IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 44, NO. 5, MAY 1997 A -Enhanced Active- 341 Bandpass Filter Ralph Duncan, Kenneth W. Martin, Fellow, IEEE, and Adel S. Sedra, Fellow, IEEE Abstract—A fully differential high-Q bandpass filter that uses lossy integrated inductors is presented. The circuit is implemented in a 0.8 m BiCMOS technology and realizes a center frequency of 750 MHz with a Q-factor that is tunable from 10 to 490 while dissipating 80–100 mW from a single 5 V supply. Since the objective of the prototype was to explore the proposed Qenhancement technique, the dynamic range is limited to 25 dB for Q = 20. (a) I. INTRODUCTION Fig. 1. T HE PROLIFERATION of gigahertz-band communication (direct broadcast satellite, cellular, and digital radio) has heightened the need for single-chip realizations of the receiver or, at the very least, the radio-frequency (RF) front-end, so as to reduce size, power consumption, and cost. A fully integrated receiver front-end effectively eliminates the bulky ceramic, crystal, or SAW filters currently used, thereby reducing power consumption, area, and assembly cost. This implies that all RF, and subsequent intermediate-frequency (IF), filtering should be integrated, without sacrificing dynamic range or receiver sensitivity, and, necessarily, the RF must employ continuoustime filters. This is because digital filters and DSP engines are relatively slow and power-hungry, and second-order effects in switched-capacitor circuits, such as clock feedthrough and poor amplifier settling, are exacerbated at high frequencies. Integrated active bandpass filters exhibit a dynamic range which, at best, is inversely proportional to their quality factor [1]–[5]. This is the result of active devices unwittingly introducing and amplifying noise, and this limitation makes designing for wide dynamic range and high difficult. Moreover, when high-frequency operation is required too, power dissipation rapidly escalates. Yet, continuous-time bandpass filters are indispensable in transceivers, and the increasing use of portable communication devices makes integration of the radio-frequency bandpass filter quite attractive. To this end, we discuss the use of lossy monolithic inductors to design a selective high-frequency bandpass filter. The role of the RF bandpass filter in a radio receiver is to provide coarse frequency selectivity; it favors the desired channel while attempting to reject large out-of-band signals. Although the required selectivity depends on the receiver architecture, this frequency discrimination must be done without adding a significant amount of noise so as not to reduce the sensitivity of the receiver. Controlled active feedback is used to compensate Manuscript received September 7, 1995; revised November 18, 1996. This work was supported by the Micronet program. This paper was recommended by Associate Editor L. A. Akers. The authors are with the Department of Electrical and Computer Engineering, University of Toronto, Toronto, Ont., Canada M5S 1A4. Publisher Item Identifier S 1057-7130(97)02968-6. (b) Q-enhanced active-RLC circuits. for the losses in the inductors, thus enhancing the of the approach has the potential filter. We show that this activeto produce less noise than other active filter design approaches, without excessive demands on the requisite amplifier. The approach, and following section introduces the activeanalyzes the circuit to define its dynamic range performance and intermodulation properties. In Section III, the design of the filter is discussed. Experimental results from a prototype circuit are presented in Section IV. II. THE ACTIVE- APPROACH The idea of activeis to realize high- natural modes tank in a feedback loop containing an by placing a passive amplifier so as to partially compensate for the losses resulting from the inductor. Typically, these inductors exhibit quality factors of 3–8 in the low gigahertz range, but a -factor as high as 10 is also possible [6]–[9]. The controlled positive is equivalent feedback required to increase the circuit to adding a variable negative resistance into the loop. This facilitates the design for a desired -factor. approaches are shown in Fig. 1. In Two activeFig. 