I thought the biggest reason for much higher switching frequencies than double the passband was still to keep response degradation from the output filter from reaching all the way down into the middle of it.
At least that's what keeps me interested in higher frequencies.
One might safely figure that if the Nyquist frequency were good enough for audio reproduction Class D power amps would have started taking over sometime in the early 80's.
At least that's what keeps me interested in higher frequencies.
One might safely figure that if the Nyquist frequency were good enough for audio reproduction Class D power amps would have started taking over sometime in the early 80's.
Last edited:
Minimum needed sampling frequency (Nyquist Theorem) has pretty much been proven by standard CD audio (44.1kHz). The problem is the filtering schemes used to make it work don't apply to class D power, at least not unless you include oversampling (jacking up the switching frequency), and even CD players didn't work very well until that was done.
Last edited:
Meanwhile, we fix a point:
The bit rate is then 44,100 samples / second x 16 bits / sample x 2 = 1,411,200 bits / s or 1.4 Mbit / s. in case of stereo signal.
Relationship Between bit depth and dynamic range is, for Each Increase in 1-bit bit depth, the dynamic range will INCREASE by 6 dB. 24-bit digital audio has to Theoretical maximum dynamic range of 144 dB, Compared to 96 dB for 16-bit;
Is clare that depth is very important. @ 24bit Resolution is 16,777,216.
I continue?
The bit rate is then 44,100 samples / second x 16 bits / sample x 2 = 1,411,200 bits / s or 1.4 Mbit / s. in case of stereo signal.
Relationship Between bit depth and dynamic range is, for Each Increase in 1-bit bit depth, the dynamic range will INCREASE by 6 dB. 24-bit digital audio has to Theoretical maximum dynamic range of 144 dB, Compared to 96 dB for 16-bit;
Is clare that depth is very important. @ 24bit Resolution is 16,777,216.
I continue?
Not entirely sure where you're headed. Either clocked (with triangle) or self oscillating are still considered linear modulators, can introduce a switching point at Any time, depending on the signal.
The problem is not sampling rate, in theory and practice the information Is there if sampled at double speed. Its the reconstruction that practically requires nearly complete removal of switching artifacts or they wind up making wierd noises in your speaker at a frequency that can be heard. It's easier to do if the switching occurs quite a bit faster than the fastest signal. At bit rate frequencies there's a possibility of more problems due to generated switching noise, not to mention loss, than would be solved by filters with smaller values. Bit rate (not technically a rate in fixed frequency sampling), but bit frequency really has no relation to class D switching speed.
The problem is not sampling rate, in theory and practice the information Is there if sampled at double speed. Its the reconstruction that practically requires nearly complete removal of switching artifacts or they wind up making wierd noises in your speaker at a frequency that can be heard. It's easier to do if the switching occurs quite a bit faster than the fastest signal. At bit rate frequencies there's a possibility of more problems due to generated switching noise, not to mention loss, than would be solved by filters with smaller values. Bit rate (not technically a rate in fixed frequency sampling), but bit frequency really has no relation to class D switching speed.
Last edited:
The human hearing has a dynamic range about 140dB and stretches over 10 octaves; theres little we can´t hear. Of course we can hear the diff between 16/44.1k and 24/192k...
Also there is a difference between theoretical conception of soundwaves which must be the basis algorithm reconstructing a digitalized signal and the real signal which is modulated by many other frequencies -especially above the limit for human hearing.
Also there is a difference between theoretical conception of soundwaves which must be the basis algorithm reconstructing a digitalized signal and the real signal which is modulated by many other frequencies -especially above the limit for human hearing.
Hi,
the speech has gone out of the question.
I asked Eva to translate the clock of the modulator on a system with AD.
just to see if I do not know etc.
400-500KHz for the class D are sufficient given the poor interpolation.
I was explaining the problem (24 bit) is inherent in the definition of sound that, despite the 16 million values, you can not get.
2Vpp assumed you have a triangle wave. 1V then it is sufficient to bring the output to Vcc rail. in this case you put 8milioni levels within one volt range?
there's no sense the 24bit. apart from the exaggerated dynamics lost.
the speech has gone out of the question.
