LINEARIZATION TECHNIQUES FOR PUSH-PULL AMPLIFIERS

Transcription

1 LINEARIZATION TECHNIQUES FOR PUSH-PULL AMPLIFIERS Rinaldo Castello Department of Electronics University of Pavia, Pavia Italy.

2 Outline of the presentation Motivation and relevant applications OL distortion and how to minimize it Output stage (push-pull) distortion Injected distortion independent of frequency response/topology CL distortion prop. to output conductance at distortion frequency Distortion vs. topology for same output stage/frequency response Input stage distortion Injected distortion depends on gain at the signal frequency Different topologies for low Input/output distortion Rinaldo Castello 1

3 Typical wireless transceivers Large number of standards Large number of bands Large number of receiver paths Large number of Components and Complex Board Rinaldo Castello 2

4 Saw-Less Transceiver Rinaldo Castello 3

5 Full Duplex Transceiver Rinaldo Castello 4

6 Baseband Requirements Baseband Stages Transmitter Side Receiver Side Rinaldo Castello 5

7 Linearity Signal Definitions At signal frequency RX Side At distortion frequency TX Side Dist. frequency Signal frequency Signal frequency Dist. frequency Rinaldo Castello 6

8 Overall Amplifiers Linearization Consider push-pull output stage Minimize open loop distortion not implicit in the push push operation Via topology optimization Minimize closed loop distortion Via topology and bandwidth optimization Rinaldo Castello 7

9 Push-Pull Output Stages Sources of OL distortion Output devices moving from linear to saturation region during signal period Crossover distortion Can be reduced e.g. preventing all transistors from shutting-off completely or turning-on on the wrong phase during transients Rinaldo Castello 8

10 Push-Pull Output Stages Cross Over Condition M1 V out I out Load V out M1 t M2 Cross-Over M1 V out Load V out M2 I out M2 t Rinaldo Castello 9

11 Push-Pull Output Stages Sources of open loop Distortion Shutting-off completely some devices caused by: circuit structure V DD V X I V DD V DD - V th M2 X M1 t V out V out V in M3 t ΔT (a) (b) Rinaldo Castello 10

12 Push-Pull Output Stages Causes of open loop distortion Shutting-off completely some devices caused by: capacive coupling V DD V X V DD C p V DD - V th M2 X M1 t C c V out V out V in M3 t ΔT (a) (b) Rinaldo Castello 11

13 Push-Pull Output Stages Eliminating Distortion Complete devices shut-off can be prevented by clamping the critical nodes: V DD V X V DD - V th I M2 X M1 t V bias M C C c V out V out V in M3 t (a) MC normally off goes on when X starts to rise (b) Rinaldo Castello 12

14 Push-Pull Output Stages Causes of distortion Turning-on some devices during the wrong phase V DD I I X M1 t C c I V out V out V in M3 M2 t (a) Vss (b) Rinaldo Castello 13

15 Push-Pull Output Stages Eliminating distortion V DD M3 V in + M1 Turning-on can be prevented by driving off the critical nodes with a large current M4 M2 V in - C c V out I Rinaldo Castello 14

16 Minimization of CL Distortion due to Push-Pull Output Stage Injected distortion depends only on output stage OL gain at signal frequency doesn t affect distortion Critical parameter for closed loop distortion output impedance at distortion frequency OL gain at distortion frequency affects distortion For given output stage and OL gain topology affects CL distortion Rinaldo Castello 15

17 Closed loop distortion for Miller vs. Nested and Multi-Nested Miller 1 Simple Miller V IN H(jω) A Cm2 gm2 V A Mout 3fIN V OUT Unity gain buffer configuration 2 Nested Miller 3 Double Nested Miller V IN V IN -1-1 Cm1 Cm0-1 Cm V A H(jω) A +- + V A H(jω) Cm2 gm2 Mout A 3fIN Cm2 gm2 V OUT Mout V OUT 3fIN Multiply open loop distortion by closed loop output impedance. Rinaldo Castello 16

18 Is topology affecting distortion? To verify if this is true make this experiment Chose the same output stage for all configurations Choose gain and bandwidth of the n stages such that the overall frequency response is the same independently of number of stages Open Loop Gain [db] 80 4s 3 s tage s 2 s tages tag es k 100k 1M 10M 100M 1G Frequency [Hz] Rinaldo Castello 17

19 Closed Loop Y OUT Y OUT = closed loop Output Admittance stages 3 stages -20 Yout [dbs] k 20dB/dec 40dB/dec 100k 1M 10M 100M 1G Frequency [Hz] 60dB/dec 4 stages Rinaldo Castello 18

20 Grounded-Out Gain (GOG) Compute output conductance injecting voltage and measuring current Critical parameter is GOG V IN Cm0 Cm1-1 Grounded Output Gain = A V A V A V IN Cm2 gm2 Mout Grounded-Out Gain [db] k 2 stages 20dB/dec 40dB/dec 100k 1M 10M 100M 1G Frequency [Hz] 3 stages 60dB/dec 4 stages Rinaldo Castello 19

