Zetex DDFA

iand

Member
2007-02-05 10:58 pm
London
Eric_R said:
Any comments on this new chipset?
Digital class-D with feedback, http://www.zetex.com/audio/audio06.asp

Looks very good for active speakers with the built in cross-over.
Are any products ready using this?

/Eric

I can't name names because of confidentiality but I know Zetex have got DDFA customers signed up for mass production this year.

From what I heard of their prototypes the sound is excellent, and everyone who's heard them seems to agree with this.

I do know that the measured performance is pretty astounding with noise and distortion more than 110dB down from full scale -- remember this is digital in not just a power amp, so it includes the functionality of a DAC (and DSP) as well.

One of their target markets is indeed active speakers, the DDFA chip includes everything needed to do all this digitally (crossover, equalisation, limiting, fault protection...).

The big issue with DDFA for high-power amplifiers (like PA which I'm interested in but is not their main target area) is the high switching speed (844kHz -- needed for digital noise shaping) which makes it very difficult to get high efficiency.

However I've been looking into this in some detail and reckon that by using fast-switching devices (+/-190V half-bridge using CoolMOS CS + series Schottky + SiC reverse diode) the loss can be kept down to a few percent even at this speed -- though it's certainly not at all easy switching 35A through 380V in less than 10ns it can be done, but extreme care is needed to avoid di/dt and dv/dt problems (nice radio transmitter...)

Ian
 
You can find it at http://www.diyaudio.com/forums/showthread.php?postid=1583852#post1583852 .
The problem comes from the infinite spectra of the PWM is being sampled by the digital amplifier on its output. It can cause spectral overlap, which appears as distortion because the distances of the non-zero energy frequencies beside the carrier harmonics are integral multiples of the modulating signal's frequecy.
 

iand

Member
2007-02-05 10:58 pm
London
Gyula said:
You can find it at http://www.diyaudio.com/forums/showthread.php?postid=1583852#post1583852 .
The problem comes from the infinite spectra of the PWM is being sampled by the digital amplifier on its output. It can cause spectral overlap, which appears as distortion because the distances of the non-zero energy frequencies beside the carrier harmonics are integral multiples of the modulating signal's frequecy.

You don't understand how all-digital PWM works in circuits like this. The digital sample-value-to-pulsewidth conversion (with about 9ns time resolution) produces quantisation error which is then noise-shaped to push it out of the audio band, but there's no intermodulation or aliasing of PWM harmonics because the input signal is already sampled and all processing is discrete-time -- there are no "infinite bandwidth" signals.

The difference between the amplifier output and an "ideal" reference DAC (<-120dB THD+N) is integrated and fed back into the PWM modulator -- this is equivalent to wideband negative feedback in an analogue amplifier.

The total noise+distortion at the DDFA amp output -- including all harmonics and spurious components -- is better than most standalone DACs, so the power amplifier function adds little or no signal degradation compared to a simple DAC.

Ian
 
You don't understand how all-digital PWM works in circuits like this. The digital sample-value-to-pulsewidth conversion (with about 9ns time resolution) produces quantisation error which is then noise-shaped to push it out of the audio band, but there's no intermodulation or aliasing of PWM harmonics because the input signal is already sampled and all processing is discrete-time -- there are no "infinite bandwidth" signals.

I suggest you to learn the very basics of describing the signals in frequency domain. Then consider that even a discrete-time signal has infinite spectra. Btw. what do you think about for example why a ZOH is needed at the output of a simple DAC?

After that I think you should make some calculation of the spectra of a non-D.C. modulated PWM signal. For example what do you think why is a Low-pass filter needed at the output of Class-D?

Please tell me how can you expect linear operation from a non-linear process?

Have you ever heard about that sampling has to be done on bandwith-limited input signal?

Have you ever watched the sway on a filtered PWM output? Do you think if you sample that then it has to be equal to the input samples only because it's sampled?

Is the sampling a wonder or some kind of magic in your opinion? Or what?
 

iand

Member
2007-02-05 10:58 pm
London
Gyula said:


I suggest you to learn the very basics of describing the signals in frequency domain. Then consider that even a discrete-time signal has infinite spectra. Btw. what do you think about for example why a ZOH is needed at the output of a simple DAC?

After that I think you should make some calculation of the spectra of a non-D.C. modulated PWM signal. For example what do you think why is a Low-pass filter needed at the output of Class-D?

