Heres a simplified LTspice version of my schematic. I've used some other components, a different current generator, ideal voltage sources and a simplified modulation topology.
I usually do all my simulations in Orcad pSpice, so I'm not that familar with the interface of LTspice. Therefore I cant verify if the simplified version acts like the original one.
Q5 and Q9 are slow, but I dosn't affect the speed of the comperator because of the configuration. The benefit is that Q5 and Q9 has a high breakdown voltage which is necessary.
Q1 and Q2 can also be slow, without affecting the speed. Therefor you can put in your favorite low noise mached BJT's here if you please.
I've included a couple of gif's of the schematic and the simulation results also.
I usually do all my simulations in Orcad pSpice, so I'm not that familar with the interface of LTspice. Therefore I cant verify if the simplified version acts like the original one.
Q5 and Q9 are slow, but I dosn't affect the speed of the comperator because of the configuration. The benefit is that Q5 and Q9 has a high breakdown voltage which is necessary.
Q1 and Q2 can also be slow, without affecting the speed. Therefor you can put in your favorite low noise mached BJT's here if you please.
I've included a couple of gif's of the schematic and the simulation results also.
Re: LTSpice component library
You'll find more info perhaps if you use "LT"
You can add models by editing component name to match the model and add the .model statement to the schematic itself.
I know you can import libraries....... somehow. I haven't found that information myself, but haven't really hunted it down. I believe it is most likely to be found in the yahoo user group for it.
Regards,
Chris
koolkid731 said:
You'll find more info perhaps if you use "LT"
You can add models by editing component name to match the model and add the .model statement to the schematic itself.
I know you can import libraries....... somehow. I haven't found that information myself, but haven't really hunted it down. I believe it is most likely to be found in the yahoo user group for it.
Regards,
Chris
Distortion due to Carrier FM
Over the past year that, I have been playing with Class D amps, I've often noticed that the carrier (modulator) is often Frequency modulated. I don’t think that I ever hear this but it certainly shows up as side bands on the audio signal. The cause of the frequency modulation is usually power supply variation due to peak audio signals. This means that the FM'ing is narrow band and should therefore fold directly down in to the audio band and mix with the audio signal. ==== Distortion====
This frequency modulation is very common on the self oscillating designs but usually not present on the fixed carrier designs, however good designs by both methods usually sound the very similar with no unusual tones or
This brings me to my dilemma, WHY isn’t Carrier frequency modulation in a class D amp audible?
All suggestions welcome, I'm especially interested in Bruno’s ideas, assuming he still follows this thread.
Classdunce
Over the past year that, I have been playing with Class D amps, I've often noticed that the carrier (modulator) is often Frequency modulated. I don’t think that I ever hear this but it certainly shows up as side bands on the audio signal. The cause of the frequency modulation is usually power supply variation due to peak audio signals. This means that the FM'ing is narrow band and should therefore fold directly down in to the audio band and mix with the audio signal. ==== Distortion====
This frequency modulation is very common on the self oscillating designs but usually not present on the fixed carrier designs, however good designs by both methods usually sound the very similar with no unusual tones or
This brings me to my dilemma, WHY isn’t Carrier frequency modulation in a class D amp audible?
All suggestions welcome, I'm especially interested in Bruno’s ideas, assuming he still follows this thread.
Classdunce
It is quite conceivable that the FM sidebands demodulate to some degree. However, since they have to mix against the carrier itself, from which they are spaced at intervals equal to the frequency of the audio signal (and its harmonics), the fold-down products will land smack-bang on the audio itself and its harmonics, not somewhere else in the spectrum. The effect will simply be some extra harmonic distortion, indistinguishable from harmonic distortion products created in more mundane ways. Since the distortion that I get in practice jives well with what I calculate without looking at carrier sideband fold-down, it must mean that it isn't a significant contributor.
Bruno,
thanks for explanation, I'm not sure that I agree completely because while it is true that the first order products fold down to the signal and its harmonics, the same cannot be said for all the second order products. With any non-linearity in the signal path the FFT gets very ugly very quickly. I guess being an ex sigma delta designer I'm always paranoid about residual energy folding down especially if I have no way to randomize it. I have spent way too many late nights chasing warbling sounds and wandering tones, even tones that magically appear and worse still disappear when you try to measure them. From my experience tones are audible up to 10dB below the systems random noise floor, so when I can see them on a spectrum analyzer I get real nervous. I’m convinced that there is something different in the classD case, I’ve heard hand waving arguments about the problem not mattering as long as each pulse is centered within its own PWM period, but that’s a crock in my opinion, because the system has memory, especially when the feedback is from after the filter.
I certainly appreciate your time and explanation,
My problem is that I don’t get to use air gapped ferrites or micrometal or anything else exotic, so ferrite soft saturation is something I have to live with and design around.
Classdunce
thanks for explanation, I'm not sure that I agree completely because while it is true that the first order products fold down to the signal and its harmonics, the same cannot be said for all the second order products. With any non-linearity in the signal path the FFT gets very ugly very quickly. I guess being an ex sigma delta designer I'm always paranoid about residual energy folding down especially if I have no way to randomize it. I have spent way too many late nights chasing warbling sounds and wandering tones, even tones that magically appear and worse still disappear when you try to measure them. From my experience tones are audible up to 10dB below the systems random noise floor, so when I can see them on a spectrum analyzer I get real nervous. I’m convinced that there is something different in the classD case, I’ve heard hand waving arguments about the problem not mattering as long as each pulse is centered within its own PWM period, but that’s a crock in my opinion, because the system has memory, especially when the feedback is from after the filter.
