You are on page 1of 5

502 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 55, NO.

6, JUNE 2008

Single Amplifier Bi-Quadratic Filter Topologies in


Transimpedance Configuration
Gaurav Chandra, Member, IEEE, Preetam Tadeparthy, Member, IEEE, and Prakash Easwaran

Abstract—Active-RC configurations, implemented as cascaded


bi-quadratic sections (biquads), are the filters of choice for low
distortion applications in the base-band. For low power and area
applications, single amplifier biquads are best, and are well studied
in literature. However, in many applications the input signal is in
current domain while the output is a voltage. Output of a current
steering digital–analog converter, and output of a down-conversion
mixer, are two such examples. In this paper, the authors propose
two generic classes of single amplifier bi-quadratic low-pass filter
topologies, achieving a second order roll-off in transimpedance
( -to- ) configuration. Both these topologies are derived with
the constraint of fully differential signal processing. The tran-
simpedance active-RC configurations presented here eliminate
redundant current–voltage–current conversion and hence provide
immense advantages in terms of noise and linearity.
Index Terms—Active filters, analog circuits, biquadratic filters. Fig. 1. Transimpedance all-pole topology with mixed feedback.

application. Most high-end communication receivers are fully


I. INTRODUCTION differential, for higher immunity to external noise coupling.
ESIGN of analog filters continues to play an important Hence, this topology would be unsuitable for high-end system
D
tion (
role in modern communications systems. For low distor-
65-dB) applications, active-RC filters are the filters
implementations.
In this work, the authors propose two new generic
of choice. In modern integrated circuits (ICs), since compo- bi-quadratic topologies that realize low-pass filtering in
nent mismatch is not so significant, cascaded implementations transimpedance active-RC configuration, while using a single
are preferred over ladder implementations due to better pro- amplifier [5]. The first topology is suited to all-pole filters
grammability options. For extremely low power and low dis- (Butterworth, Chebyshev) while the second topology imple-
tortion applications, operational amplifier (opamp) power is at ments a pair of complex conjugate zeros on the axis (notch)
a premium. Also, for higher bandwidth applications, even the along with a pair of complex poles, enabling the synthesis of the
overall area tends to be limited by the opamps. For such appli- more general class of filters (elliptical, inverse Chebyshev). All
cations, single-amplifier implementations are best, and are well the topologies presented here operate in inverting configuration,
studied in literature [1], [2]. and hence suitable for fully differential implementations. The
There are a number of applications where the input variable work presented here restricts itself to low-pass filter applica-
is a current, and would need to be converted to a voltage before tions, but the idea can be extended to other realizations.
using the conventional active-RC filters. Hence, it is desirable The layout of this paper is as follows. Section II discusses an
to obtain a transfer function that operates in transimpedance all-pole bi-quadratic realization. Section III extends the ideas in
configuration, converting the current to voltage, while also Section II to develop bi-quadratic filters with a pair of complex
achieving the desired filtering. The idea of filtering in a tran- conjugate zeros (notch). In Section IV design of a low-pass filter
simpedance configuration has been observed before [3], [4]. in a WLAN transceiver is discussed as an application. This is
However, there are certain limitations in the reported architec- finally followed by the conclusion.
tures. The architecture presented in [3] requires two opamps to
realize the bi-quad, while the circuit presented in [4] requires II. ALL-POLE BI-QUADRATIC IMPLEMENTATION
two opamps and four operational transconductance amplifiers
(OTAs). This comes with significant area and power overheads. To get a bi-quadratic topology that works in a transimpedance
The circuit presented in [4] is further limited to a single-ended configuration with an all-pole response, one needs a structure
that has a current input, voltage output, and has two poles and
no zeros in its transimpedance transfer function. Such a circuit
Manuscript received October 2, 2007; revised November 26, 2007. This paper
was recommended by Associate Editor G. Palumbo. is shown in Fig. 1. This circuit is obtained as a variation of the
G. Chandra is with Wireless Terminals Business Unit, Texas Instruments, popular single-amplifier topology known as the multiple-feed-
Dallas TX 75243 USA (e-mail: gchandra@ti.com). back (MFB) topology [2], [6]. For simplicity of drawing only
P. Tadeparthy is with the High Performance Analog Group, Texas Instru-
ments, Bangalore 560093, India.
the single-ended equivalent of the fully differential circuit is pre-
P. Easwaran is with Cosmic Circuits, Bangalore 560037, India. sented here. The capacitor is connected to input after inver-
Digital Object Identifier 10.1109/TCSII.2007.916693 sion, introducing positive feedback. This inversion is achieved
1549-7747/$25.00 © 2008 IEEE
CHANDRA et al.: SINGLE AMPLIFIER BI-QUADRATIC FILTER TOPOLOGIES IN TRANSIMPEDANCE CONFIGURATION 503

