter power is typically 50 kW to give an area coverage of just 50 km radius. 3.1.1. The Link Budget. The free-space, line-of-sight path loss is given by the equation below: Path Loss (dB) = 32.5 + 20 (log f) + 20 (log d) - (G1 + G2). [Equ. 7] Where f = frequency in MHz, d = distance in km, G1 and G2 are the antenna gains in dBi. Example: The maximum sensitivity of a receiver is -105dBm and its frequency of operation is 432MHz. The gain of the transmitter antenna is 6dBi and the receiver antenna is 2dBi. What transmitter power will be required to achieve a maximum line-of-sight range of 100km? Using Equ. 7, above: Path loss = 32.5 + 20 (log 432) + 20 (log 100) - (6 + 2) = 32.5 + 20 x 2.635 + 20 x 2 - 8 = 32.5 + 52.71 + 40 - 8 = 117.21dB It therefore follows that the minimum transmitter power to achieve -105dBm at the receiver will be 117.21 - 105 = 12.2dBm. This corresponds to a power level of just 16 milliwatts. As explained above, this assumes perfect line-of-sight conditions. In reality, for an actual range of 100km, we would not want to work on the limiting condition. For terrestrial communications, we would normally add a 10dB margin plus an additional 10dB for fades (multipath). If we now add these figures, we see that a practical transmitter power would be (12.2 + 10 +10)dBm =32.2dBm, or 1.6 watts.

Page 58 3.1.2. Multipath. In an urban environment, the RF signal is reflected by buildings and other tall structures. This means that two (or more) signals may reach the receiver antenna, having different path lengths. It should be understood that the phase of a propagated signal rotates through 360 degrees in every wavelength. If the difference in the path lengths is such that they arrive in anti-phase (180 ), then the signals will cancel. If the two signals are of comparable amplitudes, the cancellation may be complete (total loss of signal). As the frequency increases, wavelengths become shorter ( at 900MHz is just 33cm), so the receiver needs to move only a short distance to come out of the fade. See Fig. 3.1 below.

Fig. 3.1. Multipath signals at a receiver in an urban environment.

3.1.3. Dead Spots. A dead spot is where the direct signal is totally blocked and no suitable reflected signal is present. This can occur behind a tall structure, under bridges and in tunnels. A densely wooded area can also act as a total block. Cellular phones work around this problem by having many different transmitter sites , each covering a specific area, or cell (hence the term cellular), giving complete area coverage. 3.2. Modulation. Modulation may be divided into two main categories: Analogue modulation and Digital modulation. There are three basic methods of analogue modulation, AM, FM and PM, as described below. 3.2.1. Amplitude modulation (AM). This is probably the simplest form of modulation and, as the name implies, modulates the amplitude of an RF carrier signal with the modulating signal (usually audio frequencies). Advantages: 1. Simple design of receivers (although modern ICs make FM systems equally simple). 2. Low spectral bandwidth requirement. There are only two major sidebands, directly proportional to the modulating frequency. This is an advantage in the LF, MF and HF frequency bands, where bandwidth is strictly limited. Radio broadcast stations on these bands use AM. Disadvantages: 1. Poor efficiency at the transmitter. In order to prevent distortion of the modulating signal, and to avoid spurious sidebands, the transmitter PA must be operated in Class A. Also, the depth of modulation must never be allowed to approach 100%. This limits the overall power efficiency to less than 20%. 2. None of the stages in the receiver must be allowed to approach limiting, otherwise modulation will be distorted or lost. Automatic gain control (AGC) is therefore mandatory. 3. Susceptible to interference. Any atmospheric or man-made interference in the radio spectrum will be directly picked up by the receiver and superimposed upon the wanted signal.

Page 59 In early transmitter design, AM was applied directly to the final PA stage by modulating its dc supply voltage. This involved the use of a bulky and costly modulation transformer and required a high power modulator (half the power of the PA itself). Modulation was restricted to audio frequencies and, in order to reduce distortion and to prevent excessive sidebands, the modulation depth was never allowed to approach 100% (in broadcast transmitters, this figure was typically less than 50%). The overall transmitter efficiency was therefore very poor, although the use of a transformer gave some advantage in terms of PA efficiency. In later designs, modulation is applied to an earlier (low power) stage in the transmitter chain, where modulation power is very low and there is no restriction on the modulator bandwidth. However, the drawback to this technique is that all subsequent transmit driver and PA stages must be operated in linear mode (class A), otherwise the amplitude component will be lost. This in turn means that PA efficiency will be less than 25%. 3.2.2. Variants on AM (SSB systems). If we examine the AM spectrum, it may be seen that there are two sidebands, one each side of the central carrier, and the second sideband simply duplicates the information contained in the first. It is therefore apparent that one sideband could be removed, thus halving the required bandwidth.

Fig. 3.2. Spectrum of an AM carrier with a fixed modulating frequency. Linear vertical scale.

A technique used for the video signal on terrestrial TV stations, transmits a full carrier with one full sideband, and an attenuated second sideband. This is called vestigial sideband transmission and, by shaping the receiver IF response, a standard AM detector can be used to recover the video signal. It may also be noted that the carrier itself contains no useful information, other than as a frequency reference. By removing one sideband and the carrier, we can transmit one sideband only. This is known as single-sideband, or SSB. A further benefit is that the redundant sideband may now be used for a second, separate, communication channel. The two unique sidebands will be designated USB (upper sideband) and LSB (lower sideband). The use of SSB has significant advantages in terms of efficiency, but a huge disadvantage in terms of receiver complexity. Because there is no longer a carrier as a reference, it is necessary to re-insert a carrier by generating one in the receiver (using a VCO). Also, there is no longer any direct indication of received signal strength, so AGC will not work. Instead, it is necessary to measure the mean signal strength derived from the modulation component and to use this as a gain control. In some systems, an un-modulated carrier at reduced amplitude is transmitted as a reference signal for the receiver (SSB-reduced carrier) and a separate narrow-band receiver is used to extract and to reinsert the carrier. SSB is restricted mainly to long-distance communication on the HF bands, where reduced bandwidth is an advantage. All of these techniques require low-level modulation (before the transmit PA and driver stages) and linear operation through the transmit chain.

Page 60 3.2.3. Frequency modulation (FM). With FM, the modulating signal changes the carrier frequency over a pre-defined range at a rate that is directly proportional to the instantaneous modulating frequency. The maximum change in carrier frequency is known as the deviation, and the ratio of deviation to maximum modulating frequency is known as the modulation index. Advantages: 1. The transmitter PA can be operated in Class C, giving far greater power efficiency than an AM system. 2. The receiver can employ hard limiting, thus making it insensitive to external interference. No AGC is necessary, although it is sometimes employed in the early stages to increase dynamic range. 3. With a low modulation index (NBFM), the occupied bandwidth is only marginally greater than a similar AM system. 4. With a high modulation index (FM broadcast), there is a significant benefit in signal/noise ratio at the receiver output (the noise triangle effect). Radio broadcast transmitters use 75kHz deviation and a modulation index of 5. This allows high quality audio (30Hz-15kHz) and occupies 180kHz of bandwidth (from Carsons Rule below). This bandwidth is further increased to 270kHz by the addition of the 30kHz stereo sub-carrier. Carsons Rule: Required transmitter bandwidth = 2 x (F + Fm) [Equ. 8] Where F is the frequency deviation and Fm is the maximum modulating frequency. Commercial and amateur users are not allowed the luxury of such high bandwidth and, with the ever increasing demand for spectrum, channel bandwidths of just 12.5kHz are common. Using Carsons Rule in reverse, it may be seen that for a modulating frequency of 3kHz max (typical voice quality), the FM deviation must not exceed 3.25kHz. This is known as Narrow Band FM (NBFM) and it is interesting to note that the spectrum is very similar to an equivalent AM transmitter, but with much better efficiency and ease of modulation. NBFM is the preferred system for analogue communication equipment in the VHF and UHF bands (PMR, Amateur Radio, etc). 3.2.4. Modulation methods for FM. In order to produce FM, the modulating frequency must be imposed on the frequency source. The method of doing this will depend upon the nature of this frequency source. In all but the very basic (low-power) transmitters, every transmitter will derive its frequency from a highly stable source such as a quartz crystal. In modern equipment, this will more likely be a TCXO (temperature-controlled crystal oscillator), OCXO (oven-controlled crystal oscillator) or, for the most critical applications, an atomic frequency standard. 1. Direct modulation. It is possible to directly modulate an quartz crystal oscillator, but generally only when a fundamental crystal is used. This is a function of the pull ability and for fundamental crystals is about 0.1%. For a 3rd overtone this falls to 0.01%, and much less for higher overtones. Fundamental mode crystals are normally only available up to a frequency of about 40MHz, where a pull of 40kHz would be possible, although not practical for good stability. Frequency modulation is effected by using a varicap diode, usually in conjunction with an inductor, to pull the crystal (see Fig. 3.3). Precise adjustment of the centre frequency is made by the application of a dc voltage to the varicap and the modulating signal is superimposed on this voltage via an RF isolating network. Because FM is not used at frequencies below 30MHz, its usage being confined mainly to the VHF and UHF bands, one solution is to employ frequency multiplier stages after the oscillator. Thus, 18MHz followed by three doubler stages would give a final frequency of 144MHz (18 x 2 x 2 x 2). This technique also multiplies the frequency deviation by the same factor, so less pulling is needed. Just 1kHz deviation at the crystal would give 8kHz at final frequency.

