HIGH-ORDER LOWPASS AND BANDPASS ELLIPTIC LOG-DOMAIN LADDER FILTERS E.M.Drakakis†, A.J.Payne†,C.Toumazou†, A.E.J.Ng‡ and J.I.Sewell‡ † Dept. of El.&El.Eng., Imperial College, Exhibition Road, London, SW7 2BT, UK ‡Dept. of El.&El.Eng., University of Glasgow, Glasgow, G12 8QQ, UK {e.drakakis, a.j.payne, c.toumazou}@ic.ac.uk, {a.ng, j.sewell}@elec.gla.ac.uk ABSTRACT This paper reports a multitude of HSPICE simulation results corresponding to high-order (fifth, sixth and tenth) differential classA log-domain lowpass and bandpass ladder filter topologies. The designs are elliptically approximated and were obtained by means of an appropriately extended version of the XFILTER filter compiler. The non-desirable effect of the BJT finite beta is shown; circuit modifications mitigating - to a certain extent - the device nonideality-induced impact are also reported. 1. INTRODUCTION Log-domain filtering [1] is characterised by considerable mathematical burden originating from the fact that the active devices (usually BJTs) are treated as non-linear computational primitives allowing the processing of large input current signals which strongly exercise the non-linear (exponential) I-V device characteristic while preserving the input-output large-signal linearity. The mathematical complexity associated with the technique has, so far, restricted the reported externally-linear-internally-non-linear (ELIN) [2] logdomain topologies to all-pole single-ended signal-flow-graph (SFG)-based ladder designs [3], two-port cascades [4], maximally flat responses [5] or cascaded biquads with or without feedback [68]. Recent advances [9] in the development of the XFILTER CAD package [10] however, have allowed - for the first time - the automated synthesis of high-order log-domain ladder filters which comply with predetermined specifications. XFILTER offers a variety of commonly used and arbitrary frequency response approximations and matrix decompositions [10]. This progression allows for the viability of the log-domain technique to be evaluated in a systematic way. More specifically high-order differential elliptic ladder filter topologies can be studied for the first time. This paper presents a plethora of HSPICE simulation results which extend our understanding of log-domain filtering since they: a. indicate the way in which the device non-idealities and particularly the finite BJT current gain affects the intended frequency response of demanding high-order (fifth, sixth and tenth) elliptically approximated lowpass or bandpass ladder designs, and b. reveal to a certain extent the way in which these high-order designs behave in general. Possible circuit modifications (suitable for bipolar processes) mitigating the device non-ideality-induced undesirable effects on the pre-specified transfer functions are also reported. Large-signal transient simulations verifying the reported small-signal ac behaviours were also performed. 2. A FIFTH-ORDER CLASS-A DIFFERENTIAL LOW-PASS ELLIPTIC LADDER FILTER The netlist of a fifth order differential elliptically approximated ladder filter was produced by means of XFILTER. The Appendix provides part of such a netlist corresponding to the interconnection of the basic blocks termed “fde”, “fdintcap” and “fdidc” (see Fig.1); XFILTER uses these elementary blocks to realise the desired frequency responses. Voutp Voutn fde I0 I0 fdintcap Vinp Vinn I0 I0 Id C trimming current C Id fdidc trimming current Fig.1 A differential log-domain integrator [11] It should be underlined however