You are on page 1of 51
CHAPTER EIGHTEEN 18.4 18.2 18.3 18.4 18.5 18.6 18.7 18.8 Basic Proper 18.1.1 Gai 18.1.2 Desensitivity 18.1.3, The Effect of Feedback on Extraneous Signals 18.1.4 The Effect of Feedback on Distortion 18.1.5 Summary The Four Ba: Feedback Circuit Topologies and Design of Circuits Employing Node Comparison and Node Sampling 18.3.1. Analysis in Terms of Two-Ports 18.3.2 Circuit Representations for a and / 18.3.3. Design Example 18.3.4 Loop Transmission 18.3.5 Input and Output Admittances 18.3.6 Summary Analysis and Design of Circuits Employing Loop Comparison and Sampling Analy: in Terms of Two-Ports Circuit Representations for @ and / 18.4.3 Example Feedback 's Employing Node Comparison and Loop Sampling Feedback Employing Loop Comparison and Node Sampling Summary Lecture Demonstrations 18.8.1 Hum Reduction 18.8.2 Distortion Reduction nces Problems 18.1 Basic Properties of Feedback Amplifiers No amplifier is ideal, For example, no amplifier is perfectly linear, that is to say, the output voltage waveform is not an exact scaled replica of the input voltage waveform. Even if the amplifier is reasonably linear for some range of input voltage, the voltage gain of the amplifier will vary with changes in power-supply voltages or temperature because of variations in transistor characteristics with de operating point and temperature. These and many other amplifier limitations can be minimized by applying negative feedback to the amplifier, 18.1.1 Gain The concept of negative feedback as applied to amplifiers can be illustrated by means of the signal-flow block diagram shown in Fig. 18.1a or the electri- cal block diagram in Fig. 18.1b. To apply feedback to an amplifier, we must add two components, a precision attenuator, and a comparator. For simplicity we assume that the transfer functions of the three blocks are independent of frequency. Also, for the moment we suppress the issue of Input = Vi Vo Basic] Output = Va Comparator amplifier @ Vs Precision attenuator f (a) Signal-flow diagram Basic $ amplifier R, (unilateral) Vs = aVe > Passive precision attenuator so vs=f¥s (6) Electrical biock diagram Figure 18.1 Block diagrams of a feedback amplifier. 619 620 Feedback loading of the basic amplifier by the feedback network, by assuming that the basic amplifier in Fig. 18.1 has infinite input resistance and zero output resistance. Under these conditions the output voltage V, of the precision attenuator is a precisely scaled-down version of the amplifier output voltage V;. The voltage V, is compared with the input voltage V;. and the difference between the two voltages is fed to the amplifier. That is, Yy=W-V (18.1) Combining this equation with the input-output relations of the basic ampli- fier and the precision attenuator Vs = aVy (18.2) Va = fVs (18.3) and eliminating V, and Vj, we find that the voltage gain 4 of the complete feedback amplifier, often called the closed-loop gain, is Vs a : T+ af (18.4) ‘The effect of the negative feedback is now evident. If the product af is much larger than unity, Eq. 18.4 reduces to ~o 8.5 ; (18.5) ‘That is, the relationship between V; and V, depends only on f and is nearly independent of a. This is an important result because the gain a of the basic amplifier is a function of temperature, power-supply voltages, and so on, as pointed out above, On the other hand, the “gain” f can be precisely con- trolled, for the attenuator can be constructed from reliable R. L, and C ponents, the values of which are much less sensitive to operating conditions ‘The overall voltage gain 4 is approximately independent of a when a is large because the amplifier input voltage V, required to produce Vy is then the difference between two much larger voltages ; and Vs. Consequently. if a for some reason drops by a factor of 2, only a trivial change in V, is needed to double V3, thereby compensating for the loss of gain. It should be emphasized at the outset that although the present discussion is phrased in terms of voltage gain, we do not build amplifiers just to obtain voltage gain. That we can obtain with a transformer. What we are really interested in is power gain or power-handling capability— the ability to amplify a very low-power (and low-voltage) signal, so that substantial signal power can be delivered to a load, for example, a loudspeaker or a meter coil This fact should be kept in mind in the subsequent discussion. 18.1.2 18.1.3 18.1 Basic Properties of Feedback Amplifiers 621 Desensitivity Let us now calculate with more precision the sensitivity of the feedback amplifier gain 4 to changes in the gain a of the basic amplifier. Taking the differential of Eq. 18.4 (assuming small changes in a and constant f), we obtain 1 dA mar da (18.6) We can now find the fractional change in 4 from this result and Eq. 18.4 dA _ 18.7] A l+afa _ (see Problem P18.1), Thus a given percentage change in a is suppressed in the overall gain A by a factor 1 + af: For example, if the product af equals 99, then a 10% change in a gives rise to only a 0.1, change in 4 Clearly we pay a price for this improvement in amplifier performance. Comparison of Eqs. 18.2 and 18.4 indicates that the application of negative feedback has resulted in a reduction in gain, Specifically, the ratio of the gain of the overall feedback amplifier to the gain of the basic amplifier is 1 1 a (18.8) a \+af Thus the overall gain is reduced by exactly the same factor by which gain changes are suppressed. In the numerical example cited above, the feedback has suppressed changes in gain by a factor of 100, but this improvement in performance is obtained at the expense of a reduction in gain by a factor of 100. Fortunately, unstabilized gain is readily obtainable (e.g., by adding more transistor stages). Thus the trade-off of unstabilized gain for stability is not a setious limitation on the use of feedback For the polarities and interconnections specified as in Fig. 18.1, the feed- back is by definition negative or incerse if'a and fhave the same algebraic sign However, in mote complicated configurations it is not always easy to apply this criterion. A surer method of checking for negative feedback can be seen from inspection of Eq. 18.8. Specifically, if a small increase in the magnitude of ffrom zero reduces |] below Jal, the feedback is by definition negarice. The Effect of Feedback on Extraneous Signals The performance of amplifiers is often limited by the presence of extraneous signals, such as power-supply hum, thermal noise, and cross-talk from adjacent amplifiers. Under certain specific circumstances, negative feedback can be used to reduce the effect of these extraneous signals, but in many other