Please tell me how can you expect linear operation from a non-linear process?

Have you ever heard about that sampling has to be done on bandwith-limited input signal?

Have you ever watched the sway on a filtered PWM output? Do you think if you sample that then it has to be equal to the input samples only because it's sampled?

Is the sampling a wonder or some kind of magic in your opinion? Or what?

I know exactly how PWM works, but this is a sampled-data digital system *not* an analogue PWM system which can suffer from intermodulation sidebands.

The input is sampled digital data, this is upconverted and then converted to a discrete-time quantised PWM signal where the location of all the sample edges falls on a 108MHz grid. Any quantisation error due to this time resolution is then noise-shaped so it appears above the audio band.

If you look at the spectrum of the result, it's almost identical to a noise-shaped non-PWM sigma-delta DAC -- which is where the analogue signal input to a conventional power amplifier would most likely come from. The output needs lowpass filtering in exactly the same way that a conventional DAC does and for the same reasons. There are no nonlinear effects from the fast risetime edges any more than there are in the stepped output from a conventional DAC.

Before suggesting I learn more about such things you might like to consider that I've spent the last twenty years or so designing complex mixed-signal ICs, including some of the highest performance ADCs and DACs on the planet -- and I'm talking *much* higher performance than most audio here, with jitter down to a fraction of a picosecond and sampling rates up to tens of GHz...
 
So if I understand correctly, I need to buy two zetex IC's ZXCZM200 and ZXCZM800 (TQFP 176 pins), next I need 2 mosfet driver IC's, e.g. irs20957 or similar, and 4 mosfets (+ coils, caps etc) for each channel?
Maybe problem is the complexity and cost, I'm not sure if possible real massproduction of such design, especially nowadays, however, it's ok for highend stuff.
 
IVX said:
So if I understand correctly, I need to buy two zetex IC's ZXCZM200 and ZXCZM800 (TQFP 176 pins), next I need 2 mosfet driver IC's, e.g. irs20957 or similar, and 4 mosfets (+ coils, caps etc) for each channel?
Maybe problem is the complexity and cost, I'm not sure if possible real massproduction of such design, especially nowadays, however, it's ok for highend stuff.

I dont see thats as a problem. Panasonic did it with Ti chips. I think they are looking for a big compani that will put it into mass produktion. Nad is not a big company.
 
iand said:


I know exactly how PWM works, but this is a sampled-data digital system *not* an analogue PWM system which can suffer from intermodulation sidebands.

The input is sampled digital data, this is upconverted and then converted to a discrete-time quantised PWM signal where the location of all the sample edges falls on a 108MHz grid. Any quantisation error due to this time resolution is then noise-shaped so it appears above the audio band.

If you look at the spectrum of the result, it's almost identical to a noise-shaped non-PWM sigma-delta DAC -- which is where the analogue signal input to a conventional power amplifier would most likely come from. The output needs lowpass filtering in exactly the same way that a conventional DAC does and for the same reasons. There are no nonlinear effects from the fast risetime edges any more than there are in the stepped output from a conventional DAC.

Before suggesting I learn more about such things you might like to consider that I've spent the last twenty years or so designing complex mixed-signal ICs, including some of the highest performance ADCs and DACs on the planet -- and I'm talking *much* higher performance than most audio here, with jitter down to a fraction of a picosecond and sampling rates up to tens of GHz...

iand!

