Hi Lars
You always mention that some switching amplifiers wouldn't generate harmonics of the input signal (i.e. harmonic distortion) but rather noise only.
My imagination tells me that such an amp doesn't exist. While an ideal switching amplifier wouldn't introduce any error in the band of interest, a real-world one definitely will - due to timing errors, overshoot, PSU sag, nonlinearities in the "nonswitching part" of the amp etc.
The PWM amp I once developed did indeed have some HF noise that was more noticeable than THD (which was mainly 2nd order) when it was used to listen to music (which it wasn't intended for BTW). Which in turn doesn't mean that it wouldn't be cool to reduce BOTH.
The sound was nevertheless pleasing (OK, OK, OK - no parent has an ugly child....), though we could only test it in mono since there was only one amp. Stereo is usually more revealing for evaluating sound quality.
I did some simulation work on delta sigma amplifiers using a fixed sampling rate (which is the most usual implementation of a sigma-delta modualtor). The "noise" generated by the noise-shaping was mainly consisting of odd order harmonics which correlates well with a statement in a famous book:
Delta-Sigma Data Converters: by Steven R. Norsworthy, AT&T Bell Labs; Richard Schreier, Oregon State University; and Gabor C. Temes, Oregon State University
I assume your amp is working the following way: you have a 2nd order feedback-loop like the ones used for 2nd order noise shapers. Instead of a fixed sampling rate (by the use of a comparator and a subsequent clocked D-flip-flop) you just use a comparator, and the gains and time-constants of the integrators (and therefore loop-gain and phase-marging) are set in such a way that the thing oscillates by itself (feedback is taken from the output of the switching stage).
Even though I don't like self-oscillating things, I have to admit that such a topology could make a very fine amplifier.
But saying that such an amp would just give noise instead of IMD and THD, without mathematical or measurement proof (or at least some hint where the "gut feeling" comes from), is a very BOLD statement !!
Regards
Charles
You always mention that some switching amplifiers wouldn't generate harmonics of the input signal (i.e. harmonic distortion) but rather noise only.
My imagination tells me that such an amp doesn't exist. While an ideal switching amplifier wouldn't introduce any error in the band of interest, a real-world one definitely will - due to timing errors, overshoot, PSU sag, nonlinearities in the "nonswitching part" of the amp etc.
The PWM amp I once developed did indeed have some HF noise that was more noticeable than THD (which was mainly 2nd order) when it was used to listen to music (which it wasn't intended for BTW). Which in turn doesn't mean that it wouldn't be cool to reduce BOTH.
The sound was nevertheless pleasing (OK, OK, OK - no parent has an ugly child....), though we could only test it in mono since there was only one amp. Stereo is usually more revealing for evaluating sound quality.
I did some simulation work on delta sigma amplifiers using a fixed sampling rate (which is the most usual implementation of a sigma-delta modualtor). The "noise" generated by the noise-shaping was mainly consisting of odd order harmonics which correlates well with a statement in a famous book:
Delta-Sigma Data Converters: by Steven R. Norsworthy, AT&T Bell Labs; Richard Schreier, Oregon State University; and Gabor C. Temes, Oregon State University
I assume your amp is working the following way: you have a 2nd order feedback-loop like the ones used for 2nd order noise shapers. Instead of a fixed sampling rate (by the use of a comparator and a subsequent clocked D-flip-flop) you just use a comparator, and the gains and time-constants of the integrators (and therefore loop-gain and phase-marging) are set in such a way that the thing oscillates by itself (feedback is taken from the output of the switching stage).
Even though I don't like self-oscillating things, I have to admit that such a topology could make a very fine amplifier.
But saying that such an amp would just give noise instead of IMD and THD, without mathematical or measurement proof (or at least some hint where the "gut feeling" comes from), is a very BOLD statement !!
Regards
Charles
Hi Charles
Your description of my amplifier is mostly accurate.
