Aha. First you needed 2us. Now 90ps is not enough.
The 90ps figure of course is a simplification and does not take into account the effect of dither. Once you do that, the "timing resolution" of a 44.1/16 channel becomes fully (totally, etc) equivalent to that of an analog channel of the same bandwidth and SNR, quite regardless of whether sampling has been employed or whether the noise is due to quantisation or from an analogue source. Such is the beauty of nyquist/shannon.
Without dither, taking a phase measurement over multiple cycles is not guaranteed to get you more precise phase information. With dither applied, phase/frequency accuracy scale with measurement time in precisely the same way as with analogue noise.
I cannot recall arguing that a 20kHz/93dB (not 98 or 96, tri pdf dither) channel is sufficient. You may remember I added quite explicitly that the 20kHz audible limit is an oversimplification.
Besides, I couldn't argue that a channel brickwall limited to 20kHz with a 93dB SNR is enough. I am not deaf. On the other hand, the full extent of the sonic capabilities of such a channel are not to be underestimated, if done properly.
The paper by my kind competitor (I myself am at www.grimmaudio.com ) deals with how the additional bandwith/snr of high resolution audio can improve on basic red book, and whether it is such a good idea to try and keep the frequency response flat up to within an inch of nyquist.
What I am pointing out, however, is that the timing precision of a channel is not equal to one over sampling rate as you originally contended.
I am glad you too have found this out in the meantime, and agree that 500kHz is by no means a prerequisite for obtaining substantially better than 2us timing resolution.
Your changing horses from "one over sampling rate" to the "redbook or not" debate is telling. I presume you are NOT one of those persons with the unpleasant habit of changing subjects until they finally find one where the respondent fails to answer correctly, only to proclaim victory over the entire "dispute". You will understand that most disputes are communication errors in disguise. Well-informed people rarely disagree truly, but they are often in disagreement on how to formulate the thing which they are in agreement on.
Back to audio.
It should be pointed out that the 2us figure concerned a static interchannel delay error. This corresponds to a lateral displacement of a virtual source. Armed with knowledge of speed of sound, ear-to-ear distance and basic triangulation, it is left to the reader to determine what fraction of a degree this is, and how steady the listener will have had to hold his head in order to discern it.
Because of that, I believe your concerns are about dynamical errors, not statical ones. Quite different rules govern dynamical errors. The temporal precision (of single-shot events) was obtained based on the bandwidth (more precisely, frequency response) and SNR of the channel. The same temporal precision therefore calculates back simply into the same bandwidth and noise. Apart from the noise, a perfect nq/sh sampling and quantisation system has no other error mechanisms.
In reality, serious sonic degradation is to be had from sampling jitter and rounding errors in arithmetic. These are execution matters, not to be confused with the core of the debate, namely bandwidth and SNR.
Again, having heard the differences myself I am the least likely person to proclaim 20kHz and 93dB are entirely sufficient. However, once it has been determined what bandwidth and SNR are sufficient, we can simply build ourselves this channel using enough bits and samples per second as dictated by nq/sh.
I understand you have only been confronted with poorly designed class D amplifiers.
For your interest, the problem with the audio test set lies in HF demodulation in the input stages of the test set, and is indicative of poor EMC on the part of the amplifier. Additionally, APs not fitted with a pre-filter (such as the AES17 filter) encounter problems with the autoranging logic which forces the distortion detector to operate on the last few bits of the AD that sits after the RMS detector.
The cycle to cycle jitter that you see is probably due to the use of noise shaping, either by having a discrete-time control loop (Class T or the recent Yamaha MXD1) or by the use of digital PWM. Both normally place the unwanted noise outside the audio band. That this does not necessarily cause audible problems is demonstrated by noise shaped conversion systems like DSD.
Why this is so, can again be understood by seeing the transmission channel as something having a given frequency response and a given noise floor (which is not exactly white in DSD's case).
In the case of digital PWM amplifiers (ie Power DACs), I am not very enthousiastic about these either. Their total lack of error control make them colouration machines, not amplifiers worth the name. Worse still, some of the carefully shaped noise is folded back by current induced timing errors in the power stage. The existence of such devices is technically unwarranted. I think I have the right to say this, having built the best-ever in its class (PPDSD).
A good, clean analogue-controlled class D amplifier (I shan't blow my own horn, the other class D enthousiasts on DIYAudio do so already) does not have the jitters. Whatever jitter there is, comes straight from the input noise of the comparator and from opamp noise in the control loop.
With the proper tools, their audio performance can be shown to have no specific shortcomings that would set them apart from linear amplifiers. As for the sound, give me a ring when you're in Belgium, or get a pair of Hypex modules (apologies for the plug).
The 90ps figure of course is a simplification and does not take into account the effect of dither. Once you do that, the "timing resolution" of a 44.1/16 channel becomes fully (totally, etc) equivalent to that of an analog channel of the same bandwidth and SNR, quite regardless of whether sampling has been employed or whether the noise is due to quantisation or from an analogue source. Such is the beauty of nyquist/shannon.