Page 61 For a single, fixed frequency transmitter, this might be a viable solution, but a serious problem arises for a multi-channel installation. Although it is possible to switch crystals, this is not recommended and separate oscillators are preferred. However, each oscillator will require its own modulator and, as each crystal will have a different characteristic, each modulator will require manual adjustment. An alternative solution, used in early transmitters, was to use integrated phase modulation to simulate FM - see section 3.2.4 below. 2. Up Mixers. Another method of producing FM in a multi-channel transmitters is by up-mixing. A second local oscillator, operates at a fairly low frequency (say 10MHz) and is directly modulated (as described above). Channel oscillators, or a programmable frequency synthesiser, are set to operate at, say, 10MHz below the wanted frequencies. When mixed with the second LO, a modulated signal at final frequency is produced. Note that in this case there is no multiplication of the deviation. Note also that it is necessary to include a filter to remove the unwanted image (f0 - 20MHz) from the transmit output. For more on this topic, see section 3.6. 3. Synthesisers. There are two methods of producing FM with a frequency synthesiser: The first of these is to directly modulate the reference oscillator. This may be a quartz crystal or a TCXO. Where a TCXO is used, this will actually need to be a VCTCXO (Voltage Controlled TCXO). These allow the frequency to be changed over a small range by the application of an external dc voltage. A modulating signal may easily be superimposed upon this dc voltage. Unfortunately, this method has a serious limitation in that it will only work within the synthesisers loop bandwidth. As the loop bandwidth is dictated by other factors (see Part 2, section 2.3.1.), this may be less than the desired modulation bandwidth. For example, a synthesiser with a loop bandwidth of 10kHz could be successfully modulated by a typical 300Hz - 3kHz audio voice signal, but for NBFM systems, the loop bandwidth may be less than 1kHz and therefore cannot be successfully modulated. The second method is to modulate the VCO control voltage within the synthesiser loop itself. This will generally have a very high sensitivity. For example, a VCO with a 10MHz/V gain will require just 1mV to shift the frequency by 10kHz. It is usual to add a separate varicap diode, lightly coupled to the VCO to provide a less sensitive interface for modulation. Care must be taken not to introduce any unwanted noise, as this will cause residual FM and degrade the performance at the receiver. Residual FM is where there is a constant level of unwanted modulation on the signal due to noise. In the worst case, this can be of comparable amplitude to the wanted modulation. Again there is a problem in that this type of modulation only works outside the loop bandwidth, so unless the loop bandwidth is less than about 100Hz, modulation with a voice signal is not possible. Such a low loop bandwidth would have problems with settling time and may require impractically large component values in the filter. In reality, it is common to use a combination of the above methods, called dual-point modulation. Here, the modulating signal is applied to both the reference and the VCO and, by careful balance of the two inputs, it is possible to achieve linear modulation from dc to 10kHz (or more). The drawback here is that it is necessary to manually adjust the input levels to achieve balance. This is best done by using a 1kHz square-wave test signal and observing the resultant modulation on a spectrum analyser (zero scan width), an MDA (Modulation Domain Analyser) if one is available, or a suitable receiver. The level applied to the reference oscillator should initially be adjusted to give the correct deviation. The level applied to the VCO is then adjusted for the best square-wave shape. A second adjustment may be necessary for correct deviation. It may be shown that the non-linearity at the cross-over point is cancelled with this method of modulation. This method works well over a relatively small range of transmitter channel frequencies, but tends to become non-linear and distorted where the VCO is required to cover a large range of frequencies, due to the change in sensitivity of the VCO control voltage. A typical schematic for a dual-point modulator is shown in Fig. 3.3 below.

Page 62 Fig. 3.3. Basic Schematic for a Dual-Point Frequency Modulator. Capacitor values on the VCO show typical relative values for the control and modulating inputs.

3.2.5. Phase modulation (PM). Phase modulation is not used in analogue systems, there being no perceived advantage, except as an indirect method of producing FM. It may be shown that PM is analogous to FM with 90 phase-shift in the modulating frequency vector. Over a limited range then (usually only 300Hz - 3kHz audio), if a modulating signal is applied to a phase modulator via a low-pass integrating network, the resultant modulation approximates to FM. This technique was widely used in early PMR systems, before synthesisers became cheap and widely available. Here, the channel frequency was determined by a fixed quartz crystal at a sub-multiple of the final frequency, as described in section 3.2.4 above, but because there was more than one (switched) oscillator, each would have required its own modulator. The solution to this problem was to apply phase modulation instead to the first multiplier stage. With an audio integrator network, acceptable (narrow-band) FM transmission was possible. 3.3. Digital modulation. Modern communication equipment now, almost exclusively, employs digital modulation. Recent developments have seen the introduction of ever-more sophisticated modulation techniques to make the most efficient use of available bandwidth. Those described below are the most common. 3.3.1. Frequency-shift keying (FSK). This is the simplest method of digital modulation as it is widely used for data transmission. In effect, it is FM with a fixed deviation (frequency shift) corresponding to the data bits. A more sophisticated system, known as POCSAG, is used for pagers and has bi-directional frequency shifts and NRZ data. Modulation techniques are the same as for standard FM. 3.3.2. GMSK. Gaussian Mean Shift Keying is the method of modulation used on existing GSM mobile networks. In principle, it is similar to FSK but with a much higher data rate. In addition, the signal is compressed into a narrow time slot to allow multiple occupancy of the same frequency, and to allow multiplex operation for simultaneous transmit and receive functions (TDMA - Time Division Multiple Access). In order to limit spectral bandwidth, the modulating digital signal is shaped by a gaussian filter. For the same reason, the transmit envelope is shaped to comply with a carefully defined template, thereby minimising any power in the adjacent channels. 3.3.3. QPSK and 8-PSK. Quadrature Phase Shift Keying makes efficient use of the available bandwidth by limiting modulation (phase modulation) to a single quadrant in each 360 phase rotation. This allows two transmit and