1(a), the inductor current is sampled, and a voltage proportional to this current is fed back by the currentcontrolled voltage source [10]. In a dual manner, in Fig. 1(b), the capacitor voltage is sampled and a proportional current is fed back by a voltage-controlled current source [11], [12]. The controlled source in Fig. 1(a) can be realized with a transimpedance amplifier, and the controlled source in Fig. 1(b), with a transadmittance amplifier. In either circuit, and are resistive, only the quality factor is as long as affected. As expected, the resonant frequency is unchanged. It is straightforward to show that the -factor is increased according to 1057–7130/97$10.00 1997 IEEE (1a) 342 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 44, NO. 5, MAY 1997 variables, the intermediate transfer functions are found from (3) The elements of the vector represent the transfer functions from the circuit input to each state (the signal transfer of is the transfer function functions), while each from each noise source to the output (a noise transfer function). and an Assuming that the amplifier has a dc gain models its finite ostensible single-pole response ( bandwidth), the response of the amplifier is Fig. 2. A realization of Fig. 1(a). (4) (1b) is the -factor of the inductor at the center frequency where of the filter. As seen, the quality factor of the filter, in both cases, is increased by the loop gain at resonance, and the is tunable via the variable gain of the amplifiers. A possible realization of the circuit in Fig. 1(a)—the approach adopted for this paper—is shown in Fig. 2. Although the input signal can be injected anywhere in the loop of Fig. 1(a) to realize a filter with an enhanced -factor,1 feeding the signal through a capacitor allows a bandpass response to be taken from the low-impedance output of the amplifier. Referring to Fig. 1(a), observe that -enhancement will still be ) is connected to realized if only part of the capacitor (say , while the other part of the capacitor the feedback voltage is connected to ground. This makes the other portion of the tuning capacitor available to apply the input. Thus, in Fig. 2, the inductor current is sampled at the low-impedance , converted to a input of the transimpedance amplifier, voltage, and the output signal fed-back through part of the tuning capacitor. The input voltage is then fed through the other portion of the tuning capacitor. The filter in Fig. 2 can be conveniently analyzed when statespace notation is used to represent it. This facilitates the use of intermediate transfer-function analysis [13], [14] to describe network behavior. One suspects at the outset that the noise performance and sensitivities should be better than for filters: the role of the amplifier in activeis only to enhance the quality factor of natural modes that are already (or active) complex. This should be contrasted to filters where poles have to be moved from the real axis into the complex -plane. For a system represented in state-space [13] by (2) and where , , , and are real-valued matrices, are the system input and output, and contains the state 1 The point of signal-injection determines the type of transfer function realized. The circuit in Fig. 1(a) is capable of realizing high- bandpass, lowpass, and highpass transfer functions. Q For the state assignment shown in Fig. 2, the state-space description of the circuit is (5) The transfer function can be shown to be (6) where , the natural-mode polynomial, is (7) For an ideal amplifier , we see that the nominal center frequency and 3 dB bandwidth are given by (8) to denote the inductor Again, using frequency, we get at the nominal center (9) which is similar to (1). Including the finite bandwidth of the amplifier, and assum, (7) can be rewritten as ing (10) DUNCAN et al.: A Q–ENHANCED ACTIVE-RLC BANDPASS FILTER 343 where is the realized center frequency, is the realized is the closed-loop parasitic pole. By quality factor, and comparing (10) with (7), the parameters in (10) are (11) (12) (13) Not surprisingly, these equations suggest that the amplifier’s bandwidth should be considerably larger than the desired center frequency in order to get a small deviation from the ideal parameters. They also show that a large inductor quality factor is useful. This should be intuitive, since for a large starting quality factor a smaller loop gain is required to achieve a desired quality factor in the filter. An important nonideality is the finite input impedance of the transimpedance amplifier. It can be accounted for by extracting the parasitic from the nonideal circuit and lumping it together to reflect this with the nominal inductor. By redefining nonideality, all of the above equations are still applicable. A. Noise and Dynamic Range The intermediate transfer functions can be computed to be (14) of The significance of the noise transfer functions— —deserves special attention. Circuit noise sources can be regarded as auxiliary inputs to the system, and, thus, simply augment the state