Low-Power Low-Voltage Chopped Amplier with a New Class AB Output Stage for Mixed Level Applications M.A.T. Sanduleanu1 , A.J.M. van Tuijl1;2, R.F. Wassenaar1 , M.C. Lammers1 and H. Wallinga1 1 University of Twente, Department of Electrical Engineering P.O.Box 217, 7500 AE Enschede, The Netherlands tel. +31 (0)53 489 4007 fax. +31 (0)53 4892 799 2 Philips Semiconductors, Nijmegen, The Netherlands m.a.t.sanduleanu@el.utwente.nl Abstract | This paper describes the principle and namic range [1]. The solution would be to keep the the design of a 0.5m CMOS, low-power, low- largest possible swing but things get worse because voltage, chopped amplier for noise and oset re- of the conict between large swings and cascoding duction in mixed analogue digital applications. The needed to get high gains. In conventional chopper operation is based on chopping and dynamic eleopamps [2], [3], [4] for 1/f noise and oset ment matching to reduce noise and oset, without stabilized reduction, dierential ampliers are being used and excessive increase of the charge injection residual oset. It consists of a chopped transconductance bandwidth is limited to few tens of kHz. Switching stage and a new class AB stage capable of working at the dierential output will introduce most of the at 1.3V supply voltage. The main goal is to achieve switching noise and residual oset [1]. Other solutions low residual osets by chopping at high frequencies for oset reduction like ping-pong techniques have the reducing at the same time the 1/f noise of the am- disadvantage of high power consumption and linearplier. Loaded with a heavy load 32 jj 300pF it ity problems [5]. Rail to rail input stages have ohas a 91dB open loop gain and a GBW of 1.8MHz. set and linearity problems [6]. This paper presents a Simulations show a THD of -90dB for a 1k load and -83dB for a 32 load and 1KHz input signal. chopped amplier which circumvents the above menThe simulated static oset is 1.67mV. The simu- tioned drawbacks. The chopping frequency can reach lated residual oset is 450V at 10MHz chopping. MHz range without excessive increase in residual oAt 1KHz chopping, the oset becomes lower than set. The new class AB output stage allows large volt1V. age swings, low osets and high linearity. The opamp can be used wherever oset and noise specications are important. However, the chopped amplier is priI. Introduction marily meant for use as an amplier capable of driving In a CMOS technology for mixed signal applications, headphones in portable digital audio. The 1 bit outlowering the supply voltage is a must, being dictated put of the D/A converter has to be ltered in order by digital signal processing functions. The process, to get less out of band noise and to eliminate the reptuned towards digital needs becomes less and less com- etitions of the spectra due to the sampling process. patible with analog requirements. As a consequence, The analog lter has to do the job without increasing for analog functions built in digital CMOS, the sig- the in-band noise and with high linearity. Besides, nal swing reduces with power supply and so does the it has to be integrated on the same chip with the dynamic range. 1/f noise properties are worsened for D/A converter and the sampled-data FIR lter, in surface PMOST's in comparison to buried transistors a pure digital CMOS technology. The input signal for and this, under the presence of substrate bounce due the chopped opamp, consists of a baseband spectrum to low ohmic substrates which reduces further the dy- and the attenuated aliases centered around multiples 451 Proceedings of the ProRISC Workshop on Circuits, Systems and Signal Processing 1997 of the sampling frequency. In this particular design, 64 times oversampling is used, such that the eective sample frequency is 2.8MHz. Chopping at multiples of sampling frequency is an advantage, because, the spectrum is quiet, due to the sin(x)=x type of dips from sampling process. Chopping at high frequency will produce residual oset which has to be withstand by the headphone causing dc currents. The paper focuses on the design and the realization of an amplier with a class AB output stage capable of chopping up to 10MHZ, with low noise, high linearity and low residual oset. Noise at output (dBuV/root(Hz)) 452 Chopamp 40 30 20 10 0 -10 -20 -30 -40 -50 10 Vout(f) m(t) M(f) 1/n 1 2fchop fchop -1 6fchop Fig. 2. Chopped amplier spectrum 5fchop 3fchop m(t) Vin(t) 4fchop 99.43 988.6 9829 97724 1E+0 9.66E 9.61E 6 +06 +07 Frequency (Hz) m(t) A Vout(t) Snoise(f) Vnoise+offset Sthermal fchop 3fchop 5fchop Fig. 1. Chopping principle II. Chopping principle The chopping principle is depicted in