Page 63 two receive channels to exist on a single frequency in a single time slot. A further variant, giving even greater efficiency, is 8-PSK. The four quadrants are further subdivided to give 8 channels of data on a single frequency in a single time slot. One drawback of QPSK and 8-PSK (and also of CDMA, below) is the need to preserve the amplitude modulation component in the transmit signal. This requires that the transmitter must operate in Class A (linear) mode. In fact, due to strict specifications for sideband suppression in the adjacent channels, a typical transmit PA will operate at less than 20% efficiency (see paragraph 3.8.5). 3.3.4. QAM, 16-QAM, 64-QAM. Quadrature Amplitude Modulation is similar to QPSK, except that modulation is applied in the amplitude domain. In a bid to achieve ever greater spectral efficiency, some systems employ 16-QAM and even 64-QAM (64 separate sectors in the modulation vector). 3.3.5. CDMA, WCDMA. Code Division Multiple Access transmission uses coded data (which may be pseudo-random), such that only a receiver that is synchronised to the exact same code will be able to receive the signal. By creating multiple codes, many channels of data can be carried in the same transmitter bandwidth. As the data rate is increased, so the modulation bandwidth increases and may occupy several Megahertz of bandwidth (WCDMA - Wide CDMA). This is known as spread-spectrum and, for a single data channel, the signal is virtually invisible to all but a matching receiver. This technique also makes the transmission insensitive to interference or jamming and is therefore much favoured by the military services for covert operations. It is interesting to note that some CDMA receivers can operate with a negative C/N ratio (signal-to-noise) at the input, because only the wanted signal is extracted by the decoder process (known as process gain). A GPS (global positioning satellite) receiver operates with about -10dB negative C/N. 3.3.6. OFDM, OFDMA. OFDM (Orthogonal Frequency Division Multiplex) is the system used in the UK for Digital Audio Broadcasting (DAB) and Digital TV and is another modulation technique that offers highly efficient use of the available spectrum. OFDMA (Orthogonal Frequency Division Multiple Access) is used for the proposed WiMAX broadband wireless Internet system. Due to the very high bandwidth (typically 15MHz, or more), OFDMA will be restricted to microwave frequencies above 2GHz 3.3.7. ASICs for Digital Modulation. Apart from basic FSK, due to the critical and complex requirements, practically all digital systems use an ASIC (Application Specific Integrated Circuit). As the name implies, each IC is application specific and they will not be described here. Modulation usually takes place at base-band frequency (the DSP system), requiring up conversion to final transmitter frequency (see section 3.6). A further requirement is that the transmitter is switched on and off in a strictly controlled manner, according to a pre-defined mask. See more about this in section 3.8.2. 3.4. VCOs. VCOs have already been described in some detail in Part 2, section 2.2.3. The main difference is that a transmitter VCO often runs at final frequency (not with up-conversion). As with the receiver, good design is essential. Excessive phase noise will broaden the transmit spectrum to affect the adjacent channel and any unwanted noise coupled into the VCO control line will cause residual FM. Another problem with final frequency VCOs is that even a small amount of energy coupled back from the transmit power amplifier stages (or the antenna) will cause frequency pulling. This is especially true when the transmitter is rapidly switched on and off, as with many digital systems. Great care must be taken to isolate the VCO and its synthesiser loop from all other circuits. Digital data lines in close proximity to the VCO control line will couple to it, casing unwanted spurious outputs. A regulated dc

Page 64 supply with adequate decoupling, together with good RF screening, is mandatory. The RF output from the VCO should be as lightly coupled as possible to minimise the effect of frequency pulling. 3.5. Frequency Multipliers. The use of frequency multipliers in early designs has already been discussed (section 3.2.3. (1)). It is sometimes advantageous to use a frequency multiplier with a synthesised VCO. The reasons for this are twofold; first the problem of frequency pulling is eased, because the VCO and the transmit PA are no longer at the same frequency and second, the VCO noise may be less. In section 2.2.3 and it was shown that, although noise increases as a function of n2, this may be more than offset by a higher Q in the resonant circuit. For frequencies above 4GHz, multiplication may be more cost effective than the use of prescalers. Note that frequency deviation and channel spacing will also be multiplied. Non-linearity is essential in a multiplier, as it is only through deliberate distortion that harmonics of the input frequency are produced. Multiplier stages are usually limited to doublers or triplers (3x). This is because the harmonic content falls rapidly at higher order harmonics and it would be difficult to extract a useful amount of power. It is usual to employ a bipolar transistor, operating in Class B or Class AB (see Fig. 3.4 below). At extremely high frequencies, beyond 30GHz and up to 250GHz, transistors become impractical. Power varactor diodes can double or triple the final output of a transmitter. These special diodes can offer efficiencies in the order of 70% for doublers, and up to 50% for triplers. Fig. 3.4. A simple doubler circuit for 144MHz. Note that the resonant circuit at the output needs to be high Q, with a high C/L ratio for minimum loading due to coupling capacitor C2.

3.6. Up Converters. Unlike a typical receiver mixer, where the desired output frequency is usually lower than the input frequency, an up converter mixes two frequencies to give an output that is higher than either of the input frequencies. A second LO is mixed with the IF (or base-band) frequency to give the required transmitter frequency. Usually, the second LO frequency will be lower than that of the VCO, but for some applications it may be higher. For example, it is required to transmit an existing (modulated) 925MHz base-band carrier, at a new frequency of 2.45GHz. A fixed frequency second LO, operating at 1.525GHz (2.45 - 0.925), can be added to the base-band to give the desired frequency. A typical up-converter schematic is shown in Fig. 3.5. Advantages: 1. The VCO does not operate at final frequency and therefore gives isolation between the VCO and the transmitter output. 2. Allows the use of a standard (off-the-shelf) VCO, as its frequency need not be final frequency. Alternatively, an existing base-band frequency can be converted to any other frequency. 3. Modulation may be applied to the second (fixed frequency) VCO instead of the main VCO.

Page 65 Disadvantages: 1. Added cost and complexity of a second LO (or synthesiser) plus mixer. 2. The mixer output is usually at low level (say 0dBm), requiring more RF driver stages. 3. Increased spurious outputs require good filters. Fig. 3.5. Basic schematic for an up-converter and Tx filter.

FAQ 6. Should I use a frequency multiplier or an up-converter in my design? This will depend very much upon the application. For a simple fixed-frequency transmitter, using FM or FSK modulation, frequency multipliers are a simple and inexpensive option. The advantages of a frequency multiplier are discussed in section 3.5 above. For modern digital systems, modulation is at base-band (low) frequencies and linear amplification is required. For these applications, the use of an up-converter is mandatory as multipliers are necessarily non-linear. A high-level mixer gives good linearity and low IMD. 3.7. Driver stages. The driver stages are required to boost the power level from the mixer or VCO/buffer output (about 0dBm) to a sufficient level to drive the PA. Clearly, the amount of driver gain required will depend on the actual PA output power and the gain of the PA stage itself. At frequencies around 1GHz, a typical bipolar transistor PA will have a power gain of 10dB. Other technologies (FET /LDMOS) may increase this to more than 16dB. At higher frequencies, the PA gain will be even less and, at 5GHz, a single device may have a gain of only 6dB. As an example, let us say that the required transmitter output power at 925MHz is 10 watts. If the PA device has a gain of 10dB, it will require 1 watt of drive power. It follows, therefore, that for 0dBm input power (1mW) a power gain of 30dB is required from the drivers. It is important to include a band-pass filter to reduce out-of-band gain and to limit noise, which may otherwise affect adjacent channels. If up-conversion is used (see section 3.6 above), this filter must also remove any residual components of the base-band and second LO frequencies. The insertion loss of a typical filter will be 3 - 5dB and will require additional gain, making 35dB overall. With modern devices, it is possible to achieve this gain in two stages, but careful design is essential to avoid instability. A further problem arises with digital systems (briefly discussed in section 3.2.3) in that none of these driver stages must approach saturation. This means that each stage must have a high IP3, commensurate with the level of power it is handling, and that some form of gain control is mandatory. With hand-held equipment, overall efficiency is an important issue, as this will directly affect battery life.