equations. Consequently, if we add a current noise source to each node for which a KCL equation was written, and a voltage noise source to each loop for which a KVL equation was written, the following relationships ensue [14]: noise gain of a current noise source at node : in an IC realization this noise source may be the result of substrate coupling via parasitics; noise gain of a current noise source at the input of the amplifier; noise gain of a voltage noise source at the input of the amplifier; noise gain of a voltage noise source at the circuit input: this could be the result of the source resistance or input buffers (if used). In order to evaluate the output noise of the filter, we first define a measure for signal magnitudes. We choose the norm [13]. For a transfer function , this can be computed using (15) is the noise bandwidth of the specified transfer where is the peak transfer function gain. Since function, and the noise bandwidth of a second-order bandpass filter is , the -norm is easily computed.2 Suppose that has input-referred voltage noise density and input-referred current noise density . , the voltage noise resistance of the amplifier, is the value of a resistor that produces the same thermal voltage noise is defined similarly. The output power as the amplifier. noise power of the filter is, therefore, (16) and are the -norms of the noise transfer where functions for voltage noise and current noise at the input of the amplifier. This result suggests that large inductances are favored, and inductors like those in [8] can be used to advantage. From a distortion point of view, the critical signal-handling path is the input of the amplifier for it is the amplitude of its input current that sets the distortion in the amplifier. The amplifier’s input current is also the inductor current, hence the inductor current sets the maximum output voltage swing. This maximum is obtained from the norm of the signal transfer of . If denotes the rms value of function the inductor current when the amplifier is at the threshold of distorting unacceptably, the dynamic range of the filter is DR (17) The first quotient in (17) represents the dynamic range of a passive series LC tank carrying a current equal to the maximum inductor current in the active realization. It is therefore a convenient metric for assessing the performance of the active circuit. In the active realization, this metric is reduced by two factors: one is the amount of -enhancement, and the other depends on the amplifier’s noise contribution. If inductor noise dominates, then the dynamic range is inversely proportional to -enhancement; if voltage noise in the amplifier dominates (poor design or a high inductor is used), the dynamic range is inversely proportional to the , as observed in other active filter realizations [1]–[5].3 This last quotient makes sense: if one already has a high- inductor, using an active realization leads to an increase in circuit noise, and a reduction in the available dynamic range. Q 2 A highsecond-order low pass filter exhibits virtually the same noise bandwidth as a second-order bandpass filter. 3 This is because 1=r. 1 Q / 344 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 44, NO. 5, MAY 1997 the current source and the MOSFET current mirror , while current sources bias the emitter followers. Each of the MOS-FET transistors has an aspect ratio of 320/2, and each of the bipolar transistors, except the buffers and (which is a single 4 device), is a dual 4 device.4 The buffers are relatively large 15 transistors. The large baseemitter capacitance of the buffers is of little concern due to the large amount of series feedback provided by the current source loads. Degeneration resistors are used to linearise the current–voltage relationship of the current sources and to reduce the noise contribution from the base resistances of the ensure that operates in the transistors. Diodes by active mode, and positive feedback is effected through interchanging the output nodes of the amplifier. This circuit uses two 5 nH inductors, with a measured of about 2 at 1 GHz ( 16 ), a nominal 1 pF tuning 0.6. The capacitors were implemented capacitance, and using double polysilicon layers. No allowance was made for tuning the center frequency of filter, although this can be accomplished using an impedance multiplier as discussed in [12] and [18]. Fig. 3. The Q-enhanced bandpass filter. A. Amplifier Noise B. Intermodulation Distortion The choice of amplifier directly impacts the dynamic range of the filter since the amplifier is the dominant source of both noise and distortion. Suppose that the amplifier in Fig. 2 