I asked Eva to translate the clock of the modulator on a system with AD.
just to see if I do not know etc.
400-500KHz for the class D are sufficient given the poor interpolation.
I was explaining the problem (24 bit) is inherent in the definition of sound that, despite the 16 million values, you can not get.
2Vpp assumed you have a triangle wave. 1V then it is sufficient to bring the output to Vcc rail. in this case you put 8milioni levels within one volt range?
there's no sense the 24bit. apart from the exaggerated dynamics lost.
AP2, what I mean is that your example does not make sense. Your sample audio signal has strong components above the clock frequency added on purpose to create non linearity, and we already know they are aliased back to the 0-fsw range. Removing these components more efficiently improves linearity more than increasing fsw. That's because the higher the fsw, the more critical timing errors from real-world circuits become.
Juhleren, phase shift and group delay are not distortion. Negative feedback can fully compensate for phase shift in the output filter making it appear as a constant delay across the audio band of a few tens of microseconds, which is harmless. However, linearity errors from a simple triangle wave modulation scheme (plus PSRR errors) can only be attenuated with negative feedback but not eliminated. On the other hand, non-linear modulators just prevent the non-linearity from happening in the first place. UcD style self-oscillating modulators are a good example: They give far better linearity than any triangle wave scheme operating at the same idle frequency, they compensate for power supply variations in a cycle per cycle basis, and they cancel output filter phase shift and peaking at the same time. Leaving the modulator free to place switching points at any time (rather than using a fixed clock) is part of the trick.
It's like the difference between modern CPUs and old ones. The newer ones can execute code faster than old CPUs running at the same Mhz clock because they can complete more instructions per clock cycle.
We already gave up on using higher Mhz clocks for CPUs and now are focused on doing more and more work per clock cycle and paralleling CPU cores, which also requires a re-thinking of software algorithms.
Same applies to class D. Going higher than 300-400khz is pointless. You can already do wonders at 300khz with the right modulation scheme.
Juhleren, phase shift and group delay are not distortion. Negative feedback can fully compensate for phase shift in the output filter making it appear as a constant delay across the audio band of a few tens of microseconds, which is harmless. However, linearity errors from a simple triangle wave modulation scheme (plus PSRR errors) can only be attenuated with negative feedback but not eliminated. On the other hand, non-linear modulators just prevent the non-linearity from happening in the first place. UcD style self-oscillating modulators are a good example: They give far better linearity than any triangle wave scheme operating at the same idle frequency, they compensate for power supply variations in a cycle per cycle basis, and they cancel output filter phase shift and peaking at the same time. Leaving the modulator free to place switching points at any time (rather than using a fixed clock) is part of the trick.
It's like the difference between modern CPUs and old ones. The newer ones can execute code faster than old CPUs running at the same Mhz clock because they can complete more instructions per clock cycle.
We already gave up on using higher Mhz clocks for CPUs and now are focused on doing more and more work per clock cycle and paralleling CPU cores, which also requires a re-thinking of software algorithms.
Same applies to class D. Going higher than 300-400khz is pointless. You can already do wonders at 300khz with the right modulation scheme.
Last edited:
Eva,Can you put me in the wrong. haha... this is fantastic!
Have I not interested now , in development of modulator scheme, TRW or UDC.
work on my amp and psu is finished, I'm working on 8 channels video (8 graphics processors "ARM").
well, independently from the amplifier, you manage to translate the example of the modulator to an AD converter? This was the question for you. I do not care if it is better TRW or UCD. (except that you know how I think about selfoscillant) Linearity very poor, I will not repeat the same things.
they needed me to explain all the amp? FBN etc?.
sometimes, you seem a locked disk.
Have I not interested now , in development of modulator scheme, TRW or UDC.
work on my amp and psu is finished, I'm working on 8 channels video (8 graphics processors "ARM").
well, independently from the amplifier, you manage to translate the example of the modulator to an AD converter? This was the question for you. I do not care if it is better TRW or UCD. (except that you know how I think about selfoscillant) Linearity very poor, I will not repeat the same things.
they needed me to explain all the amp? FBN etc?.
sometimes, you seem a locked disk.