g.1. Here, the input signal is multiplied with a rectangular signal m(t) with unity amplitude and 50% duty-cycle. As a result, the signal is modulated at odd harmonics of the chopper frequency because, the chopping signal has spectral contributions only at odd harmonics. The signal will be amplied and modulated back, leaving spectral contributions at even harmonics of the chopper frequency. The strength of the modulation signal decreases with 1=n where n is the harmonic number. Therefore, chopping is not producing important foldover eects as in the sampling case. Oset and 1/f noise are modulated at odd harmonics leaving the baseband free of 1/f noise. In the ideal chopping case, the bandwidth of the amplier should be innity. As long as this is true, multiplying the signal twice with m(t) will reconstruct the input signal ideally. If the bandwidth of the amplier is limited, the result is a high frequency residue centered around the even harmonics and the signal in the baseband is attenuated. The eect of chopping on 1/f noise can be seen from g.2 where the measured low frequency spectrum of a chopped amplier is shown. Here, the chopping frequency is 1MHz. Two situations are considered: an unchopped version and a chopped one. The residual noise, left at low frequencies, is just the thermal noise. A necessary condition to cancel out the 1/f noise can be derived from g.3. This condition would be: fchop BWsignal + fcorner (1) Snoise(f) Sthermal fchop fchop-fcorner BW signal f Fig. 3. Baseband spectrum The eect of chopping on white and 1/f noise can be quantied by considering the power spectral density of the noise at the output as shown in g.4. A nite amplier bandwidth has been assumed. The normalized power spectral density of the white noise remains unchanged, as long as, the bandwidth of the amplier is about ten times larger than the chopping frequency. Under this condition, the normalized power spectral density of the 1/f noise remains nite at low frequencies. In g.4 the constant kF is the 1/f noise constant and depends on the process. As a conclusion, if we want to chop at high frequencies, we need a large amplier bandwidth compared to chopping frequency in order to reduce 1/f noise and to let the thermal noise unchanged. Low-Power Low-Voltage Chopped Amplier with a New Class AB Output Stage 453 III. Basic Circuit VDD In conventional choppers, the signal is being transposed at the input of the dierential pair, amplied and demodulated back at the output nodes as shown in g.5. Switching at high impedance nodes would be disadvantageous due to limited bandwidth of the amplier. In this approach, high frequency chopping is not possible and therefore, this method is limited to few tens of KHz. Besides, the switching noise is directly coupled to the output. Because we have switches in the middle of the supply voltage, for low voltage applications charge pumps are needed. This is to ensure that all switches are rmly open and/or closed. The basic circuit, shown in g.6, comprises an input modulator, a PMOS dierential pair, current sources and a low voltage, high bandwidth cascoded mirror, to perform a dierential to single ended conversion. The second chopper would transpose again the signal at low impedance nodes and demodulates back the signal, canceling out the oset of the bottom transistors. The oset and noise from the current sources would be canceled out by the third chopper which matches dynamically the two transistors on top. There are no consequences on the signal due to the third chopper. The benet of chopping at low impedance nodes comes from the unlimited bandwidth of the basic amplier. Therefore, we can chop at much higher frequencies where the only limitation would be the charge injection residual oset. In plus, the cascode transistors would provide low-pass ltering for the high frequency spectral contributions coming from chopping. In this approach, charge pumps are not needed if the common mode voltage is well chosen and switching is close to the supply rails. The output node, used for Miller compensation, lters out the undesired high frequency spectral components from switching, delivering to the output stage an oset/noise free voltage. S1/F /k T F Stherm al/S0 1 1.2 0.861 1.1 0.5 0.9 BW/fchop 0.826 0.0 0 5 10 0 f/fchop 5 10 Fig. 4. Noise PSD after chopping CM FEEDBACK VB2 OUT IN VB3 VB4 VSS Fig. 5. Conventional choppers VDD VB1 VB2 OUT IN VB3 VSS Fig. 6. Basic principle IV. The Class AB output stage The chopped amplier is divided in two parts: the gain stage and the output stage capable of driving low-ohmic loads. The output stage is shown in g.7 and consists of an input common-mode current source of the pair