Page 66 3.8. The Power Amplifier. The last active device in the transmitter chain is generally known as the Power Amplifier (PA). This device (sometimes there are two, as in a balanced amplifier) provides the specified RF output power to the antenna. Transmitter output power is generally defined as the power fed to a resistive dummy load, connected at the antenna port. Inevitably, there will be losses between the PA and the antenna port (PIN switch and/or filter/duplexer). A good design will make every effort to minimise these losses, but the actual PA power may need to be 2 - 3dB higher than the specified output power. Thus, for 10 watts output, the PA will need to deliver 16 - 20 watts. Matching. In order to deliver maximum power, the device should be matched at input and output. Figures will be given on the manufacturers data sheet for input and output impedances under different operating conditions. As these impedances are usually complex, it is important to remember that the matching network must be the conjugate of the given figures. For example, a transistor having a Zout = 5 + j3 needs a matching impedance of 5 - j3. The output will generally be matched to 50, but sometimes it is convenient to match the input directly to the driver impedance (non-50). In calculating values for the matching network, it is helpful to use a Smith Chart and there are several useful programs that will do this on your PC. Another factor to be considered is the Q of the matching network. For a single frequency, the Q can be quite high, but where a transmitter is required to operate over a wide band of frequencies, the Q will need to be low. It is no use matching the transmitter at band centre if there is a serious mismatch at the band extremities. Low Q is achieved by using two or more sections in the matching network. The effect of this can easily be seen on the Smith Chart, when Q curves are displayed. A link to a free program, QuickSmith, is given in section 3.17 (Appendices). At microwave frequencies, lumped elements (capacitors, inductors) become unsuitable as tuning components and are used primarily as chokes and bypasses. Matching, tuning, and filtering at microwave frequencies are therefore accomplished with distributed (transmission-line) networks (see Fig. 3.8). It is common to use a transmission line between the device and load to provide the desired matching value. A stub that is a quarter-wavelength at the frequency of interest and open at one end provides a short circuit. Similarly, a quarter-wavelength shorted at one end provides an open circuit. Stubs that are less than a quarter-wavelength behave as capacitors. Example: A PA output device has an output impedance of 6 + j4. For maximum power, the device must therefore be matched with 6 - j4. A fixed capacitor (to ground) will provide the -j4 term, so the resistive load should be 6. If the supply voltage is 12V, what is the maximum RF output power for Class A operation? If the RF load is 6 and the output impedance is 6, the effective resistive impedance is 12. The input power will therefore be: E2/R = (12)2/12 = 12 watts The rms value of the output waveform will be Vs x 1/22 = 12 x 0.3536 = 4.243V The power delivered to the load is therefore E2/R = (4.243)2/6 = 3.0 watts. This is one quarter of the total dc input power, making the efficiency of the stage just 25%. This ignores saturation voltage at the output device and is the best possible for Class A . In a practical output stage, even lower efficiency is probable. PA Modules. For frequencies above 120MHz, it has become increasingly common to use a PA module. These offer a compact and convenient solution and are readily available for most of the standard frequency bands in the VHF/UHF ranges and with output power in the range 1 - 30 watts. Manufacturers include Mitsubishi, Philips, Motorola and Celeritek. Although discrete PA devices are still available, their use is now mainly limited to base stations, where size is less important and much higher power may be required. Fig. 3.6 shows an actual PA module for 2.4GHz.

Page 67 Advantages: 1. Guaranteed performance. 2. Extremely compact size. 3. 50 input and output impedances (requires no matching). Disadvantages: 1. May not be available for less popular frequency bands. 2. Single source manufacturer. 3. Can be more expensive than discrete equivalent. Fig. 3.6. A Celeritek PA module for 2.4GHz (on the right-hand section of the board), together with its driver, filter and other associated components. Note that this module is rated for 1 watt output power, but for linear operation it was backed off to just 200mW. Example: The required output from a transmitter is 10 watts at 1.28GHz, operating in linear mode. Preferred solution: Mitsubishi PA Module type RA181213G. 3-stage MOSFET RF Power Module. Frequency Range = 1.24 - 1.3GHz. Pout = 18W. Pin = 0.2W. Efficiency = 20% (typical). Zin/Zout = 50. Vsupply = 12.5V The data sheet for this device shows that it will easily deliver the required output power. Linear mode is achieved by controlling the RF input drive power and the gate voltage. Drive requirements: The input power for 10 watts out will be about 100mW (20dBm), but a filter should be included before the PA and will probably add 3dB of insertion loss. The pre-driver stage should therefore deliver +23dBm of power. With a VCO output of 0dBm, it is possible to get 23dB gain from a single stage at this frequency.

3.8.1. PA Efficiency and Class of operation. The bias conditions for the PA stages and the mode of operation are generically defined by class. Definition of the various classes is as follows: Class A. When the active device is biased for linear operation, such that any small change at its input causes a corresponding, but much larger change at its output, this is defined as Class A operation. When this is applied to a PA stage, a constant high current will flow through the device. Maximum power will be when the load equals the (resistive) output impedance of the device and the peak-to-peak output voltage is equal to the supply voltage (less any volt-drop across the active device itself). Maximum stage efficiency is 25% and, for good linear performance, may drop to 15 - 20%. Class B. For Class B operation, the active device is biased at the cut-off point (i.e. zero current) and conducts only during the positive half of the drive cycle. In the absence of a tank circuit, this would act like an inverting half-wave rectifier and only negative-going half cycles would appear at the output. The

Page 68 tank actually comprises a matching network with a Q value greater than unity and preferably at least 3. To recover the missing half cycle, the tank allows the supply voltage to over-swing by an amount equal to the negative excursion, thus effectively doubling the supply voltage and hence the available RF output voltage. Since the actual dc supply does not change and average current remains the same, it follows that the PA efficiency is doubled to 50%. Again, this is an ideal figure and actual efficiency may be more like 45%. Class AB. In Class AB, the active device is biased such that it is just turned on, but the quiescent current is very much lower than for a Class A amplifier. Class AB is not linear, and so could not be used where a linear amplifier is required. Its benefit is that it is more efficient than a Class A stage, but requires less drive power than Class B or Class C. Class C. Class C operation is very similar to Class B, except that the active device is biased beyond cut-off. With discrete silicon transistors, this condition is conveniently achieved by simply omitting any dc bias components and returning the base-drive input to ground through an RF choke, or resistor. The drive voltage will cause base current to flow during the positive half-cycle and the rectifying action of the base-emitter junction will result in a negative dc voltage on the base. The angle of conduction will depend upon the amplitude of the drive voltage and the value of resistance in the return path. Thus, a large value of resistor would cause a large negative bias and the transistor would conduct only on the tips of the drive waveform, resulting in little or no output power. A low value of resistor will result in a larger angle of conduction and, with zero ohms, the bias will effectively be the fixed Vb-e of the transistor itself (about 0.7V). For adequate drive to achieve the desired power output, the conduction angle should not be less than 60, where a PA efficiency of up to 70% is possible. See Fig. 3.7 below for a typical circuit. Class C output stages are quite common for FM (or FSK) transmitters in the VHF and low UHF bands, but as the operating frequency is increased, it becomes difficult to achieve sufficient gain in the PA stages without using some forward bias. For very high power transmitters using vacuum tubes, a fixed negative supply voltage is required to achieve Class C operation. Fig. 3.7. A Class C power amplifier for the 440MHz band. Self-bias is produced by conduction at the base-emitter junction and is proportional to the drive current and the value of base resistor.

Page 69 Fig. 3.8. A 2-stage PA for 1.8GHz, using FETs and transmission line matching elements. Note that this is not operating in Class C - these devices require a negative gate voltage for normal conduction.

Class D. A Class-D PA uses two or more transistors as switches to generate a square-wave at the transmitter frequency. A series-tuned output filter passes only the fundamental-frequency component to the load. Current is drawn only through the transistor that is on, resulting in a theoretical 100% efficiency for an ideal PA. If the switching is sufficiently fast, efficiency is not degraded by reactance in the load, but practical PAs suffer from losses due to saturation, switching speed, and output capacitance. Finite switching speed causes the transistors to be in their active regions while conducting current. Output capacitances must be charged and discharged once per RF cycle, resulting in power loss that is proportional to, and increases directly with frequency. Class-D PAs with power outputs of 100 W to 1 kW are readily implemented at HF, but are seldom used above lower VHF because of the losses associated with output capacitance. Class E. Class E employs a single transistor operated as a switch. The load voltage waveform is the result of the sum of the dc and RF currents charging the load shunt capacitance. For optimum performance, the PA voltage drops to zero and has zero slope just as the transistor turns on. The result is an ideal efficiency of 100%, elimination of the losses associated with charging the load capacitance in class D, reduction of switching losses, and good tolerance of component variation. Variations in load impedance and shunt susceptance cause the PA to deviate from optimum operation, but the capability for efficient operation in the presence of significant drain capacitance makes class E useful in some applications. High-efficiency HF PAs with power levels to 1 kW can be implemented using low-cost MOSFETs intended for switching rather than RF use. Class E has been used for high-efficiency amplification in the PA for a 900MHz CDMA handset, using a single GaAs-HBT RFIC that includes a single-ended Class-AB PA. A typical PA module produces 28 dBm (631 mW) at full output with a typical PA efficiency of 35 50%. A useful (free) design program may be found at .