by flipping the output wires of the fully differential circuit. The Since for a biquad, the two important parameters are and ,
transfer function is we will consider the pole- sensitivity and pole-frequency
sensitivity . The sensitivities with respect to passives (each
individual element) as well as active (opamp gain-bandwidth
product) are important. However, for integrated circuit applica-
(1) tions the most important parameter is opamp excess phase [8],
which leads to a corner frequency and pole- of which limits the power dissipation. Hence, only the active sen-
sitivities are discussed in detail. The passive sensitivities would
(2) be identical to the well known MFB biquad, and the reader is
referred to [9]–[12] for a detailed treatment.
Assuming an integrator model for the opamp with a gain-
And pole- factor
bandwidth product , following expressions were obtained for
the active sensitivity:
(3)
(7)
The negative term in the denominator in (3) can be controlled
to obtain very high- values. For low- implementations, the (8)
capacitor may be omitted.
An important consideration for any transimpedance config- For the case , and assuming , (7)–(8) simplify
uration is the input impedance behavior. The proposed circuit to
achieves ideally zero input impedance at dc, as seen from Fig. 1,
since the capacitors convey no current and hence there is no cur- (9)
rent through resistor . However, at higher frequencies the input
impedance will be finite due to the capacitive current through
the capacitors and . At very high frequencies, though, the
capacitor will provide a low impedance path again. Thus, the
input impedance is bandpass in nature. The transfer function,
when evaluated, confirms the observation (10)

While, the expression for maximum input impedance simplifies


(4) to
The input impedance can be reduced, though, by skewing the
resistor ratios. From the above equation, it can be seen that the (11)
maximum input impedance is seen at the corner frequency of
the filter, and is equal to To reduce peak input impedance, a large value of is desired,
which degrades the sensitivities. Predictably, the tradeoff is
(5) worse for high- values. As an example, the sensitivity be-
havior against peak input impedance, normalized to , is
plotted in Fig. 2, for and . For the sake of com-
It must be noted that for , this reduces to . pleteness, the passive sensitivities and (sensitivity to
Introducing positive feedback degrades the input impedance ) are also plotted. Gain-bandwidth product of the opamp, ,
behavior. is kept as , which is typical of filter designs. In practice, a
For a filter, the dc transimpedance gain is fixed by the suitable design point can be achieved, as will be demonstrated
system requirements. Thus, to improve the input impedance be- in Section IV.
havior, can be made much smaller compared to . However,
this impacts the sensitivity of the filter, as will be elaborated later III. BI-QUADRATIC IMPLEMENTATION WITH A NOTCH
in this section.
The topologies presented in Section II have limited applica-
tion in filter design, as these do not have a pair of complex con-
A. Sensitivity Analysis
jugate zeros on the axis (notch). For the synthesis of more
In the previous section, we observed that the input impedance general filters like elliptical and inverse Chebyshev, a notch is
of the proposed transimpedance filter is not zero, but can be con- also required. In this section, we will develop a topology that
trolled. In this section we will find that this comes at the cost of has a pair of complex poles as well as a notch.
higher sensitivity, although a suitable tradeoff may be obtained. To get a notch, the input current needs to see zero transmis-
We use the most commonly used definition of sensitivity, which sion at a purely imaginary frequency. One can utilize the prop-
is the Bode sensitivity, defined in [7] as erties of a perfectly balanced Twin-T network to achieve this
objective. For a balanced Twin-T, the currents through the two
paths are completely out of phase at a purely imaginary fre-
(6) quency, and hence it can provide a transmission zero on -axis.
Embedding a Twin-T network in the forward current path for the
504 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 55, NO. 6, JUNE 2008

assumption, for , the approximate expressions for the


pole-frequency , pole- value , and the notch frequency
are

(16)

(17)

(18)

Alternatively, for a given corner frequency and pole- factor,


the component values can be solved exactly using a computer
program.
Just like the circuit discussed in Section II, this circuit also
shows zero input impedance only at dc, and the impedance is
bandpass in nature. If one ignores the higher order pole, the
expression of the peak input impedance (normalized to tran-
simpedance ) is found to be
Fig. 2. Sensitivity variation with different values of input impedance.