circumstances, feedback produces no improvement whatsoever. We shall differentiate between these extremes by means of a few simple examples 18.1 Basic Properties of Feedback Amplifiers 623 We have already discussed (in Chapter 16) an apparent contradiction to this statement. In Section 16.1.4, feedback was introduced around a three- stage amplifier to make the operating point of the output transistor stable in spite of changes in Vy,. A careful examination of Fig. 16.5 reveals that the “extraneous signal” source Vp; and the source of desired signals V; are in effect both connected to the amplifier input, because both give rise to a base current V/R,. The improvement in signal-to-noise ratio, or more accurately the signal-to-drifi ratio, in this amplifier occurs because the signal is ac whereas the drift is essentially de. Thus it is possible by adding capacitor C, to the feedback network to apply the feedback selectively, that is, we have large amounts of feedback at de and no feedback at signal frequencies. Next let us examine the effect of feedback if the extraneous signal is being introduced at some point other than the amplifier input. A general representa- tion of this case is shown in Fig. 18.34. Here we assume that we have two amplifiers a, and a). Amplifier a is assumed to have associated with it an extraneous signal V,, whereas amplifier a, is assumed to be noise-free. Again using superposition, we find the output voltage V, to be aya, a Taf * Tr aah" cay Vv Thus the signal-to-noise ratio at the amplifier output is -_ yaa (18.12) 4 Va a ae + + + | “i a V5 a a $ Va t (a) Vn -\s + vs « RS Va a Figure 18.3 Effect of feedback on extraneous signals introduced at some point other than at the input. 624 Feedback which is a factor of a, better than would be achieved with amplifier a, alone (Fig. 18.3b). Thus if it is possible to build an amplifier a, that does not have the same inherent problem of extraneous signals as amplifier a. then it is possible to improve the output signal-to-noise ratio of a, by means of feedback This condition implies that the extraneous signal F, in Fig. 18.3 cannot be thermal noise associated with the amplifier input, because amplifier a, would then suffer from the same problem as ay A practical example that can be adequately represented by Fig. 18.3 is the problem of power-supply hum associated with the output stage of an audio amplifier. Because such an output stage often operates with collector cur- rents in the ampere range, it is expensive to provide adequate filtering for the de power supplying this stage. On the other hand, because the preceding stages in the amplifier operate at much lower signal levels, power-supply filtering here is relatively inexpensive. Hence it is reasonable to represent the output power amplifier by the block a, in Fig. 18.3 and assume that Vy, represents power-supply hum. Amplifier a, is then a second amplifier which operates at much lower signal levels from a well-filtered power supply. To be specific, assume that the power amplifier a, has a voltage gain of unity but a large power gain (which is why it is needed) and that V, is a 120-cycle hum signal of 1-volt amplitude. The exact point at which the extraneous signal is introduced in a, is not important in this discussion, hence for simplicity we assume that it enters at the input of a3. If the desired signal V; is also of I-volt amplitude, then when a, is operated alone (Fig. 18.3b), the output signal-to-hum ratio is unity. If now a hum-free amplifier a, of gain 100 and a feedback network with fequal to one are added, then from Eq. 18.11, the output voltage will still contain about 1 volt of signal when the feedback network is used and the I-volt input signal is applied to the comparator, but the hum signal will be reduced by a factor of 101. This reduction occurs because the voltage V; in Fig. 18.34 now contains a hum component exactly inverted with respect to the hum from ¥,, and this component almost cancels the hum. The results discussed above can be readily verified by observing the wave~ forms in the simple audio amplifier shown in Fig. 18.4a. When the power amplifier stage is operated with no feedback by placing the switch in position 1, the voltage ¥, is sinusoidal, whereas the output voltage V4 contains a large unwanted hum component, as can be seen from Fig, 18.4b. When feedback is applied around the output stage and the low-level amplifier, (in this case a 709 operational amplifier) by moving the switch to position 2, we see from Fig. 18.4c that the hum disappears in the power-amplifier output voltage Vs, but the input Vs to the power amplifier now contains an inverted hum signal required to cancel the hum.! The amplitude of the desired signal at the input to the power amplifier is almost the same with or without feedback. (See Problem P 18,25.) "This discussion ean be used as the basis ofa simple yet very effetive lecture demonstration ection 188. 626 Feedback 18.1.4 The Effect of Feedback on Distortion In the preceding discussion we have assumed that all circuits in the block diagrams had linear transfer functions. We now examine the case where the amplifier is nonlinear. Specifically let us assume that the relation between the de output voltage and the de input voltage of the basic amplifier in Fig. 18.1 is as shown in Fig. 18.5. Because the assumed transfer characteristics are incrementally linear, we can apply the analysis in Section 18.1.1 separately ‘over each of the linear ranges. On this basis we would expect that the overall transfer curve of the feedback amplifier would be substantially more linear than the transfer curve of the basic amplifier, because we have already shown that feedback tends to suppress changes in basic amplifier gain. Figure 18.5 Transfer function of basic amplifier. To be more specific, let us assume that for |v] less than I volt, the transfer curve of the basic amplifier has a slope of 1000, while for |v3| between 1 and 3,5 volts, the slope is 100, as shown in Fig. 18.5. Because of the nonlinearity in the circuit, it is necessary to carry out the analysis in the time domain rather than the frequency domain. Thus we write Eqs. 18.1, 18.2, and 18.3 explicitly in terms of incremental variables in the time domain: v3(t) = av,{t) (18.13) valt) = fos (18.14) v,(t) = 0y(t) = v4) (18.15) Hence x(t) = 7-20) (18.16) lta 18.1 Basie Properties of Feedback Amplifiers 627 If we now assume that f = 0.1, then from Eq. 18.16, for |v| less than | volt, the incremental output voltage is 100004() 1 + (1000)(0.1) and for |v3| between I and 3.5 volts, the slope of the transfer curve, that is, the incremental gain, is oy) = = 9.9040) Thus the application of negative