The Impossible Power Amplifier

or what shall we do with these transistor modules

The Incident

Once upon a time, I discovered a box of these transistor modules:

(click on image for larger view)

Their datasheet showed some impressive ratings:

Collector current: 150 A

Breakdown voltage: 1000 V

Power dissipation: 1000 W

As they are relicts from ancient times of power electronics, those aren't IGBT modules but BJT. One module containing a half bridge made of two transistors and two free wheeling diodes. Three of the modules make up the output stage of a typical variable frequency drive or power inverter. To achieve the high current rating, the modules use a three transistor darlington configuration. Even then, according to the datasheet, there's quite an amount of base drive required. Regarding the target application, the transistors are intended to be used in switching mode.

The Idea

Although these transistors are clearly made for high power switching applications, would it be possible to operate them in the linear mode? I found another datasheet, indicating the SOA for the transistors: There's no SOA limit but the maximum power dissipation. Quite impressive for a BJT. There's also a chart showing the Hfe over collector current, indicating a range from 500 to 1000 for a collector current in the low range. Looks like a decent starting point.

Found some other useful stuff:

A large heatsink with fans attached

Power supplies rated 750W, 59V:

Vintage components like these

and many others

So what? We've got a 1.5kW +/- 59V power supply, power transistors, and a suitable heatsink?
Let's build a power amplifier!

It's a long way round

What looks quite straightforward at first thought brings a lot of trouble and pitfalls. Especially if you try to use these power modules to build a linear amplifier:

The transistors are triple darlingtons for high collector currents - which means there's quite an amount of base charge stored inside, among all the other trouble associated with darlington stages

Both legs are NPN, so one cannot build the output stage as a simple complementary emitter follower configuration

The upper transistors emitter and the lower transistors collector terminals are connected internally, so there's no way to insert a resistor into the emitter of the upper transistor

Stop oscillating!

As you should know, any kind of emitter follower backs up as an oscillator circuit if you don't have a suitable base resistor installed. With these modules, things get even worse: It is very hard to operate them in linear mode without causing parasitic oscillations. A simple base resistor fixes the oscillations, well - sometimes it does, sometimes not. I managed to operate the upper transistor as a voltage follower (including a simple OP-AMP / small signal transistor circuit to compensate for the base voltage drop) into a resistive load. Then I started to add the lower transistor to the circuit in various configurations, starting from a simple constant current sink circuit to some kind of push-pull configuration driven by photocouplers. At some extend, testing different load configurations, I always ended up with either a nice oscillator circuit or some other kind of uncontrollable load dependencies.

Simulation of some circuit variants also kept ending up in oscillations or non-useful dependencies of gain from load conditions. Some examples of the non-working circuits:

Breakthrough!

The most promising anti-oscillating circuit turned out to be a variation of this one:

The key to success is the parallel L/R combination of the 1k5 resistor and 2.2mH inductor. Using these components to drive the base terminals of the power modules, oscillations were gone. The inductive component is high impedance for high frequencies, taking out the high frequency gain from the emitter follower. The resistor is for damping the inductor. A single base resistor of 1k5 would dampen the emitter follower to stop oscillating, but would not provide enough base drive. The inductor is low impedance for low frequency, providing plenty of base drive.

The PNP transistors driving the base (via the L/C) are configured as level shifters in common base configuration. The Op-Amp output provide base current through the 120R resistors through the PNP transistors into the power module base terminals. The Op-Amps are configured as a phase-splitter, providing normal and inverted drive for the upper resp. lower transistor.

Basically, the output stage now is a push-pull current sink/source configuration, controlled by the input voltage to the phase splitter Op-Amps.

Another Op-Amp acts as a PI controller to control the current set point according to the difference of input voltage and output voltage (including some gain).

Less distortion

The above configuration either works as a class B amplifier, showing heavy crossover distortion or in a non-controlled manner in class AB, depending on the setting of the bias potentiometer. Using the potentiometer, one can inject some idle current into the output stage, leading to less crossover distortion. The amount of idle current is not stable over any kind of variation of environmental parameters like temperature. Main reason is the temperature dependency of the base-emitter voltage and the base terminal beeing shunted by a resistor to emitter. With temperature variation, the base-emitter drop changes, causing a varying voltage over the rather stable resistor, causing this resistor to shunt more or less drive current away from the transistors base. So using a constant base drive current still results in a varying collector current over temperature. The variation of collector current is quite large caused by the triple darlington configuration.

Less temperature dependency

There's no way to put a sense resistor into the emitter of the upper leg transistor, as one would usually do to stabilize the idle current in a common push-pull class AB amplifier. The emitter of the lower leg transistor is available, since the whole driver circuit uses a floating +/- 12V supply scheme (floating around the output potential), this emitter potential is way off in terms of voltage. One could place a sense resistor there, still it wouldn't be simple to close the loop. Neither useful, since that's only one half of the output, we'd require sensing in both legs. Nor this would be the usual way to place the resistor, in a quasi-complimentary configuration using power NPNs only, one would place the lower sense resistor into the collector (which is not available here).