Page 70 Class F. Class F boosts both efficiency and output by using harmonic resonators in the output network to shape the drain waveforms. The voltage waveform includes one or more odd harmonics and approximates a square wave, while the current includes even harmonics and approximates a half sine wave. Alternately (inverse class F), the voltage can approximate a half sine wave and the current a square wave. As the number of harmonics increases, the efficiency of an ideal PA increases from 50% toward 100% (e.g., 70.7, 81.65, 86.56, 90.45 for two, three, four, and five harmonics, respectively). The required harmonics arise naturally from non-linearity and saturation in the transistor. While class F requires a more complex output filter than other PAs, the impedances at the virtual drain must be correct at only a few specific frequencies. A variety of modes of operation in-between classes C, E, and F are possible. 3.8.2. Linearity Techniques and Spectral Purity. With the exception of GMSK (GSM mobile network) and FSK, digital modulation techniques usually require linear output at the PA. This is because the modulation spectrum contains components in both the frequency and the amplitude domains. However, another and equally important consideration is that non-linearity creates unwanted sidebands in adjacent channels. For any multi-channel system, there are strict limitations on the permitted level of adjacent channel power, which would otherwise degrade the performance for other users. Even with normal Class A operation, there is sufficient non-linearity to fail these specifications, and it is necessary to further reduce output power and efficiency in the interests of spectral purity. This means that a PA device (or module) having a specified RF output power of, say, 50 watts may need to be run (backed-off) at only 10 watts to ensure the required linearity. With such poor efficiency, any means of increasing this efficiency is advantageous. Although there are well-understood methods for improving PA efficiency through bias point setting, these methods are not compatible with complex modulations used in current digital wireless devices. One such technique is known as Load-pulling. In normal operation, maximum power is delivered to the load when the load and output impedances are the same. However, we are no longer talking about normal operation, because the output power has been deliberately reduced to well below that which is available. In these circumstances it has been found that, for a given current through the PA device, more power may be delivered to the load when there is an intentional mismatch. This is called loadpulling and the degree of mismatch must be discovered experimentally, as it will depend upon the device in use. Fig. 3.9 below shows the spectrum of a correctly designed transmitter (with digital modulation) and that of an over-driven transmitter. It may be seen that for the Ideal PA, very little power is produced in the adjacent sidebands. The over-driven PA produces a totally unacceptable level of power in the adjacent sidebands. A realistic specification might call for better than 50dBc adjacent channel power. Fig. 3.9. Typical transmitter output spectrum, showing the relative levels of power in the adjacent channels for an ideal PA, versus one that is overdriven.

Page 71 Another successful technique, known as Polar Modulation, uses intelligent software (DSP) to split the signal into two paths. See section 3.9 for more on this. Envelope shaping. Most digital systems require that the transmitter output is switched rapidly on and off, corresponding to different time slots. This appears as a step-function, which is potentially rich in harmonics and would generate a large amount of energy in adjacent channels. The resulting output spectrum would be similar to the over-driven PA, shown above. In order to prevent this, the transmit envelope must be shaped according to a pre-defined mask. There must be no sudden changes, and all transitions must follow a smooth curve. Rise time must be rapid enough, so that the transmitter has reached full power before the first data bit occurs and the fall time must ensure that the transmitter is fully switched off before the end of its time slot. Envelope shaping is achieved by careful control of the RF drive levels and will usually be implemented by a D-A converter and a voltage-controlled attenuator. The required levels during each part of the transmitter envelope are pre-programmed by the system microprocessor. Fig. 3.10. A typical controlled envelope shape for a transmitter output (Output power dBc versus time ms).

3.9. Linearisation Techniques. Portable wireless devices are becoming smaller, lighter, and more powerful with each new generation of cell phone technology. The output power amplifier (PA) stage consumes a big part of the battery capacity in a portable device. Therefore, by reducing consumption in the PA stage, designers can achieve a profound impact on overall device battery life. Fortunately, there are methods available to combine baseband signal processing with basic analog/RF signal processing to linearise the PA and still maintain relatively high efficiency. Feedback linearises the transmitter by forcing the output to follow the input. It can be applied either directly to the RF amplifier or indirectly to the modulation envelope or phase components. In RF feedback, a portion of the RF-output signal from the amplifier is fed back to and subtracted from the RF-input signal without detection or down-conversion. The delays involved must be small to ensure stability, and the loss of gain at RF is a significant design issue. The use of RF feedback in discrete circuits is usually restricted to HF and lower VHF frequencies, but it can be applied within MMIC devices well into the microwave region. Envelope feedback reduces distortion associated with amplitude non-linearity. It can be applied to either a complete transmitter or a single PA. The RF input signal is sampled by a coupler and the envelope of the sample is detected. The resulting envelope is then fed to one input of a differential amplifier, which subtracts it from a similarly obtained sample of the RF output. The difference signal, representing the error between the input and output envelopes, is used to drive a modulator in the main RF path. This modulator modifies the envelope of the RF signal, which drives the RF PA. The envelope of the resulting output signal is, therefore, linearised to a degree determined by the loop gain of the feedback process. For a BJT amplifier in which amplitude non-linearity is dominant, two-tone IMD is typically reduced by 10 dB.

Page 72 Polar Modulation. Polar modulation is one of the more common techniques in use today, that allows designers to build power amplifiers with efficiencies as high as 60 percent, thus reducing overall system power. The polar loop overcomes the fundamental inability of envelope feedback to correct for AM-PM distortion by adding a phase-locked loop to the envelope feedback system. Envelope detection and phase comparison generally take place at the IF. For a narrow-band VHF PA, the improvement in two-tone IMD is typically around 30 dB. The envelope bandwidth must be at least twice the RF bandwidth, but the phase bandwidth must be at least ten times the RF bandwidth. One form of modulation that is not impaired by non-linear amplification is phase modulation (PM). It therefore follows that a PM modulated signal can be faithfully reproduced with good efficiency. With polar modulation, designers can separate the AM component out of the waveform, amplifying only the remaining phase modulated signal. The AM component is then re-inserted after the PA stage to restore the modulation back to its original form. To recover the AM component, the input bias on the PA can be optimised such that the PM signal is hard limited, basically pushing the PA into a Class E state (switched mode). With the PM component hard limited, the fundamental output power of the PA can be made proportional to the square of the drain bias voltage. The drain voltage can then be modulated with the AM component to effectively recombine the AM information with the PM information. In this way, the intended modulation can be reconstructed with low distortion, while maintaining very high PA power efficiency. See Fig. 3.11. Fig. 3.11. Block diagram of a simple feed forward polar modulator.

The polar modulation technique shown above is a type of feed-forward linearisation. To make the system function properly, the gain and delay characteristics of the PA must be understood so the magnitude and phase paths can be properly synchronized for minimum modulation distortion at the output. Digital signal processing technology allows the PA characteristics to be stored in memory to account for variations with temperature and voltage (bias). A more advanced topology employs a feedback loop with the PA output magnitude. This control loop accounts for variations in PA gain due to temperature and any non-linearity in the magnitude feed forward path. Sigma-delta modulation is used in the feedback path to provide a stable control voltage to the PA drain bias for magnitude reconstruction. In this configuration, the PA is driven with +10 dBm, driving the PA with the PM component as a modulated RF signal. The operating efficiency of the PA under these conditions is around 60 percent. See Fig. 3.12.

Fig. 3.12. Block diagram of a feedback polar modulator.

