Audibility of DAC Phase Noise.

[Blumlein 88]

No, it's not sarcasm. ChatGPT doesn't write sarcasm.

Oh so now you come in with the sarcasm.

Here is the response from Perplexity.ai about ChatGPT.

Conclusion​

While ChatGPT can certainly produce sarcastic remarks, the effectiveness and clarity of sarcasm can vary based on context and user interaction. If you want to see sarcasm in action, feel free to provide a scenario, and I can whip up a sarcastic response for you!

 

[Beave]

Hmm, that makes me wonder what each AI system would say about the other ones when queried. Could we get them to trash talk each other? Make insinuations of impropriety?

 

[Petrushka]

I've owned a couple devices of his. I'd have no complaint with his work. It flat out performed in exemplary fashion at all times. In all ways.

But right at the beginning he describes hearing over a 45khz capable ribbon a difference in a 5 khz squarewave filtered to 100 khz response and one filtered to 35 khz response. Everyone heard it he said. More than FR they all seemed to hear rate of change he said. Well, okay, but they would have heard the 5 khz fundamental, and very likely the 15 khz 3rd harmonic. The filter change to 35 khz response would have lowered volume of the 5 khz a little, and would have lowered the volume of the 15 khz 3rd harmonic quite a bit. No surprise it sounded obviously different, and it wasn't about rate of change. It was just about a FR which inadvertently changed levels in what could be heard. Oops!

There would have been a .6 db difference in loudness in the two signals and a different level mix between the 3rd harmonic larger than this.

My words of wisdom: HiFi is 85% frequency response, and much of the rest is about level differences.

Frequency response is about level.

 

[tubesguy]

Yes, there's no way to listen to the turntable alone. The system itself wasn't great, and in no small part due to the Wilson speakers. The lobing from the crossover was clearly audible as was the peaky treble ("low level detail retrieval").

Might have been a faulty Vendetta phono preamp?

 

[Wanman]

I am curious the technical reasons for your personal view?

Happy to make an attempt at explaining my opinion on the matter: It is in part because I am working with a discrete resistor DSD-only dac. Because of the large output step size (0 or 1), clock phase noise can have bigger effect on the integrated output than would be the case for a multibit dac. It looks to me like phase noise can have a substantial effect on some small details of sound, such as low level localization cues. Of course, phase noise can be kept very low with careful enough design, but doing it well can be perhaps too costly for most of the consumer dac market. Anyway, in the process of investigating the perceptual effects of lower and lower phase noise (down to the SOA level), I noticed what it sounds like to me personally as phase noise level is changed in this particular type of dac. To me signal-correlated noise doesn't sound like a noise because it the noise has become a low level part of the audio signal itself. I suspect that noise masks what would otherwise be low level audible details of a well made CD recording. IMHO, many dacs do that to some extent or another. It takes very high resolution FFTs to see small-ish changes in spectral line noise skirts, but I believe there does appear to be some measurable correlation with subjective perception. Probably I am failing to effectively write of the above in the form of a personal opinion, but that's what it is. In no way do I claim the above opinion to be proven fact.

 

[SIY]

Might have been a faulty Vendetta phono preamp?

LOL, the regulators on those things go out at the drop of a hat.

At some point, I'll be back up there (need to visit the great Tenuta's in Kenosha) and I have a package for you...

 

[SIY]

In no way do I claim the above opinion to be proven fact.

Anyway, in the process of investigating the perceptual effects

You aren't investigating perceptual effects. That takes things like controls. Get some data instead of spending energy weaving stories.

 

[DonH56]

Happy to make an attempt at explaining my opinion on the matter: It is in part because I am working with a discrete resistor DSD-only dac. Because of the large output step size (0 or 1), clock phase noise can have bigger effect on the integrated output than would be the case for a multibit dac.

A one-bit DAC for DSD input? Assuming a conventional oversampled approach it should be no more sensitive to phase noise than other DACs, and have the advantage that high-frequency noise is filtered. Also not sure what the resistors are doing; you just need a switch and output filter.

It looks to me like phase noise can have a substantial effect on some small details of sound, such as low level localization cues.