(19)

Assuming , this can be simplified to

(20)

Equation (20) is similar to (11), with a dependence of


on the component skew.
Another method to get a better range is to provide capac-
itive feedback at the bridge-T network instead of the input net-
work. This avoids the capacitive loading at the input network.
This circuit is shown in Fig. 4. Because the total capacitance
seen by the Twin-T network is still , the notch is preserved,
but pole -factor can now be adjusted to arbitrarily high values.
The parameter can be varied to obtain arbitrary values of .
Fig. 3. Transimpedance topology with a notch and pair of complex poles. The circuit can be used without inversion to obtain conventional
negative feedback, to lower the , instead of enhancing it. For
circuit shown in Fig. 1 leads to the circuit shown in Fig. 3. Here, the circuit in Fig. 4, the denominator polynomial (13)–(14) are
is a design parameter that can take any real value, and is used modified to
to adjust the of the circuit. The transfer function of this circuit
is (21)

(12) (22)

where The expression for peak input impedance is also modified to

(13) (23)

(14) For both the topologies presented in this section, the input
(15) impedance behavior can be controlled by skewing the com-
ponents (making ). This comes at the cost of higher
The transfer function is actually third order, with a notch and sensitivity, just like the circuits discussed in Section II. This
three poles. Two of these are complex conjugate poles, whose tradeoff is now discussed.
-factor can be controlled by adjusting the parameter and the
capacitor . The inversion shown in the figure is optional. A A. Sensitivity
positive value of (without inversion) can be used to reduce Active sensitivity is now discussed for the circuits shown in
the of the circuit, while the negative value can be used to Figs. 3 and 4. With the opamp being modeled as an integrator,
enhance the . It can be shown that the third (real) pole lies the transfer functions are computed and sensitivities evaluated.
at a much higher frequency and can be ignored. With such an The expressions for are presented below.
CHANDRA et al.: SINGLE AMPLIFIER BI-QUADRATIC FILTER TOPOLOGIES IN TRANSIMPEDANCE CONFIGURATION 505

Fig. 4. Alternate approach to achieve better Q-range.


For the circuit in Fig. 3 (using )

(24) Fig. 5. Sensitivity plots for the circuit in Fig. 3.

For the circuit in Fig. 4 (splitting )

(25)

The expressions for , for the circuits shown in Figs. 3 and 4,


are presented in (26), (27). For the circuit in Fig. 3 (using ),
we obtain (26), shown at the bottom of the page. For the circuit
in Fig. 4 (splitting )

(27)
The expressions are tedious, but presented here for the sake of
completeness.
The sensitivities for the circuit in Fig. 3 is plotted against
varying input impedance values, are shown in example Fig. 5.
The figures correspond to , and .
Just like the sensitivity plots shown earlier in Fig. 2, it is Fig. 6. Variation with pole- .Q
seen that the active sensitivity degrades if low values of input
implementations, characteristic of mixed feedback structures.
impedance are desired. The sensitivity plots for the circuit in
Also, the implementation shown in Fig. 4, with the split ca-
Fig. 4 are not shown here, but are similar.
pacitor , is slightly better for high- implementations. But
Fig. 6 illustrates the effectiveness of mixed feedback, by plot-
the circuit in Fig. 3, with the positive feedback capacitor ,
ting the active sensitivity variation for a range of pole- values.
performs better for low- configurations. Based on the sensi-
The sensitivities of both the circuits are plotted, as against the
tivity analysis presented here, an appropriate topology can be
case where no mixed feedback is used. The sensitivities are
selected.
plotted for and . Only the sensi-
tivity is plotted, the results are similar for . From Fig. 6,
it is evident that without the enhancement techniques, the sensi- IV. DESIGN APPLICATION
tivity goes up quadratically with value, approaching very high The ideas developed in the previous two sections are now ap-
values for high- implementations. However, the proposed cir- plied to a design application to illustrate the impedance tradeoff
cuits achieve asymptotically linear dependence on for high- discussed earlier. The application considered here is the receive

(26)
506 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, VOL. 55, NO. 6, JUNE 2008

of concern. For the current filter implementation, it was found


that the average impedance was sufficiently low to suppress the
output stage nonlinearity.
It is interesting to note that the tradeoff is slightly worse for
the biquad with notch, as compared to an all-pole implemen-
tation. The higher sensitivity of the structure is the cost paid
for implementing a notch. The operating points for the cur-
rent design examples are marked in the figure. It is evident that
compared to the asymptotic values of minimum sensitivity, the
degradation in sensitivity is not very significant.

