
TL/H/8499
A
L
National Semiconductor
Linear Brief 52
Robert A. Pease
December 1980
A Low-Noise Precision
Op Amp
It is well known that the voltage noise of an operational
amplifier can be decreased by increasing the emitter current
of the input stage. The signal-to-noise ratio will be improved
by the increase of bias, until the base current noise begins
to dominate. The optimum is found at:
I
e(optimum)
e
KT
q
h
FE
r
s
where r
s
is the output resistance of the signal source. For
example, in the circuit of Figure 1, when r
s
e
1 k
X
and
h
FE
e
500, the l
e
optimum is about 500
m
A or 560
m
A.
However, at this rich current level, the DC base current will
cause a significant voltage error in the base resistance, and
even after cancellation, the DC drift will be significantly big-
ger than when l
e
is smaller. In this example, l
b
e
1
m
A, so
l
b
c
r
s
e
1 mV. Even if the l
b
and r
s
are well matched at
each input, it is not reasonable to expect the l
b
c
r
s
to track
better than 5 or 10
m
V/
§
C versus temperature.
A new amplifier, shown in Figure 2, operates one transistor
pair at a rich current, for low noise, and a second pair at a
much leaner current, for low base current. Although this
looks like the familiar Darlington connection, capacitors are
added so that the noise will be very low, and the DC drift is
very good, too. In the example of Figure 2, Q2 runs at
l
e
e
500
m
A and has very low noise. Each half of Q1 is
operated at 11
m
A
e
l
e
. It will have a low base current
(20 nA to 40 nA typical), and the offset current of the com-
posite op amp, I
b1
–l
b2
, will be very small, 1 nA or 2 nA.
Thus, errors caused by bias current and offset current drift
vs. temperature can be quite small, less than 0.1
m
V/
§
C at r
s
e
1000
X
.
The noise of Q1A and Q1B would normally be quite signifi-
cant, about 6 nV/
Hz, but the 10
m
F capacitors completely
filter out the noise. At all frequencies above 10 Hz, Q2A and
Q2B act as the input transistors, while Q1A and Q1B merely
buffer the lowest frequency and DC signals.
this amplifier is predicted to be 1.4 nV/
Hz, which is quite
small compared to the Johnson noise of the 1 k
X
source,
For audio frequencies (20 Hz to 20 kHz) the voltage noise of
4.0 nV/
Hz. A noise figure of 0.7 dB is thus predicted, and
has been measured and confirmed. Note that for best DC
balance R6
e
976
X
is added into the feedback path, so
that the total impedance seen by the op amp at its negative
input is 1 k
X
. But the 976
X
is heavily bypassed, and the
total Johnson noise contributed by the feedback network is
below
(/2
nV/
Hz.
To achieve lowest drift, below 0.1
m
V/
§
C, R1 and R2
should, of course, be chosen to have good tracking tempco,
below 5 ppm/
§
C, and so should R3 and R4. When this is
done, the drift referred to input will be well below 0.5
m
V/
§
C,
and this has been confirmed, in the range
a
10
§
C to
a
50
§
C.
Overall, we have designed a low-noise op amp which can
rival the noise of the best audio amplifiers, and at the same
TL/H/8499–1
V
OUT
j
(n
a
1) V
IN
a
V
OS
c
(n
a
1)
a
(l
b2
b
l
b1
)
c
r
s
c
(n
a
1)
a
V
noise
c
(n
a
1)
a
i
noise
c
(r
s
a
R
IN
)
c
(n
a
1)
FIGURE 1. Conventional Low-Noise Operational Amplifier
C
1995 National Semiconductor Corporation
RRD-B30M115/Printed in U. S. A.