I don't use a D-flip flop to control timing, since it would add rounding errors to my timing. (Quantization noise).
The topology that sigma-1 (one for you Guido T) people refer to as Class D, uses a triangular wave for input to the comparator. This allows for the comparator's CMRR (in dB) to be translated into harmonic distortion.
Assuming the comparator, operates at 0 mV on both input terminals, the CMRR can not translate into distortions, but only the noise will modulate with the amplified signal.
This is what i mean, in real life - depending on the type of comparator and integrator used - you may of course get HD anyway. That is why my claim is that
'SOME types of PWM amplifiers don't produce HD, but instead random noise'.
Your description of my amplifier is mostly accurate.
I don't use a D-flip flop to control timing, since it would add rounding errors to my timing. (Quantization noise).
The topology that sigma-1 (one for you Guido T) people refer to as Class D, uses a triangular wave for input to the comparator. This allows for the comparator's CMRR (in dB) to be translated into harmonic distortion.
Assuming the comparator, operates at 0 mV on both input terminals, the CMRR can not translate into distortions, but only the noise will modulate with the amplified signal.
This is what i mean, in real life - depending on the type of comparator and integrator used - you may of course get HD anyway. That is why my claim is that
'SOME types of PWM amplifiers don't produce HD, but instead random noise'.
O.K.
I can agree that you have one source of distortion less, since you are not relying on the linearity of a triangle-generator/comparator topology.
Regards
Charles
P.S.: Ever tried to take feedback from the output filter ? I know that you use a core with an air-gap (I used the same principle with my amp) to reduce nonlinearity but it might still be advantageous to include the filter into the feedback loop.
I can agree that you have one source of distortion less, since you are not relying on the linearity of a triangle-generator/comparator topology.
Regards
Charles
P.S.: Ever tried to take feedback from the output filter ? I know that you use a core with an air-gap (I used the same principle with my amp) to reduce nonlinearity but it might still be advantageous to include the filter into the feedback loop.
Hi again Charles
I have made a THD measurement on an amplifier of the type we are discussing. Below shown is the THD with an RC output filter (to measure THD of the modulator and amplier alone). The amplitude is equivalent to 10W RMS.
The THD is close to the limit of my THD analyzer, around 0.001%, as can be seen mainly 2nd order.
I have made a THD measurement on an amplifier of the type we are discussing. Below shown is the THD with an RC output filter (to measure THD of the modulator and amplier alone). The amplitude is equivalent to 10W RMS.
The THD is close to the limit of my THD analyzer, around 0.001%, as can be seen mainly 2nd order.
Attachments
Charles: My above measurements would surely support your suggestion to take the feedback signal AFTER the output coil. However i think it would be sacrificing some of the inherent stability towards capacitive and inductive loads.
I know of one company doing the feedback-after-the-choke thing, the result of that is also pretty good.
When it comes to feedback loops, though i have an idea of keeping them as short as possible, and with as low delay as possible, in order to keep IMD down.
I know of one company doing the feedback-after-the-choke thing, the result of that is also pretty good.
When it comes to feedback loops, though i have an idea of keeping them as short as possible, and with as low delay as possible, in order to keep IMD down.
With one of my first lab setups about 15 years ago the sonic difference between both types of feedback takeoff was like day and night (guess which one was better ?).
In the end I just didn't use feedback takeoff after the filter because I didn't have the knowledge (and the ideas), to make such a thing stable for a wide range of load conditions.
Regards
Charles
In the end I just didn't use feedback takeoff after the filter because I didn't have the knowledge (and the ideas), to make such a thing stable for a wide range of load conditions.
Regards
Charles
I wouldn't describe the difference as 'night and day'.
One thing is a resistor load and a sinewave signal, another is a music signal, and a complex speaker load. Things behave differently.