Page 73 3.10. Transmitter Output Filters. The purpose of the transmitter output filter is to remove all spurious frequencies outside the allocated system bandwidth. These will usually be transmitter harmonics, local oscillator products (where an up-converter is used) and any other spurious frequencies that may find their way into the transmitter chain. As this filter is directly in the output path, it is very important that it has a low insertion loss (less than 3dB) and that it can handle the high power. In a mobile phone, the filter will be part of the duplexer/diplexer. As was discussed in Part 1, section 1.3, there are many different types of filter, dependent upon the frequency of operation and architecture, and the same criteria apply to the transmitter output filter. In a receiver, low insertion loss is important because it directly adds to the noise figure. Insertion loss in a transmitter represents loss of output power. Where discrete components are used in a filter, it is important to remember that these components must be capable of handling the RF currents involved. For example, in a 10 watt transmitter, the RF current will be 0.447 amps and 0603 type chip inductors would be entirely unsuitable here. Fig. 3.13 below shows a low-pass filter (LPF) for 900MHz (component values become impractical above 1GHz ). Fig. 3.14 shows its predicted response. It also shows an Elliptic Filter, which can be very useful in some applications. Fig. 3.13. A low-pass filter and an elliptic filter for 900MHz. The elliptic filter can be very useful for suppression of a troublesome harmonic or spur at a specific frequency. Close tolerance components are required for the series elements for matching response. Capacitor values must include board strays and actual (real) values will be less than those shown.

Fig. 3.14. Frequency response of the two filters in Fig. 3.13. The elliptic filter (trace 2) has better than 80dB rejection at 1.25GHz , compared with 23dB for the LPF, but has less rejection at higher frequencies.

Page 74 Fig. 3.15. A typical microstripline filter. This would be etched on a pcb with a solid ground plane on the next layer. The wide sections of line act as capacitors (to ground) and the narrow lines act like (but not the same as) inductors. 3.12. PIN Switches. The PIN diode is a special kind of diode that is used for switching RF signals. Unlike a normal diode that switches off abruptly when the polarity is reversed, the PIN diode is very slow to turn off. This means that when it is biased on by a continuous forward dc current, it will exhibit a constant (low) forward resistance to an applied RF signal, even though the signal is changing polarity. In this respect it acts like a physical switch and can be used to switch between transmit and receive functions in a simplex system, as well as other RF switching functions. One possible configuration for a Tx/Rx switching circuit is shown in Fig. 3.16 below. Fig. 3.16. PIN diode switch for Tx/Rx function. The two inductors act only as RF chokes to isolate the dc switching circuit. Operation is as follows: Q1 on /Q2 off (Tx enable), current flows through L1, D1, L2 and R1 to enable the transmit path. The PIN diode current is set by the value of R1 and the voltage across R1 (about +9V) appears as a negative bias on D2, fully isolating the receive path. Q2 on /Q1 off (Rx enable), current flows through R9, D2, L2 and R1 to enable the receive path. As the receiver power will be very small, the current through D2 (set by R1+R9) can be significantly less to conserve power. Note that, although the PIN diodes chosen must be capable of handling the RF current, the forward dc current through the diodes has no direct relationship with the transmit power and needs only to hold the diodes in their low resistance state. Often it is necessary to short-circuit the receiver input during the transmit function, using an additional PIN diode, otherwise enough signal can reach the receiver input to cause heavy saturation. Although this is not a problem in itself, it will take a finite time for the receiver to recover, giving an undesirable delay before the receiver will function. Alternative configurations are possible where the shorting function is performed without the need for an additional diode. PIN switches are also used in other applications, such as switched attenuator pads or path selection.

Page 75 3.13. Isolators and Circulators. An RF isolator is a two-port passive device made of magnets and ferrite material which is used to protect other RF components from excessive signal reflection. An RF circulator is a three-port passive device used to control the direction of signal flow in a circuit. To understand how these components control the signal flow, think of a cup of water into which you place a spoon and stir in a clockwise motion. If you sprinkle some pepper into the cup and continue to stir, you will notice that the pepper easily follows the circular motion of the water. You can also see that it would be impossible for the pepper to move in a counter-clockwise direction because the water motion is just too strong. The interaction of the magnetic field to the ferrite material in isolators and circulators create magnetic fields similar to the water flow in the cup. The rotary field is very strong and will cause any RF signal at one port to follow the magnetic flow to the adjacent port, but not in the opposite direction. Fig. 3.17 shows the schematics for a circulator and an isolator. Notice how an isolator is a circulator with the third port terminated. The arrows represent the direction of the magnetic fields and the signal when applied to any port of these devices. Fig. 3.17. If a signal is connected at port A, and port B is well matched, the signal will exit at port B with very little loss (typically 0.4dB). If there is a mismatch at port B, the reflected signal from port B will be directed to port C. An important consideration when specifying an isolator or circulator is to ensure that the device has adequate isolation between the ports for your given application. The amount of isolation is directly affected by the load at port 3. If the match on port 3 is poor, isolation may be less than 10 dB, but if the match is improved by using a good quality termination, then isolation will be better than 20 dB. Isolators and Circulators are generally available in the frequency range 0.8 - 8GHz. 3.13.1. Applications. The output of a transmitter will ideally be matched to its antenna at all times, but if there is a serious mismatch (e.g. the antenna is broken or disconnected), all of the power will be reflected back to the PA device. This could result in overheating of the PA and possible failure. An isolator in the transmit path will divert the reflected power to the termination load instead of the PA device. Note however that the termination load must be capable of handling the reflected power, which may be as much as the rated PA output power. Another good reason for using an isolator in the PA output is to avoid Reverse IMD. This is where two or more transmitters are co-sited (as with mobile phone base stations) and a strong signal from an adjacent transmitter feeds back to the PA device in the first transmitter. This signal can modulate the output of the PA, giving rise to unwanted inter-modulation products (IMD) that will appear as spurious outputs. The inclusion of an isolator will prevent reverse IMD. A Circulator is commonly used as a duplexer (a transmitter and receiver sharing an antenna). Figure 2 in Fig. 3.18 below, shows that when a transmitter is connected at port A, the output goes directly to the antenna at port B with typically less than 0.4dB loss. The isolation due to the circulator means that only a small signal reaches port C, the receiver input. However, as with the PIN switch circuit described in the previous section, the attenuated transmitter signal may still be enough to saturate the receiver and a shorting path (probably using a PIN diode) must also be provided at the receiver input.

Page 76 Figure 3 in Fig. 3.18, shows a test bench application where an isolator is placed in a measurement path so that any mismatch in the device under test (DUT) will not reflect back to the signal source. Fig. 3.18.

3.14. The Directional Coupler. The directional coupler comprises two parallel transmission lines, separated by a small gap. It may be shown that the signal coupled from the first line into the second is related to the length of the lines and the width of the gap, and that the phase relationship is 180. Up to a maximum frequency of about 2GHz, a directional coupler can be made on standard FR4 board, but tends to occupy a large amount of board space. For 0.5mm thickness board, a 50 transmission line will be approximately 1mm wide and the minimum (practical) gap is 0.1mm. To obtain a useful coupling factor, the lines need to be at least 30mm long. It is better to buy in a ready-made unit with a defined performance and which will be physically much smaller. A typical coupling factor would be 10:1 (20dB). Applications include VSWR measurement and power monitoring, especially in a feedback power control circuit. See Fig. 3.20 below. Fig. 3.19. Schematic for a directional coupler. Forward power (IN to OUT) will produce a voltage at the CPL FORWARD terminal, whilst reflected power will produce a voltage at CPL REVERSE. For measurement purposes, all ports must be terminated with 50. Fig. 3.20. A power control loop using a directional coupler. Note the use of a shottky diode detector.