feedback has reduced a 10 to 1 change in the basic amplifier incremental gain to a 9.9 to 9.1 change over the same range of output voltage swing (see Fig. 18.6), Clearly the price for this improved linearity is gain reduction. But in many instances this is not a serious problem. The lost gain can be recovered by adding a new amplifier at the input of the feedback amplifier, and this new amplifier, because it does not have to handle large signal amplitudes, can presumably be designed to introduce almost no distortion. So we reduce the amplification function of the large-signal portion of the system in order to improve its large-signal fidelity and then transfer the job of providing gain to the small-signal portions of the system, where the distortion is much less. In this type of problem, then, inverse feedback does lead to improvement in overall system linearity, without sacrifice of system gain. It is very important to note, however, that the feedback becomes i creasingly less effective as the incremental gain of the basic amplifier becomes 628 Feedback smaller, As an extreme example, when the basic amplifier discussed above saturates for |v3| greater than 3.5 volts, the incremental gain a goes to zero; hence incrementally both af and A also go to zero, as can be seen in Fig. 18.6. Thus there is a limit to the amount of nonlinearity that the feedback can correct. Every amplifier has fundamental “hard” limits on the output signal levels that it can produce. These hard limits, which arise because transistors in the amplifier cut off, saturate, or otherwise become grossly nonlinear, cannot be relieved by feedback Le 10@ Audioamp Speaker LE bitrain low-level amplifier ay Power amplifier a2 (2) Transfer curve of the power amplifier with no feedback (e) Teanster curve with feedback Figure 18.7 Demonstration of the effects of feedback on output-stage distortion, A simple circuit that illustrates the effect of feedback on distortion is shown in Fig. 18.7a. Because the amount of distortion is related to the output signal level, to obtain a fair intercomparison it is important that the output signal with feedback be made equal to the output signal with no feedback. This is accomplished by switching the feedback as shown in Fig. 18.1 Basic Properties of Feedback Amplifiers 629 18.7a. With the switch in position 1, the power amplifier a has no feedback, and the 709 operational amplifier with feedback has unity gain. When the switch is moved to position 2, overall feedback is applied around the low- level amplifier (the 709 operational amplifier) and the power amplifier, so that the overall gain is now unity. An oscilloscope photograph of the transfer curve without output-stage feedback, v3 versus r, (or rs), is shown in Fig. 18.7b. The nonlinearity around the origin arises because neither output transistor conducts if vs is less than 0.4 volts peak. For large v3, the transfer curve flattens because fp falls at high currents, and finally because the transistor saturates. The transfer curve with overall feedback (switch in position 2) is shown in Fig. 18.7c The improvement in linearity is obvious. (2) Some feedback (4) Same feedback a8 (<), but larger signal input Figure 18.8 Oscilloscope photographs showing the effect of feedback on output-stage distortion, Shown in Fig, 18.8 are oscilloscope photographs of the input and output waveforms to the power amplifier, with and without the feedback, assuming in all cases a sinusoidal drive. The distortion in the output signal, so clearly visible in Fig. 18.8a, has virtually disappeared in Fig. 18.8c. To produce this nearly sinusoidal output signal, a dramatic change in the input voltage vs has been required, as can be seen from the figure, Because the nonlinearity is still present in the power amplifier a3. it has been necessary to predistort the 18.1.5 18.2 630 Feedback waveform at the input to a2, in order to achieve an almost sinusoidal output signal. This has important consequences in terms of the design of the low- level amplifier which drives the power amplifier : the output stage of the low- level amplifier must be designed to handle signal amplitudes much larger than normal to allow for this predistortion. ‘Note that in Fig. 18.8d the power amplifier has been driven into saturation, ‘The low-level amplifier tries valiantly to compensate for this, but it too eventually saturates. (For further details on the use of this circuit for a lecture demonstration, see Section 18.8.2.) Summary We have seen that when negative feedback is applied to an amplifier, the amplifier is desensitized, in that its gain is more constant than before, extraneous signals are suppressed under certain conditions, and distortion is reduced. We shall subsequently see that the bandwidth of the amplifier is also increased by feedback, and the input and output impedances are altered, These advantages are all obtained at a price: the voltage gain is reduced in direct proportion to the amount of desensitivity. However, voltage gain is usually easy to obtain, so that trading voltage gain for desensi- tivity is often a good bargain. Several practical points deserve emphasis at this juncture. First, feedback stabilizes gain and reduces distortion only because it makes the gain depend on the properties of certain passive components in the network rather than the active components, These passive components are much more “ideal”” than active devices such as transistors; they are much more linear and much more constant in value in spite of changing temperature, humidity, operating current, and similar factors. Second, voltage gain is easy to obtain, power-handling capability is not Stages in which power-handling capability is not a major issue can be made linear and free from certain extraneous signals such as power-supply hum. to a much greater extent than stages in which power is an important issue. Thus we are willing to increase the voltage gain of a circuit by adding a low- level transistor stage if such a modification will improve the properties of the power-handling part of the circuit. The Four Basic Feedback Circuit Topologies In the preceding section the basic properties of feedback amplifiers were studied in terms of simple block diagrams. We now relate these block diagrams to actual transistor circuits, and develop ways of identifying the transfer functions a and f by inspection. Because we wish eventually to include capacitive effects in order to discuss the frequency response of feed- back amplifiers, all remaining calculations in this chapter will be in terms of complex amplitudes of voltage and currents. that is, in the frequency domain Thus we must assume from here on that all