In fact i started out with air coils, which have very low harmonic distortion, however as it turned out a Ferrite coil sound much better ( at least if you keep it big enough so it won't saturate
)
One thing is a resistor load and a sinewave signal, another is a music signal, and a complex speaker load. Things behave differently.
In fact i started out with air coils, which have very low harmonic distortion, however as it turned out a Ferrite coil sound much better ( at least if you keep it big enough so it won't saturate
)I assume this is due to two things:
EMC: your coil was huge and I have a distinct feeling that this might have affected the circuit.
Skin effect: due to the greater wirelength than a ferrite coil has.
Regarding my experiences: The test setup was definitely not HiFi but the increase in sound quality was remarkable when the feedback was taken from the filter instead of the switching stage. I was driving a "real" speaker that wasn't even impedance corrected. When I removed the load however, the thing started to oscillate like hell (it even made FM reception impossible !).
For a really good amp the difference might not be as high but still worthwile.
Regards
Charles
That's what member subwo1's straightforward and clever circuit (within this thread http://www.diyaudio.com/forums/showthread.php?s=&threadid=14847&highlight= ) is based upon.
Regards
Charles
Regards
Charles
honored
My sincerest thanks for the compliment, Charles.
Hi Lars, my philosophy in design is that more complexity is not always best overall. The circuit design is rooted in my earlier experimentation with PWM amps, when I learned a lot about things to avoid. So, since the onset of my incapacitation and not having access to the workbench, I worked with the simulator to create the circuit in post #474 of the mentioned thread. I cannot be sure that I will have the chance to prototype it, but sure hope to.
The goal of the circuit is not to compete with the purest sounding class A topologies but to seek good overall reproduction with simplicity and efficiency from a fairly small circuit. I am not talented with computers, so I did not even simulate the circuit for distortion but concentrated on its general behavioral characteristics. My personal approach to audio reproduction is to bi-amplify, using an efficient but powerful subwoofer amp having a decent damping factor along with a lower power analog amp for the rest. Also, I prefer crossing-over the sub amp to the lower power one at a frequency higher than is usually done, say, somewhere between 150 and 200 hz.
My sincerest thanks for the compliment, Charles.
Hi Lars, my philosophy in design is that more complexity is not always best overall. The circuit design is rooted in my earlier experimentation with PWM amps, when I learned a lot about things to avoid. So, since the onset of my incapacitation and not having access to the workbench, I worked with the simulator to create the circuit in post #474 of the mentioned thread. I cannot be sure that I will have the chance to prototype it, but sure hope to.
The goal of the circuit is not to compete with the purest sounding class A topologies but to seek good overall reproduction with simplicity and efficiency from a fairly small circuit. I am not talented with computers, so I did not even simulate the circuit for distortion but concentrated on its general behavioral characteristics. My personal approach to audio reproduction is to bi-amplify, using an efficient but powerful subwoofer amp having a decent damping factor along with a lower power analog amp for the rest. Also, I prefer crossing-over the sub amp to the lower power one at a frequency higher than is usually done, say, somewhere between 150 and 200 hz.
OK.....
couldn't resist to simulate a self-oscillating switching amp with a double-integrator feedback loop.
I furthermore couldn't resist to take feedback from the output filter. So the first integrator is in fact a PID. I didn't fiddle around with the time-constants or gains very much. In fact when the thingie didn't want to oscillate after a few tries, I boldly added some hysteresis to the switching stage, although I know it schould be doable otherwise.
Anyway, the simulation results look quite promising.
Here is the circuit diagram.
Regards
Charles
couldn't resist to simulate a self-oscillating switching amp with a double-integrator feedback loop.
I furthermore couldn't resist to take feedback from the output filter. So the first integrator is in fact a PID. I didn't fiddle around with the time-constants or gains very much. In fact when the thingie didn't want to oscillate after a few tries, I boldly added some hysteresis to the switching stage, although I know it schould be doable otherwise.
Anyway, the simulation results look quite promising.
Here is the circuit diagram.
Regards
Charles
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.