Page 113 - The Art and Science of Analog Circuit Design

One Trip Down the 1C Development Road

also generate serious noise problems and cause the tail pole to move in,
reducing the speed of the amplifier. It was time to retreat to the current
feedback approach and see how good I could make it.
The current feedback topology has very low feedthrough when oper-
ated inverting, so I started with that approach. Unfortunately the feed-
through was not as good as I expected and I started looking for the cause.
The source of feedthrough was found to be the emitter-base capacitance
of the current steering transistor coupling signal into the pre-distortion
diode that was holding the transistor off. Unfortunately the off diode was
high impedance (no current in it) so the signal then coupled through the
collector base capacitance of the steering transistor into the collector,
where it was not supposed to be. Since the steering transistors had to be
large for low R b and R e, the only way to eliminate this problem was to
lower the impedance at the bases of the steering transistors.
What I needed was four buffer amplifiers between each of the four
pre-distortion diodes and the current steering transistors. To preserve the
pre-distortion diodes' accuracy, the input bias current of the buffers
needed to be less than one microamp. The offset of the buffers had to
be less than a diode drop in order to preserve the input stage common
mode range so that the circuit would work on a single 5V supply. Lastly,
the output impedance should be as low as possible to minimize the
feedthrough.
The first buffer I tried was a cascode of two emitter followers, as
shown on the left in Figure 8-11. By varying the currents in the followers
and looking at the overall circuit feedthrough, I determined that the out-
put impedance of the buffers needed to be less than 75O for an accept-
able feedthrough performance of 60dB at 5MHz. I then tried several
closed loop buffers to see if I could lower the supply current. The circuit
shown in Figure 8-11 did the job and saved about 200 microamps of
supply current per buffer. The closed loop buffer has an output imped-
ance of about TO, that rises to 65Q at 5MHz. Since four buffers were
required, the supply current reduction of 800 microamps was significant.
At this point it became obvious to me that for the feedthrough to be
down 60dB or more, the control circuitry had to be very accurate. If the
full scale voltage was 2.5V and the control voltage was 0V, the offset
errors had to be less than 2.5mV for 60dB of off isolation. Even if 1
trimmed the 1C to zero offset, the system accuracy requirement was still
very tough. I therefore wanted to come up with a circuit that would in-
sure that the correct input was on and the other input was fully off when
the control was close to zero or full scale. I thought about adding inten-
tional offset voltage and/or gain errors to the V-to-I converters to get this
result, but it didn't feel good. What was needed was an internal circuit
that would sense when the control was below 5% or above 95% and force
the pre-distortion diodes to 0% and 100%. Since the diodes were fed
with currents, it seemed that sensing current was the way to go,
Since the currents that feed the pre-distortion diodes corne from iden-
tical mirrors, I wanted to see if I could modify the mirrors so that they

96