that the netlist provided in the Appendix is indicative only, corresponds to an ideal case (beta >> 1, capacitor values in the range of nFs) and simply aims to serve as an example of the functionality of the extended XFILTER code. The simulation results which follow though, were obtained using commercially available process parameters. When the basic blocks fde, fdintcap and fdidc are assembled [11] a differential log-domain integrator is produced. As it will become clear, a considerable improvement of the simulated responses can be achieved once the diode-connections are substituted for BJTs (as shown in Fig.1). This simple buffering scheme seems to somewhat mitigate the betainduced errors of the topology (1/β errors at the buffering nodes now become proportional to 1/β2). Fig.2 reveals in a comparative way the effect of beta upon the fifth order elliptic response (the rest of the nominal transistor parameter values were kept unaltered when beta was changed): higher beta values lead to responses closer to the ideal one. However, when the buffering connections are introduced, a considerable improvement can be observed when beta=200 (see Fig.3): the passband is situated close to the ideal one whereas the notch depth increases (compare with Fig.2). Curves (1) and (2) correspond to different dc bias current values present at the input (denoted hereafter as D). For curve (1) D=85µA whereas for curve (2) D=8.5µA. Fig.4 illustrates in detail the passband ripple both for the ideal case and when beta=200 (curve (1)); small (∼ 16nAs) trimming currents fed into the integrating nodes (see Fig.1) can compensate for the Q-degradation as shown by curve (2) in the same figure. Fig.5.a shows an example of tuning via current (one octave up in frequency (curve (3)) and octave down (curve (2)) for a response with a cut-off frequency of 1MHz (curve (1)); curve (4) of the same figure corresponds to curve (1) when compensated by means of small trimming currents (∼ 0.2µA). Figure 5.b shows the output current THD levels for curve (4) of Fig.5.a when the modulation index M.I. (defined as the ratio of the peak amplitude to the dc bias current D) varies between 10% and 90% for three different frequencies (0.1, 0.5 and 0.9 MHz) situated in the passband; 1000 points per period were used with the simulation accuracy set to the maximum allowable level. The results can be characterised as satisfactory; for input M.I. of 50 to 60% the THD distortion level remains below -60dBs. -40 THD (dBs) -50 -60 -70 100KHz -80 500KHz -90 900KHz -100 0 10 20 30 40 50 60 70 80 90 100 M .I.(%) Fig. 5.b Passband THD level (dBs) for curve (4) of Fig.5.a 3.SIXTH AND TENTH ORDER CLASS-A DIFFERENTIAL BAND-PASS ELLIPTIC LADDER FILTERS The beta-induced error reduction scheme mentioned previously does seem to offer a viable solution when the capacitor and bias current value spreads are low; for the fifth order LP elliptic design for example, the capacitor and current spread values were 2.5 and 17 respectively. In contrast, when high spread designs are generated, buffering does not seem to improve the frequency response. More specifically: considering a sixth-order bandpass elliptic ladder response with approximate capacitor spread value of 10.3 and current spread value higher than 100 and referring to Fig.6, it can be easily deduced that despite the incorporation of buffering no improvement can be observed (for clarity the ideal response is presented shifted in the frequency domain). On the other hand, for the case of lower capacitor