amplifiers are operated in the linear mode. 18.2 The Four Basic Feedback Circuit Topologies 631 In addition to a nplification (or attenuation), two signal processes are represented in the general feedback block diagram (Fig. 18.1a). These are sampling of an output voltage or current to feed a signal to the precision attenuator, and comparison of either voltages or currents at the input to derive a signal to feed to the basic amplifier. It follows that the general block diagram can be realized with transistor circuits in four distinct and topo- logically identifiable forms. These are shown in Figs. 18.94, b, ¢, and d, It turns out to be much easier when dealing with actual circuits to differentiate among these types first by means of the circuit topology rather than what variable is sampled or compared. Thus we focus attention on the nature of the interconnections of the source, amplifier, and feedback network at the input and of the load, amplifier, and feedback at the output. In Fig. 18.94 and b, the circuit topology at the input is clearly loop in character, that is, the source, the basic amplifier, and the feedback network are connected in a loop by the comparator. On the other hand, in Figs. 18.9¢ and d, we have node topology at the input, because the source, the basic amplifier, and the feedback network share a common node pair. Thévenin equivalent sources are shown when loop comparison is employed, and Nor- ton equivalent sources are given for node comparison, because this choice, although not essential, will greatly simplify the analysis later on, In a similar vein, there are two topologically distinct methods of sampling the signal at the output of the amplifier. In Figs. 18.9a and d the topology is Joop in character, in that the basic amplifier output, the load, and the feed- back network are connected in a loop. In these cases the feedback network samples the output or load current. In Figs. 18.9 and ewe have node topology, because load, amplifier output, and feedback network share a common node pair. Thus the feedback network samples output or load voltage. It is vitally important that these identifications be made at the outset of any analysis or design, because the basic characteristics of the feedback amplifier, such as input impedance and output impedance, depend critically on the type of comparison and/or the type of sampling employed. Transistor amplifier examples that correspond closely to each of the four two-port diagrams in Fig. 18.9 are shown in Fig, 18.10. The identification of node or loop comparison can be made once the source terminals, the basic amplifier input terminals, and the feedback network input terminals? have been identified. Similarly we can determine whether node or loop sampling has been employed at the output once we identify the load, the basic amplifier output, and the feedback network output terminals. Although the identific tion of the general topology of the feedback is reasonably clear in each amplifier in Fig. 18.10, in fact only Fig. 18.10c is unambiguously related to its counterpart in Fig. 18.9, However, we shall see later that if certain key * For convenience in the two-port analysis. we designate the source end of the feedback network as the input and the load end as the output, in spite of the fact that the important flow of signals in the feedback network is from right to left 632 Feedback Rs "© Basic |Comparator| ampifier Feedback network (a) Loop comparison, loop sampling Basic Comparator amplifier Feedback network (6) Loop comparison, nade sampling ask Comparator amps | Feedback network (c) Node comparison, node sampling t + Basic Comparator| ae Feedback network (d) Node comparison, loop sampling Figure 18.9 The four basic configurations of two two-ports for a feedback amplifier. The polarities show the normal reference directions for terminal voltages at the ports, These con- figurations are sometimes referred to in the literature as series-series. series-shunt, shunt- shunt, and shunt-series feedback 18.2. The Four Basic Feedback Circuit Topologies 633 (a) Loop comparison, loop sampling (b) Loop comparison, node sampling Yo (4) Node comparison, loop sampling Figure 18.10 Transistor feedback amplifier examples that correspond closely to the two-port diagrams in Fig. 18.9. 18.3 18.3.1 634 Feedback assumptions are made, the remaining circuits in Fig. 18.10 also reduce to their counterparts in Fig. 18.9. Although the discussion in this chapter will be confined to the four basic types of feedback connections, it is important to point out that combinations of two or more of these methods may also be used. In addition, it is often desirable to use a “nested” feedback configuration, in which the “basic amplifier” is itself composed of a more basic amplifier and a feedback net- work, Fortunately, in both cases the analysis methods discussed in this chapter can be applied with only minor modification. Analysis and Design of Circuits Employing Node Comparison and Node Sampling As longs we suppress the issue of loading the basic amplifier by the feedback network (precision attenuator), it is possible to make a clean distinction between the basic amplifier and the feedback network, as we did in Figs. 18.1a and b. In many feedback amplifier circuits, however, the passive elements in the feedback network represent significant impedances connected in series or in parallel with the basic amplifier. Thus it is no longer possible to have a basic amplifier “box” whose gain is completely independent of the passive elements in the feedback network. We can of course analyze any specific feedback circuit by direct application of Kirchhoff’s voltage and current laws, ignoring the feedback aspects altogether. Feedback, after all, is an extremely helpful, but not absolutely essential, point of view. We resort to such direct analysis methods to verify approximate solutions in certain difficult cases. It is obviously better, however, to identify if possible an a circuit and an f circuit whose transfer functions are appropriate for use in the basic feedback relation a lear (8.17) Fortunately such an identification can be made on the basis of three key assumptions about the network parameters. Even more important, it is possible to identify by inspection that portion of the complete feedback amplifier circuit that forms the a circuit. Clearly this identifi s central to the issue of rapid analysis and design of feedback circuits. Analysis in Terms of Two-Ports Let us now identify the a and f transfer functions for the simplest of the four circuits shown in Fig. 18.9 or Fig. 18.10—the circuit employing node com- parison and node sampling (Fig. 18.9¢ or Fig. 18.10c). It is clearly preferable to make this identification in terms of the general two-port representation in Fig. 18.9¢ rather than some specific amplifier such as the one in Fig 18.10c. Therefore we adopt the following procedure 1. We evaluate the overall gain of the general network in Fig. 18.9¢, casting the result in the “feedback form” (Eq. 18.17). 