I certainly appreciate your time and explanation,
My problem is that I don’t get to use air gapped ferrites or micrometal or anything else exotic, so ferrite soft saturation is something I have to live with and design around.
Classdunce
Idle tones are different in a self oscillating classD than a classic switched cap sigma delta. The problem is more like a continuous time sigma delta, but even in this respect there seems to be a major difference.
Part of the difference is that in a sigma delta DAC your source is sampled at a certain frequency so any frequency error or jitter creates distortion relative to the sampled time of the source.
This makes a ClassD amp more like a Sigma delta ADA (ADC-DAC). The only experience that I have with ADA's was with a hearing aid that I designed a few years back. The output was a classD amp but the input was a classic 2nd order switched cap sigma delta. In this case the frequency had to be fixed because all the digital filters that post processed the sound assumed a fixed sample frequency.
The hand waving argument (for letting the carrier frequency vary) goes something like this:
Because your input sampling and output sampling have the same period there is no error for a perfect pulse centered PWM. If the carrier frequency is over 500Khz and centered than you can easily achieve -90dB 3rd harmonic for the digital PWM output, so the error is small anyway.
Trouble is it looks like **** after the filter, especially into a real speaker load (around -45dB 3rd). So the feedback term is basically the same as adding a precorrection signal of (input referred)amplitude -45dB to the input signal. If you do this analysis with the sample frequency fixed than the math is quite simple and the spectral compounds are easy to predict. However if you let the frequency vary than your simple equation becomes very complex because the energy of the feedback term is effectively multiplied by the frequency error term and added to the feedback term. When the Delta Freq and feedback error are both small errors (less than 0.1%), than the added compounds all fall below -110dB for a -3db input signal. The delta Freq term typically scales with output amplitude so at small signal the inharmonic spurs are below -110dB to -120dB (relative to full scale)
Unfortunately in cheap systems the Inductor filter is very non-linear and the supply variation is large with audio amplitude (up to 20%) in my case. Because of this you get large Frequency errors and large amplitude errors, so the FFT has inharmonic spurs all over the place at amplitudes of only -60dB. Fortunately they scale with signal amplitude so the "tones" are around -110dB for a -60dB signal.
What is interesting, is that I cant hear these "tones" my feeling is that this occurs because the tones are non-stationary, which makes them more like a sweeping tone in a sigma delta. In my world this is the good news / bad news problem. Initially everything looks great but given enough product volume someone always finds a certain input stimuli which results in the sweeping tone being audible. My customer always DEMANDS a fix yesterday and a screen test for the “faulty” parts = VERY VERY bad news for volume production.
Enough of the rave, I’ve got real work to do..
Classdunce
Part of the difference is that in a sigma delta DAC your source is sampled at a certain frequency so any frequency error or jitter creates distortion relative to the sampled time of the source.
This makes a ClassD amp more like a Sigma delta ADA (ADC-DAC). The only experience that I have with ADA's was with a hearing aid that I designed a few years back. The output was a classD amp but the input was a classic 2nd order switched cap sigma delta. In this case the frequency had to be fixed because all the digital filters that post processed the sound assumed a fixed sample frequency.
The hand waving argument (for letting the carrier frequency vary) goes something like this:
Because your input sampling and output sampling have the same period there is no error for a perfect pulse centered PWM. If the carrier frequency is over 500Khz and centered than you can easily achieve -90dB 3rd harmonic for the digital PWM output, so the error is small anyway.
Trouble is it looks like **** after the filter, especially into a real speaker load (around -45dB 3rd). So the feedback term is basically the same as adding a precorrection signal of (input referred)amplitude -45dB to the input signal. If you do this analysis with the sample frequency fixed than the math is quite simple and the spectral compounds are easy to predict. However if you let the frequency vary than your simple equation becomes very complex because the energy of the feedback term is effectively multiplied by the frequency error term and added to the feedback term. When the Delta Freq and feedback error are both small errors (less than 0.1%), than the added compounds all fall below -110dB for a -3db input signal. The delta Freq term typically scales with output amplitude so at small signal the inharmonic spurs are below -110dB to -120dB (relative to full scale)
Unfortunately in cheap systems the Inductor filter is very non-linear and the supply variation is large with audio amplitude (up to 20%) in my case. Because of this you get large Frequency errors and large amplitude errors, so the FFT has inharmonic spurs all over the place at amplitudes of only -60dB. Fortunately they scale with signal amplitude so the "tones" are around -110dB for a -60dB signal.
What is interesting, is that I cant hear these "tones" my feeling is that this occurs because the tones are non-stationary, which makes them more like a sweeping tone in a sigma delta. In my world this is the good news / bad news problem. Initially everything looks great but given enough product volume someone always finds a certain input stimuli which results in the sweeping tone being audible. My customer always DEMANDS a fix yesterday and a screen test for the “faulty” parts = VERY VERY bad news for volume production.
Enough of the rave, I’ve got real work to do..
Classdunce
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.