and current value spread designs, the desirable effect of buffering can be clearly noticed again as shown in Fig.7 which illustrates a sixth-order bandpass elliptic response with approximate capacitor and current spread values of 1.5 and 31 respectively; referring to the same figure, the passband loss of curve (2) is rectified and the position of the transmission zeros becomes clearer. Fig. 8 illustrates in detail a similar sixth-order bandpass elliptic response tuned via current to a higher center frequency; extensive simulations revealed that the “tilt” of the pass-band can be mitigated by fine tuning the current value of a specific fde block. In a similar way, the transmission zeros can also be leveled-up. Fig.9 deals with a third elliptic bandpass design of sixth-order characterised by even lower capacitor and current spread values (1.4 and 15 respectively): the desirable impact of buffering is verified once more. The slight tilt of the passband can be corrected by means of a combined effect of small trimming currents and small resistors connected in series with the integrating capacitors. Curves (2) and (3) of Fig. 10 show the effect of the presence of trimming currents only whereas curves (4) and (5) of the same figure illustrate the impact of series-connected resistors only; the response (6) is obtained for a specific trimming current-series resistor value combination which corrects the passband tilt. Large-signal transient simulations showed that no severe penalty was paid as far as linearity (IMD product levels) is concerned. Table I provides indicative simulation results; for two fundamental tones of 0.98 and 1.02 MHz (situated 2% apart from the center frequency) with a combined M.I. corresponding to 86%, the IMD product levels were ≈-40dBs. For convenience Table II summarises the capacitor and bias current spread values of the sixth-order bandpass topologies presented so far. M.I. (%) IMD(dBs) 28 -56 57 -44 86 -40 Table I IMD product levels for the 6th-order BP response of Fig.10; fundamental tones: 0.98 and 1.02 MHz capacitor current BP filter spread spread Fig.6 10.3 >100 Figs.7,8 1.5 31 Figs.9,10 1.4 15 Table II Capacitor and bias current spread values for the 6th-order bandpass designs Fig.11 considers the frequency response of a tenth order differential bandpass elliptic ladder topology for both ideal and realistic process parameters. The design is characterised by a capacitor value spread of 1.58 and a bias current spread value of 15; the center frequency equals to 1MHz, the passband extends to 200KHz, the passband ripple amounts to 0.5dBs whereas the stopband rejection is more than 70dBs. The deviation from the ideal case is disappointing when beta-compensation is not incorporated. However, as illustrated in Fig.12, an appropriate combination of trimming currents and series connected resistors (found by means of exhaustive simulations) in conjunction with a limited trimming (less than 5% of the nominal value) of a specific fdintcap block capacitor pair value, can lead to a considerable overall improvement for a similar design. Table III provides indicative simulation results; for two fundamental tones of 1.6 and 1.7 MHz (situated 3% apart from the center frequency which equals 1.65MHz) with a combined M.I. corresponding to 80%, the output current IMD levels were ≈ -38dBs. M.I. (%) IMD(dBs) 40 -44 60 -41 80 -38 Table III IMD product levels for the 10th-order BP response of Fig.12; fundamental tones: 1.6 and 1.7 MHz 4. SUMMARY This paper dealt with the presentation