AD8307
Rev. D | Page 11 of 24
Further analysis shows that right up to the point where the input to
the first cell is above the knee voltage, VOUT changes by (A 1)EK
for a ratio change of A in VIN. This can be expressed as a certain
fraction of a decade, which is simply log10(A). For example,
when A = 5, a transition in the piecewise linear output function
occurs at regular intervals of 0.7 decade (log10(A), or 14 dB
divided by 20 dB). This insight allows the user to immediately
write the volts per decade scaling parameter, which is also the
scaling voltage, VY, when using base 10 logarithms, as
()
)
(
log
1
10 A
E
A
V
in
Change
Decades
V
in
Change
Linear
V
K
IN
OUT
Y
=
(4)
Note that only two design parameters are involved in determining
VY, namely, the cell gain A and the knee voltage, EK, while N,
the number of stages, is unimportant in setting the slope of the
overall function. For A = 5 and EK = 100 mV, the slope would be
a rather awkward 572.3 mV per decade (28.6 mV/dB). A well
designed log amp has rational scaling parameters.
The intercept voltage can be determined by using two pairs of
transition points on the output function (consider
Figure 24).
The result is
())
1
/
1
(
+
=
A
N
K
X
A
E
V
(5)
For the case under consideration, using N = 6, calculate VZ =
4.28 μV. However, be careful about the interpretation of this
parameter, because it was earlier defined as the input voltage at
which the output passes through zero (see
Figure 21). Clearly,
in the absence of noise and offsets, the output of the amplifier
chain shown in
Figure 23 can be zero when, and only when,
VIN = 0. This anomaly is due to the finite gain of the cascaded
amplifier, which results in a failure to maintain the logarithmic
approximation below the lin-log transition (labeled 1 in
Figure 24).
Closer analysis shows that the voltage given by Equation 5
represents the extrapolated, rather than actual, intercept.
DEMODULATING LOG AMPS
Log amps based on a cascade of A/1 cells are useful in baseband
applications because they do not demodulate their input signal.
However, baseband and demodulating log amps alike can be
made using a different type of amplifier stage, called an A/0 cell.
Its function differs from that of the A/1 cell in that the gain
above the knee voltage EK falls to zero, as shown by the solid
line in
Figure 25. This is also known as the limiter function, and
a chain of N such cells are often used to generate hard-limited
output in recovering the signal in FM and PM modes.
01
08
2-
02
5
SLOPE = A
SLOPE = 0
OU
TP
U
T
AEK
0
EK
INPUT
A/0
tanh
Figure 25. A/0 Amplifier Functions (Ideal and Tanh)
Devices, Inc., communications products incorporating a logarith-
mic IF amplifier all use this technique. It becomes apparent that
the output of the last stage can no longer provide the logarithmic
output because this remains unchanged for all inputs above the
limiting threshold, which occurs at VIN = EK/AN1. Instead, the
logarithmic output is now generated by summing the outputs of
all the stages. The full analysis for this type of log amp is only
slightly more complicated than that of the previous case. It is
readily shown that, for practical purposes, the intercept voltage,
VX, is identical to that given in Equation 5, while the slope
voltage is
()
A
AE
V
K
Y
10
log
=
(6)
Preference for the A/0 style of log amp over one using A/1 cells
stems from several considerations. The first is that an A/0 cell
can be very simple. In the AD8307, it is based on a bipolar
transistor differential pair, having resistive loads, RL, and an
emitter current source, IE. This exhibits an equivalent knee
voltage of EK = 2 kT/q and a small signal gain of A = IERL/EK.
The large signal transfer function is the hyperbolic tangent
(see the dashed line in
Figure 25). This function is very precise,
and the deviation from an ideal A/0 form is not detrimental. In
fact, the rounded shoulders of the tanh function result in a
lower ripple in the logarithmic conformance than that obtained
using an ideal A/0 function.
An amplifier composed of these cells is entirely differential
in structure and can thus be rendered very insensitive to
disturbances on the supply lines and, with careful design, to
temperature variations. The output of each gain cell has an
associated transconductance (gm) cell that converts the differen-
tial output voltage of the cell to a pair of differential currents,
which are summed simply by connecting the outputs of all the
gm (detector) stages in parallel. The total current is then converted
back to a voltage by a transresistance stage to generate the
logarithmic output. This scheme is depicted in single-sided