After some adjustment of the scope settings, and after waiting for a quiet point in the music, got this:
Once again it is after the DC blocking caps and with the transformer loading the dac output.
EDIT: Also found the FFT display lags the channel 1 trace. If set to single sweep mode, the FFT display will update a couple of times after channel 1 yellow trace displays. To get this image required putting the scope directly into STOP mode at the right time. In that case the FFT display does not continue to update.
Once again it is after the DC blocking caps and with the transformer loading the dac output.
EDIT: Also found the FFT display lags the channel 1 trace. If set to single sweep mode, the FFT display will update a couple of times after channel 1 yellow trace displays. To get this image required putting the scope directly into STOP mode at the right time. In that case the FFT display does not continue to update.
Last edited:
Mark,
I’m fine with all the things you’re doing, but at least I try to understand what’s going on.
At this moment the most interesting thing that I’m looking forward to hear from you is the effect what the proper termination of your interlink will bring.
That’s the sole background of my questions referring to HF noise.
Hans
I’m fine with all the things you’re doing, but at least I try to understand what’s going on.
At this moment the most interesting thing that I’m looking forward to hear from you is the effect what the proper termination of your interlink will bring.
That’s the sole background of my questions referring to HF noise.
Hans
HQPlayer manual states this:
Fifth order modulators are more suitable for DACs that have simple analog reconstruction filters. Seventh order modulators provide better technical performance, but also put more demands on the DAC's analog reconstruction filter.
So comparing HQPlayer with 7th order modulator to PCM2DSD which has 5th order modulator is not exactly apples-to-apples comparison. With Marcel's original filter HQPlayer with 7th order modulator would probably work better.
Fifth order modulators are more suitable for DACs that have simple analog reconstruction filters. Seventh order modulators provide better technical performance, but also put more demands on the DAC's analog reconstruction filter.
So comparing HQPlayer with 7th order modulator to PCM2DSD which has 5th order modulator is not exactly apples-to-apples comparison. With Marcel's original filter HQPlayer with 7th order modulator would probably work better.
Finally, You start to see something closer to what I have observed.
Attachments
would like to point out that these shaped noise floors are depending principally on the modulator algorithm, not the filtering quality in the audio band, neither from presence or absence of the audio signal.
And as visible, it can have a noise peak level above the floor close to 60/70db high, and in a ~wide band, so not insignificant energy.
And this is inherent in DSD processing, (single bit) - if you don't see it then it's a problem of the measurement.
With multibit noise shaping the peaking can be less, the distribution different, but not absent altogether
And as visible, it can have a noise peak level above the floor close to 60/70db high, and in a ~wide band, so not insignificant energy.
And this is inherent in DSD processing, (single bit) - if you don't see it then it's a problem of the measurement.
With multibit noise shaping the peaking can be less, the distribution different, but not absent altogether
Its pretty hit or miss with this scope whether it shows the 'correct' shape or not. It really depends if I get lucky when I hit STOP. Out of a few tries I might get one that doesn't look too flat. Also, the scope doesn't trigger properly in NORMAL mode, so its just free running unless I set the sweep rate to capture individual RF pulses at the dac output. However, at that sweep rate it doesn't make a very nice looking FFT down where we expect to see the dip before the noise starts rising.
Oh, well. Maybe I'll get a more suitable RF spectrum analyzer at some point.
Hans:
I think there is some difference in sound with your termination, but I'm not prepared to say whether it will stay or go at this point.
Last edited:
Mark,
I’m glad you hear a difference with termination, it had to be the case with your setup.
Since this termination becomes effective in the ultrasonic range, it can be the prove that reflections are taking their share.
The problem can be though that previously found improvements are now again to be evaluated.
You could also try to remove the 270pF that you reported to have installed at the Firdac side.
This will slightly improve FR at 20Khz.
Hans
I’m glad you hear a difference with termination, it had to be the case with your setup.
Since this termination becomes effective in the ultrasonic range, it can be the prove that reflections are taking their share.
The problem can be though that previously found improvements are now again to be evaluated.
You could also try to remove the 270pF that you reported to have installed at the Firdac side.
This will slightly improve FR at 20Khz.
Hans
Couple of comments from my side:
0. It seems to me that the transformer is Mark's main filter. At least he had no difficulty measuring the ultrasonic noise at the transformer input.
1. Sigma-delta modulators are normally either chaotic or at the edge of chaos (Lyapunov exponent of 0), so their output signals are normally never periodic. In fact, if they were, they couldn't contain noise, only tones.
2. The modulator algorithm has the largest impact on the out-of-band quantization noise spectrum, but the shape of the quantization noise spectrum does often depend on the signal as well, especially above a quarter of the DSD rate. A typical example are the idle tones that get frequency modulated by the desired signal in a single-bit sigma-delta modulator.
3. Compared to a straightforward single-bit sigma-delta, my algorithm with the embedded pulse width modulator with random rotation makes the ultrasonic noise stronger, but less dependent on the signal. I suspect the fancier DSD512 modulators of HiQPlayer do something similar.
4. As Mark sometimes complains about signal-dependent noise, I was surprised that he seems to prefer relatively straightforward modulator algorithms. Hence my curiosity about the ultrasonic noise spectrum. Of course ultrasonic noise only matters if it gets converted into the audio band somehow, but there are plenty second-order effects that can do just that.