That would have to be seriously high correlated phase noise at low frequencies. If true you need a better clock source and/or clock recovery circuit. Lots of off-the-shelf PLLs to do the trick. It could also be noise and distortion injected from your switch. But I do not know what "looks like" means; the noise should be easily measured with a spectrum analyzer or phase noise test rig. I would use an SA or a good 'scope as you can better see exactly the noise (jitter, whatever) involved, especially using a 'scope with decent jitter package. The SA will tell you spectral information and (with the right option) generate the phase noise plot, but a 'scope is more likely to provide jitter separation so you can better isolate cause and effect. My last work involved analyzing nearly 30 different types of jitter using a 'scope and some post-processing (mine and others' -- I am not the best programmer by a long shot).

Of course, phase noise can be kept very low with careful enough design, but doing it well can be perhaps too costly for most of the consumer dac market.

Test results indicate the consumer market has solved that problem for a couple of decades or more now. Most DAC chips these days include integrated clock recovery and regeneration that pushes clock noise (phase and otherwise) well below audibility.

Anyway, in the process of investigating the perceptual effects of lower and lower phase noise (down to the SOA level), I noticed what it sounds like to me personally as phase noise level is changed in this particular type of dac. To me signal-correlated noise doesn't sound like a noise because it the noise has become a low level part of the audio signal itself. I suspect that noise masks what would otherwise be low level audible details of a well made CD recording. IMHO, many dacs do that to some extent or another. It takes very high resolution FFTs to see small-ish changes in spectral line noise skirts, but I believe there does appear to be some measurable correlation with subjective perception. Probably I am failing to effectively write of the above in the form of a personal opinion, but that's what it is. In no way do I claim the above opinion to be proven fact.

Signal-correlated noise sounds more like correlated jitter than noise per se. That is different than phase noise, or rather could be consider part of the integrated phase noise of the clock, but sounds like you need to get a 'scope or analyzer that does jitter separation so yo can see what's gong on. You could inject phase noise into your DAC at various levels so you can measure and hear the result. Best if you have a SW-controlled test rig that can enable blind trials.

Again, it seems like it would have to be very high noise level to mask audible low-level detail. I have read that over and over on audiophile sites and in reviews but things like lsb-level dither and noise are way, way below audibility.

 

[Robin L]

Might have been a faulty Vendetta phono preamp?

The version I heard was with the CTC Blowtorch preamp. Curl told me he wasn't happy with the Wilson WATTs. That would be the weak link in the system, along with the room they were playing in. When I was speaking of "low level resolution" it wasn't about a treble lift, but about the way all the separate elements of sound were clearly audible without smearing. In any case I'll admit that I haven't had as much exposure to properly set up turntable rigs as others here and leave it at that.

 

[Wanman]

I would use an SA or a good 'scope as you can better see exactly the noise (jitter, whatever) involved, especially using a 'scope with decent jitter package.

Yes, I am familiar with those methods. However, scope jitter packages are not generally good at measuring close-in phase noise, say 1-10Hz offset from the carrier (at dac clock frequencies). PN measurement equipment can do it but is expensive, although I have PN plots for my SOA clocks. Hard to measure some of correlated noise effects at the output of dac though. For IC dacs substrate coupled noise can also be an issue; at least that is reportedly some of the motivation for why AKM separated AK4191 and AK4499EX into two chips. Once again, I am trying to keep this in the realm of opinion. In particular, because I am new around here. Don't want to get into into trouble for what might be taken as, or perceived as, being disrespectful. Probably I should stop here.

 

[DonH56]

Yes, I am familiar with those methods. However, scope jitter packages are not generally good at measuring close-in phase noise, say 1-10Hz offset from the carrier (at dac clock frequencies). PN measurement equipment can do it but is expensive, although I have PN plots for my SOA clocks. Hard to measure some of correlated noise effects at the output of dac though. For IC dacs substrate coupled noise can also be an issue; at least that is reportedly some of the motivation for why AKM separated AK4191 and AK4499EX into two chips. Once again, I am trying to keep this in the realm of opinion. In particular, because I am new around here. Don't want to get into into trouble for what might be taken as, or perceived as, being disrespectful. Probably I should stop here.

At the risk of asking a question well beneath you, have you looked at the power supply noise at the output stage? Early designs suffered from supply noise since it directly modulates the output level(s).

 

[SIY]

The version I heard was with the CTC Blowtorch preamp.

If memory serves, that's just a line amp. In any case, that was always in his system when I was up at his place.

 

[Robin L]

If memory serves, that's just a line amp. In any case, that was always in his system when I was up at his place.

Out of curiosity, when did this happen?

 

[SIY]

Numerous times in the early 2000s. I left CA in late 2008, so it was before then.

I still have the HP3581A he sold me, complete with a masking tape label with his name and phone number. I got a lot of use out of that until soundcards got better and easier to use.

