Semiconductor companies are now introducing fully-differential amplifiers. These diff-amps are differential not only at their input but also at their output, doubling output range. These input and output circuits have closed paths not shared by other signals through a common ground node. This circuit (and hence signal) isolation improves signal integrity. By keeping both input and output circuits complete in themselves, ground is only important for dc analysis and range determination. So what do these diff-amps have to do with race cars?

State-of-the-art or "leading-edge" technology is often developed in adventurous settings. New automotive ideas, such as overhead cams and multi-port valving, are refined and characterized by applying them in demanding situations that have no significant effect on present company business. And that setting is the race track, where technical concepts compete for performance advantages. The winning ideas go on to become integrated into products, where they then help compete for market share.

Where is the "race track" setting in electronics? Although the electronics industry lacks a glamorous event like the Indy 500, exciting developments occur in several branches of the field. For instance, during WW II, the Radiation Laboratory at MIT turned out some excellent results in the development of radar. Electronikers can admire, even to this day, the set of "Rad Lab" volumes from that era, containing good, solid presentation of theory backed up by implemented electronics. An earlier example: when Vladimir Zworykin developed television at RCA, this was an impressive breakthrough entirely unlike incremental improvements of next-generation commodity products.

In the 1950s, the development of the laboratory-quality cathode-ray oscilloscope at Tektronix by Howard Vollum, Jack Murdock, Cliff Moulton, John Kobbe (to whom one Tek legend attributes the invention of the JK flip-flop), Bill Polits and other highly creative engineers led to a most desirable technical environment for the motivated designer. With over 70 % of the oscilloscope market - a market driven by technological advances - and with founders that were inventive engineers themselves, Tek was an engineering-driven company, an idea-advancing enterprise.

Although Tek and H-P are the outstanding examples of test and measurement (T&M) instrument companies, it is generally the case that high-performance T&M is the "race track" of the electronics industry. After all, one must be able to measure the behavior of circuits under development, and test equipment circuitry must be that much better to measure it. No wonder that in T&M equipment interesting circuits are found. And this leads to diff-amps.

Oscilloscopes have used fully-differential amplifiers for decades. (Why did it take so long for them to appear as commercial ICs?) The typical examples are found in vertical amplifiers. These scope subsystems amplify the probe voltage by precise gains before applying the amplified waveform to the vertical deflection plates of the CRT. And except for the first stage, which is driven by a ground-referenced probe, they are fully differential amplifiers, through and through. To demonstrate, let's look at part of the vertical amplifier of a Tek T935A 35 MHz scope – now obsolete, 1970s-vintage, and low-cost. The input buffer amplifier stage is shown below, scanned from the manual. (And by the way, the old Tek "instruction manuals" as they were called, contained schematics which were works of art, unparalleled by EDA CAD drawings of today - the price of progress!)

The very first stage consists of JFETs Q4222A and B. The probe waveform is input to the gate of Q4222B. With the other JFET below it, a ´1 buffer amplifier is formed, with near-zero offset voltage between input and output. This is accomplished by using matched JFETs, and using the lower one as a current source. Its gate is connected to the –8V supply, and whatever V_GS results due to the drop across R4225 (the 20 W resistor in its source) corresponds to a drain current that flows through the JFET above it. The JFETs are matched, and the upper JFET will then have the same V_GS. The corresponding 20 W resistor, R4224, lower-terminal voltage is consequently the same as the input gate voltage. Some current of the upper JFET is gained as base current of Q4232, but it is minor and the matching is quite good.

This amplifier drives a full-diff-amp at the second stage, consisting of Q4232 and Q4234. Only the upper BJT (Q4232) is driven by the waveform to be amplified, while the lower input - at the base of Q4234 - is ac grounded to the scope probe circuit ground, thus completing the return of the input circuit. Because vertical amplifiers (like all amplifiers) have input offset error, the otherwise unused input is used for offset-error adjustment, which in oscilloscope language is dc balance. The word balance is a hint that scope amplifiers are heavily differential and that the two sides of the amplifier must be made to operate with the same dc conditions.

The output of stage 2 is also differential. This stage is only an emitter-follower, with no voltage gain, but it is needed to present a high input impedance to the JFET buffer while driving stage 3 with a low impedance. In other words, it presents a voltage-source to the next stage. At its differential output, however, the input waveform is not yet differentially balanced because the emitter-followers have no gain interaction between them and no splitting of the input waveform between them occurs. Stage 2 is differential only in that it has 2 inputs and two outputs. With no voltage gain, the input difference voltage is the output difference voltage.

The succeeding three stages to the delay line are shown below, a continuation of the same amplifier.

Q4258 and Q4268 form a fully differential amplifier, with shared emitter resistance R4254, a 63.4 W, 1 % resistor. Resistors R4257 and R4267 go to the –8V supply and, being much larger than R4254, function as current sources to the BJT emitters.

The waveform at the base of the upper BJT is divided through the emitter circuit and shared (nearly) equally with the lower BJT so that at the load resistors, balanced waveforms appear, having equal magnitude and opposite polarity. If R4254, or R_E, were split into two series resistors of value R_E/2 each, then their midpoint would be a virtual ground for a balanced-input diff-amp. But for this stage, half the magnitude of the input waveform, which is only applied to the upper BJT, will appear instead.

The next stage (Q4274, Q4284) is the second half – the common-base stage – of a complementary cascode amplifier. It is fully differential, as is the final common-emitter stage (Q4276, Q4286).

To calculate the differential voltage gain of the complementary cascode stage, note that the emitter dynamic resistances of Q4274, Q4284, which shunt the resistors R4271, R4281 (both 825 W), are much smaller, so that most of the dynamic current from Q4258, Q4268 flows through Q4274, Q4284 to develop a voltage across R4273, R4283 (both 499 W). The purpose of these resistors is to provide emitter bias current to the common-base stage. The stage gain is determined largely by the collector load resistors and the emitter resistor R4254:

where upper and lower voltages are denoted by subscripts u and l. Their differences are the input and output differential voltages. Each BJT contributes to the total gain – hence the ´2 before the BJT gain in A_v. Because R_E (R4254) is so close to the value of the dynamic emitter resistance of the BJT, r_e, a better gain approximation adds 2×r_e to R_E in the denominator of the gain equation, where

at room temperature. Then A_v @  –12.9, with 3.72 mA of emitter current for each BJT. Neglected is the loading of the input impedance of the next stage on the load resistors. Do you suppose the amplifier designer was shooting for a gain of –10?

Fully differential monolithic amplifiers are now appearing, such as the ADI AD8138, to drive high-resolution ADCs and for other high-performance (high speed and precision) amplifier applications. Their predecessors can be found in typical oscilloscope circuitry over the last few decades.

Ó Dennis L. Feucht, 2001