M14 and M15, with active load, and the class AB control. The output transistors M35, M36 are driven in phase from high impedance nodes. To control the output currents, a feedback control has been chosen. A scaled copy of the current in the output devices ows through M33 and M37. The two copies are forced in the transistors M34 and M32. The feedback loop will enforce the condition: VGS34 + VGS32 = Ebiasn (2) If the transistors M34 and M32 are in weak inversion and forward saturated (VDS UT ) then, the same 454 Proceedings of the ProRISC Workshop on Circuits, Systems and Signal Processing 1997 1.6m VDD Ibias ID_M36 1.4m M36 M37 1.2m M10 M9 ID_M35 1.0m Ebiasp 800.0u VOUT Ebiasn M15 M14 600.0u Rload M34 400.0u 200.0u IMIN M32 Iin M33 M35 IMIN VSS Fig. 7. Class AB output stage condition can be rewritten as: (mI36 ? IMIN )(mI36 ? IMIN ) = KW2 U (E ? 2V ) 2 (3) T n T biasn 0.0 -60.0m -40.0m -20.0m 0.0 20.0m 40.0m 60.0m V OUT Fig. 8. The class AB control of the output transistors the condition (2) can be rewritten as: s IMIN + Ebiasn ? 2VT = 2 (mI36 ? s 2 (mI35 ? IMIN (5) where n is the slope factor, UT is the thermal voltage, V. Circuit priciple represents the gain factor, m the scaling factor and KW is a constant. In most cases, the constant KW The circuit consists of a chopped transconductance stage and a class AB output stage. The nal version can be approximated with: of the circuit is depicted in g. 9. The dierential KW = 20 (n ? 1) (4) input signal is being transposed by the input chopper M3, M4, M5, M6, modulated and amplied by a PMOS dierential pair M1 and M2 working in weak The transistors M32 and M34 are considered identical inversion. The large bandwidth mirror M7, M8, M13 and at the same temperature. One can see that for a and M14,15 used for subtraction, is being chopped at given temperature, the product of the left hand side low impedance nodes. This operation demodulates terms is constant. The class AB behavior has been back the signal and modulates the noise and oset at simulated, as shown in g.8. If only the product rule odd multiples of the chopper frequency. The cascoded would have been implemented, the residual current is bias sources on top M43, M44, M52 and M24,25 are not limited. Hence, if one of the transistors would dynamically matched by using a third chopper out of conduct a lot of current, the other one works at very the signal path. The single ended signal current acts low currents and the gate-source voltage associated as a common-mode current of the dierential pair M14 with it becomes low. This is undesirable as long as the and M15 being splitted in two branches. Due to the circuit that drives the gates of the output transistors equal splitting, the output transistors M36 and M35 should have some 500div600mV voltage room to keep are driven in phase. The only sources of oset and all of the transistors in saturation. noise remains the output stage and mismatches beIn this approach there is a well dened residual cur- tween the two signal branches. There are few reasons rent in the output transistors. The two extra cur- to desire a high gain for the transconductance stage. rent sources IMIN subtract a small current from the For low ohmic loads, the output stage attenuates the copied output currents coming from M33 and M37. signal up to -12dB. Hence, the open loop gain of the This causes in all situations a shift upwards for the amplier drops down. If the transconductance stage current owing in the output transistors. The quies- has large gain, the oset and 1/f noise generated in cent current in the output stage is about 300A. If the output stage can be neglected. However, in modthe transistors M34 and M32 are working in strong ern processes, the output resistance of the transistors inversion, there is also a class AB control, this time, is very low. Simple cascoding does not oer a solu- Low-Power Low-Voltage Chopped Amplier with a New Class AB Output Stage 120/1.6 M47 120/1.6 60/1.6 M46 4/0.5 M44 4/0.5 1/10 120/1.6 120/1.6 M45 M43 M49 455 1/10 M42 M41 1/10 M40 25/1 25/1 160/0.5 M39 M38 M37 M36 2720/0.5 Fold=60 4/0.5 M50 M48 4/0.5 M51 120/1.6 60/1.6 M53 60/1.6 M24 350/1 Fold=6 M52 M0 30uA 120/1.6 120/1.6 M25 350/1 Fold=6 M22 M23 120/1.6 M20 300/1 Fold=8 300/1 Fold=8 M1 120/1.6 M21 C1 5nA 5nA 40nA 40nA 5pF M2 M4 2/0.5 M3 2/0.5 M5 2.5pF 2/0.5 C2 2/0.5 M6 M18 M19 70/1 70/1 10uA M17 M16 50/1 M54 50/1 M15 M14 M13 70/1 35/1 2/0.5 35/1 M10 27.6/15 M34 M9 2/0.5 10uA M11 2/0.5 2/0.5 160/0.5 Fold=6 M32 M12 M7 105/1 M26 M8 105/1 80/1 M27 M28 10/1 10/1 M31 160/0.5 M33 Fold=6 M29 M30 80/1 M35 1000/0.5 Fold=26 80/0.5 50/150/1 Fig. 9. Circuit diagram tion in this case. Besides, we want large swings