18.3 Circuits Employing Node Comparison and Node Sampling 635 2. We identify from this result the a and f transfer functions. 3. We make appropriate approximations so that a and f have simple interpretations in terms of the circuit, thereby enabling us to find a and f by inspection directly from a circuit without carrying out the complete analysis in each case. To carry out the general analysis, we must first decide which of the six possible two-port linear network descriptions (y, z, h, etc.) is the most appro- priate representation for the two-ports in Fig. 18.9c. Clearly, with node comparison we should select the description in which the source and the input elements of the two-port are connected at one common node pair. Similarly, with node sampling, we want a two-port description with one common output node pair. The y-parameter description, involving short- circuit admittance parameters y;. ¥,. yy. and y,, meets these conditions. On this basis the complete “node-node” feedback amplifier (Fig. 18.9¢ or Fig. 18.10c) can be represented as the parallel interconnection of two y-parameter two-ports, as shown in Fig. 18.11. To distinguish between thes two-ports, a second subscript has been added to the y parameters —an a for basic amplifier parameters and an f for parameters associated with the feedback network. Basic amplifier _ : a Yon —-~-44 Figure 18.11 Representation in terms of y parameters of a feedback amplifier with node sampling and node comparison. The analysis of the circuit in Fig. 18.11 is straightforward. The node equations are 1, = (G+ Via + Visi + Vou + YepK (18.18a) 0 = (a + yp hi + (Gr + Yea + Yo (18.18b) To simplify the algebra, we define two admittances Y= G+ Yat My (18.19) 636 Feedback and Gi+ Yoa + Yor (18.20) Solving for V, after substituting Eqs. 18.19 and 18.20 into Eqs. 18.18, we obtain ¥, =e + => ta * ss) (18.21) ss HY — Osa + Ye Vea + Yes) ‘This equation can be manipulated into the “feedback form” (Eq. 18.17) ina number of different ways, but the most useful result for our purposes is obtained by dividing numerator and denominator of Eq. 18.21 by the product vy, (18.22) This equation tions: n the form of Eq. 18.17 if we make the following identifica- Yea * Meg = — le oot 18.23 a vy, ( ) and f= rat Yep (18.24) Meaningful circuit interpretations for a and f result only when we make three fundamental but eminently reasonable approximations. First. we must assume that the basic amplifier, and not the feedback network, supplies the gain in the system. Since the basic amplifier is designed to have substantial gain, whereas the feedback network is some sort of passive attenuator, this assumption is easy to justify. In terms of the y parameters, this approximation constrains the relative size of the two forward-transfer parameters y;, and yr, Weal >> Wes! (18.25) Second, we must assume that most of the signal fed back from output to input flows via the feedback network rather than backward through the basic amplifier. In y-parameter terms, this implies a constraint on the reverse- transfer parameters y,_ and Yy¢! Wed € esl (18.26) A third condition, which we shall need in the next section, is that the basic amplifier be unilateral enough so that changes in its load admittance do not 18.3.2 18.3 Circuits Employing Node Comparison and Node Sampling 637 appreciably affect its input admittance, the same condition required in the tuned amplifier discussion, Section (7.1.4. Thus in y-parameter terms, I patra << {YY (18.27) When the constraints of Eqs. 18.25 and 18.26 are applied to Eqs. 18.22, 18.23, and 18.24, we obtain Npa XY, Consequently, by comparison with Eq. 18.17, ~ Ste My - SS ee eee —— (18.29) “SVM (G+ Vn + iMG + Yon + Sos) ’ f= Sey (18.30) Note that whereas certain parameters of the feedback network, specifically vip and y,,, are contained in the a transfer function, the f transfer function now depends solely on the parameters of the passive attenuator. It should be clear from Section 18.1 that for successful operation of a feedback amplifier, the f circuit must hace this property. Circuit Representations for and / To avoid the unpleasant task of having to calculate the y parameters for each circuit under consideration, it is necessary to find simple circuit interpreta- tions for the a and f transfer functions specified by Eqs. 18.29 and 18.30. To this end we note first that the circuit shown in Fig. 18.12b has a transfer function equal to © TNe 1, Ye — Syate ee) (This can be derived by direct calculation or by setting yy, and y,, to zero in Eqs. 18.18 and 18.21.) Thus this circuit will have the a transfer function prescribed in Eq. 18.29, provided that the amplifier is unilateral enough so that the y,}4 term in the denominator can be neglected, that is, provided Eq. 18.27 is satisfied. Recall from the tuned amplifier discussion that the product [}'j.Jya] is usually smaller for a cascade of stages than it is for any one stage alone. Thus the condition above will be valid for a multistage amplifier, even if it is doubtful for each individual stage. By comparing the circuit in Fig. 18.12h to the complete feedback amplifier circuit, repeated for convenience in Fig. 18.12a, it is possible to deduce simple rules for forming the a circuit for any feedback amplifier that employs node comparison and node sampling (a) Complete feedback amplifier, y-parameter representation wade | [E0183] [4 Yor ‘Ae i 1 a= 2 paaea) OM) (Gaagea) (®) The corresponding a circuit Figure 18.12 Formation of the a circuit in y-parameter terms, 1. Identify the various components of the feedback amplifier: the source, the load, the basic amplifier, and the feedback network, as discussed in Section 18.2 2. Form the a circuit by augmenting the basic amplifier by the elements viz and G, on the input and y,, and G,, on the output. That is, all loading effects of the source, the feedback network, and the load are associated with the a circuit. The required augmentation of the basic amplifier can be found without calculating any y parameters and, in fact, without any calculation what- soever. Specifically, the proper loading on the a-circuit inpur node produced by the source, the basic amplifier, and the feedback network is that which remains after shorting the output node of the complete feedback amplifier to destroy the feedback. Similarly the proper loading on the