Without dither, taking a phase measurement over multiple cycles is not guaranteed to get you more precise phase information. With dither applied, phase/frequency accuracy scale with measurement time in precisely the same way as with analogue noise.
I cannot recall arguing that a 20kHz/93dB (not 98 or 96, tri pdf dither) channel is sufficient. You may remember I added quite explicitly that the 20kHz audible limit is an oversimplification.
Besides, I couldn't argue that a channel brickwall limited to 20kHz with a 93dB SNR is enough. I am not deaf. On the other hand, the full extent of the sonic capabilities of such a channel are not to be underestimated, if done properly.
The paper by my kind competitor (I myself am at www.grimmaudio.com ) deals with how the additional bandwith/snr of high resolution audio can improve on basic red book, and whether it is such a good idea to try and keep the frequency response flat up to within an inch of nyquist.
What I am pointing out, however, is that the timing precision of a channel is not equal to one over sampling rate as you originally contended.
I am glad you too have found this out in the meantime, and agree that 500kHz is by no means a prerequisite for obtaining substantially better than 2us timing resolution.
Your changing horses from "one over sampling rate" to the "redbook or not" debate is telling. I presume you are NOT one of those persons with the unpleasant habit of changing subjects until they finally find one where the respondent fails to answer correctly, only to proclaim victory over the entire "dispute". You will understand that most disputes are communication errors in disguise. Well-informed people rarely disagree truly, but they are often in disagreement on how to formulate the thing which they are in agreement on.
Back to audio.
It should be pointed out that the 2us figure concerned a static interchannel delay error. This corresponds to a lateral displacement of a virtual source. Armed with knowledge of speed of sound, ear-to-ear distance and basic triangulation, it is left to the reader to determine what fraction of a degree this is, and how steady the listener will have had to hold his head in order to discern it.
Because of that, I believe your concerns are about dynamical errors, not statical ones. Quite different rules govern dynamical errors. The temporal precision (of single-shot events) was obtained based on the bandwidth (more precisely, frequency response) and SNR of the channel. The same temporal precision therefore calculates back simply into the same bandwidth and noise. Apart from the noise, a perfect nq/sh sampling and quantisation system has no other error mechanisms.
In reality, serious sonic degradation is to be had from sampling jitter and rounding errors in arithmetic. These are execution matters, not to be confused with the core of the debate, namely bandwidth and SNR.
Again, having heard the differences myself I am the least likely person to proclaim 20kHz and 93dB are entirely sufficient. However, once it has been determined what bandwidth and SNR are sufficient, we can simply build ourselves this channel using enough bits and samples per second as dictated by nq/sh.
I understand you have only been confronted with poorly designed class D amplifiers.
For your interest, the problem with the audio test set lies in HF demodulation in the input stages of the test set, and is indicative of poor EMC on the part of the amplifier. Additionally, APs not fitted with a pre-filter (such as the AES17 filter) encounter problems with the autoranging logic which forces the distortion detector to operate on the last few bits of the AD that sits after the RMS detector.
The cycle to cycle jitter that you see is probably due to the use of noise shaping, either by having a discrete-time control loop (Class T or the recent Yamaha MXD1) or by the use of digital PWM. Both normally place the unwanted noise outside the audio band. That this does not necessarily cause audible problems is demonstrated by noise shaped conversion systems like DSD.
Why this is so, can again be understood by seeing the transmission channel as something having a given frequency response and a given noise floor (which is not exactly white in DSD's case).
In the case of digital PWM amplifiers (ie Power DACs), I am not very enthousiastic about these either. Their total lack of error control make them colouration machines, not amplifiers worth the name. Worse still, some of the carefully shaped noise is folded back by current induced timing errors in the power stage. The existence of such devices is technically unwarranted. I think I have the right to say this, having built the best-ever in its class (PPDSD).
A good, clean analogue-controlled class D amplifier (I shan't blow my own horn, the other class D enthousiasts on DIYAudio do so already) does not have the jitters. Whatever jitter there is, comes straight from the input noise of the comparator and from opamp noise in the control loop.
With the proper tools, their audio performance can be shown to have no specific shortcomings that would set them apart from linear amplifiers. As for the sound, give me a ring when you're in Belgium, or get a pair of Hypex modules (apologies for the plug).
Thanks for your detailed response. As someone who doesn't care to develop anything too close to what is already marketed, I'm becoming interested in the idea of building a class D amp using feedback ideas such as you propose implemented in part with vacuum state devices, and perhaps switching with soic low charge injection mosfets to boost the switch frequency as high as practicable - I don't mind giving up a little efficiency if I can minimize signal impulse overshoot.