AP2: You are lost again. We are neither talking about A/D nor about D/A conversion. We are talking about PWM.
Do I have to teach you Fourier series too? To demonstrate that the frequency content of 44100Hz Pulse Width Modulation is exactly perfect audio up to 22050Hz plus the 44100Hz carrier residual and its harmonics at n*44100Hz? And nothing else. No more components. No distortion. It could go directly to a voice coil if it wasn't for EMI.
This applies to UcD "variable frequency PWM" too, which is just delta sigma modulation, the approach that gives the best error cancellation, and the way the high performance A/D and D/A that you use work.
In the case of delta sigma A/D, variable frequency fixed-length*amplitude-product pulses are produced on demand to exactly match the average value of the input signal (feedback loop), then the pulses are counted to calculate their density and generate the digital word.
In the case of delta sigma PWM (UcD), variable frequency variable length pulses are produced on demand (accounting for supply voltage changes on a pulse by pulse basis) to exactly match the average of the input signal (feedback loop). Then the pulses are low-pass filtered (for EMI) and applied to the voice coil.
Do I have to teach you Fourier series too? To demonstrate that the frequency content of 44100Hz Pulse Width Modulation is exactly perfect audio up to 22050Hz plus the 44100Hz carrier residual and its harmonics at n*44100Hz? And nothing else. No more components. No distortion. It could go directly to a voice coil if it wasn't for EMI.
This applies to UcD "variable frequency PWM" too, which is just delta sigma modulation, the approach that gives the best error cancellation, and the way the high performance A/D and D/A that you use work.
In the case of delta sigma A/D, variable frequency fixed-length*amplitude-product pulses are produced on demand to exactly match the average value of the input signal (feedback loop), then the pulses are counted to calculate their density and generate the digital word.
In the case of delta sigma PWM (UcD), variable frequency variable length pulses are produced on demand (accounting for supply voltage changes on a pulse by pulse basis) to exactly match the average of the input signal (feedback loop). Then the pulses are low-pass filtered (for EMI) and applied to the voice coil.
Last edited:
Of course? Why 'of course'?
You say so, but can you provide a link to a blind study that shows it to be the case? Why not read the Wikipedia article on SACD. SACD was designed to exceed the performance of CD but no-one has shown a detectable difference in blind tests.
I would be very interested to see a study that shows people's hearing exceeds the range encompassed by CDs, and I have tried to find one without success.
w
>Do I have to teach you Fourier series too? To demonstrate that the frequency content of 44100Hz Pulse Width Modulation is exactly perfect audio up to 22050Hz plus the 44100Hz carrier residual and its harmonics at n*44100Hz? And nothing else. No more components. No distortion. It could go directly to a voice coil if it wasn't for EMI.
I think that's all probably true (assuming a theoretically perfect modulator), except the last sentence. Have you ever built and listened to a ~40 kHz PWM amp and listened to it? I built a small one that ran at 65k with just a bit of series inductor at the output and it had problems, though I didn't measure it, that sounded like intermodulation or subharmonics in the upper audio band that were extremely irritating. If you could really pull that off with exceptionally low distortion you could have brutally efficient full band high power amplifiers. Even with worries about EMI, if it actually worked you'd probably see active speakers where the amplifier wiring were only a few inches. But, it doesn't work, not for hi-fi anyway. All the "filterless" chips (circuits) on the market (which typically run in the 100's of kHz anyway) don't claim to get better than a tenth or so % distortion and far below maximum power at that.
I think that's all probably true (assuming a theoretically perfect modulator), except the last sentence. Have you ever built and listened to a ~40 kHz PWM amp and listened to it? I built a small one that ran at 65k with just a bit of series inductor at the output and it had problems, though I didn't measure it, that sounded like intermodulation or subharmonics in the upper audio band that were extremely irritating. If you could really pull that off with exceptionally low distortion you could have brutally efficient full band high power amplifiers. Even with worries about EMI, if it actually worked you'd probably see active speakers where the amplifier wiring were only a few inches. But, it doesn't work, not for hi-fi anyway. All the "filterless" chips (circuits) on the market (which typically run in the 100's of kHz anyway) don't claim to get better than a tenth or so % distortion and far below maximum power at that.
