At my university we had a course about transistor physics, modelling, biasing, feedback, etc. ending with the construction of a power amplifier.
We managed to build a working one, but with less than stellar performance. So after the project was done I started designing a better one, learning form our mistakes. I devoured Distortion In Power Amplifiers and started designing.
After testing lots of different VAS stages and output stages in LTspcice, I came up with the following schematic, which turns out to be quite similar to the blameless one. Interestingly, on its own the cascoding VAS had way less distortion than the darlington one, but with global feedback they were about the same due to the larger gain in the darlington.
Not pictured is the power supply which was made by half-wave rectifying a 25VAC voltage into two 4700uF caps, giving +/- 32VDC to power LM7[89]24 regulators.
It was an interesting experience. Looking at the final blameless schematic (figure 33) there were some parts I did not understand, so I omitted them. Sometimes I'd hit a problem, find a solution, and realise I just added that component I did not understand before.
After construction on stripboard, I initially burned a fuse in the transfomer due to a short. Then I powered it with a 50 ohm series resistor to limit current and probed around until I fixed all the mistakes.
After I got it working I was amazed by the volume and clarity compared to the uni project. Except there was quite a bad hum. My initial plan was to make a star ground, which I read a few things about(in our course we used nice clean 10V lab supplies), but this turned out to be hard on stripboard. But after rerouteing some ground wires so that the AC supply did not "cross" the signal path, I got rid of the hum.
So I did some measurements and everything seemed fine. Until I removed the current limiting resistor, after which it started to oscillate on the positive side. This seems kind of strange because my dominant pole capacitor is already quite big, it only happens when I remove the resistor, and only on the positive side of the input.
I could add more (some designs include additional caps around the output transistors) or bigger capacitors, but as can be seen in the bode plot, gain already starts to taper off at 20KHz (the MyDAQ won't do higher frequencies, sorry) so and advice on how to fix the stability issues is appreciated.
Furthermore, there are still a few parts in figure 33 that don't make sense to me. Mainly the bias of the current source C11, R21 R22, the extra resistor R14 in the rubber diode, together with adjacent R12, R24, and finally the filter at the output.
One of these sections could very well be the solution to my problem, but I don't want to blindly copy anything.
The output filter seems to be essentially a low-pass with a cut-off somewhere outside the audible range, which might hide the oscillation from the speaker, but that seems bad engineering to rely on.
We managed to build a working one, but with less than stellar performance. So after the project was done I started designing a better one, learning form our mistakes. I devoured Distortion In Power Amplifiers and started designing.
After testing lots of different VAS stages and output stages in LTspcice, I came up with the following schematic, which turns out to be quite similar to the blameless one. Interestingly, on its own the cascoding VAS had way less distortion than the darlington one, but with global feedback they were about the same due to the larger gain in the darlington.
Not pictured is the power supply which was made by half-wave rectifying a 25VAC voltage into two 4700uF caps, giving +/- 32VDC to power LM7[89]24 regulators.
It was an interesting experience. Looking at the final blameless schematic (figure 33) there were some parts I did not understand, so I omitted them. Sometimes I'd hit a problem, find a solution, and realise I just added that component I did not understand before.
After construction on stripboard, I initially burned a fuse in the transfomer due to a short. Then I powered it with a 50 ohm series resistor to limit current and probed around until I fixed all the mistakes.
After I got it working I was amazed by the volume and clarity compared to the uni project. Except there was quite a bad hum. My initial plan was to make a star ground, which I read a few things about(in our course we used nice clean 10V lab supplies), but this turned out to be hard on stripboard. But after rerouteing some ground wires so that the AC supply did not "cross" the signal path, I got rid of the hum.
So I did some measurements and everything seemed fine. Until I removed the current limiting resistor, after which it started to oscillate on the positive side. This seems kind of strange because my dominant pole capacitor is already quite big, it only happens when I remove the resistor, and only on the positive side of the input.
I could add more (some designs include additional caps around the output transistors) or bigger capacitors, but as can be seen in the bode plot, gain already starts to taper off at 20KHz (the MyDAQ won't do higher frequencies, sorry) so and advice on how to fix the stability issues is appreciated.