is ideal except for being weakly nonlinear. For a given choice of amplifier, distortion of the filter is related to that of the amplifier in a simple manner. Because the gain of the bandpass filter is relatively constant for frequencies close to the center frequency, the filter behaves as a simple positive-feedback amplifier for in-band signals. So we should expect the filter’s third-order intermodulation distortion to be worse than that of the amplifier, and it can be shown to be (18) This is not a surprising result, albeit an unattractive one. It speaks directly to the required linearity of the amplifier, and to the need for low circuit noise if high dynamic range is to be realized along with a high quality factor. III. THE FILTER To investigate the utility of the proposed -enhancement scheme, a simple prototype circuit was designed and fabricated using a 0.8 m BiCMOS technology [15], [16]. The circuit is shown in Fig. 3, and is a direct implementation of the circuit in Fig. 2. The amplifier is a resistively loaded Gilbert quad, [17], with the output nodes buffered by the emitter followers , . Thus, the midband gain resistance is defined by the 180 load resistors and the ratio of the collector current and . By varying either of or , -control of can be effected. Biasing for the input diodes is provided by The (differential) input-referred noise of the amplifier, together with the appropriate noise gains, determine the output noise contribution of the filter. Referring to Fig. 3, we see that for a balanced circuit the noise contribution from the current is common-mode, and is rejected by the circuit, so source and we will ignore it. Similarly, noise resulting from establishes a common-mode component on the gates of and which, ultimately, is rejected by the circuit. The equivalent voltage noise power spectral density at the input of the amplifier can be shown to be (19) , and are the base resistance, collector where , is the resistance, and transconductance of the devices, and current-gain of the quad. Due to the large transconductance of , as compared to [the ratio is ], the noise contribution of the MOSFET’s is minuscule. From (19), for a low-noise filter it is imperative that the base resistance of the input transistors be minimized. In the BiCMOS technology used by us, the bipolar base resistance was quite high ( 100 ) as the bases were not realized using self-aligned polysilicon, so appreciable noise resulted. In addition, the input-referred current noise power is easily found, and is (20) where is the collector current in each transistor comprising the input diodes. 4 The unit BJT is 4 2 0.8 m2 . DUNCAN et al.: A Q–ENHANCED ACTIVE-RLC BANDPASS FILTER 345 Fig. 5. A high- Q bandpass filter. ratios, the simulated value was used to deduce internal signal levels. IV. MEASURED RESULTS Fig. 4. Layout plot of the bandpass filter. B. Distortion in the Amplifier Distortion in this amplifier arises mostly from device mismatches and the presence of parasitic base and emitter resistances. These nonidealities result in parabolic and cubic nonlinearities [17], respectively, but even-order distortion is of secondary concern in a fully balanced configuration. Parasitic base and emitter resistances are more troublesome because they affect the third-order intermodulation distortion. If these in resistances are modeled as a single resistor the emitter of each transistor, the third-order intermodulation distortion can be expressed as (21) , , and where is the amplitude of the signal current in is emitter–area ratio of to . As seen, intermodulation distortion depends on three factors: the first is the voltage modulation at the input of the amplifier; the second is the current modulation of the diodes; and the third is the current density in the devices. That is, if the current gain is equal to the emitter-area ratio of the devices (the devices are operating with equal current densities), no distortion is introduced by the finite emitter resistances. This latter condition is unlikely to occur since the variable current gain is used to tune the factor of the filter. Thus, good linearity comes at the expense of increased power and circuit area. The layout of the circuit is shown in Fig. 4. The active 740 m. The area of the circuit is approximately 860 m inductors are readily apparent at the left of the figure, with the amplifier and capacitors occupying much of the central portion of the picture. A. Tuning 180 The test setup comprised the filter, a pair of 50 power splitters, and four 100 k potentiometers that allowed external control of the bias current of every transistor in the signal path. One