V. CONCLUSION
The authors have presented two bi-quadratic active-RC
filter topologies in transimpedance ( -to- ) configuration. The
topologies are derived with the constraint of fully differential
implementation. Though the topologies do not offer ideally
zero input impedance at all frequencies, the effect can be
Fig. 7. Tradeoff between peak input impedance and sensitivity. minimized at the cost of slightly higher sensitivity. The tradeoff
is explained analytically and quantitatively through a design
filter for a 802.11b/g wireless local area network (WLAN) di- application. A current mode interface enables lower noise and
rect-conversion transceiver [13]. For this application, a fifth- a significant reduction in the voltage swing at the mixer/filter
order inverse Chebyshev filter was implemented, with two pairs interface, resulting in significant linearity improvements.
of complex poles, one real pole and one notch. The first biquad is
implemented in a trans-impedance configuration, with a corner REFERENCES
frequency MHz and . The notch frequency is [1] A. S. Sedra and P. O. Brackett, Filter Theory and Design: Active and
MHz. It is optional to put the notch in the first stage. For Passive. Champaign, IL: Matrix, 1978.
the desired corner frequency and pole- , the following compo- [2] B. Carter, “More filter design on a budget,” Texas Instruments, Dallas,
nent values were obtained: k , , k , TX, Tech. Rep. SLOA096, 2001.
[3] A. Carlosena and E. Cabral, “Novel transimpedance filter topology for
pF, pF. instrumentation,” IEEE Trans. Instrum. Meas., vol. 46, pp. 862–867,
Above values correspond to the choice . This leads 1997.
to an acceptable input impedance behavior with minimal impact [4] S. Minaei et al., “Transimpedance type multifunction filter without
using any external passive elements,” in Proc. 25th Midwest Symp. Cir-
on sensitivity. These component values lead to a total capaci- cuits Syst. (MWCAS), Aug. 2002, pp. 633–636.
tance of 49 pF and output referred noise of 68 V. This was [5] G. Chandra, P. Tadeparthy, and P. Easwaran, “High order tran-
sufficient to meet the target specifications. simpedance filter with a single operational amplifier,” U.S. patent
Similarly, the circuit shown in Fig. 4 was implemented to 10/711724, Sep. 30, 2004.
[6] B. Carter, “A differential opamp circuit collection,” Texas Instruments,
obtain a notch in the transfer function. The following component Dallas, Tech. Rep SLOA64, 2001.
values were computed by a numerical solver: , [7] M. S. Ghausi, “Analog active filters,” IEEE. Trans. Circuits Syst., vol.
k , pF, pF, , CAS-31, no. 1, pp. 13–31, Jan. 1984.
[8] R. Schaumann et al., “Effects of excess phase in multiple-feedback
These values lead to the desired pole-zero locations, and an active filters,” IEEE. Trans. Circuits Syst., vol. CAS-27, no. 9, pp.
additional pole at 116 MHz, which indeed can be ignored 967–970, Sep. 1980.
as was suggested initially. The noise performance of this circuit [9] K. Laker and M. Ghausi, “Synthesis of a low-sensitivity multiloop
is very similar to the earlier circuit. feedback active RC filter,” IEEE. Trans. Circuits Syst., vol. CAS-21,
no. 2, pp. 252–259, Feb. 1974.
It was noted earlier that the circuits proposed here do not offer [10] K. Laker and M. Ghausi, “Corrections to synthesis of a low-sensitivity
zero input impedance, but the impact can be minimized at the multiloop feedback active RC filter,” IEEE. Trans. Circuits Syst., vol.
cost of slightly higher sensitivity. Fig. 7 shows the tradeoff be- CAS-21, no. 6, p. 811, Nov. 1974.
[11] E. Moustakas and S. Chan, “Sensitivity considerations in a mul-
tween peak input impedance and the sensitivity, for both the tiple-feedback universal active filter,” IEEE. Trans. Circuits Syst., vol.
topologies. These were obtained by choosing the opamp gain- CAS-24, no. 6, pp. 695–703, Jun. 1977.
bandwidth product ( ) to be 100 MHz (ten times the filter [12] K. Laker et al., “Multiple-loop feedback topologies for the design of
corner frequency). It must be noted that for multitone signals low-sensitivity active filters,” IEEE. Trans. Circuits Syst., vol. CAS-26,
no. 1, pp. 1–21, Jan. 1979.
in WLAN, lower frequencies are significantly attenuated and it [13] A. Kamath, G. Chandra, and P. Easwaran, “Reducing noise and distor-
is the average impedance and not the peak impedance, that is tion in a receiver system,” U.S. patent 7 177 613, Feb. 13, 2007.

You might also like