of HSPICE simulation results corresponding to differential class-A fifth-order lowpass, sixth-order bandpass and tenth-order bandpass ladder log-domain filters. In every case the elliptic approximation was used. The filters were synthesised by virtue of an appropriately extended version of the XFILTER CAD package. Examples of the impact of the finite BJT current gain were provided. Moreover, simple circuit modifications leading to improved responses were reported. Test-chips are currently being laid-out. 5. ACKNOWLEDGEMENTS Financial support from the Engineering and Physical Sciences Research Council (GR/M45368) is gratefully acknowledged. 6. REFERENCES [1] D.R.Frey, “Exponential State-Space Filters: a generic currentmode design strategy”, IEEE Trans. CAS-I, vol.43, no.1pp. 3442, 1996 20 beta=2000 beta=200 10 0 -10 ideal response TF (dBs) -20 -30 -40 -50 -60 -70 -80 2 10 3 4 10 5 10 10 Freq (Hz) 6 7 10 10 8 10 Fig.2 High-beta values lead to behaviours close to the ideal response; capacitors in the range of hundreds of pFs; current source values in the range of tens of µAs; power supply levels: ±1.65V 20 10 ideal response (2) 0 (1) -10 -20 TF (dBs) iinp vdd 17 dc= 8.49e-05 ac= 4.2e-05 0.0 iinn vdd 18 dc= 8.49e-05 ac= 4.2e-05 180 xfdintcap1 1 2 fdintcap c= 4.9e-08 xfdintcap2 3 4 fdintcap c= 8.6e-08 xfdintcap3 5 6 fdintcap c= 3.5e-08 xfdintcap4 7 8 fdintcap c= 8.6e-08 xfdintcap5 9 10 fdintcap c= 8.2e-08 xfdidc6 11 12 fdidc io= 7e-05 xfdidc7 13 14 fdidc io= 1e-04 xfdidc8 15 16 fdidc io= 5.7e-05 xfdidc9 1 2 fdidc io= 4.7e-05 xfdidc10 5 6 fdidc io= 2.9e-05 xfde11 7 8 1 2 fde io= 1e-04 xfde12 7 8 3 4 fde io= 6.9e-05 xfde13 9 10 3 4 fde io= 1e-04 xfde14 9 10 5 6 fde io= 1e-04 xfde15 1 2 8 7 fde io= 6.8e-05 xfde16 3 4 8 7 fde io= 1e-04 xfde17 3 4 10 9 fde io= 1e-04 xfde18 5 6 10 9 fde io= 6.1-05 xfde19 13 14 12 11 fde io= 6.1e-05 xfde20 11 12 14 13 fde io= 5.9e-06 xfde21 15 16 14 13 fde io= 1.8e-05 xfde22 13 14 16 15 fde io= 4.2e-05 xfde23 13 14 2 1 fde io= 2e-05 xfde24 11 12 4 3 fde io= 8.3e-06 xfde25 15 16 4 3 fde io= 2.6e-05 xfde26 13 14 6 5 fde io= 4.2e-05 xfde27 7 8 11 12 fde io= 1e-04 xfde28 7 8 13 14 fde io= 4.9e-05 xfde29 9 10 13 14 fde io= 7e-05 xfde30 9 10 15 16 fde io= 1e-04 xfde31 1 2 12 11 fde io= 4.7e-05 xfde32 5 6 16 15 fde io= 2.9e-05 xfde33 17 18 1 2 fde io= 8.5e-05 xfde34 17 18 11 12 fde io= 8.5e-05 q35 17 17 0 0 nbjt q36 18 18 0 0 nbjt q37 19 5 0 0 nbjt q38 20 6 0 0 nbjt eoutp 19 0 vcvs vdd 0 1.0 eoutn 20 0 vcvs vdd 0 1.0 .probe ac i(eoutp) i(eoutn) -30 -40 -50 -60 -70 -80 2 10 3 10 4 10 5 10 Freq (Hz) 6 7 10 10 8 10 Fig.3 The buffering improves (compare with Fig.2) the desired response when beta=200; power supply levels: ±1.65V 4 3 2 (2) (with trimming currents) 1 TF (dBs) [2] Y.Tsividis, “Externally Linear Time Invariant Systems and their application to companding signal-processors”, IEEE Trans. CAS-II, vol.44, no.2, pp. 65-85, 1997 [3] V.W.Leung and G.W.Roberts, “Effects of Transistor Nonidealities on High-Order Log-domain ladder Filter Frequency Responses”, IEEE Trans. CAS-II, vol.47, no.5, pp. 373-387, 2000 [4] G.van Ruymbeke, C.C.Enz, F.Krummenacher and M.Declerq, “A BiCMOS programmable continuous-time imageparameter method synthesis and voltage-companding technique”, IEEE JSSC, vol.32, no.3, pp.377-387, 1997 [5] M.N.El-Gamal and G.W.Roberts, “Very High-Frequency Logdomain Bandpass Filters”, IEEE Trans. CAS-II, vol.45, no.9, pp. 1188-1198, 1998 [6] E.M.Drakakis, A.J.Payne and C.Toumazou, “Multiple Feedback Log-domain Filters”, in Proc. IEEE ISCAS98, vol.1, pp. 