Last edited:
What you experienced is self disturbance due to EMI 
That's why an output filter is required, but the purpose is to attenuate EMI gently, not to completely eliminate carrier residual (benign thing). Note that in my experience preventing self disturbance in class D is not a trivial task even with plenty of output filtering. Also, how did you generate the PWM timing? It's not a trivial task either, I suppose it was open loop. btw: Yes, when I transitioned from sinewave inverters into class D I did some 65khz experiments.
And yes, I'm developing a series of class D modules for powered speakers with the highest power density and efficiency currently on the market, and they are going to be unrivaled for a long time (until some archaic designer minds finally evolve, so I'm actually hurting myself when I push them to evolve). The subwoofer version is already being produced and sold. I can't give more details due to non-disclosure agreements, I can only say that home theater guys are loving it.
That's why an output filter is required, but the purpose is to attenuate EMI gently, not to completely eliminate carrier residual (benign thing). Note that in my experience preventing self disturbance in class D is not a trivial task even with plenty of output filtering. Also, how did you generate the PWM timing? It's not a trivial task either, I suppose it was open loop. btw: Yes, when I transitioned from sinewave inverters into class D I did some 65khz experiments.And yes, I'm developing a series of class D modules for powered speakers with the highest power density and efficiency currently on the market, and they are going to be unrivaled for a long time (until some archaic designer minds finally evolve, so I'm actually hurting myself when I push them to evolve). The subwoofer version is already being produced and sold. I can't give more details due to non-disclosure agreements, I can only say that home theater guys are loving it.
Last edited:
I used a crystal oscillator divided down and buffered into a switched complimentary current source to generate the triangle, used an AD790 for the comparator. Yes it was open loop, but I noticed the ratty sound went away with a real output filter and the switching frequency rasied above 300kHz. I started playing with class D protos on a breadboard and quickly found it impossible to do much without soldering to a blank board (ground plane) dead bug style. I have no doubt that an excellent woofer amp can be made with any switching frequency slightly above the audio band, but honestly for tweeters I haven't been able to do anything that beats class A output circuits and lots of negative feedback. I've been somewhat captivated by the performance of the new EPC devices for their speed capability and have been trying to design around those because they make the driver stage much easier. But if you can make an switch amp run so slow and cover tweeter frequencies very well it would, without a doubt, change how people think and do business in audio amps. It might even have an adverse effect on the advance of switching power transistors.
Hi,
Eva,is incredible, even if my English is not good.
I asked you to simply translate (only the clock). not the "PWM" function.
just to see if you understood, What really are the triangular 330Khz.
ok, I give up, do not want to know.
For the rest, I think you love UCD scheme only because all trim with FBN instead of solving problems and therefore know more deeply.
This is my opinion. Or show me that your dear UCD, does not increase the THD in proportion to the percentage of modulation.
is not acceptable 0.1% or greater than 75% modulation. (if an audiophile amplifier area).
goes round and round we go to the same point.
I agree that is enough for a good resolution 330Khz.
Regards
Eva,is incredible, even if my English is not good.
I asked you to simply translate (only the clock). not the "PWM" function.
just to see if you understood, What really are the triangular 330Khz.
ok, I give up, do not want to know.
For the rest, I think you love UCD scheme only because all trim with FBN instead of solving problems and therefore know more deeply.
This is my opinion. Or show me that your dear UCD, does not increase the THD in proportion to the percentage of modulation.
is not acceptable 0.1% or greater than 75% modulation. (if an audiophile amplifier area).
goes round and round we go to the same point.
I agree that is enough for a good resolution 330Khz.
Regards
I sold all of the Class D amplilfiers that I had simply because they can't compete with Class A or even a Class A/B design. For non believers, I had an office system with either Bel Canto, Audio Zone or NuForce amps and a AVI Class A/B Integrated. All I had to do was play the any of the Class D amps for a bit and the person auditioning the system always thought that it sounded great until I hooked up the AVI. It's like night and day. Class D does not have the capability to reproduce sound the way a Class A or A/B does. I my opinion.
Jeff
Jeff
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.