Stability tests. (25-Oct-2007.)
The MJR-7 circuit was found to be unstable with a range of low value capacitor loads. This was first confirmed by an improved simulation, which also showed that one method of avoiding the problem is to add 100nF across the amplifier output. This appears to be an effective solution in practice, but an alternative is to change the resistor in series with the 220pF compensation capacitor. To find the optimum value the method I used was to use a 1k variable resistor, and find the maximum and minimum resistance for stability at various load capacitance values. The results are shown as a graph:
The vertical scale is the resistance in series with the compensation capacitor, and the horizontal scale is load capacitance in nF. The two black traces are the resistance limits between which stability is maintained. With the original 100R resistor, the line shown dotted, instability occurs between 1.1nF and 3.6nF loads. Changing the resistor value to 220R, shown as a solid line, there is stability at all capacitor load values. It may be a good idea to also add the 100nF across the output, even if this is no longer essential, it adds extra security and will also help reject any interference picked up by the speaker cables.
A small inductive load can have a parallel resonance with the 100n near the unity gain frequency, so there can still be problems. My simulations suggest that this option will work, but the stability margins are still not as good as I would like. So far my simulations seem to suggest worse stability problems than are found in my real amplifier, so maybe this is reason to be optimistic. There is evidence of some variation in stability with different transistors and layout, so for anyone not able to test for stability with capacitive loads including the 100n at the output is a good idea.
The reason for an upper limit on the series resistor is that increasing it too far increases the unity gain frequency of the amplifier to 10MHz or more, where other sources of phase shift become significant. Another component we could change is the 7R5 in parallel with the output inductor. I now think my reluctance to reduce this, although not completely thought out, was a good choice. The inductor damping is wrong for high capacitor loads, but for the low values where instability was a problem the 7R5 is not too bad.