I have already designed control for mixed-signal systems for many purposes, even for digital amplifiers. And moreover I can model these kinds of systems analitically. I have also reached very impressive results of THD for DDFA, e.g.: no harmonics above the sub-120 dB noise floor, and I exactly know how can it be reached. So I state the Zetex chip can do that but not on that way you think. And that way is far beyond of taking some periodic samples and fed a digital counter acting as a comparator with them to get some less-resolution PWM, and apply noise-shaping to avoid low SNR.
Your statement that these systems are all-digital amplifiers is not true. When you said there are no sidebands you thought it because you considered it to be a purely digital system. I have also designed controls for all-digital processes. That is a completely different field because there the spectra is periodic and you have to take care about the range of [-Fs/2;Fs/2]. E.g. this is why a low-pass filter is needed at the input of every A/D performing baseband sampling. But the Digital Amplifiers are NOT all-digital systems. Your statement would be true if you could make dirac-impulses in discrete time to the speaker instead of PWM and could sample them with an A/D. Signals in discrete time systems have values only in discrete times.
I know exactly how PWM works, but this is a sampled-data digital system *not* an analogue PWM system which can suffer from intermodulation sidebands.
PWM will only have less resolution with limited time-resolution this is why noise-shaping is needed. Used with noise-shaping it only approximates the analog PWM with hold at its input. There will be no black magic if you feed a less-resolution PWM with periodically sampled signal. The result will be a distorted, higher-noise output in the baseband, and carrier harmonics with their sidebands above. If you state something you should prove that. You wrote you are senior in Chip Design (not in DSP design). could you show us some analitical calculations on the method of carrier and sideband-less PWM you have stated ? I think it could cause a new industrial revolution because then every D/A should be converted to PWM because it's efficiency and unbeatable quality. And you could prove that square-wave = sine-wave, which could be more hard to the reality. This Zetex chip is not a part of a Matrix scene. But I guess it should be if it could operate with this performance on the way you stated. If I tell you that you will meet with the messiah tomorrow, I think it would not be true, so I don't tell. I don't know, I don't tell, I don't lie only for a hype.
The difference between the amplifier output and an "ideal" reference DAC (<-120dB THD+N) is integrated and fed back into the PWM modulator -- this is equivalent to wideband negative feedback in an analogue amplifier.
This completely means nothing except the error signal can be sampled with better resolution because of the analog substraction. This is because there are no twenty-some bit low-delay A/Ds with few-hundred kHz sampling frequency capability. The "ideal" DAC doesn't produce the harmonics caused by the PWM out-of-baseband components so those are driven with the subraction to the feedback A/D. There's something that you didn't find and a mixed-signal system is not an all-digital system.

So could I please you show us some provement of the digitally controlled PWM comes without carrier components and sidebands as you stated? And could you explain how can a Direct Digital Feedback Amplifier (DDFA) be an all-digital system? Do you know what does analog and digital terms mean, what is the difference?

P.S.: The financial crysis is estimated to be over in this autumn so I think Zetex will have more orders.
 

iand

Member
2007-02-05 10:58 pm
London
Gyula said:


iand!

I have already designed control for mixed-signal systems for many purposes, even for digital amplifiers. And moreover I can model these kinds of systems analitically. I have also reached very impressive results of THD for DDFA, e.g.: no harmonics above the sub-120 dB noise floor, and I exactly know how can it be reached. So I state the Zetex chip can do that but not on that way you think. And that way is far beyond of taking some periodic samples and fed a digital counter acting as a comparator with them to get some less-resolution PWM, and apply noise-shaping to avoid low SNR.
Your statement that these systems are all-digital amplifiers is not true. When you said there are no sidebands you thought it because you considered it to be a purely digital system. I have also designed controls for all-digital processes. That is a completely different field because there the spectra is periodic and you have to take care about the range of [-Fs/2;Fs/2]. E.g. this is why a low-pass filter is needed at the input of every A/D performing baseband sampling. But the Digital Amplifiers are NOT all-digital systems. Your statement would be true if you could make dirac-impulses in discrete time to the speaker instead of PWM and could sample them with an A/D. Signals in discrete time systems have values only in discrete times.

PWM will only have less resolution with limited time-resolution this is why noise-shaping is needed. Used with noise-shaping it only approximates the analog PWM with hold at its input. There will be no black magic if you feed a less-resolution PWM with periodically sampled signal. The result will be a distorted, higher-noise output in the baseband, and carrier harmonics with their sidebands above. If you state something you should prove that. You wrote you are senior in Chip Design (not in DSP design). could you show us some analitical calculations on the method of carrier and sideband-less PWM you have stated ? I think it could cause a new industrial revolution because then every D/A should be converted to PWM because it's efficiency and unbeatable quality. And you could prove that square-wave = sine-wave, which could be more hard to the reality. This Zetex chip is not a part of a Matrix scene. But I guess it should be if it could operate with this performance on the way you stated. If I tell you that you will meet with the messiah tomorrow, I think it would not be true, so I don't tell. I don't know, I don't tell, I don't lie only for a hype.

This completely means nothing except the error signal can be sampled with better resolution because of the analog substraction. This is because there are no twenty-some bit low-delay A/Ds with few-hundred kHz sampling frequency capability. The "ideal" DAC doesn't produce the harmonics caused by the PWM out-of-baseband components so those are driven with the subraction to the feedback A/D. There's something that you didn't find and a mixed-signal system is not an all-digital system.

