You are on page 1of 8

Analog Integrated Circuits and Signal Processing, 21, 109±116 (1999)

# 1999 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands.

Class AB Output Stages of the Integrated Ampli®ers

VADIM IVANOV
Burr-Brown Corp., Tucson, AZ

Received November 12, 1997; Accepted April 24, 1998

Abstract. General approach to the design of the class AB output stages of bipolar and CMOS ampli®ers based on
common-mode negative feedback is discussed. Frequency compensation of the result multiloop system is provided
by feed forward links. Effectiveness of the proposed method is shown on designs of rail-to-rail bipolar and CMOS
output stages, all-NPN output stage, output stage of the hearing aids IC. Common design mistakes are shown using
industry-standard ICs.

Key Words: operational ampli®er, class AB, frequency compensation, minimum selector

1. Introduction shown in Fig. 2: common-collector (drain), common-


emitter (source) and mixed (all-NPN or N-channel).
One of the most important parts of the integrated Consequently, it has two controlled variables: output
operational or power ampli®er is the output stage. voltage Vout , de®ned by the input signal, and quiescent
Output stage should provide good linearity and high- current Iq ˆ min…I1; I2†, de®ned by internal feed-
frequency behavior combined with high power backs and structure of the stage. The goal of the class
ef®ciency and low current consumption. Good AB stabilization is the preserving of good linearity
linearity means stable gain of the output stage in and speed combined with power ef®ciency by keeping
spite of output voltage (current). Ideal and typical Iq within certain limits.
dependencies of the input-referred error of the Typical dependence of the Iq from output current is
ampli®er from the output signal is shown in Fig. 1. shown in Fig. 3. Because transconductance of bipolar
Any sharp change of the error curve means drop of and, in ®rst approximation, of CMOS transistors
the ampli®er open-loop gain at this point and de®ned by its collector (drain) current, value of Iq
nonlinear distortions, especially non-desirable for with zero output current de®nes lowest gain and
the audio applications. Difference in error curve capacitive load capability of the stage (introduced by
angle for sinking and sourcing of the output current the capacitive load time constant is T ˆ CL =gm [1]).
mean second harmonic in signal distortions. Iqmin is not so important and should be kept non-zero
For low frequencies those errors can be negligible mainly to prevent saturation or complete shut off of
due to the high enough open-loop gain and small the stage transistors.
absolute value of the error; for frequencies close to the Any class AB circuit in every operating point can
unity-gain bandwidth of the ampli®er these non- be described by the graph in Fig. 4 [2], as shown on
linearities become more important. As the result, in examples in part 3 of the paper. General properties of
distortion-sensitive applications like audio ampli®ers this differential structure with common-mode feed-
bandwidth exceeds signal bandwidth in 1000 and back are described in part 2 (full and mathematically
more times which subsequently leads to excess power correct analysis is done in [2]); knowledge of these
consumption, complicated technology, etc. properties clears drawbacks of the existing decisions
Output stage of OpAmp consists of two power ( part 3) and helps to generate a set of new circuits for
transistors; main topologies of the output stages are the output stages ( part 4). Any feedback loop has a

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
110 V. Ivanov

Fig. 1. Input error versus output voltage.


Fig. 3. Iq …Vout †.
potential stability problem; recommendations on the
compensation of these local feedback loops are in part This represents most general caseÐin existing
5 of the paper. circuits part of these links can be absent.
Output current of the stage is equal

I1 ÿ I2 ˆ …A1B1 ‡ A2B2†Vin ÿ YD…B1E1 ÿ B2E2†


…1†
2. Differential Structure with Common-Mode
Feedback where Y is generally the complex non-linear
minimum-like function of I1, I2.
Output stage can be abstracted to differential structure It can be seen, that conditions for ideally linear stage
with common-mode feedback as shown in Fig. 4 [2], are B1E1 ˆ B2E2 and …A1B1 ‡A2B2† ˆ const.
where X1, X2Ðcurrent (or more rarely voltage) on It is relatively easy to achieve E1 ˆ E2 ˆ E and
the inputs of each of output transistors, B1, B2Ðgain A1 ˆ A2 ˆ A. Gain B1, B2 of output transistors is
of output transistors, F1, F2Ðcurrent-dependent usually proportional to their transconductance,
transfer coef®cients of the current measurement/ another words, to the current for bipolars or to
minimum selector circuit, D and E1, E2Ðgain in sqrt(current) for CMOS (without use of the local
class AB stabilization loops. Gain B1, B2 is usually feedbacks for gain control). In area around zero output
proportional to the transconductance of the output current of both transistor currents are close to the Iq ,
transistors Q1, Q2. and it is possible to get B1 ˆ B2 ˆ B and stage gain is
equal to …AB1 ‡ B2† ˆ AB.