a-circuit ow:put node produced by the load, the basic amplifier, and the feedback network is that which remains after shorting the input node of the complete feedback amplifier to destroy the feedback. An important feature of this feedback Figure 18.13 18.3 Circuits Employing Node Comparison and Node Sampling 639 formulation is that the a circuit contains the complete basic amplifier without modification. Thus the formation of the a circuit amounts to nothing more than adding to the existing basic amplifier certain admittances normally as- sociated with the feedback network. This feature is emphasized in Fig. 18.13a. Lie + Ge Soy Basic amplifier | Yor S Gt gis Feircuit (c) Complete feedback amplifier Formation of a and f circuits. It turns out to be impossible to obtain for the f transfer function a simple circuit interpretation similar to the a circuit deduced above. The problem is that on the basis of the simplifying assumptions on page 636, the f transfer function (Eq. 18.30) is by definition y,,, the short-circuit reverse transfer admittance of the feedback network. Because we have already accounted for the effects of yj, and y,, by including them in the a circuit, the fcircuit turns out (o be a pure transadmittance, with zero input and output admittances, Clearly no physical network element associated with the feedback network has these properties. Thus we must be content to calculate y,, for the feedback 18.3.3 640 Feedback network, and represent the f circuit as an ideal dependent current source ¥,;V;, as shown in Fig. 18.13b. The parameter y, is calculated in accordance with the standard y-parameter rules: short the input of the feedback network, apply a voltage source V at the output terminals, and calculate the current J, flowing in the input short, as shown in Fig. 18.135, The a and f circuit of Figs. 18.13a and b have been reconnected in Fig. 18.13c to reestablish the complete feedback amplifier. Note that because of the way the f circuit has been defined, it does not load the a circuit in any way, nor is the f circuit transfer function changed by loading effects of the a circuit. Thus by forming the a and f circuits in the particular manner des- cribed above, we have in effect manipulated the general node-node feedback amplifier circuit into a form that can again be described by the original block diagrams of Fig. 18.1 Design Example Let us now apply these results to the specific node-comparison node- sampling feedback circuit shown in Fig. 18.14a (essentially a repeat of Fig. 18.10, The complete midfrequency incremental model is shown in Fig. 18.146, Note that to conform with the node topology at the input, the input source has been transformed to a Norton equivalent. Following the procedure outlined in Section 18.3.2 for finding the a circuit, we include the source resistance, the load G,,, and the transistor amplifier in the a circuit, as shown in Fig. 18.14c, The loading effect of the feedback network at the input of the a circuit is found by shorting the output node in Fig. 18.14b to ground; hence we see a conductance G, to ground. In a similar manner we find the output loading by shorting the input node in Fig. 18.14 to ground; hence we see an additional conductance G, to ground at the output of the a circuit. The incremental gain of the a circuit v5 a can now be calculated by inspection. The f transfer function is calculated by shorting the input node in Fig 18,14b, and applying a source V, at the output node, as shown in Fig. 18.14d. Thus for this circuit hog, (18,32) The resulting f circuit is shown in Fig. 18.14e. These results can be verified readily by direct calculation from Fig. 18.14b (see Problem 18.2). Suppose that we wish to design a feedback amplifier of the type shown in Fig. 18.140 to have a closed-loop voltage gain V,/V, of — 100 and a desensi- tivity to basic amplifier parameter changes of about 70. Assume that R, = 50 ohms. First we must translate these specifications into requirements on 18.3 Circuits Employing Node Comparison and Node Sampling 641 [evs (4) Circuit for calculating F (e) f circuit Figure 18.14 Node comparison, node sampling feedback amplifier. Shading indicates basic amplifier Af, and a. For this type of feedback 4 is not a voltage gain, but rather a transimpedance : A 2 ¢ — 3 eo eery (18.33) Thus —100 x 50 = —S000 ohms 642 Feedback From Eq. 18.7 the desensitivity is 1 to be “about 70.” Hence from Eq. | transimpedance, must be af, and the specifications call for this the basic amplifier “gain,” that is, “ a= 72 = ACL + af) = (—5000)(70) = —350kohm This corresponds to a voltage gain of the a amplifier of Vo Vo - —350 kohm - Vv, IR, 0.05kohm An alternative calculation is _ gee 7 + af) = Also, from Eq. 18.33, because a is large, the feedback transfer function must be 100)(70) = —7000 = —0.2mmho or, more accurately, 0.197 mmho —350 kohm — With all of the basic feedback parameters defined, we can now proceed with the detailed design. In this example, the value of G, is uniquely determined by f (see Eq 18.32): G, = ~f = 0.2mmho The parameters of the b mplifier can be chosen on the basis of the a circuit incremental model shown in Fig. 18.14c. The gain of such a circuit has already been calculated in Section 16.1.1 (see Eq. 16.2), except for the minor modification that in the present problem the feedback conductance G, appears in parallel with G, on the input and G, on the output. The design of this amplifier for prescribed gain has already been discussed in Section 16.1.2 and will not be repeated here. In fact, the three-stage design outlined in that section comes close to meeting our present specifications. Using Eq. 16.2 (modified to include the effect of G,), the parameter values listed in Table 16.1 (page 560), and resistor values R, = 500 ohms, Ry» = 200 ohms, R,, = 100 ohms, R,3 > Ry, Gp = 0.2 mmho, we obtain a= . = —354kohm (see Problem 18.3). This completes the midfrequency design.* ® Although the above calculations are correct as far as they go, we shall find in Chapter 19 that if three identical transistors are used, the feedback amplifier as designed here is hopelessly unstable. That is, the amplifier has natural frequencies in the right half plane, hence exponentially growing natural modes. 