Consoles 841inexorably toward the point where the total amplifier
and network phase shift reaches 180° at the inverting
input (that’s a full 360° total), and the circuit oscillates.
The frequency at which it oscillates is inversely relative
to the capacitance value. It isn’t unusual, with small
values, to find oscillations right at the edge of the
high-frequency sensitivity of an oscilloscope. Hanging a
long piece of wire on the amplifier output (especially
shielded cable with its high shield to inner capacitance)
is a surefire guarantee of instability for this very reason.
It has the added complication that there is a measure of
inductance there, too. It is conceivable that a long cable
might start to look like a mismatched tuned stub at a
frequency where the amplifier still has some gain,
creating a creditably good, stable RF generator.
What this extra resistance-capacitance output circuit
is in effect doing is to add dramatically to the transit
time of the amplifier where actually the termination
problem is creating far more delay than could possibly
exist within the device itself. That the cures for the two
ills are similar shouldn’t be a surprise. Fortunately, a
simple fix for this instability is to buffer away the load
from the output feedback termination with a small
resistor of typically 33– 150:. This usually does it, but
at the expense of head room loss due to the attenuation
from the buffer resistor against the load termination.
Provided the load is greater than about 2 k: which it
would really have to be in order to prevent getting close
to current drive saturation in the IC output stage, this
head room loss should be well less than 1 dB. A better
way is to buffer off with a small inductance, giving
increasing isolation with frequency; a phase-shifting
characteristic opposite to that of the (normally) capaci-
tive load provides a total termination that is phase
constant at the higher frequencies. At the lower audio
frequencies, of course, the inductive reactance is very
low, and the load sees the very low dynamic output
impedance of the amplifier. The buffering inductance
becomes virtually transparent.
Both of these techniques also provide a measure of
protection against the possibility of RF signals finding
their way into the amplifier by means of rectification in
the output stage or inverting input. Very often output
stages are more prone to RF field detection than inputs.
Some devices with a quite low output impedance
before applied feedback (i.e. those with unbuffered,
complementary emitter-follower output stages) are not
likely to be fazed as much by these effects (pun totally
intentional) but it is just as well to design in these
considerations habitually. Emergency replacement,
device upgrades or IC internal design changes can
evoke this problem unintentionally.25.7.9 Compensating Op-AmpsOp-amps generally have a couple of pins dedicated for
compensation, which can be taken as a less than subtle
message from the manufacturer that their product isn’t
stable under certain conditions of usage and needs
external kludging. Usually this is at low closed loop
gain where the bandwidth is at its most extreme. The
classic solution is to shrink the bandwidth of the ampli-
fier by slowing the amplifier down. Among other
things, this wrecks the slew rate that’s been hand-
somely paid for.
The most ordinary means of slowing down the
devices is to slug an internal gain stage, leaving the
other stages intact. On the bright side, if it is this internal
gain stage around which the external compensation
capacitor is hung that is tending toward instability, the
capacitor should cure it. Sadly, it rarely is that stage. If a
previous stage, say, the input differential amplifier, is
unstable, all the capacitor will do is slow up the ampli-
fier and reduce the slew rate to the extent that the oscil-
lation is no longer visible at the output. It does not cure
the instability. It’s still in there, hiding. Often the only
external manifestations are supposed dc offset voltages
that won’t go away and a poor-sounding amplifier.
There is a moral to this tale of compensation: don’t
use op-amps that require compensation if at all avoid-
able. Stability should be ensured by the circuit as a
whole, and if speed is to be preserved, the op-amp
should not be used below the gain at which it’s happily
stable. Compensation achieves stability by masking a
symptom and not by tackling the cause.
The previous precautions, in addition to the feedback
phase-leading capacitor, are now required circuit prac-
tice for using the newer, fast devices in many op-amp
configurations. It should be said here that because there
is no facility for implementing phase leading around theFigure 25-20. Output impedance as part of the feedback
loop.VinVoTL071 Output resistance
included in amplifier220 pF10 dB gain
setup1.8 k 73.9 k 7
Cable
capacitance
etc.128 7