at the output nodes of the transconductance stage in order VI. Simulations to ensure that the output transistors are driven out of saturation. A gain boosting circuit M16, M17, M22 The circuit has been simulated and the results are and M23 has been added to the cascode transistors shown below. For a 32 load, the maximum outM18, M19, M20 and M21. It's current consumption put voltage is close to the rails within 360 millivolts. is limited to few nanoamperes and the noise and oset of the gain boosting transistors is limited by the large areas used. At the output node, we can go as low as VGS 16 + VGS 18 ? VT . Because the gain boosting 1.8MHz transistor works at low current, in weak inversion, the gate-source voltage of M16 is close to VT and therefore we can go as low as 600mV from the VSS rail. The same applies to the PMOS counterparts. The gate of the transistor M14 is connected to a constant voltage and the dierential amplier will enforce a constant voltage at the gate of M15. The gain boosting circuit limits the minimum supply voltage at about 1.4V : 100.0 (LIN) 50.0 0.0 DB(Vout) -50.0 CL =5pF -100.0 CL =300p -150.0 100.0m 10.0 1.0 1.0k 100.0 100.0k 10.0k 10.0M 1.0M (LOG) VDD;min = VGS16 + VDSsat;18 + jVGS22 j + VDSsat;20 j Fig. 10. Open loop gain 1.0G 100.0M F The open loop gain of the circuit is presented in 456 Proceedings of the ProRISC Workshop on Circuits, Systems and Signal Processing 1997 g. 10 under the condition of a heavy load of 32 jj5pF and 32 jj300pF load capacitance, respectively. The opamp has a gain bandwidth product of 1.8 MHz, 91 dB low frequency gain and the phase margin is 87 degree and 83 degree respectively. However, by steering the output transistors in triode region, to have a close to rail output, instability can occur. In order to prove that the circuit is stable in all possible situations, a transient analysis has been performed for dierent amplitudes at the output as shown in g.11. 2.0 1.5 1.0 500.0m VOUT 0.0 -500.0m -1.0 -1.5 -2.0 40.0u 80.0u 120.0u 160.0u 200.0u 60.0u 100.0u 140.0u 180.0u 220.0u T Fig. 11. Transient response VII. Conclusions A chopped amplier with a new class AB stage has been presented. It's principle is based on chopping and dynamic element matching to reduce 1/f noise and oset without excessive increase of the charge injection residual oset. The main goal of the circuit it is to achieve low noise and residual osets by chopping at high frequencies. It can drive low ohmic loads without stability problems with high gain and high linearity, meeting specications for high end audio applications. The circuit has been designed in a 0.5m digital CMOS process. A summary of performance based on simulations is presented below. AV 0Lj32 jj300pF 91 dB GBW 1:8 MHz THDj1k ;1kHz > 90 dB THDj32 ;1kHz > 83 dB VOS;STATIC < 1:67 mV VOS jfchop = 10MHz < 450 V VOS jfchop = 1kHz < 1 Vp VNOISE;THERMAL 37 nV/ Hz SUPPLY 3:3 V 10% VDD;MIN 1:4 V P 1:56 mW Technology 0:5m, 2PS,3AL, CMOS Harmonic distortion depends on the load and frequency. This has been presented in g. 12. At 1 kHz frequency, the THD can be lower than -90 dB for a References 1 k load and signal amplitudes close to 3 Vpp. For a 32 load, the THD is lower than -83 dB for sigSanduleanu, B. Nauta and H. Wallinga, "Lownal amplitudes close to 2pVpp. The noise and oset [1] M.A.T. power, low-voltage chopped transconductance amplier for simulations shows 37nV/ Hz input referred noise and noise and oset reduction", Proc. ESSCIRC'97, Southamp1.67mV static osets. Chopping at 10MHz, the residton, UK, Sept. 1997, pp. 204-207. ual oset is 450V but at 1KHz chopping, the residual [2] A. Bakker and J.H. Huijsing, "A CMOS chopper opamp with integrated low-pass lter", Proc. ESSCIRC'97, oset is lower than 1V. -40 -50 -60 )Bd( DHT 1 kΩ load 32 Ω load -70 -80 -90 -100 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 Output amplitude (Volt) Fig. 12. THD vs. Amplitude Southampton, UK, Sept.1997, pp. 200-203. [3] K.C. Hsieh, P.R.Gray, D.Senderowicz, D.G. Messerschmitt, "A low-noise chopper stabilised dierential switched capacitor ltering technique", IEEE J.Solid-State Circuits. vol. SC-16, no.6, Dec.1981, pp.708-714. [4] C. Menol and Q. Huang, "A low-noise CMOS instrumentation amplier for thermoelectric infrared detectors" IEEE J.Solid-State Circuits. vol. 32, no.7, July 1997, pp. 968-976. [5] I.E. Opris and G.T.A. Kovacs, "A rail to rail ping-pong opamp", IEEE J.Solid-State Circuits. vol. 31, no.9, Sept. 1996, pp.1320-1324. [6] W.C. Wu, W.J. Helms, J.A. Kuhn and B.E. Byrkett, "Digital-compatible high-performance operational amplier with rail to rail input and output ranges", IEEE J.SolidState Circuits. vol. 29,n0.1, Jan.1994, pp.63-66.