Furthermore, there are still a few parts in figure 33 that don't make sense to me. Mainly the bias of the current source C11, R21 R22, the extra resistor R14 in the rubber diode, together with adjacent R12, R24, and finally the filter at the output.
One of these sections could very well be the solution to my problem, but I don't want to blindly copy anything.
The output filter seems to be essentially a low-pass with a cut-off somewhere outside the audible range, which might hide the oscillation from the speaker, but that seems bad engineering to rely on.
Attachments
Well, it's not exactly blameless. All the blame is on me though. For starters I did not tune crossover distortion yet. I turned it up too high once and that looked good but melted everything. I believe the peaks in the FFT are mainly that, crossover. In simulation all the harmonics are -150dB or less, so that's not too bad.
I also far from eliminated all the sources of distortion Douglas Self mentions that are due to construction and not due to the schematic. I just barely got rid of mains hum, and that's it for layout.
Again, I could blindly copy a working amplifier design and build it, but that is not the point. I want to learn and understand. I could study your design, and maybe understand and learn a bit, but right now I'd rather focus on fixing the flaws of my own design.
I also far from eliminated all the sources of distortion Douglas Self mentions that are due to construction and not due to the schematic. I just barely got rid of mains hum, and that's it for layout.
Again, I could blindly copy a working amplifier design and build it, but that is not the point. I want to learn and understand. I could study your design, and maybe understand and learn a bit, but right now I'd rather focus on fixing the flaws of my own design.
Not asking you to copy anything.
But , biasing the beta enhancement semi and other operating points brings
this topology to a "tipping point".
A little bit outside these operating points equals mediocre performance.
Just getting close will drop THD20 below 20ppm.
This will hold true without TMC , or any other enhancement.
This is why they call this "blameless".
PS - also the least likely to oscillate , with even 68pf miller across the
active darlington VAS.
Base stoppers on the CCS's and the clamp semi (Q28) are just to eliminate
residual ringing at overload.
OS
But , biasing the beta enhancement semi and other operating points brings
this topology to a "tipping point".
A little bit outside these operating points equals mediocre performance.
Just getting close will drop THD20 below 20ppm.
This will hold true without TMC , or any other enhancement.
This is why they call this "blameless".
PS - also the least likely to oscillate , with even 68pf miller across the
active darlington VAS.
Base stoppers on the CCS's and the clamp semi (Q28) are just to eliminate
residual ringing at overload.
OS
That web page you've read is a nice summary of amplifier distortions, but it's nowhere near enough to design and build a stable amp with good performance. Get Self's book, or Cordell's, or Slone's, or all of them, and devour them too. Read papers by Leach, Cherry, Linsley Hood, Hawksford... Search this forum, it's a goldmine.
A good paper on stability is this one. You can't stabilize a real amplifier in simulation, but it's a good tool to learn about the subject, play with component values, see how they affect stability... I just simulated your circuit (using MJL2119x output transistors) and it looks like it's very, very over-compensated: I get a ULGF of 65kHz and a gain @ 20kHz of ~10dB, so your oscillation problem clearly can't be fixed by brute force compensation, and there's no way that you're even going to approach Blameless performance with that scheme. A Cdom=100p will give you a theoretical ULGF of ~650kHz and a gain @ 20kHz of ~30dB, which is conservative but more like it. Other problems I see:
- You use small signal transistors for everything but the output. To keep it simple, use something like BD139/140 for the VAS CCS, Vbe multiplier, 2nd VAS transistor and drivers.
- You've left the 100R degeneration resistors in the input stage but increased the tail current a lot, not that you don't see IPS's with 10mA of tail current, but ~20:1 degeneration (which is what you have) is unusually high. I'd make R5=220R.
- R19 is normally ~470-2k2, 1k being typical. For R7 Cordell likes to use the equivalent of 10:1 degeneration, which would be ~22R in this case. Also there should be a resistor between ground and the collector of Q9, a typical value being 1k too.
- You may get away without the output filter with a dummy load, but to drive a real loudspeaker, not using it is asking for trouble.