So could I please you show us some provement of the digitally controlled PWM comes without carrier components and sidebands as you stated? And could you explain how can a Direct Digital Feedback Amplifier (DDFA) be an all-digital system? Do you know what does analog and digital terms mean, what is the difference?

P.S.: The financial crysis is estimated to be over in this autumn so I think Zetex will have more orders.

I didn't say that the DDFA is an all-digital amplifier, there's no such thing when the final output is analogue. What I meant was that the input is digital and all the signal processing and PWM conversion is digital discrete-time, the final PWM output stage and the error feedback ADC (see below) are obviously analogue.

In a discrete-time system like the Zetex DDFA the ideal values of all the outputs are known at the clock frequency (108MHz); the modulator takes a discrete-time digital sampled-data input at the input rate (e.g. 44.1kHz, 48khz, 88.2kHz, 96khz...) and first sample rate converts this to the PWM frequency (about 850kHz). This high-resolution 850kHz signal is then converted to a 108MHz PWM signal (128 possible pulse widths per PWM cycle) and the resulting quantising error is fed back into the noise shaper.

Mathematically this is the same as an oversampled 7-bit sigma-delta DAC running at 850kHz, except that the 7 bit amplitude control for each pulse is done by a PWM signal not a PAM signal (multibit DAC); the area under each pulse and its time position are exactly known, assuming a perfect output stage. The noise shaping pushes the inband quantisation noise down to well below -120dB, just like any other high-resolution oversampled DAC.

Of course the signal is still a PWM signal at 850kHz not a sample-and held PAM signal, but the spectrum inband (up to 20kHz) is identical -- see the plots on the Zetex website. And after an analogue LPF the waveform looks analogue too, just like a single-bit sigma-delta DAC after filtering.

So unless you're saying that all DACs can't produce an analogue output after a reconstruction filter -- which would be a pity, since nearly all DACs on the market do exactly that -- then the DDFA PWM output after filtering is equivalent to a conventional DAC. If you don't believe this I suggest you take it up with Mr. Nyquist...

You don't need a 20+ bit ADC with hundreds of kHz sampling rate for the feedback digitiser, because this only digitises the integrated error between the actual power stage PWM output and the "ideal" PWM reference DAC.

The analogue error integration (in the analogue chips) means that any edge distortion effects are included, there's no lower-rate sampling as such in the ADC which can cause aliasing effects, the totla area under each pulse is what is being measured. This feedback is indeed analogue, but doesn't limit the performance because only the small residual error is converted.

The digital circuits then predistort the input to the PWM modulator which drives the actual power output stage to compensate for the timing and amplitude errors in it, such that it has the same performance as the reference DAC and the error signal tends to zero.

The end result is an amplifier with performance at the speaker output pretty much as good as the best DACs available have at the DAC output -- not my opinion here, this has been demonstrated by both measurements and listening tests. In fact, Zetex are also applying just the modulator chip as an 8-channel DAC, with performance up there with the best on the market.

I'm not guessing here about how this works, I designed the low-jitter (picoseconds) clock and PWM output stages that produce this reference PWM DAC output :)
 

iand

Member
2007-02-05 10:58 pm
London
IVX said:
So if I understand correctly, I need to buy two zetex IC's ZXCZM200 and ZXCZM800 (TQFP 176 pins), next I need 2 mosfet driver IC's, e.g. irs20957 or similar, and 4 mosfets (+ coils, caps etc) for each channel?
Maybe problem is the complexity and cost, I'm not sure if possible real massproduction of such design, especially nowadays, however, it's ok for highend stuff.

If this was the case I doubt that the big interest in DDFA would be in the home cinema market in the Far East, which is not exactly insensitive to price.

I guess that the added cost in the signal path (multiple chips, drivers, MOSFETs) is more than compensated for by the reduction in power supply cost because of the very high PSRR over the entire audio bandwidth.

Also bear in mind that the integrated driver/MOSFET chips are quite cheap, but separate drivers and MOSFETS can be even cheaper due to the optimised processes which can be used for each -- more bits, but even lower cost if that's your main drive.
 
I guess that the added cost in the signal path (multiple chips, drivers, MOSFETs) is more than compensated for by the reduction in power supply cost because of the very high PSRR over the entire audio bandwidth.

From the way you describe the working principle of this modulator/error-correction topology one can probably not go too cheap with the supply or the PSRR will suddenly drop significantly.

Regards

Charles