When playing with one of the amplifiers in Bob Cordell's book, I found a surprising case of instability in a double-EF output stage.
The amplifier is the example from the book, page 63. It normally uses the not too fast MJL2119x transistors. Because I saw a problem with robustness against the output series resonance at high frequencies, I tried to use the faster MJL3281/1302. To my surprise, the output stage became completely unstable, bursting into oscillation of about 8MHz during the positive half-wave. This can be seen below in instability.png. If one introduces base-stopper resistors of 10R, the circuit again is stable, whereas 2R are not enough. The quiescent current was about 50mA as in the case with MJL2119x.
Although not that extreme, I know this from experience. The sound of an amplifier may change more than reasonable, if one changes a little bit the base-stopper resistors or changes a little bit the unity gain crossover of the global NFB. For another project, I therefore decided to completely abandon darlingtons in the output stage.
One can always question the accuracy of a simulation model that does not even include stray capacities and inductances and that heavily depends on the transistor models. Nevertheless it would be nice to have a better way to evaluate the OPS stability than looking at the transient response.
After all, a multiple-EF contains a tight feedback loop around the transistors that should be accessible to analysis via Bode diagram. The three pictures show a proposal for that.
Below, ef-mjl21194.png tries to analyse the loop of the original OPS. In order to break the intrinsic feedback loop, the emitter current of Q2 is measured by R3 and translated into the original current by G1, leaving the emitter of Q2 at almost constant potential. (This even more correctly would be possible with a 0V-voltage source in place of R3 and a current-dependent current source instead of G1, depending on the current through the former voltage source)
Thus, Q1 and Q2 do not "see" the feedback signal. From 20mA to 2A output current (stepping of IE), the unity gain crossover appears to be in the region 20 to 30 MHz, phase margin is always larger than 90 degrees. (BTW, the value of R7 - VAS output resistance - has no influence at all).
The difference between ef-mjl3281.png and ef-mjl3281-10R.png is the base stopper resistor R2 in the latter. One sees lower unity gain crossover freqencies and better phase margins. Nevertheless, even the former Bode diagram would suggest stability.
Are there any proposals how to better evaluate the OPS stability and the "distance to the disaster"? I have the impression that my simple approach does fail. One point is of course that only one half of the OPS is included into the model. But in the positive half wave, the negative OPS should not contribute to the behaviour.
Does anybody have a hint how one should evaluate the intrinsic loop of multiple-EF output stages in the frequency domain?
BR,
Matze
The amplifier is the example from the book, page 63. It normally uses the not too fast MJL2119x transistors. Because I saw a problem with robustness against the output series resonance at high frequencies, I tried to use the faster MJL3281/1302. To my surprise, the output stage became completely unstable, bursting into oscillation of about 8MHz during the positive half-wave. This can be seen below in instability.png. If one introduces base-stopper resistors of 10R, the circuit again is stable, whereas 2R are not enough. The quiescent current was about 50mA as in the case with MJL2119x.
Although not that extreme, I know this from experience. The sound of an amplifier may change more than reasonable, if one changes a little bit the base-stopper resistors or changes a little bit the unity gain crossover of the global NFB. For another project, I therefore decided to completely abandon darlingtons in the output stage.
One can always question the accuracy of a simulation model that does not even include stray capacities and inductances and that heavily depends on the transistor models. Nevertheless it would be nice to have a better way to evaluate the OPS stability than looking at the transient response.
After all, a multiple-EF contains a tight feedback loop around the transistors that should be accessible to analysis via Bode diagram. The three pictures show a proposal for that.
Below, ef-mjl21194.png tries to analyse the loop of the original OPS. In order to break the intrinsic feedback loop, the emitter current of Q2 is measured by R3 and translated into the original current by G1, leaving the emitter of Q2 at almost constant potential. (This even more correctly would be possible with a 0V-voltage source in place of R3 and a current-dependent current source instead of G1, depending on the current through the former voltage source)
Thus, Q1 and Q2 do not "see" the feedback signal. From 20mA to 2A output current (stepping of IE), the unity gain crossover appears to be in the region 20 to 30 MHz, phase margin is always larger than 90 degrees. (BTW, the value of R7 - VAS output resistance - has no influence at all).
The difference between ef-mjl3281.png and ef-mjl3281-10R.png is the base stopper resistor R2 in the latter. One sees lower unity gain crossover freqencies and better phase margins. Nevertheless, even the former Bode diagram would suggest stability.
Are there any proposals how to better evaluate the OPS stability and the "distance to the disaster"? I have the impression that my simple approach does fail. One point is of course that only one half of the OPS is included into the model. But in the positive half wave, the negative OPS should not contribute to the behaviour.
Does anybody have a hint how one should evaluate the intrinsic loop of multiple-EF output stages in the frequency domain?
BR,
Matze
.] I will give Bob a hint, maybe he can add the capacitor in his next edition.