- Adjust the output stage bias for a voltage drop across R8 of ~26mV.
With these changes you'll get a more reasonable starting point. It may still oscillate, veroboard and long air wires don't help, so you may need to rethink your layout, use base stoppers, add a small (10-33p) cap between VAS collector and ground, add some lead-lag across the feedback resistor, bc caps on the drivers... Read and learn about supply decoupling and grounding layout too.
I'd start by simulating it, playing with those components, seeing how they affect DC operation point, stability and THD, getting a better feel for what they do. When you run a .tran simulation, plot the currents of all the nodes of all the transistors, this is usually illuminating. See how it clips, both positive and negative, and what the dissipation is in each transistor. Oh, and did I mention "search the forum"? 😉
Cheers,
Cabirio
Edit: forgot to mention: for good thermal stability, mount the Vbe multiplier on top of one of the output transistors. Also, in the picture I see four output transistors for what looks like a single channel, are you using two paralleled ones per rail?
A good paper on stability is this one. You can't stabilize a real amplifier in simulation, but it's a good tool to learn about the subject, play with component values, see how they affect stability... I just simulated your circuit (using MJL2119x output transistors) and it looks like it's very, very over-compensated: I get a ULGF of 65kHz and a gain @ 20kHz of ~10dB, so your oscillation problem clearly can't be fixed by brute force compensation, and there's no way that you're even going to approach Blameless performance with that scheme. A Cdom=100p will give you a theoretical ULGF of ~650kHz and a gain @ 20kHz of ~30dB, which is conservative but more like it. Other problems I see:
- You use small signal transistors for everything but the output. To keep it simple, use something like BD139/140 for the VAS CCS, Vbe multiplier, 2nd VAS transistor and drivers.
- You've left the 100R degeneration resistors in the input stage but increased the tail current a lot, not that you don't see IPS's with 10mA of tail current, but ~20:1 degeneration (which is what you have) is unusually high. I'd make R5=220R.
- R19 is normally ~470-2k2, 1k being typical. For R7 Cordell likes to use the equivalent of 10:1 degeneration, which would be ~22R in this case. Also there should be a resistor between ground and the collector of Q9, a typical value being 1k too.
- You may get away without the output filter with a dummy load, but to drive a real loudspeaker, not using it is asking for trouble.
- Adjust the output stage bias for a voltage drop across R8 of ~26mV.
With these changes you'll get a more reasonable starting point. It may still oscillate, veroboard and long air wires don't help, so you may need to rethink your layout, use base stoppers, add a small (10-33p) cap between VAS collector and ground, add some lead-lag across the feedback resistor, bc caps on the drivers... Read and learn about supply decoupling and grounding layout too.
I'd start by simulating it, playing with those components, seeing how they affect DC operation point, stability and THD, getting a better feel for what they do. When you run a .tran simulation, plot the currents of all the nodes of all the transistors, this is usually illuminating. See how it clips, both positive and negative, and what the dissipation is in each transistor. Oh, and did I mention "search the forum"? 😉
Cheers,
Cabirio
Edit: forgot to mention: for good thermal stability, mount the Vbe multiplier on top of one of the output transistors. Also, in the picture I see four output transistors for what looks like a single channel, are you using two paralleled ones per rail?
Last edited:
That is a lot to digest. I'll look at it when I have more time. The Self book is definitely on my wishlist.
For now I can say that I solved the oscillation problem by increasing the output capacitor on the voltage regulators. The regulators are the other 2 "transistors" you are seeing. So just 2 output transistors.
I'm currently listening to my amp on a real loudspeaker. Uhm... I should put the filter on top of the list of things to look into I guess. The point is that high frequencies can kill the tweeter, right?
For now I can say that I solved the oscillation problem by increasing the output capacitor on the voltage regulators. The regulators are the other 2 "transistors" you are seeing. So just 2 output transistors.
I'm currently listening to my amp on a real loudspeaker. Uhm... I should put the filter on top of the list of things to look into I guess. The point is that high frequencies can kill the tweeter, right?
- Status
- Not open for further replies.