Analog Devices AN-202 Application Notes

AN-202

A

a

One Technology Way • P.O. Box 9106 • Norwood, MA 02062-9106 • 781/329-4700 • World Wide Web Site: http://www.analog.com

APPLICATION NOTE

An IC Amplifier User’s Guide to Decoupling, Grounding,

and Making Things Go Right for a Change

By Paul Brokaw

“There once as a breathy baboon

Who always breathed down a bassoon,

For he said "It appears

that in billions of years

I shall certainly hit on a tune”

(Sir Arthur Eddington)

This quotation seemed a proper note with which to begin
on a subject that has made monkeys of most of us at one
time or another. The struggle to find a suitable configu­ration for system power, ground, and signal returns too
frequently degenerates into a frustrating glitch hunt.
While a strictly experimental approach can be used to
solve simple problems, a little forethought can often
prevent serious problems and provide a plan of attack if
some judicious tinkering is later required.

The subject is so fragmented that a completely general
treatment is too difficult for me to tackle. Therefore, I’d
like to state one general principle and then look a bit more
narrowly at the subject of decoupling and grounding as
it relates to integrated circuit amplifiers.

. . . Principle: Think—where the currents will flow.

I suppose this seems pretty obvious, but all of us tend to
think of the currents we’re interested in as flowing “out”
of some place and “through” some other place but often
neglect to worry how the current will find its way back to
its source. One tends to act as if all “ground” or “supply
voltage” points are equivalent and neglect (for as long
as possible) the fact that they are parts of a network of
conductors through which currents flow and develop
finite voltages.

In order to do some advance planning it is important to
consider where the currents originate and to where they
will return and to determine the effects of the resulting
voltage drops. This, in turn, requires some minimum
amount of understanding of what goes on inside the cir­cuits being decoupled and grounded. This information
may be lacking or difficult to interpret when integrated
circuits are part of the design.

Operational amplifiers are one of the most widely used
linear lCs, and fortunately most of them fall into a few

classes, so far as the problems of power and grounding
are concerned. Although the configuration of a system
may pose formidable problems of decoupling and signal
returns, some basic methods to handle many of these
problems can be developed from a look at op amps.

OP AMPS HAVE FOUR TERMINALS

A casual look through almost any operational amplifier
text might leave the reader with the impression that an
ideal op amp has three terminals: a pair of differential
inputs and an output as shown in Figure 1. A quick review
of fundamentals, however, shows that this cannot be
the case. If the amplifier has an output voltage it must be
measured with respect to some point . . . a point to
which the amplifier has a reference. Since the ideal op
amp has infinite common-mode rejection, the inputs are
ruled out as that reference so that there must be a fourth
amplifier terminal. Another way of looking at it is that if
the amplifier is to supply output current to a load, that
current must get into the amplifier somewhere. Ideally,
no input current flows, so again the conclusion is that a
fourth terminal is required.

Figure 1. Conventional "Three Terminal" Op Amp

A common practice is to say, or indicate in a diagram,
that this fourth terminal is “ground.” Well, without get­ting into a discussion of what “ground” may be, we can
observe that most integrated circuit op amps (and a lot
of the modular ones as well) do not have a “ground”
terminal. With these circuits the fourth terminal is one or
both of the power supply terminals. There is a tempta­tion here to lump together both supply voltages with the
ubiquitous ground. And, to the extent that the supply
lines really do present a low impedance at all frequencies
within the amplifier bandwidth, this is probably reason­able. When the impedance requirement is not satisfied,
however, the door is left open to a variety of problems
including noise, poor transient response, and oscillation.

REV. B

AN-202

DIFFERENTIAL-TO-SINGLE-ENDED CONVERSION

One fundamental requirement of a simple op amp is that an
applied signal that is fully differential at the input must be
converted to a single-ended output. Single-ended, that is,
with respect to the often neglected fourth terminal. To see
how this can lead to difficulties, take a look at Figure 2.