5. I haven't a clue what to conclude.
0. It seems to me that the transformer is Mark's main filter. At least he had no difficulty measuring the ultrasonic noise at the transformer input.
1. Sigma-delta modulators are normally either chaotic or at the edge of chaos (Lyapunov exponent of 0), so their output signals are normally never periodic. In fact, if they were, they couldn't contain noise, only tones.
2. The modulator algorithm has the largest impact on the out-of-band quantization noise spectrum, but the shape of the quantization noise spectrum does often depend on the signal as well, especially above a quarter of the DSD rate. A typical example are the idle tones that get frequency modulated by the desired signal in a single-bit sigma-delta modulator.
3. Compared to a straightforward single-bit sigma-delta, my algorithm with the embedded pulse width modulator with random rotation makes the ultrasonic noise stronger, but less dependent on the signal. I suspect the fancier DSD512 modulators of HiQPlayer do something similar.
4. As Mark sometimes complains about signal-dependent noise, I was surprised that he seems to prefer relatively straightforward modulator algorithms. Hence my curiosity about the ultrasonic noise spectrum. Of course ultrasonic noise only matters if it gets converted into the audio band somehow, but there are plenty second-order effects that can do just that.
5. I haven't a clue what to conclude.
We just go with what sounds best in a given situation. It could be that there are other problems with more fancy modulators that create non-well-defined problems. Don't know. Would have to listen and evaluate perceptually.
Also, noticed that a lot of people online are talking about choosing HQ player algorithms depending on the particular dac one has. IOW, they are using the algorithms to compensate for differences/imperfections in the dacs.
What we would like to do is try to find a sound an average audiophile and or reviewer type listener tends to prefer. Space is a big deal for some people. I will say it can be a bit addicting once someone gets enough exposure to it. Then people notice when its missing or not as good on another system. There are other things that matter too. We try to discriminate those things and search for perceptual optimizations. Vinyl recordings of master tapes are one thing used for reference during that process. It should sound natural, spacious, musical, and like a real un-amplified acoustic performance as much as possible. Maybe the end result is not everyone's cup of tea (can't please everyone), but IME people do tend to like the sound of it once they hear it. Biggest problem is cost. Minimizing cost for a given level of perceptual performance looks to be a whole lot of work.
Last edited:
Sorry Hans and all, what I wrote as point 3 is terribly inaccurate.
The total power of the out-of-band quantization noise is essentially the same as for a straightforward modulator, but there is more noise ending up in the first few hundred kilohertz above the audio band, so you see more noise at the low-pass filter output. It is indeed less dependent on the desired signal.
The total power of the out-of-band quantization noise is essentially the same as for a straightforward modulator, but there is more noise ending up in the first few hundred kilohertz above the audio band, so you see more noise at the low-pass filter output. It is indeed less dependent on the desired signal.
Using different modulators depending on which DAC is used (as recommended in HQPlayer manual) makes perfect sense. E.g. with properly chosen modulator it is possible to get better results (sound & performance) from ESS DACs with DSD than with PCM. So all these people choosing modulators based on their DACs just go with what sounds best for them. I'm sure they know what sounds best to them better than anybody else.
Good to know that DSD is becoming more accepted as a useful thing.
Would like to just mention that although people do know if they like something, the liking of something is often not random. Humans sometimes tend to have common subjective preferences, whether they are fully aware of it or not (e.g. for food taste; something which is well known for certain base-level flavors, and sometimes focused upon as a worrisome health related issue).
When it comes to audio reproduction, it may be there there are various statistical/systematic preferences at different levels (not necessarily base-level) that many people tend to share. Doesn't necessarily have to be limited to things that are most commonly measured. As Geddes points out, there is still much that is not known about human hearing.
Just saying, there could potentially be an objectively better dac to satisfy systematic aural preferences common among humans. Objectively better in the statistical sense, that is (meaning that very many people prefer it once they hear it, and that it does not tend to become boring/uninteresting over time). At a minimum it would require that human preference is not random, which it IMHO it probably isn't.
Would like to just mention that although people do know if they like something, the liking of something is often not random. Humans sometimes tend to have common subjective preferences, whether they are fully aware of it or not (e.g. for food taste; something which is well known for certain base-level flavors, and sometimes focused upon as a worrisome health related issue).
When it comes to audio reproduction, it may be there there are various statistical/systematic preferences at different levels (not necessarily base-level) that many people tend to share. Doesn't necessarily have to be limited to things that are most commonly measured. As Geddes points out, there is still much that is not known about human hearing.
Just saying, there could potentially be an objectively better dac to satisfy systematic aural preferences common among humans. Objectively better in the statistical sense, that is (meaning that very many people prefer it once they hear it, and that it does not tend to become boring/uninteresting over time). At a minimum it would require that human preference is not random, which it IMHO it probably isn't.
Last edited:
Interesting, they must have done something wrong when designing the on-chip modulator then.
With PCM input ES9038 & ES9039 have lots of high order artefacts even within audio bandwidth. With DSD input there is much less high order stuff (or it is buried below noise floor). Noise floor is very similar with PCM and DSD.
According to datasheet block diagrams ES9038 & ES9039 DACs have volume control on DSD as well so this probably means DSD is internally converted to PCM at some stage. Maybe this conversion does some filtering as well.
According to datasheet block diagrams ES9038 & ES9039 DACs have volume control on DSD as well so this probably means DSD is internally converted to PCM at some stage. Maybe this conversion does some filtering as well.