Next problem: Using emitter resistors to set the idle current results in quite a large temperature dependency with darlingtons. Improper temperature compensation of the bias voltage for the output transistors easily causes thermal runaway leading to auto destruction in simple built power stages using power darlington transistors. I've got triple darlingtons here which would be even more unstable over temperature, so one more reason to avoid this classical biasing scheme.

Second Breakdown!

No, not the transistors. Rather a second breaktrough for the circuit design.

Using a bunch of spare closed loop current transducers (these well known blue blocks from a well known swiss manufacturer), I was able to do some analog computation that yields the actual idle current through the transistors. Measuring the sum of the current going "into" the power module (collector of upper transistor and emitter of lower transistor) minus the rectified current "leaving" the module (output current) results in the current going straight "through" the module. Which is the actual idle current. This goal can be achieved using one current sensing module and two power diodes, or using to sensors and doing the math in small signals. The latter one is at the expense of one more current sensor module, but avoids the additional distortion introduced by the current steering power diodes. So I went for the second module after some experiments using diodes and one current sensor module.

Now, we've got a working amplifier circuit:

U4A is a simple differential input buffer, providing a large common mode range. The common mode range is extended by adding some noise gain, using R27 and R39. Its output signal is filtered by lowpass R36 and C4, to limit the input slew rate. U4B is a PI control amplifier, with its input node beeing the input signal and the fed back output signal through R45 and R46. The ratio of R37 to R45/R46 provides gain.

U5A and U5B form the phase splitter, providing inverted and non-inverted base drive waveforms. R55 is nothing but a current limiting resistor in case one of the outputs goes below GND. R33 resp. R51 set the base current for the power transistors. Q2 and Q3 provide the necessary level shifting. Zener D11 sets the driver circuit GND potential to about 6.9V above the power stage output potential, to provide headroom for the base drive through Q2. The driver power supply voltages are floating at -5.1V and +18.9V around the output potential. The gain stage negative feedback is taken from output GND.

Current transducer U3 is sensing the sum of the power stage supply current. One can simply achieve this by putting two wires in opposite direction through the sensor. The other transducer (U6) senses the output current.D1 and D2 put a rectified version of its output current into the summing amplifier U1A, the input sense current is subtracted through D3 and D4. D3 and D4 aren't necessary at all, I left them there for their aesthetic value. U1A provides a voltage proportional to the actual idle current at its output node. RV1 provides offset adjustment for U6, whereas RV2 is used to set the idle current by offsetting U3s output. The control amplifier U1B drives a current into the LED of photocoupler U2. A proportional current appears at Pin 5 of U2, driving base current through Q1 into the base terminals of both power transistors, summed into the phase splitters output by R25 and R26. U1B controls this additional base current to the idle current set point.

Isolated power supply

The amplifier gain and driver stage requires a floating +/- 12V power supply. I've built one using a bunch of discrete components, providing up to 7 isolated and regulated +/- 12V outputs.

This is a simple Royer oscillator, using a bunch of 1:1 transformes, rectifier diodes and 7812 / 7912 type linear post regulators. This circuit isn't particular good at efficiency, it just provides the necessary low noise power supplies. Each amplifier requires two isolated +/- 12V supplies, the final box will have three amplifier channels, so there's one more +/- 12V supply left for some common input stage circuitry.

Three phase output

Having a working amplifier circuit, I added some stuff to it: Offset adjustment, this is required since the input buffer creates a large offset voltage due to its high noise gain, and an additional current output mode, using a seperate current sensing resistor and control amplifier. I built three of them and placed them all on top of the large heatsink:

Some testing of a single channel, pushing 38Vrms into a 8 Ohms load resistor:

Rescued an enclosure from beeing scrapped, all the stuff fits in:

Input related stuff

To have an easy way to generate a three phase voltage, I made an input circuitry common to all three channels. It consists of a switching matrix and a phase shifter. The switching matrix allows to route the amplifier inputs to independent input connectors or to use a three phase voltage generated by the phase shifter. The phase shifter requires a single phase input voltage, which is to be taken from one of the input connectors.

The phase shifter takes its input signal and generates a pair of two signals that are 90 degrees out of phase by routing the signal through two chains of all-pass filters. There's an easy to use designer for this purpose at Tonnesoftware: QuadNet, that does all the necessary component calculations. Given the inphase and quadrature (0 degrees and 90 degrees) signal, it a simple job of two summing amplifiers to create a three-phase system (0 degrees, 120 degrees and 240 degrees). This particular phase shifter network is designed to work from 10Hz to 1000Hz, so one can simulate a three-phase system within this frequency range.