Fig. 2. Topologies of output stage. Fig. 4. General signal graph of the output stage.

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
Class AB Output Stages of the Integrated Ampli®ers 111

For the big output currents minimum-selector


circuit breaks one of the links F1, F2, for example if
I1 4 I2 F1 ˆ 0, F2 ˆ 1, so Y ˆ 12. Current 12 is
keeping equal to Iq with possible error less than
Vin A/DE. If B2F2DE 1 (otherwise loop gain is not
suf®cient and class AB operation can not be kept), and
gain of the ampli®er is:
Iout *I1 ˆ 2B1Vin A …2†
If I1 5 I2
Iout ˆ 2B2Vin A
Fig. 5. Graph of the non-symmetrical output stage.
Summarizing:
± any output stage can be generalized to the structure achieve superior linearity. Typical dependence of the
in Fig. 4; input error versus output voltage is shown in Fig. 7.
± ideally linear stage needs matching Second drawback of such stages is that long gain
chain when sourcing current leads to additional
E1 ˆ E2; delays, complicated compensation and loss of the
A1 ˆ A2 ˆ const ampli®er speed.
most difficult B1 ˆ B2 ˆ const To linearize such stages, the non-linear local feed-
back can be used as in OP296 [5] (Fig. 8), as shown in
± at any operating point the loop gain of at least one structure in Fig. 5 by dashed line. Non-linearity of this
of the feedback loops F-D-E-B should be 4 1 to
keep class AB operation.
Structures of the existing class AB circuits can lack
some of the links of Fig. 4 and mentioned above
conditions could be not satis®ed; it means various
drawbacks, as shown in the next part. Designer's
decision is which drawbacks are acceptable.
Understanding of properties of the basic structure
can help to generate may be complicated, but more
accurate and fast circuits.

3. Structures of Class AB Stages and Their


Properties

First variation of basic structure is shown in Fig. 5 and


describes the widely used stage shown in Fig. 6; it can
also characterize output stages of MC33502 [3] and
many other industry circuit decisions for OpAmps and
power ampli®ers.
It differs from Fig. 4 by the removal of E2 and A1
links. Links F1 and F2 are non-linear, strong for small
currents and weakening with the rise of current.
Structure is inherently non-symmetrical and has
different signal gain for sinking and sourcing of
current. As a result, it is not possible in principle to Fig. 6. All-NPN stage.

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
112 V. Ivanov

Fig. 7. Input error vs. Vout .

local feedback should be matched with F2. It improves


to some extension symmetry of error curve by
decreasing of the gain for sourcing currents (one non-
linearity compensates another non-linearity), but does
not help to ®x speed problems. Fig. 9. Output stage with local feedbacks.
Another way to provide class AB operation is the
non-linear local feedbacks which keep minimum point close to the Vout ˆ 0 or with small output current
current of the each output transistor stable regardless both local feedbacks are active and open-loop gain of
of input signal, as was ®rst suggested by D. Monticelli ampli®er drops down (Fig. 11), causing zero-cross
(Fig. 9) [4]. Modi®cations of this circuit are used in distortions of the signal.
numerous OpAmps, like LMC604x [5], LM646x, etc.
Structure of those stages is shown in Fig. 10.
Both gain links are independent and each one has
non-linear local feedback keeping minimum current 4. Class AB Stages with Common-Mode
equal to Iq . As soon as this local feedback becomes Feedback
active, input signal does not change current of this
transistor any more. Historically ®rst and still most popular class AB
It is easy to achieve equal and big gain when output stage (Fig. 12) is consistent with basic structure
sourcing and sinking current. However, in transition in Fig. 4. It's error versus output voltage is shown in
Fig. 13. Stage gain rises with output current due to
increase of transconductance of output transistors
with current (2); gain in zero point and zero-cross
distortions are de®ned by the transconductance when
output transistors are biased by Iq. As shown in Fig.
13, error curve gets smoother and gain variation
decreases with increase of Iq . For the same reason

Fig. 10. Graph of the output stage with independent local


Fig. 8. Use of the local feedback to linearize stage. feedbacks.