Page 77 The schematic in Fig. 3.20 shows a Power Control Loop for active control of the transmitter output power. In the example shown, a Mitsubishi PA module was used and has a convenient power control feature. Note the use of a shottky diode as the signal detector - actually a dual type HSMS2815. Unlike a standard silicon diode, which has a conduction threshold voltage of about 600mV, a shottky diode has a conduction threshold of less than 150mV. As a power control detector, this means that it can operate down to much lower levels than would otherwise be possible. The range has been further increased by the application of a small forward bias through R1, R6. In this application, the coupled line is not terminated with 50 (in the forward direction) because we need to maximise the voltage at the detector. For precise measurement of forward and reverse power (SWR measurements), a connectorised and calibrated line would normally be used (see section 3.16 on measurement techniques). At frequencies below about 300MHz, the Bird Thru line SWR meter gives a simple measurement of forward and reverse power, using a range of plug-in sensors. This is in fact a directional coupler. Directional couplers are also a convenient means of monitoring the output of a transmitter. A 20dB coupler will give a sample of the output signal that can be directly connected to most test equipment, without the need for additional attenuators. SWR. (Standing Wave Ratio). To calculate SWR from the forward and reverse power measurements: SWR = (Pf + Pr)/(Pf - Pr) [Equ. 9] where Pf is the measured forward power and Pr is the measured reverse power. Remember that this is a ratio. I am never quite sure which way round this should be ...sometimes it is given as (for example) 2:1 and sometimes as 1:2. Example 1: A 10 watt transmitter is connected to an antenna that is not optimised for the frequency in use. Using a directional coupler, the measured forward power is 6.2 watts. The measured reverse power is 3.1 watts. Note that the sum of forward and reverse power may not necessarily add up to the expected 10 watts, because the PA is no longer correctly matched to the load. Using Equ. 9: SWR = ( Pf + Pr)/(Pf - Pr) = (6.2 + 3.1)/(6.2 - 3.1) = 9.3 /3.1 SWR = 3.0 .which would normally be stated as 1:3 This would be rather unsatisfactory, as one-third of the available transmitter power is lost. Example 2: The SWR of an antenna is given as 1:1.5 at the frequency of interest. What will be the reflected power, assuming that the forward power is 10 watts? So, using x as the unknown in the equation: 1.5 = (10 + x)/(10 - x) then, 10 + x = 15 - 1.5x and 2.5x = 5 Therefore x = 5/2.5 = 2 Hence the reflected power will be 2 watts. Most standard antennas are not much better than this and a SWR of 1.5 is considered quite good. To convert SWR to dB (as in Return Loss): Power dB = 10 x log [(SWR +1)/(SWR - 1)] [Equ. 10]

Example 3: If the SWR is 1:1.5, what power does this represent as a return loss? dB = 10 x log [(1.5 + 1)/(1.5 - 1)] = 10 x log (2.5/0.5) = 10 x log 5 = 6.99dB This agrees with Example 2. 6.99dB is a factor of 5 (approx.), and 2 watts is one-fifth of 10 watts.

Page 78 3.15. Power Supply Management. Because power is routed to all of the active devices, it follows that unwanted signals can be coupled through the power supply. Great care should be taken to ensure adequate decoupling, both at the RF frequencies and, where applicable, switching frequencies. It is common to see triple decoupling, a low value (self-resonant) for the RF, an intermediate value (typically 0.1uF) for switching and other non-RF signals, and a large value (typically 10uF or more) for LF decoupling. Voltage regulator ICs generally specify a capacitor of about 0.1uF to be sited as close as possible to the regulator input. This is essential to prevent oscillation. It is advisable to route power lines as far as possible away from circuits such as synthesisers and VCOs. For stable performance, separate voltage regulators are mandatory for these circuits and it should be understood that some voltage regulators are themselves noisy. For example, the old 78L05 type regulators (surprisingly) have a better noise performance than many of the later types. There will be high RF currents in and around the PA area and it is advisable to isolate this as far as possible from other areas of the board. Good grounding is essential (multiple vias), not only for a low impedance to ground, but also to help conduct heat to the copper ground plane. The main supply voltage should be carried on separate tracks, free of spur tracks to other circuit areas (see Figures 3.21 and 3.22 below).

Fig. 3.21. Schematic showing correct routing for power supply and decoupling components.

Fig. 3.22. The schematic below shows how not to route the power supply and decoupling components. RF decoupling will be ineffective unless it is close to the point that needs decoupling. Heavy current in the PA supply feed will couple to other devices connected to the same line.

Page 79 3.16. Measurement Techniques. The following measurements are those most commonly performed on a complete transmitter system. Note that a working impedance of 50 is assumed throughout. Unless you are working only with low power transmitters (less than 1 watt), a mandatory item of test equipment will be a high power fixed in-line attenuator (see Fig. 3.23). Most test equipment will be damaged by high levels of RF power and an attenuator will reduce the output power level to a safe value. A 20dB attenuator will reduce 10 watts to 100mW and 30dB will reduce 100 watts to 100mW. At this level, small low power attenuators can be added as required; a further 10dB will reduce the power to 10mW (+10dBm), which is a safe level for all common test equipment. Fig. 3.23. A 30dB, 100 watt fixed in-line attenuator.

Most equipment designers will be working to an ETSI or IEEE specification and the test requirement and method of measurement will already be defined. The following measurement techniques are therefore intended as general guidance only. 3.16.1. Power Output. To measure power output, an obvious choice might seem to be a Power Meter. However, this is of very limited value and, unless it is a specialist instrument with an external trigger and sample-hold feature, it can only measure average power. In a digital system, the transmitter is only enabled for a short time and the average power is unrepresentative. Power output will usually be measured on a Spectrum Analyser. Many of the modern instruments have built-in specialist programs and masks for digital systems. As described above, the signal must first be attenuated to a suitable level (not more than +20dBm to avoid the possibility of overload). The centre frequency of the analyser should be set to the expected transmit frequency and the vertical and horizontal scales adjusted for best display. For a digital system , an external trigger must be used to synchronise the analyser with the transmitter burst. The peak amplitude of the displayed signal will be a measure of the maximum output power, but remember to add on the attenuator value. For example, if a 20dB attenuator is used and the reading on the Spectrum Analyser is +13dBm, then actual power output will be 13 + 20 = 33dBm (2 watts). 3.16.2. Harmonics and Spurious. Again, the instrument of choice will be a Spectrum Analyser, preferably with a frequency range up to at least 10GHz. It is important that the analyser is not overloaded or it will generate harmonics of its own. A maximum input of +10dBm is recommended. Initially, the analyser will be set for wide scan so that any large signals can be identified. A more careful analysis should then be done, with the spectrum divided into convenient bands and with the signal averaging function enabled to show up any non-constant signals. For digital systems, harmonics of the transmitter output are best identified by synchronising the scan with the external trigger and then setting the analyser frequency to the expected value. Remember that you need to add the value of the external attenuator to all measurements.

Page 80 3.16.3. Adjacent Channel Power. For most digital systems, the adjacent channel power is directly affected by the linearity of the PA stages and any significant output in adjacent channels may readily be seen (and measured) on the spectrum analyser. The externally triggered and synchronised set-up used for power measurement can be adjusted to show adjacent channel power (see Fig. 3.9.). Some analysers can do this automatically. For a narrow-band (simplex) system it is rather more difficult, as the channels will be too narrow for resolution by the filters in the spectrum analyser. In this application it is necessary to use a good channel filter to remove the on-channel signal. As channel filters are designed for the receiver IF (say 10.7MHz), it is necessary to first convert the transmitter output frequency down to IF, using a passive mixer and a low-noise signal generator. A typical set-up is shown in Fig. 3.24 below. Fig. 3.24. A typical test set-up for the measurement of adjacent channel power. Transmitter under test.


Mixer Low noise Signal Generator.

Band-pass (channel) filter.