Hi,
(Actually this would be a good thing for a new thread)
Vacuum devices are nice in that they potentially switch very, very fast. For that reason I've always been curious to know if this theoretical advantage could be materialised in a practical circuit. I hope you will be the one to finally push that through.
Just a few thoughts:
Use of tubes in class D has been limited to anode supplies in AM transmitters (1-quadrant operation). The most obvious problem is that a conducting tube is still a diode, which means you'll need to add a diode in antiparallel to it. Class D amps for loudspeaker use operate in all four quadrants, so the power devices operate in reverse conduction nearly half the time.
If the nonlinearity produced around zero anode current is small wrt the rail voltage, impact on THD is low, but given that even a 6550 "fully on" has around 100 ohms plate resistance, care will have to be taken to match plate resistance with the resistance of the antiparallel diode.
In short, a number of problems are certain to pop up, some tough but none should be impossible to solve.
On the other hand, it would be informative to have a look at the "state of the art" (ie ucd and mueta). None of these have overshoot, because their frequency response is very benign (no bumps, no sharp corners). Before that I too thought that decent audio performance could only be had at very high switching frequencies.
In any case, good luck
(Actually this would be a good thing for a new thread)
Vacuum devices are nice in that they potentially switch very, very fast. For that reason I've always been curious to know if this theoretical advantage could be materialised in a practical circuit. I hope you will be the one to finally push that through.
Just a few thoughts:
Use of tubes in class D has been limited to anode supplies in AM transmitters (1-quadrant operation). The most obvious problem is that a conducting tube is still a diode, which means you'll need to add a diode in antiparallel to it. Class D amps for loudspeaker use operate in all four quadrants, so the power devices operate in reverse conduction nearly half the time.
If the nonlinearity produced around zero anode current is small wrt the rail voltage, impact on THD is low, but given that even a 6550 "fully on" has around 100 ohms plate resistance, care will have to be taken to match plate resistance with the resistance of the antiparallel diode.
In short, a number of problems are certain to pop up, some tough but none should be impossible to solve.
On the other hand, it would be informative to have a look at the "state of the art" (ie ucd and mueta). None of these have overshoot, because their frequency response is very benign (no bumps, no sharp corners). Before that I too thought that decent audio performance could only be had at very high switching frequencies.
In any case, good luck
It won't let me vote, so I'll reply here to show my support.
I am interested in class-D, but have not got around to attempting to design one yet. As an engineer I am of course always looking to improve efficiency, and something which can theoretically reach 100% seems... desirable.
One thing that strikes me is that using a PSU seems redundant. Could an off-line SMPS and class-D amplifier not be combined? If it's worth doing then I expect someone has already done it.
I am interested in class-D, but have not got around to attempting to design one yet. As an engineer I am of course always looking to improve efficiency, and something which can theoretically reach 100% seems... desirable.
One thing that strikes me is that using a PSU seems redundant. Could an off-line SMPS and class-D amplifier not be combined? If it's worth doing then I expect someone has already done it.
Hi, thanks!
http://www.diyaudio.com/forums/showthread.php?threadid=14847&highlight=
http://www.diyaudio.com/forums/showthread.php?threadid=25005&highlight=
http://www.diyaudio.com/forums/showthread.php?threadid=36349&highlight=
Can we have an anolog digital amplification forum now?
Just kidding.. I'd vote for "Class D" as a title though.
Cheers!
(deleted the one link it was already recommended)
http://www.diyaudio.com/forums/showthread.php?threadid=14847&highlight=
http://www.diyaudio.com/forums/showthread.php?threadid=25005&highlight=
http://www.diyaudio.com/forums/showthread.php?threadid=36349&highlight=
Can we have an anolog digital amplification forum now?
Just kidding.. I'd vote for "Class D" as a title though.
Cheers!
(deleted the one link it was already recommended)
Hi!
Thanks for the new forum ! Great !
Please updated this page:http://www.diyaudio.com/ ;-)
Even Mr Pass states that most of the amps dicussed here are not "digital"!
Why not just plainly: "Switchmode", SMPS could go in there also !
/
edit: OK- Maybe it dosn't qualify to be on the front page yet?
Thanks for the new forum ! Great !
Please updated this page:http://www.diyaudio.com/ ;-)
Even Mr Pass states that most of the amps dicussed here are not "digital"!
Why not just plainly: "Switchmode", SMPS could go in there also !
/
edit: OK- Maybe it dosn't qualify to be on the front page yet?
Hello again,
Let's not forget this one
http://www.diyaudio.com/forums/showthread.php?s=&threadid=40388
About the title.. I don't really care what it's called but is "Class T" not more of a marketing scheme than an actual class of amplifier, and so should it not be removed?
Thanks,
Chris
Let's not forget this one
http://www.diyaudio.com/forums/showthread.php?s=&threadid=40388
About the title.. I don't really care what it's called but is "Class T" not more of a marketing scheme than an actual class of amplifier, and so should it not be removed?
Thanks,
Chris
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.