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
Class AB Output Stages of the Integrated Ampli®ers 113

Fig. 11. Error vs Vout.

similar error curve have all stages based on structure


of Fig. 4 as far as gain of both output transistors (B1,
B2 links in Fig. 4) is not stabilized by the local
feedbacks. Fig. 13. Error vs Vout.
Various realizations of the structure of Fig. 4 for
rail-to-rail bipolar and CMOS ampli®ers were voltage drop on the resistor in the emitter (source) of
invented by Prof. J. Huijsing and his group [7±10]. the transistor connected in parallel (Q20 in Fig. 14),
For the straight forward realization of structure of which is equal to:
Fig. 4 following circuit blocks are necessary:
1. Sensors of the current of output transistors. Those I IR
sensors should be linear for small currents close to VR ˆ Vt ln & if VR  Vt
NI N
the Iq ; linearity in the big current area is not
important. Some variants of current measurement
are shown in Fig. 14. For measurements can be NÐratio of emitter sizes.
used: 2. Circuit of minimum-like function of I1, I2. In
base-emitter (gate-source) voltage: circuit in Fig. 12 function I1 I2 is used. In Fig. 15
few simplest minimum selectors are shown: diode
r minimum selector, diode minimum selector pro-
I L
Vbe ˆ Vt ln e or Vgs ˆ Vt ‡ k ID viding voltage gain and CMOS minimum selector
Is W [11].
3. Two feedback ampli®ers (for each of output
transistors). Gain of differential stage is usually
suf®cient to provide the effective class AB
stabilization feedback.
Examples of output stages designed by strait

Fig. 12. Common-collector output stage. Fig. 14. Current sensing.

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
114 V. Ivanov

Fig. 15. Minimum-selection circuits.

realization of structure of Fig. 4 and using described


above blocks are shown in Figs. 16±19.
In Fig. 16Ðrail-to-rail output stage of the bipolar
OpAmp [12]. Sum of the base-emitter voltages and Fig. 17. Rail-to-rail CMOS output stage.
consequently product of currents of output transistors
is stabilized here. Emitters of transistors of A1, A2 1-V class AB output stage for the hearing aids
differential stages are connected, ensuring identity of ampli®er with differential output (usually these amps
the A1,A2 gain. Emitters of E1, E2 ampli®ers are also are loaded on transformer) is shown in Fig. 19.
connected for the same reason. These are few examples of the variety of output
CMOS rail-to-rail output stage with minimum stages which can be designed using described
selector circuit is shown in Fig. 17. Identity of A1, A2 approach.
and E1, E2 gain can be achieved by appropriate
choice of the transistor's W/L and/or IA , IE currents.
Output stage on Fig. 18 with all NPN output 5. Frequency Compensation of the Class AB
transistors is not rail-to-rail, but still viable for high- Local Feedback Loops
frequency and high-current operational and power
ampli®ers due to better speed and current capabilities All class AB stages which are build using structure of
of NPN transistors for most processes. Fig. 4 contain at least 2 feedback loops. For system
For simplicity, ampli®ers A1, A2 replaced by the stability both loops should be frequency compensated
single transistor Q3; it leads to the slight non- at all operating points.
symmetry of the stage error. Among variety of compensation methods most
More symmetrical stage with the same structure for
high frequency applications is described in [14].

Fig. 16. Bipolar rail-to-rail output stage. Fig. 18. All-NPN class AB stage.