317-330 [7] E.M.Drakakis and A.J.Payne, “Leapfrog Log-domain Filters”, in Proc. IEEE ICECS 1998, vol.2, pp.385-388 [8] D.R.Frey, “Log-domain Filtering: an approach to current-mode filtering”, IEE Proc. Part-G, vol.140, pp.406-416, 1993 [9] A.E.J.Ng, J.I.Sewell, E.M.Drakakis, A.J.Payne and C.Toumazou, “A Unified Matrix Method for systematic Synthesis of Log-domain Ladder Filters”, submitted to IEEE ISCAS 2001 [10] A.E.J.Ng and J.I.Sewell, XFILT reference manual, ver. 3.1. Dept.of El. and El.Eng., Univ.of Glasgow, 1997 [11] F.Yang, C.Enz and G.van Ruymbeke, “Design of Low-Power and Low-Voltage Log-domain Filters, in Proc. IEEE ISCAS96, vol.1, pp. 117-120 APPENDIX A sample XFILTER netlist corresponding to a fifth-order lowpass elliptic design: 0 ideal response -1 -2 (1) -3 -4 3 10 4 10 Freq (Hz) Fig.4 The application of small trimming currents (∼16nAs) can lead to considerable improvement in the passband; power supply levels: ±1.65V 10 10 ideal response (4) (with trimming currents) 0 0 -10 (2) with buffering (3) (1) -10 -20 without buffering TF (dBs) TF (dBs) -30 -20 -40 -30 -50 -60 -40 -70 -50 -80 -60 5 10 6 7 10 -90 5 10 8 10 10 6 Fig.5.a Tuning of 1MHz response (curve(1)) via current; capacitors in the range of tens of pFs; current source values in the range of tens of µAs; power supply levels: ±1.65V 7 10 Freq (Hz) Freq (Hz) 10 Fig.9 Very low spread sixth-order bandpass elliptic design verifying the desirable impact of buffering. Operational conditions same as in Figs.7 and 8 5 20 shifted ideal response (80nA,0 Ohm) (3) 0 (60nA,0 Ohm) (2) (50nA,12 Ohm) (6) (0nA,0 Ohm) (1) 0 (4)(0nA,10 Ohm) (5)(0nA,20 Ohm) -5 without buffering -40 TF (dBs) TF (dBs) -20 -60 -10 with buffering -80 -15 -100 -120 1 10 2 3 10 10 4 5 10 6 10 7 10 -20 8 10 6 10 Freq (Hz) 10 Freq (Hz) Fig.6 Sixth-order bandpass elliptic response; capacitors in the range of tens of pFs; current value spread > 100 10 0 10 -10 TF (dBs) Fig.10 Effects of small trimming currents and small resistors connected in series with the integrating capacitors for the response of Fig.9 (a) -20 0 ideal response -30 -40 -10 without buffering -50 -20 -60 -70 3 10 4 -30 10 Freq (Hz) (1) with buffering 0 TF (dBs) -10 -40 TF (dBs) 10 -50 (b) -60 -20 (2) without buffering -30 -70 -40 -50 -80 -60 -90 -70 6 10 Freq (Hz) Fig.7 Low-spread sixth-order elliptic response; (a): shifted ideal frequency response (b): improved response when buffering is applied; capacitors in the range of tens of pFs; current source values in the range of µAs; power supply levels: ±1.65V -100 5 10 7 10 Fig.11 Very low spread 1MHz tenth-order bandpass elliptic response; capacitor values in the range of tens of pFs and current source values in the range of µAs; power supply levels: ±2V 10 0 5 -2 TF (dBs) (a) TF (dBs) 6 10 Freq (Hz) -4 -6 0 (a) -5 -10 -8 -15 -10 2.5 3 3.5 Freq (Hz) 4 -20 6 10 4.5 Freq (Hz) 6 x 10 0 0 -10 -20 (b) TF (dBs) TF (dBs) -20 -30 -40 (b) -40 -60 -50 -80 -60 -70 6 10 7 10 Freq (Hz) Fig.8 Sixth-order bandpass elliptic response with a center frequency of 3.4MHz and a passband of ∼340KHz. (a):passband detail (b):the complete response; buffering is applied; spread values and operational conditions same as in Fig.7 -100 5 10 6 10 Freq (Hz) 7 10 Fig.12 A beta-compensated tenth-order bandpass elliptic response with small trimming currents (∼90nAs) and small resistors (∼11Ω) connected in series with the integrating capacitors; operational conditions as in Fig. 11; center frequency=1.65MHz, passband≈400KHz. (a):passband detail (b): the complete response