21 Closed Loop Y OUT vs Grounded-Out Gain Yout [dbs] k 20dB/dec 2 stages 3 stages 4 stages 40dB/dec 60dB/dec 100k 1M 10M 100M 1G Frequency [Hz] G r o u n d e d - O u t G a i n [ d B ] k 20dB/dec 40dB/dec 60dB/dec 100k 1M 10M 100M 1G F r e q u e n c y [H z] Yout and NormalizedGOG [db] k 2 stages 20dB/dec 40dB/dec 100k 1M 10M 100M 1G Frequency [Hz] 3 stages 60dB/dec 4 stages Rinaldo Castello 20

22 Closed Loop Linearity HD3 = Third Harmonic Tone at 3f IN for an input tone at f IN Consistent with output conductance plot. Rinaldo Castello 21

23 Multi-stage topology Unity Gain Bandwidth Increasing number of stage generally forces to use smaller bandwidth for power consumption and stability reasons GOPG N=4 GOPG N=4 N=3 N=3 N=2 N=2 f f (a) b) is the typical situation and the distortion advantage is partially lost Rinaldo Castello (b) 22

24 Output Stage Distortion Conclusions Injected distortion depends only on output stage Critical parameter for closed loop distortion is GOG at frequency of distortion Shape and bandwidth of GOG are critical GOG for Multi-Miller has: 20( N -1) db/dec slope with N = # of stages same unity gain frequency as OL gain Rinaldo Castello 23

25 CL distortion due to input stage For a given input and output stage OL gain at signal freq. affects injected distortion OL gain at distortion frequency has no effect To minimize distortion from input stage Maximize open loop gain in front of the output stage at the frequency of the signal Via topology and bandwidth optimization Rinaldo Castello 24

26 OL response/gog effect on distortion OL bandwidth Affects CL distortion due to both input and output stage Shape of OL response/gog GOG at frequency of distortion affects output stage generated distortion OL gain at frequency of signal affects first stage distortion injection Rinaldo Castello 25

27 Extending GOG for same OL response Two Stage Amplifier cascode Miller compensated Miller capacitance returned to low impedance node GOG behavior is the same but bandwidth is larger than simple Miller by a factor CM/Cgs1 Rinaldo Castello 23

28 Extending OL bandwidth Three Stage nested Miller Amplifier with DC Feedforward C 1 V in A 1 A 2 A 3 C 2 V out Open Loop Gain [db] A f C 1 halved distorsion A f f (a) (b) Multipath nested Miller Can control very accurately the doublet separation, is only limited by matching Rinaldo Castello 27

29 Frequency Response Shaping Two Stage Amplifier with Pole-Zero in First Stage C 2 V in A 1 A 2 A 3 C 1 V out Open Loop Gain [db] A f -40dB/dec A f Signal BW f (a) (b) Slope of two in overall FR Rinaldo Castello 28

30 Frequency Response Shaping Miller amplifier with high pass feedback Slope of two in overall FR Rinaldo Castello 29

31 Frequency Response Shaping Three Stage Miller Amplifier with DC Feedforward C 2 V in A 1 A 2 A 3 C 1 V out Open Loop Gain [db] A f -40dB/dec A f Signal BW f (a) -40dB/dec (b) Slope of two in overall FR w Z = g M4 g M51 g M52 C Z w UGBW = g M52 C M3-20dB/dec Rinaldo Castello 30

32 Frequency Response Shaping Three Stage Amplifier with Pole-Zero in First Stage Slope of two in overall FR Top circuit equivalent to the one below but implemented with 1 less transconductor Rinaldo Castello 31

33 Frequency Response Shaping Three Stage Amplifier with 2 Pole-Zero Stages Slope of three in overall FR No Complex Zeros Rinaldo Castello 32

34 Bandwidth for Different Loads Push-pull operation makes output gain smaller than one at peak voltage swing Both effects can degrade stability forcing small bandwidth Returning the compensation capacitance to different nodes drastically changes behavior Rinaldo Castello 33

35 Miller with different (R, C) load Unity gain Bandwidth reduced for small resistive load Output pole enters in band for large capacitive load 80 typ RL low Magnitude [db] E+3 1E+4 1E+5 1E+6 Frequency [Hz] 1E+7 1E+8 Rinaldo Castello 1 P1 R2 ( C2 + gmp CA CL + CA) gmp gmp Z1 - P2 CA CL GAIN = gm2 gmp R2 RL GBW gm2 C2 + CA CA + gmp RL 34

36 Cascoded Miller with different (R,C) load Compensation out of signal path Bandwidth constant with load Second pole out of band for large capacitive load LHZ in frequency response 80 typ RL low Magnitude [db] E+3 1E+4 1E+5 1E+6 1E+7 1E+8 Frequency [Hz] GBW gm 2 C A Z1 gm a C A Rinaldo Castello 35

37 Three Stage Nested Miller topology Behavior for Varying Loads (R) Effect on stability of changing the input node of C2 Rinaldo Castello 36

38 Frequency Stability Three Stage inner Miller compensation typ RL low 1.5 Vout [V] E E E E E E-05 Time [s] Rinaldo Castello 37

39 Frequency Stability 3 stage inner cascode Miller compensation Vout [V] typ RL low E E E E E E-05 Time [s] Rinaldo Castello 38