18.3 Circuits Employing Node Comparison and Node Sampling 643. To verify this design, we calculate the desensitivity and overall gain that will be achieved when the 5-kohm resistor is connected from output to input around the basic amplifier, as shown in Fig. 18.14a. The desensitivity is 1+ af =1 + (—354)(-0.2) = 718 and an overall gain with feedback is 354 A= Foe = -494 koh Note that because we have employed node comparison and node sampling, what has actually been stabilized in this amplifier is the transimpedance V,/1,, and not the voltage gain V,/V,. If R, is a reliable parameter, this is no problem, because Z (18.34) In this example, the voltage gain is 7 AG, = (—4.94)(20) = —99 Equation 18.34, when combined with the relationship (18.35) gives rise to an alternative point of view that is often useful for rapid circuit calculations. Specifically. from Eqs. 18.34 and 18.35, the overall voltage gain is just the ratio of two resistor values: A R y ke (18.36) The validity of this simple relationship can be seen from Fig. 18.14a. Because the three-stage amplifier has a very large voltage gain, the voltage V, at the base of the first transistor must be very small: fractions of a millivolt if V, is a few volts. If the amplifier has substantial feedback, V, will also be much smaller than V,. Thus to a first approximation assume that = 0 (18.37) Hence 1, x0 (18.38) The voltage V, must adjust itself to satisfy these conditions. From Eq 18.37, the voltage drop across R, must be V,, and the drop across R, must be ¥,. Also, because f,, ~ 0, these two resistor currents must be equal and 18.3.4 644 Feedback opposite R, R, oo Hence Eq. 18.36. This simplified approach is particularly useful in dealing with operational amplifiers (see Fig. 16.16), because such amplifiers often have voltage gains as high as 10° or 10°. In addition, they usually have a sufficiently low output impedance that the loading effects of the feedback network can be neglected. See Problems P.18.22 and P.18.25, Loop Transmission It is clear from the preceding section that one important parameter in feed- back amplifier calculations is the desensitivity 1 + af, which can often be approximated for large desensitivity as just af Algebraically, the latter quantity in y-parameter terms is the product of Eqs. 18.29 and 18.30 Vode Sia of =a = IVa Ts Via + dip + GIVoe + (18.40) Yay + Ge But for the same reasons as before, it is helpful to have a simple circuit interpretation of this product, in addition to the circuits we already have for aand alone. Because the f circuit does not load the a circuit, as pointed out above, it is possible to cascade the a and f circuits to obtain a new network whose transfer function is the product of the individual transfer functions, as given in Eg. 18.40. Such a cascade connection is shown in Fig. 18.15a, Thus, the transfer function af, often called the loop transmission, is the current gain 1,/I, measured with the feedback destroyed by disconnecting the f circuit at the amplifier input and shorting the now free input terminals of the f circuit. To emphasize that almost no circuit modifications are needed to make this calculation, we have in Fig. 18.15h redrawn the circuit of Fig, 18.1Sa with the feedback y-parameter box restored to its original condition ‘The important conclusion is that the transfer function 1,/1, is the same for both circuits. Thus to calculate the loop transmission of a node-node circuit, use the following procedure: 1, Disconnect the feedback cireuit at the amplifier input node. 2. Add an admittance yj, to the amplifier input 3. Short the input terminals of the feedback network, and calculate the current that flows in this short in response to a current /, at the amplifier input. As has already been pointed out, when the loop transmission is posi- tive, that is, when the phase of af is zero, the feedback is by definition negative. To illustrate this method of finding the loop transmission, we calculate af for the feedback amplifier designed in Section 18.3.3, Starting from the I 18.3 Circuits Employing Node Comparison and Node Sampling 645 + Basic amplifier Vo @ circuit ag ~ feireuit fon in terms ofthe a and f circuits + Basic : amplifier GS Vo Figure 18.16 Calculation of af for circuit in Fig, 18.14. 646 © Feedback incremental model for the complete feedback amplifier, Fig. 18.140, we construct the circuit to calculate af (Fig. 18.16) by the following steps 1. Disconnecting the feedback resistor G, at the amplifier input and shorting the feedback resistor to ground, 2. Adding a conductance G, across the amplifier input. A calculation of ,/I, will agree with the product of the aand ffactors obtained separately in Section 18.3.3 (see Problem 18.4). 18.3.5 Input and Output Admittances In addition to the desensitivity issues discussed in Section 18.1, feedback is often employed to change the existing input or output impedance of the basic amplifier. It turns out that the effect of feedback on the input impedance is uniquely related to the type of input comparison employed, whereas the effect on the output impedance is uniquely related to the type of output sampling Clearly the easiest way of calculating the input admittance with node com- parison is to apply a test voltage V; to the amplifier input, and calculate the resultant current J,. An appropriate y-parameter circuit is shown in Fig. 18.17. Note that in accordance with previous assumptions, we have neglected hk ve Gs vi ia A+ vz Ol yy Vo Figure 18.17 Calculation of input admittance. Vea 8nd yr. On this basis the current I, is 1, = VG. + Via + Yu) + YpVo (18.41) But because we neglect yj, ali Gi + You + Yor v, (18.42) Accordingly, from Eqs. 18.41 and 18.42, we find that the total input current his Ves¥ fa Via + Yip — 18.43 oF Yee Sis G+ You + Yori : : gure 18.18 18.3 Circuits Employing Node Comparison and Node Sampling 647 Therefore, the input admittance of the complete feedback amplifier is, from. Eqs. 18.43 and 18.40. 1 = Yn = (G+ Yip + Mig + af) (18.44) Hence the input admittance is increased from the value it would have with “no feedback” (in the sense y,, = 0) by the same factor 1 + af that reduces the gain. Of course, normally the input admittance of the amplifier system would be defined as what the source sees, and G, is part of the source. So we should set G, = 0 in Fig. 18.17 and Eq. 18.44 to get the proper value of ¥,, for the amplifier system alone. But note that G, appears twice in Eq. 18.44, directly in the first factor and buried in af in the second factor. Thus to obtain the correct answer by this method, the loop transmission af must be recalculated for G, = 0. An easier way to find the input admittance of the feedback amplifier alone is simply to subtract G, from the result given by Eq. 18.44. Either method gives the same result, of course. 