V+

–IN
+IN

CURRENT

MIRROR

OUTPUT

V–

Figure 2. Simplified “Real” Op Amp

The signal flow illustrated by Figure 2 is used in several
popular integrated circuit families. Details vary, but the
basic signal path is the same as the 101, 741, 748, 777,
4136, 503, 515, and other integrated circuit amplifiers. The
circuit first transforms a differential input voltage into a
differential current. This input stage function is represented
by PNP transistors in Figure 2. The current is then con­verted from differential to single-ended form by a current
mirror that is connected to the negative supply rail. The
output from the current mirror drives a voltage amplifier
and power output stage that is connected as an integrator.
The integrator controls the open-loop frequency response,
and its capacitor may be added externally, as in the 101, or
may be self-contained, as in the 741. Most descriptions of
this simplified model do not emphasize that the integra­tor has, of course, a differential input. It is biased positive
by a couple of base emitter voltages, but the noninverting
integrator input is referred to the negative supply.

It should be apparent that most of the voltage difference
between the amplifier output and the negative supply
appears across the compensation capacitor. If the negative
supply voltage is changed abruptly, the integrator ampli­fier will

force

the output to follow the change. When the
entire amplifier is in a closed-loop configuration the
resulting error signal at its input will tend to restore the
output, but the recovery will be limited by the slew rate
of the amplifier. As a result, an amplifier of this type may
have outstanding low frequency power supply rejection,
but the negative supply rejection is fundamentally limited
at high frequencies. Since it is the feedback signal to the
input that causes the output to be restored, the negative
supply rejection will approach zero for signals at frequen­cies above the closed-loop bandwidth. This means that
high-speed, high-level circuits can “talk to” low-level
circuits through the common impedance of the negative
supply line.

Note that the problem with these amplifiers is associated
with the negative supply terminal. Positive supply rejection
may also deteriorate with increasing frequency, but the
effect is less severe. Typically, small transients on the

positive supply have only a minor effect on the signal
output. The difference between these sensitivities can
result in an apparent asymmetry in the amplifier transient
response. If the amplifier is driven to produce a positive
voltage swing across its rated load, it will draw a current
pulse from the positive supply. The pulse may result in a
supply voltage transient, but the positive supply rejection
will minimize the effect on the amplifier output signal. In
the opposite case, a negative output signal will extract a
current from the negative supply. If this pulse results in a
“glitch” on the bus, the poor negative supply rejection will
result in a similar “glitch” at the amplifier output. While a
positive pulse test may give the amplifier transient response,
a negative pulse test may actually give you a pretty good
look at your negative supply line transient response, instead
of the amplifier response!

Remember that the impulse response of the power supply
itself is not what is likely to appear at the amplifier.
Thirty or forty centimeters of wire can act like a high Q
inductor to add a high-frequency component to the normally
overdamped supply response. A decoupling capacitor
near the amplifier won’t always cure the problem either,
since the supply must be decoupled to somewhere. If the
decoupled current flows through a long path, it can still
produce an undesirable glitch.

Figure 3 illustrates three possible configurations for nega­tive supply decoupling. In 3a, the dotted line shows the
negative signal current path through the decoupling and
along the ground line. If the load “ground” and decoupled
“ground” actually join at the power supply, the “glitch”
on the ground lines is similar to the “glitch” on the nega­tive supply bus. Depending upon how the feedback and
signal sources are “grounded,” the effective disturbance
caused by the decoupling capacitor may be larger than the
disturbance it was intended to prevent. Figure 3b shows
how the decoupling capacitor can be used to minimize dis­turbance of V– and ground buses. The high-frequency
component of the load current is confined to a loop that
does not include any part of the ground path. If the ca­pacitor is of sufficient size and quality, it will minimize the
glitch on the negative supply without disturbing input or
output signal paths. When the load situation is more com­plex, as in 3c, a little more thought is required. If the ampli­fier is driving a load that goes to a virtual ground, the actual
load current does not return to ground. Rather, it must be
supplied by the amplifier creating the virtual ground as
shown in the figure. In this case, decoupling the negative
supply of the first amplifier to the positive supply of the
second amplifier closes the fast signal current loop with­out disturbing ground or signal paths. Of course, it is
still important to provide a low impedance path from
“ground” to V– for the second amplifier to avoid disturb­ing the input reference.