power splitter was used at the input to achieve single-ended-to-differential conversion, and the other at the output to provide differential-to-single-ended conversion. A very high- response is shown in Fig. 5. The measured center frequency and were 740 MHz and 494, respectively. The relatively low center frequency is, in a large part, caused by significant parasitic wiring inductance (a layout error) introduced by the connection from the inductors to the amplifier. This parasitic increased the nominal circuit inductance by approximately 20%. Fig. 6 illustrates the tuning of the -factor, and also the second-order dependence of the center frequency on the is tuned from 20 to about 120, the factor. In the plot, as center frequency decreases from 768 to 741 MHz. This is attributed to the change in the operating point of the amplifier (and hence its frequency response) as the bias currents change to tune the quality factor. The flattening of the responses away from the center frequency is attributed to a combination of noise from the test setup and the input-signal coupling through the circuit board. C. The Output Driver B. Linearity and Dynamic Range driver is a Darlington, emitter-degenerated difThe 50 ferential pair. It nominally dissipates 55 mW, and supports an output swing of 400 mV. The driver exhibited a simulated insertion loss of 18 dB, and since this is defined by resistor 20, used equalAn intermodulation distortion test, for amplitude signals placed at 765.1 and 775.3 MHz. The output response is plotted in Fig. 7. For an input of 40 dBm, the filter exhibited a signal-to-distortion ratio of 36 dB. The 346 Fig. 6. Tuning the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 44, NO. 5, MAY 1997 Q-factor. Fig. 8. Jump resonance for Q = 32. V. CONCLUSION A high-frequency, high- bandpass filter operating at 750 MHz was designed and implemented using inductors with factors of 2 at 1 GHz. The circuit was fabricated in a 0.8 m BiC-MOS technology and demonstrates the potential of integrating a low-noise, high- 1 GHz bandpass filter. Although the dynamic range of the filter was rather limited, this was primarily due to the linearity limitations of the amplifier used. As discussed, the linearity of the requisite amplifier deserves careful attention, but for a fixed- application, the Gilbert amplifier can still be used to advantage. The noise performance of the amplifier as well needs to be improved. ACKNOWLEDGMENT Fig. 7. Intermodulation distortion for Q = 20. poor distortion behavior is attributed to the nonlinearity of the amplifier. Noise from the circuit board, coupled with the low gain of the output drivers ( 18 dB when driving 50 ), resulted in the test setup dominating the output noise. In spite of this limitation, the spurious-free dynamic range for this experiment was estimated as 25 dB, with an internal third-order intercept point of 8 dBm. C. Jump Resonance The folding of intermodulation distortion products back on the fundamental typically results in gain-compression in amplifiers. In a bandpass filter, however, the phase of the cubic with respect to the fundamental changes as the input frequency is swept through resonance [19], [20]. This gives rise to gain-expansion (larger gain) below resonance and gaincompression above resonance. This increase in gain below resonance potentially leads to instability and is illustrated in Fig. 8. In the filter, it is speculated that it is caused by the strong cubic nonlinearity of the amplifier. For the response shown, the filter- is set to 32 and the input power increased from 50 to 20 dBm. This phenomenon severely limited the signal handling capability of the filter. The authors are indebted to M. Snelgrove for the use of the resources at Carleton University, and to P. Lauzon for the layout of the circuit board. Discussions with J. Long and M. Copeland were much appreciated. REFERENCES [1] A. Abidi, “Noise in active resonators and available dynamic range,” IEEE Trans. Circuits Syst., vol. 39, pp. 296–299, Apr. 1992. [2] G. Groenewold, “The design of high dynamic range continuous-time bandpass filters,” IEEE Trans. Circuits Syst., vol. 38, pp. 838–852, Aug. 1991. [3] B.-S. Song and P. R. Gray, “Switched-capacitor high-Q bandpass filters for IF applications,” IEEE J. Solid-State Circuits, vol. SC-21, pp. 924–933, Dec. 1986. [4] L. Toth, V. Gopinathan, N. G. Maratos, and Y. P. Tsividis, “Bounds on noise in integrated active-RC and MOSFET-C filters,” in Proc. ISCAS, 1993, pp. 1255–1258. [5] Y.