 

[Wanman]

At the risk of asking a question well beneath you, have you looked at the power supply noise at the output stage? Early designs suffered from supply noise since it directly modulates the output level(s).

Not beneath me at all. You are right on, IMHO. Only in this case its the final clocking drive circuitry power to the output array. The output array itself has clean, low noise power and approximates the interleaving of two RTZ DSD dacs to help cancel out errors, as described in an expired patent. The clock drive circuitry also has clean, low noise power, but the PDN design for both pieces of circuitry is critical. If I clock in a very slightly less clean way there are obvious degradations to the sound, including blurred out low level details of various types. The surface layer of sound is still intact, it doesn't have symptoms of audible masking. Maybe only 40dB down or somewhere like that where low level detail goes missing. Exact effect depends on clocking PDN properties. At least, that's how I am able to observe audible effects in this set of experiments. Just one way to affect/degrade phase noise, again only IMHO.

 

[DonH56]

Not beneath me at all. You are right on, IMHO. Only in this case its the final clocking drive circuitry power to the output array. The output array itself has clean, low noise power and approximates the interleaving of two RTZ DSD dacs to help cancel out errors, as described in an expired patent. The clock drive circuitry also has clean, low noise power, but the PDN design for both pieces of circuitry is critical. If I clock in a very slightly less clean way there are obvious degradations to the sound, including blurred out low level details of various types. The surface layer of sound is still intact, it doesn't have symptoms of audible masking. Maybe only 40dB down or somewhere like that where low level detail goes missing. Exact effect depends on clocking PDN properties. At least, that's how I am able to observe audible effects in this set of experiments. Just one way to affect/degrade phase noise, again only IMHO.

Thanks.

Would NRZ DACs help? One less level to deal with, but comes with its own set of problems...

Your troubleshooting seems to implicate the clock, so either the clock signal itself is not clean enough, or the switches (etc.) are too sensitive to clock noise. More gain in the drivers? Higher slew? At least if it's 40 dB down it should be easier to find. When I was a repair tech decades ago, it wasn't the blown amps that were a pain, it was the ones with 0.01% THD instead of 0.001%.

 

[Wanman]

Its not exactly that I have trouble. The dac works fine as-is. There are two related questions though. First is whether the high cost of making the dac can be reduced without reducing its desirable qualities? Second is whether it can be made even better? Right now the dac is plenty good enough aside from cost.

Actually the reason I started describing the dac and its sensitivity to excessive phase noise was to explain how its possible to compare the sound of a very clean dac to one with more audio signal-correlated noise. You probably know all about this already, but just for the record the mechanism that produces correlated noise in this case is the convolution of very close-in phase noise (1-10Hz offset, or less) with the audio signal. The intermodulation products are almost exactly in very narrow frequency band centered around an audio test signal spectral line in an FFT. That's because phase noise has a 1/f profile, so its highest amplitude is at very low offset frequencies from the carrier. Thus the noise appears as a thickening around the base of the spectral line of a test signal.
This is how the noise can move up and down at some low level, always following the amplitude of the audio signal.

The above explaination is to follow up on what I suggested in an earlier post which was to the point that someone may be hearing noise without knowing they are hearing it (if that makes sense in the context of how correlated noise can be produced and then exist in music reproduction, perhaps to be heard in some way or not). Some people may consider the idea controversial, is all. IMHO its more or less similar to some of the things ESS talked about in their modulator seminar from several years ago. For reference, it is at: https://www.yumpu.com/en/document/read/23182504/noise-shaping-sigma-delta-dacs-ess-technology-inc

 

[SIY]

First is whether the high cost of making the dac can be reduced without reducing its desirable qualities?

That's trivially easy. You can VASTLY reduce the cost and improve the performance. Assuming that's what your goal actually is.

 

[Wanman]

Actually I wouldn't mind talking to DonH56 some more. He seems to understand more about these things than most folks.

 

[Wanman]

IMHO, improving imaging performance would be quite difficult. I have owned Benchmark DAC-3, Topping D90. Have auditioned several other commercial dacs as well. Have built my own ESS, AKM, and Rohm based dacs. Quite a bit of experience with what they can and can't do. Imaging historically seems to be a very hard one to get right. Maybe that's because although it is known what humans use as cues for localization of a sound source in space, but there are not industry standard methods for measuring those cues. Could in principle be done with modern microphones that can capture air motion and pressure coming from multiple directions at once.