The key to understanding decoupling circuits is to note
where the actual load and signal currents will flow. The
key to optimizing the circuit is to bypass these currents

–2–

REV. B

AN-202

LOAD

–V

LOAD LOAD

CURRENT

MIRROR

V+

V–

–
+

V–

IMPEDANCE

around ground and other signal paths. Note, that as in
Figure 3a, “single point grounding” may be an oversim­plified solution to a complex problem.

PNP

OUTPUT

TRANSITOR

POWER

GROUND

V–

LOAD

LOAD GROUND

SIGNAL

CURRENT

LOOP

POWER
SUPPLY
TERMINAL

Figure 3a. Decoupling for Negative Supply Ineffective

PNP

OUTPUT

TRANSITOR

CIRCUIT

COMMON

V–

DECOUPLING

CAPACITOR

LOAD

SIGNAL

CURRENT

LOOP

Figure 3b. Decoupling Negative Supply Optimized for
“Grounded” Load

V+

HIGH

PNP

OUTPUT

TRANSITOR

FREQUENCY

SIGNAL

CURRENT

PATH

jingle to a small damped signal at the op amp supply termi­nal. The residual has larger

low

-frequency components,

but these can be handled by the op amp supply rejection.

Figure 4. Damping Parallel Decoupling Resonances

FREQUENCY STABILITY

There is a temptation to forget about decoupling the nega­tive supply when the system is intended to handle only
low-frequency signals. Granted that decoupling may not
be required to handle low-frequency signals, it may still
be required for frequency stability of the op amps.

Figure 5 is a more detailed version of Figure 2, showing the
output stage of the lC separated from the integrator (since
this is the usual arrangement) and showing the negative
power supply and wiring impedance lumped together as a
single constant. The amplifier is connected as a unity gain
follower. This makes a closed-loop path from the amplifier
output through the differential input to the integrator input.
There is a second feedback path from the collector of the
output PNP transistor back to the other integrator input.
The net input to the integrator is the difference of the
signals through these two paths. At low frequencies this
is a net, negative feedback. The high-frequency feedback
depends upon both the load reactance and the reactance
of the V– supply.

V–

V–

NPN OUTPUT
TRANSITOR

Figure 3c. Decoupling Negative Supply Optimized for
Virtual Ground” Load

Figures 3b and 3c have been simplified for illustrative
purposes. When an entire circuit is considered, conflicts
frequently arise. For example, several amplifiers may be
powered from the same supply, and an individual de­coupling capacitor is required for each. In a gross sense
the decoupling capacitors are all paralleled. In fact, how­ever, the inductance of the interconnecting power and
ground lines convert this harmless-looking arrange­ment into a complex L-C network that often rings like the
“Avon Lady.” In circuits handling fast signal wavefronts,
decoupling networks paralleled by more than a few centi­meters of wire generally mean trouble. Figure 4 shows
how small resistors can be added to lower the Q of the
undesired resonant circuits. The resistors can generally
be tolerated since they convert a bad high-frequency

REV. B

–3–

Figure 5. Instability Can Result from Neglecting
Decoupling

When the supply lead reactance is inductive, it tends to
destabilize the integrator. This situation is aggravated
by a capacitive load on the amplifier. Although it is difficult
to predict under exactly what circumstances the circuit will
become unstable, it is generally wise to decouple the nega­tive supply if there is any substantial lead inductance in the
V– lead or in the common return to the load and amplifier
input signal source. If the decoupling is to be effective, of
course, it must be with respect to the actual signal returns,
rather than to some vague “ground” connection.

POSITIVE SUPPLY DECOUPLING

Up to this point we have not considered decoupling the posi­tive supply line, and with amplifiers typified by Figures 2