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
Class AB Output Stages of the Integrated Ampli®ers 115

7. Conclusion

General method of the class AB circuits design based


on general differential structure with common-mode
feedback has been proposed.
Effectiveness of the method is shown on few non-
trivial examples of the variety of class AB stages
which can be generated using described approach.
The same methodology can be used for the design not
only of half-bridge or differential stages, but also for
bridge, 3-phase and other stages with unlimited
quantity of outputs and internal variables, for class
AB input stages, etc.
Method also empowers designer to estimate draw-
Fig. 19. Differential output stage.
backs of the existing decisions without their detailed
analysis and, most important, indicates means for the
circuit improvement.

useful for this case is the compensation by the low- References


gain but fast feedforward links [13,12]. Circuit
diagram of the stage in Fig. 16 with addition of feed 1. E. Vittoz, ``Low-power low-voltage limitations and prospects
in analog design.'' Analog circuit design. Low-power Low-
forward link for compensation of the high-side loop is
voltage Integrated ®lters and Smart power. Kluwer Boston,
shown in Fig. 20 (low-side loop of this stage does not MA 1994.
need compensation). Function of forward link 2. V. N. Ivanov and V. V. Ivanov, ``
provided if I2 > I1 due to the fact that speed and '' (Power integrated ampli®ers). Rumb: St.-
transconductance of the bipolar transistors are Petersburg, 1987 (in Russian).
3. R. Grif®th et al., ``A 1V BiCMOS Rail-to-rail ampli®er with n-
proportional to the collector current.
channel depletion mode input stage.'' Digest ISSCC'97
Similar compensation needs stage in Fig. 18. pp. 352±353.
4. D. Monticelli, US pat. 4,570,128 ``Class AB output stage with
large swing.''
5. F. Goodenough, ``Rail-to-rail op amps use depletion-mode
PMOSFETs.''//ED, p. 127, 1992.
6. J. Butler, US pat. 5,521,553 ``Method for controlling an output
stage of a bipolar micropower rail-to-rail ampli®er.''
7. W. C. M. Renirie, K. J. de Langen, and J. H. Huijsing, ``Parallel
Feedforward Class-AB Control Circuits for Low-voltage
Bipolar Rail-to-Rail Output Stages of Operational
Ampli®ers.'' Analog ICs and Signal Processing 8, 37±48,
1995.
8. J. Fonderie and J. Huijsing et al., ``Operational ampli®er with
rail-to-rail input and output ranges.'' IEEE J. Solid-State
Circuits SC-24, pp. 1551±1559.
9. J. Fonderie and J. Huijsing, Design of low-voltage bipolar
operational ampli®ers. Kluwer, Boston MA, 1993.
10. J. H. Huijsing and D. Linebarger, ``Low-voltage operational
ampli®er with rail-to-rail input and output ranges.'' IEEE J.
Solid-State Circuits SC-20, pp. 1144±1150, 1985.
11. E. Seevinc, ``CMOS translinear circuit.'' Proc. of Workshop
Advances in Analog Circuit Design. Lausanne, 1996.
12. V. Ivanov, ``A micropower precision single-supply instrumen-
Fig. 20. Feed forward link for compensation of the high-side tation ampli®er.'' Digest ISSCC-97 pp. 354±355.
class AB loop. 13. J. Huijsing et al., ``Low-voltage low-power OpAmp based

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset
116 V. Ivanov

ampli®ers.'' In: Analog integrated circuits and signal proces- From 1980 till 1991 he was design engineer and
sing. 8, pp. 49±6, Kluwer, Boston MA. researcher in ``Electropribor'' Research Institute, St.
14. K. de Langen et al., ``A 1-GHz bipolar class AB operational
ampli®er with multipath nested Miller compensation for 76 dB Petersburg, Russia, working on analog electronics and
gain.'' IEEE J. of Solid-State Circuits 32, (4), 1997. ASICs in power, sensor and RF areas with OES
Electronics, St. Petersburg, Russia. From 1991 to
1995 he designed various ASICs in power, sensor and
RF areas with OES Electronics, St. Petersburg, Russia.
In 1996 he joined Burr-Brown Corp., Tucson, AZ, as a
design engineer.
He has co-authored two books on analog IC design
and holds 3 U.S. and more than 40 USSR author
certi®cates and Russian patents on inventions in
analog circuit techniques. His interests include
utilization of the control theory methods for the
analog IC design.

Vadim Ivanov received his degree in electrical


engineering in 1980 from St. Petersburg Electrical
Engineering University, Russia.

M9331 Kluwer Academic Publishers Analog Integrated Circuits and Signal Processing (ALOG) Tradespools Ltd., Frome, Somerset

You might also like