-T. Wang and A. Abidi, “CMOS active filter design at very high frequencies,” IEEE J. Solid-State Circuits, vol. 25, pp. 1562–1573, Dec. 1990. [6] J. R. Long, “High-frequency integrated circuit design in BiCMOS for monolithic timing recovery,” Master’s thesis, Elect. Eng. Dept., Carleton Univ., Canada, 1992. [7] N. M. Nguyen and R. G. Meyer, “Si IC-compatible inductors and LC passive filters,” IEEE J. Solid-State Circuits, vol. 25, pp. 1028–1030, Aug. 1990. [8] Y.-C. Chang et al., “Large suspended inductors on silicon and their use in a 2 m CMOS RF amplifier,” IEEE Electron Device Lett., vol. 14, pp. 246–248, May 1993. [9] K. Negus et al., “Highly-integrated transmitter RFIC with monolithic narrowband tuning for digital cellular handsets,” in ISSCC Dig. Tech. Papers, Feb. 1994, pp. 38–39.. DUNCAN et al.: A Q–ENHANCED ACTIVE-RLC BANDPASS FILTER [10] R. Duncan, K. Martin, and A. Sedra, “A Q-enhanced active-RLC filter,” in Proc. ISCAS, 1993, pp. 1416–1419. [11] Y. Tsividis, “Integrated continuous-time filter design,” in Proc. CICC, 1993, pp. 6.4.1–6.4.7. [12] S. Pipilos and Y. Tsividis, “Design of active RLC integrated filters with application in the GHz range,” in Proc. ISCAS, 1994, pp. 645–648. [13] W. M. Snelgrove and A. S. Sedra, “Synthesis and analysis of state-space active filters using intermediate transfer functions,” IEEE Trans. Circuits Syst., vol. CAS-33, pp. 287–301, Mar. 1986. [14] G. W. Roberts, “Generalization and applications of the intermediate function technique,” Ph.D. dissertation, Elect. Eng. Dept., Univ. Toronto, Canada, 1989. [15] R. Hadaway et al., “A sub-micron BiCMOS technology for telecommunications,” J. Microelec. Eng., vol. 15, pp. 513–516, 1991. [16] “Design rules and process parameters for CMC 0.8-m BiCMOS, a version of NTE BAT-MOS,” Rep. ICI-040R0, Canadian Microelectron. Corp., Kingston, Ont., Canada. [17] B. Gilbert, “A new wideband amplifier technique,” IEEE J. Solid-State Circuits, vol. 3, pp. 353–365, Dec. 1968. [18] R. Duncan, “Active-RLC filters in silicon,” Ph.D. dissertation, ECE Dept., Univ. Toronto, Canada, 1995. [19] M. Snelgrove, private communication. [20] J. Cherry, “Distortion analysis of weakly nonlinear filters using Volterra series,” Master’s thesis, Elect. Eng. Dept., Carleton Univ., 1994. [21] A. Gelb and W. Vander Velde, Multiple-Input Describing Functions and Non-Linear System Design. New York: McGraw-Hill, 1968, ch. 3. Ralph Duncan received the B.A.Sc., M.A.Sc., and Ph.D. degrees in electrical engineering from the University of Toronto, Toronto, Canada, in 1989, 1991, and 1995, respectively. He is currently with Maxim Integrated Products, Sunnyvale, CA, as member of the technical staff in the Signal Processing and Converters group. 347 Kenneth W. Martin (F’91) received the B.A.Sc., M.A.Sc., and Ph.D. degrees from the University of Toronto, Canada, in 1975, 1977, and 1980, respectively. From 1977 to 1978, he was a member of the Scientific Research Staff at Bell Northern Research, Ottawa, Canada, where he conducted some of the early research in integrated switched-capacitor networks. Between 1980 and 1992 he was, consecutively, an assistant, associate, and full professor at UCLA. In 1992, he accepted the endowed “Stanley Ho Professorship in Microelectronics” at the University of Toronto. He has also been a consultant to many high technology companies including Xerox Corp., Hughes Aircraft Co., Intel Corp., and Brooktree Corp. in the areas of high-speed analog and digital integrated circuit design. He has ongoing research programs in the areas of analog CMOS and BiCMOS systems, high-speed GaAs MESFET and HBT circuits, and digital-signal-processing algorithms for fixed and adaptive filters. Professor Martin was appointed as the Circuits and Systems IEEE Press representative (1985 to 1986). He was selected by the IEEE Circuits and Systems Society for the Outstanding Young Engineer Award that was presented at the IEEE Centennial Keys to the Future Program in 1984. He was elected by the Circuits and Systems Society members to their administrative committee (ADCOM 1985–1987). He served as an Associate Editor of the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS from 1985 to 1987, and has served on the technical committee for many International Symposia on Circuits and Systems. He is currently an Associate Editor for the PROCEEDINGS OF THE IEEE, chairman of the IEEE CAS Guillemin–Cauer Best Paper Awards Committee, and a member of the IEEE Circuits and Systems Board of Governors (1995 to 1997). He was awarded a National Science Foundation Presidential Young Investigator Award that ran from 1985 to 1990. He was a corecipient of the Beatrice Winner Award at the 1993 ISSCC. Adel S. Sedra (M’66–SM’82–F’84), photograph and biography not available at the time of publication.