Spectrum Analyser


Notes: 1. The channel filter should be a high quality type with good out-of-band attenuation and must be correctly terminated (this will probably not be 50, so a matching network must be included). 2. The mixer should be a level 6 (+6dBm) type and must be correctly terminated at the IF port. 3. The signal generator should be a low-noise type and its output set to a level of +6dbm. 4. The attenuator in the transmitter path should be such that not more than 0dBm level reaches the mixer input. Procedure: The transmitter is set to transmit with normal modulation on mid-band channel frequency and the signal generator frequency is set to (transmitter frequency - IF) to produce an output from the mixer at IF (10.7MHz, or other). The spectrum analyser is set to display the IF output signal and the level is set to reference level (full screen). The Delta Marker function should be used, if available. Now increase the signal generator frequency by an amount of one channel-space (e.g. 12.5kHz). The spectrum analyser display will now show the output level in the adjacent channel and the delta marker will give a direct readout in dBc (normally this should be better than -50dBc). This will be the upper adjacent channel. To display the lower adjacent channel, reduce the signal generator frequency by an amount of two channel-spaces. This measurement may also be repeated at the band edges (lowest and highest channel frequencies). 3.16.4. Transient Adjacent Channel. This measurement will only be required on a simplex system and is a measure of the instantaneous output in an adjacent channel when the transmitter is switched on and off. The test set-up is the same as in Fig. 3.24 above, but the transmitter is un-modulated and the spectrum analyser is set for zero scan width and a slow scan (say 5 seconds). Some means will be required to switch the transmitter on and off manually, at about a 0.5 second rate. Procedure: With the transmitter set on mid-band channel frequency but un-modulated, set the signal generator to (transmitter frequency - IF), as before. Adjust the spectrum analyser display to reference level and set the delta marker. Now increase the signal generator frequency by an amount of one channel-space. The trace should fall to less than the level measured for adjacent channel power, as there is no longer any modulation. Switch on the peak hold function and switch the transmitter on and off as rapidly as possible. If transient adjacent channel interference is present, it will be displayed as a series of peaks across the width of the screen. The amplitude of these peaks is a measure of the transient adjacent channel interference and should normally be better than -45dBc.

Page 81 3.16.5. Frequency Pulling. Just as transient adjacent channel performance is normally associated only with simplex systems, so frequency pulling is associated only with duplex and digital systems, involving the use of one or more synthesiser loops. In Part 2 of this book it was shown that a synthesiser is extremely sensitive to any disturbance in the control loop voltage. Due to the large change in supply current as the transmitter is turned on and off, such disturbances can easily be introduced in the form of transient spikes, causing an instantaneous change in frequency. Frequency shifts can also occur due to changes in load on the VCO, or if RF from the transmitter output is allowed to feed back into the synthesiser section. This phenomenon is known as frequency pulling and must be within prescribed limits (dependent on specifications). It is possible to measure frequency pulling with a spectrum analyser on zero scan and slope detection, but this is unreliable and inaccurate. To make a proper measurement, an MDA (Modulation Domain Analyser) or a VSA (Vector Signal Analyser) will be required. These instruments work in a frequency/time domain rather than a frequency/amplitude domain. A trigger signal will be required to coincide with the switching function on the transmitter and the change in output frequency will be displayed as a step, usually followed by an exponential recovery time dictated by the loop bandwidth of the synthesiser. 3.16.6. Reverse IMD. Reverse IMD can occur when a very strong interfering signal is fed back from the antenna to the PA stages, where it can cause spurious modulation of the carrier (IMD). This situation can arise when two or more transmitters are co-sited and the energy from one antenna can couple into another. As described in section 3.13.1, an isolator or a circulator in the PA output path will greatly reduce the probability of reverse IMD. In order to measure reverse IMD, the set-up shown in Fig. 3.25 will be required. Transmitter Fig. 3.25. under test. Set-up for measurement of Reverse IMD. Signal Generator. Directional coupler 20dB


Spectrum Analyser


Notes: 1. This is one application where a directional coupler is used. You could also use a circulator in this application. When connected as shown, it will monitor the forward power plus induced IMD from the transmitter under test. The output from the test RFPA will not be coupled out to the spectrum analyser because it is in the reverse direction. 2. The RFPA (RF Power Amplifier) must be capable of providing a continuously variable power output and should provide sufficient power to meet IMD levels after 20dB of attenuation. 3. The 20dB attenuator in the transmitter path must be able to handle the full RF power from both the transmitter and the RFPA. Procedure: The transmitter under test should be un-modulated and its output (from the directional coupler) displayed on the spectrum analyser. Set this level to reference level. Now set the frequency from the signal generator to some convenient offset frequency (say transmitter + 1MHz). Gradually increase the output level from the signal generator (and hence the RFPA) until sidebands appear at both sides of the displayed transmitter output. When these sidebands reach a level of -30dBc (or some other level if specified), this is the IMD threshold. Without changing the settings, you must now reverse the outputs from the directional coupler to display the output from the RFPA. Measure this level and add 20dB for the coupler factor to give the figure for Reverse IMD.

Page 82 3.16.7. Modulation. As digital modulation becomes increasingly complex, the need for specialist test equipment becomes almost mandatory. A Vector Signal Analyser can display 8-PSK signals and can measure EVM, but new requirements are being added year on year. Some of these can be covered by software upgrades and most of the latest test equipment is software-driven. Such equipment is very expensive (typically $50,000 or more). Analogue modulation and simple digital modulation, such as FSK, are much easier to measure and again the preferred instrument is a spectrum analyser. AM. In section 3.2.2, Fig. 3.2 it was shown that for a single, fixed modulating frequency, two sidebands are produced at a distance from the carrier equal to the modulating frequency. On a linear vertical scale, with 100% modulation depth these sidebands will be half the amplitude of the carrier. However, this is not a satisfactory method of measurement. In order to measure the modulation depth it is necessary to reduce the display bandwidth until a solid (in-filled) cone is seen. As the modulation depth is decreased, an unfilled area will appear under the cone, corresponding to the modulation trough. The trough will become progressively bigger as modulation is reduced. The relationship between the hump and the peak amplitude is a fair measure of modulation depth. The following expression applies: Modulation % = [(Vp + Vt)/(Vp - Vt)] x 100 where Vp is peak amplitude and Vt is trough amplitude. FM and FSK. To set the deviation, first set the spectrum analyser to display the transmitter frequency without modulation (remember to include a suitable attenuator between the transmitter and the analyser). Set the carrier amplitude to about 90% of screen height and set a scan width appropriate to the expected deviation (e.g. for 3kHz, a 10kHz scan would be appropriate). With a fixed sinusoidal modulation, the spectrum analyser will display a flat-topped cone as shown in Fig. 3.26 below. The distance between the peaks is the double-sided FM deviation. Deviation is normally quoted as single-sided, so actual deviation is half of this. Modulation level may be adjusted to give the required peak deviation. For square-wave or FSK modulation, the in-fill will not be present and just two peaks will be seen. To view the modulation, leave the other settings unchanged and set the analyser for zero scan and 1dB/div vertical scale. Now use the fine frequency control on the analyser to offset the frequency just enough to bring the display to half screen height. The FM will now be slope detected by the analyser filter. Use the video trigger mode and adjust the trigger level and the scan time to synchronise the demodulated output. Non-linearity in the slope detection is partially compensated by using a 1dB/div vertical scale, rather than linear. The above settings are especially useful in setting up dual-point modulation, as described in section 3.2.4 (3). The two adjustments for modulation level are set interactively to give the best square wave signal. Fig. 3.26. Typical display of FM with sinusoidal modulation. For FSK (square-wave) the in-fill will not be present and just two distinct peaks will be displayed.

Page 83 3.17. Appendices. 3.17.1. Attenuator Values. The following resistor values may be used to make a 50 ohm Pi attenuator (see Fig. 3.27). Fig. 3.27. Attenuator Pads. Note: Resistor values given are the nearest E24 standard value. Actual attenuation will therefore be approximate. 3dB pad 6dB pad 10dB pad 20dB pad R1 300 R1 150 R1 100 R1 62 R2 18 R2 36 R2 68 R2 240 R3 300 R3 150 R3 100 R3 62 3.17.2. Varicap Diodes. A range of hyper-abrupt varicap tuning diodes, available from Alpha Industries.

Fig. 3.28. This shows that with a typical range of control voltage from just 0.5 - 2.5V, a capacitance range of 1 - 6pF (SMVl247) up to 10 - 60pF (SMV1255) is possible.

3.17.3. Free Design Software: At the time of writing, the following software is available for FREE download. Appcad 3.0.2 - A useful design program if you have nothing better, but take care - it is very limited! Quick Smith - An excellent interactive Smith Chart program. Design guide for Class E power amplifiers - by Jim Tonne EasyPLL (Part of the Webench program) - This program is not actually downloadable and must be used online. Although designed around National synthesiser ICs, results are equally applicable to most other synthesisers.
Copyright (2007) on this e-book is owned by Davood Mirzahosseini. If you wish to reproduce any of this material, kindly first ask permission by contacting me at: