Hello,
Thinking that there may be some interest within the forum, I’ve attached (I hope!) a file that describes the bias control system that I developed for my latest audio power amplifier project (BEG MB1). It provides excellent performance.
I apologize for the document’s length but you only need to read the first three paragraphs to get a basic understanding of how it works. The subsequent paragraphs in the main section provide implementation and performance details. The appendices discuss calibration and some additional design considerations.
I hope you find it interesting!
Bruce
Thinking that there may be some interest within the forum, I’ve attached (I hope!) a file that describes the bias control system that I developed for my latest audio power amplifier project (BEG MB1). It provides excellent performance.
I apologize for the document’s length but you only need to read the first three paragraphs to get a basic understanding of how it works. The subsequent paragraphs in the main section provide implementation and performance details. The appendices discuss calibration and some additional design considerations.
I hope you find it interesting!
Bruce
Very nice article Bruce, well researched, designed and written. Chapeau!
Too bad it is in the open now - it might have been a nice article for AudioXpress ;-).
As can be seen in some of the graphs, bias stability with power output bursts is always a challenge.
I have a question about the distortion measurements.
The distortion profile versus output power is rather flat which is unusual, but could be a feature of your amp.
Did you do any distortion measurements with the bias controller disabled, to isolate any possible contribution by it?
Jan
Too bad it is in the open now - it might have been a nice article for AudioXpress ;-).
As can be seen in some of the graphs, bias stability with power output bursts is always a challenge.
I have a question about the distortion measurements.
The distortion profile versus output power is rather flat which is unusual, but could be a feature of your amp.
Did you do any distortion measurements with the bias controller disabled, to isolate any possible contribution by it?
Jan
Jan,
First, let me thank you for the kind words.
Regarding your question about the amplifier’s distortion profile, measurements with the bias controller disabled are problematic. As no parts of the bias generator are mounted on the main heatsink, the bias drifts rapidly during sweeps. That said, from quick spot measurements, I do not believe that the bias controller changes the profile significantly.
Were you focused on the low power end? I should have been more clear that I plotted THD, not THD+N, so we are only looking at harmonics. Thus, the curves do not rise from noise at the low end.
The flatness at the high power end (up to clipping) was somewhat of a surprise to me. This says that the dominant harmonics rise at the same rate as the fundamental. I was thinking that the rate is usually higher. Note, however, that the distortion doubles at all levels each time the load resistance is cut in half. Distortion vs. frequency is also flat, due, no doubt, to the uniform level of feedback.
The “MB1” is a low feedback, wide open-loop bandwidth design. The gain stages were linearized open loop by strong degeneration and keeping the load currents small compared to the bias. Perhaps you, or some of the other very smart members, could improve my understanding of the distortion mechanisms at work here.
Bruce
P.S. I plan on opening a thread on the amplifier soon.
Yes, wide open loop BW could have that effect.
Looking forward to your amp design.
(Or have it published at AudioXpress ;-).
Jan
Looking forward to your amp design.
(Or have it published at AudioXpress ;-).
Jan
Not an expert at all in this field but found this old 1996 amp design equipped with a non-linear bias control.
Thanks, Jacques. I had not seen this article before. It reminds me of Edmond Stuart's Auto Bias project. Old? Sadly, I was looking into bias regulation techniques 10 years earlier! 🙂
Nice project!
I'm gonna throw some info you could find useful:
Best temperature sensor: MicroMELF diode soldered on the collector tab of the power device. Inconvenient, because it sits on the power rail.
Schematic with this sensor: it makes a temperature-dependent current that tweaks the bias controller. Note the driver scheme in the schematic didn't work out, I don't recommend it. But the temperature sensing part worked very well. About as good as the ThermalTraks.
All these schemes have the same flaw: they try to tweak the voltage in real time, which always ends up adding more distortion than the distortion it's supposed to remove. This criticism does not apply to Bruce's way of keeping the voltage between the power device bases constant on an audio timescale, and tweaking it slowly as temperature changes. That works.
It could mean the main distortion mechanism is in the frontend, or somewhere that isn't the output stage, and thus doesn't depend that much on output level. If you want to measure just the output stage, you can do so open loop by connecting the signal source to the input of your output stage. If your source has a driving impedance that is close to the output impedance of your VAS, it should give a usable measurement.
I'm gonna throw some info you could find useful:
Best temperature sensor: MicroMELF diode soldered on the collector tab of the power device. Inconvenient, because it sits on the power rail.
Schematic with this sensor: it makes a temperature-dependent current that tweaks the bias controller. Note the driver scheme in the schematic didn't work out, I don't recommend it. But the temperature sensing part worked very well. About as good as the ThermalTraks.
All these schemes have the same flaw: they try to tweak the voltage in real time, which always ends up adding more distortion than the distortion it's supposed to remove. This criticism does not apply to Bruce's way of keeping the voltage between the power device bases constant on an audio timescale, and tweaking it slowly as temperature changes. That works.
It could mean the main distortion mechanism is in the frontend, or somewhere that isn't the output stage, and thus doesn't depend that much on output level. If you want to measure just the output stage, you can do so open loop by connecting the signal source to the input of your output stage. If your source has a driving impedance that is close to the output impedance of your VAS, it should give a usable measurement.
Excellent write-up!
No expert on the matter, but I've always wondered why we don't put bias under the control of some small uP. Just about every input necessary could be boiled down by software into an appropriate bias setting. As an analog guy I'm terrified of putting digital anything in a circuit, but heck, it's a power amp. I can't imagine there would be much problem.
No expert on the matter, but I've always wondered why we don't put bias under the control of some small uP. Just about every input necessary could be boiled down by software into an appropriate bias setting. As an analog guy I'm terrified of putting digital anything in a circuit, but heck, it's a power amp. I can't imagine there would be much problem.
I wrote about this a few yrs ago. You use a standard Vbe multiplier to provide 1st order bias control and the uC to do the 2nd order bit. I have some thoughts about it scrawled in a notebook somewhere. Maybe a project for a rainy day
🙂
🙂
I played with another all analog control loop approach back in 1990. It used what I called a "minimum peak detector" across the output transistor emitter resistors to provide the error signal for the control loop. This worked, but not well so I abandoned it. I revisited this a couple of years ago and came across a patent from 1991 (attached), which used a very similar approach to mine! Perhaps I should have considered a patent 🙂. Even with modern parts this approach has issues. Now, a very high speed ADC monitoring the voltage across these resistors and a FPGA or microprocessor would seem to have promise. Bits and software are not really my things, however.
The bias loop in my 1996 article was based on Frans Tol's and Johan H. Huijsing's idea from 1976, although my non-linear network was novel as far as I know, and there are many implementation details different. I don't know any pre-1976 articles about class-AB biasing with non-linear common-mode loops. If anyone else does, I would be very interested to know.
You may also be interested in @IanHegglun's thread about bias loops: https://www.diyaudio.com/community/...ching-auto-bias-power-amp.375141/post-6735925
Marcel,
I enjoyed your 1996 article. It's fun seeing really different approaches to solving a problem. In addition to the bias control, the clipping detector is clever. Not all topologies have the "free" IPS collector. I have always used a differential amplifier across the input bases to monitor the error voltage. It is, of course, quite small with high feedback amplifiers and changes linearly with input level. At the onset of clipping, it rises very steeply and this is easily detected by a comparator. This approach works very well but uses more parts. I, too, have extended the clipping indication for easy observation, though I use a 555 timer. In my latest amp, I used the extended pulse period (500 ms) to detect excessive clipping for disconnecting the speaker.
Bruce