1 y4e¥4 + Gy Vi Sra | You Gy Ve stVo 7 { So Calculation of output admittance. To calculate the output admittance, apply a test voltage V; to the output terminals as in Fig, 18.18 and find the resultant terminal current J,. (See Problem P18.5). The result is Yes fa Te = Vives + You + Gy Gombe Ty + Via + Yi (18.45) and thus, from Eqs. 18.45, 18.29, and 18.30, the output admittance of the complete feedback amplifier is wt = (Gy + Yoq + Yop + af) (18.46) 648 Feedback Hence the output admittance is also increased from its value “without feedback” (in the sense Yzz = 0) by the same factor 1 + af that reduces the gain. The output admittance is normally defined without including G,. So G, should be either placed equal to zero in Eq. 18.46 (and again this must be done in af as well as in the first factor) or simply subtracted from ¥,, given by Eq. 18.46. To illustrate these concepts, we calculate the input and output admittance of the amplifier designed in Section 18.3.3. The input admittance of the a circuit alone is, from Fig. 18.14¢, 1 Yona = Gy + Gp + ——— ner which in this design is, from Section 18.3.3 and Table 16.1, 1 Yine = 20402 + 55-95 10.58 mmho So with feedback the input admittance according to Eq. 18.44 is increased to Yonal + af) = (20.58)(71.8) = 1480 mmho or R 0.675 ohm: ‘This value of input admittance includes the effect of G,. However, the shunt- ing effect of G, is very small in this example. Specifically, the input admittance of the feedback amplifier alone is Y= Yin — Gs 1460 mmho, or R = 0.685 ohm (see Problem P18.6). The output admittance of the a circuit alone is, from Fig. 18.14¢, Youta = Gr + Gr =02+10 = 10.2 mmho So with feedback the output admittance in accordance with Eq. 18.46 increases to Your = Youall + af) = (10.2)(71.8) = 733 mmho 18.3.6 18.3 Circuits Employing Node Comparison and Node Sampling 649 or Ray = 1.36 ohms The output admittance of the amplifier alone, exclusive of G,,, is Y= You = Gx = 733 ~ 10 = 723 mmho or R = 1.38 ohms Summary Before discussing the other three basic feedback forms, let us summarize the important points developed so far concerning the application of feedback to transistor amplifiers. First, it is not possible in general to break up a transistor feedback amplifier into two mutually independent “boxes” with mutually independent transfer functions, as we did in the idealized examples in Section 18.1 (see, for example, Fig. 18.1). However, we have shown that we can derive by inspection an a circuit and an fcireuit which possess many of the desirable properties of the boxes in Fig, 18.1. The a and f circuits are by definition those circuits whose transfer functions appear in the basic feedback expres- sion a t+af The a circuit contains the basic gain-producing portion of the complete amplifier, that is, the “basic amplifier” part of our original black box realiza- tion (Fig. 18.1), But the @ circuit departs from the ideal basic amplifier concept in that it also includes the loading effects of source, output load, and feedback network. Thus the a-circuit transfer function is dependent to a greater or lesser extent on the elements in the feedback network The f circuit in this realization turns out to be a pure dependent source, with no resistances associated with it. The “gain” parameter in the depen- dent source is a function solely of the passive elements in the feedback net- work and is independent of the unreliable active elements in the circuit Furthermore, because the f circuit is lossless, it does not load the a circuit Thus the fcircuit possesses the two most important properties of the precision attenuator in our black box idealization (Fig. 18.1). Given a set of design parameters—for example, the closed-loop gain A and the densensitivity 1 + af—it is relatively simple to find the values of a and f required to meet the specification. First, a= All + af) 18.4 18.4.1 Figure 18.19 650 Feedback If ais large, then pm desensitivity a Analysis and Design of Circuits Employing Loop Comparison and Sampling We turn now to a second feedback amplifier example, that which uses loop comparison and loop sampling, as shown in Figs. 18.94 and 18.104. For- tunately, we can make extensive use of the general y-parameter results in Sections 18.3.1 and 18.3.2, because from a network point of view, the present case of loop-loop feedback is clearly the dual* of the node-node case con- sidered in Section 18.3. Thus we can minimize the mathematics and concen- trate on the only new issue, that of manipulating an actual circuit of this type. such as Fig. 18.10a, into the feedback form. In fact, the reader who is facile enough with the theory of linear two-ports and the principle of duality to be willing to accept a simple interchange of letters (y + 2) as a proof can turn directly to Section 18.4.2, Analysis in Terms of Two-Ports The proper two-port configuration to use in analyzing the loop-loop feed- back circuit of Fig. 18.9a is clearly the one that has current as the independent variable at both input and output. This is the z-parameter formulation, involving four open-circuit impedance parameters z), z,, Zy. and z,. To this, end, two series-connected z-parameter two-ports are shown in Fig, 18.19, Ry Re, Vs Feedback network ‘rameter representation of a feedback amplifier employing loop comparison and loop sampling “See Reference 18.1, p. 42. 18.4.2 18.4 Circuits Employing Loop Comparison and Sampling 651 As before, we add second subscripts a and f to distinguish between basic amplifier parameters and feedback network parameters. Following a procedure exactly analogous to that in Section 18.3.1, we assume that the basic amplifier has all the gain, that is, Lz > lZpyl (18.47) and the feedback network has all the feedback val & [2 (18.48) Finally, we assume that the basic amplifier is nearly unilateral: lZpaZeal < 1ZiZo) (18.49) where 2a t 24 +R (18.50) Zoa + Zor + Ry (18.51) Z, Clearly, to preserve the loop topology at the input we must use a Thévenin equivalent source, a voltage source in series with R,. Also, with loop sampling we should calculate the output current. On this basis, we find that (18.52) Hence (18.53) (18.54) (see Problem P18.7). Circuit Representations for a and f If the basic amplifier is unilateral enough to satisfy the inequality in Eq. 18.49, the a circuit shown in Fig. 18.20a has the a transfer function specified by Eq. 18.53 (see Problem P18.8). Thus the rules for forming the a circuit for a given feedback amplifier of this type are as follows: 1. Identify the source, the load, the basic amplifier, and the feedback network by the methods discussed in Section 18.2 2. Augment the basic amplifier with all loading effects of the source, the load, and the feedback network. To find the proper loading to be added to the input loop, open the output loop of the complete feedback amplifier to destroy the feedback. To find the loading to be added to the output loop, open the input loop of the complete feedback amplifier to destroy the feed- back.

You might also like