Is DSD noise audible?

[Scytales]

Something similar had happened to me last year after an earache: male baritone singer voices singing fortissimo were distorted and only in one ear!

 

[SoundsGood64]

MaxwellsEq said:
If there is a conversion to PCM anywhere in the path, the ultrasonic noise may not be suppressed and can be quite high level.

It seems to me that, to the contrary, the decimation of DSD in PCM would suppress the ultrasonic content of DSD better than anything else, because any signal above the Nyquist frequency (which is relatively low in any PCM format compared to DSD) have to be sufficiently suppressed in order to avoid aliasing of said frequencies in the pass-band. Surely any designer of digital decimation filter know that, because it has been pretty much the textbook method the obtain PCM from delta-sigma modulated signals for decades !

An illustration of that kind of suppression of DSD ultrasonic content when DSD is decimated in low-sample rate PCM can be seen in NTTY's review of the Denon DCD-SA1 when the optional decimation filter is enable:

I have a different interpretation of what @MaxwellsEq stated, which is that when DSD is transcoded to PCM, in some cases the ultrasonic noise inherent in DSD is filtered out using a low pass filter in the digital domain (as I believe it should be) and other times it is not (and gave an example of some PCM files that had been originally from DSD still having the ultrasonic noise). Whether such filtering occurs is a question that has come up for me when I use a Sony UBP-x800m2 that sends 176.4khz PCM (transcoded from DSD64) out via HDMI. Does that PCM stream have the ultrasonic noise removed with a low pass filter in the digital domain? (Does anyone know?) Because if the ultrasonic noise isn't filtered out, it would be present in the PCM stream up to around 88khz. The examples of DSD to PCM on all these SACD players NTTY analyzed were probably transcoded to 88.2khz PCM, inferring this because the noise in the PCM signal starts declining at around half that.

With regard to the SACD DSD64 signal above from Denon DCD-SA1, the ultrasonic noise rises after 20khz to 25khz, reaches a maximum around 72k or so, and then slowly declines. NTTY recorded a similar pattern for SACD playback from Oppo 95, Yamaha CD-S2000, and Pioneer BDP-LX58. In all of these cases the noise begins to decline after around 70khz. Is this decline because of an analog FIR filter (low pass) in the DAC chip that is on for DSD input? I think "yes" because the DSD1792 DAC spec sheet (the DAC in the DCD-SA1) shows 4 FIR filter options (and Denon probably picked one), but asking here to ensure I understand this correctly.

 

[MaxwellsEq]

I have a different interpretation of what @MaxwellsEq stated, which is that when DSD is transcoded to PCM, in some cases the ultrasonic noise inherent in DSD is filtered out using a low pass filter in the digital domain (as I believe it should be) and other times it is not (and gave an example of some PCM files that had been originally from DSD still having the ultrasonic noise

Correct. I have some "hi rez" PCM 24/96 tracks which have an upside down "U" of ultrasonic noise which have clearly been created from a DSD master, but without a proper filter. The ultrasonic noise can peak as high as the levels in the audible midrange.

I can't hear it, obviously and at those frequencies I suspect my crossover and tweeters are blocking it. However, Bruno justified some of the Purifi amplifier changes to improve stability in just this case. Which suggests that older Class D amplifiers (and class A/B amplifiers with insufficient GBW before feedback) may be unstable playing this stuff.

 

[voodooless]

Pretending that ultrasonic noise on DSD output is normal is the similar to thinking a car with 3 wheels is fine to drive: it’s flawed to begin with. That doesn’t mean the concept of a car is flawed, it just means the implementation is.

 

[Scytales]

Correct. I have some "hi rez" PCM 24/96 tracks which have an upside down "U" of ultrasonic noise which have clearly been created from a DSD master, but without a proper filter. The ultrasonic noise can peak as high as the levels in the audible midrange.

A decimation filter being also a low-pass filter (to avoid aliasing), I think that a reduction of the sample rate to 44.1 or 48 kHz instead of 88.2 or 96 kHz would take care of the ultrasonic noise level that bothers you. I guess there are many software solutions to do that work on digital files.

 

[MaxwellsEq]

A decimation filter being also a low-pass filter (to avoid aliasing), I think that a reduction of the sample rate to 44.1 or 48 kHz instead of 88.2 or 96 kHz would take care of the ultrasonic noise level that bothers you. I guess there are many software solutions to do that work on digital files.

Yes, I can, and do fix it. I can fix it because I know it is there because I check my tracks.

There may be some people who unwittingly download Hi Rez content direct from legitimate publishers, don't test and so are unaware that it has this flaw and who would not know how to fix it.

To clarify - it's probably not a big issue in most cases, but it may be an issue with amplifiers with marginal stability at high frequencies.

 

[MaxwellsEq]

Yes, I can, and do fix it. I can fix it because I know it is there because I check my tracks.

There may be some people who unwittingly download Hi Rez content direct from legitimate publishers, don't test and so are unaware that it has this flaw and who would not know how to fix it.

To clarify - it's probably not a big issue in most cases, but it may be an issue with amplifiers with marginal stability at high frequencies.

Here is an example. This is Zubin Mehta's version of Mars from Holst's The Planets at 192kHz. I've made no edits.

This is during a loud part:

This is a quieter section. The ultrasonic noise is louder than all but the lowest frequencies:

 

[dualazmak]

My post #532 on my project thread would be somewhat of your reference and interest, I assume.
- Summary of rationales for "on-the-fly (real-time)" conversion of all music tracks (including 1 bit DSD tracks) into 88.2 kHz or 96 kHz PCM format for DSP (XO/EQ) processing: #532

 

[Multicore]

Abstract: The problem was a short term eustachion tube disorder which has been resolved, and I no longer hear the distortion.

Good thing I read the whole post because on the basis of the abstract I thought it was a problem with a noisy tube.

 

[SoundsGood64]

Here is an example. This is Zubin Mehta's version of Mars from Holst's The Planets at 192kHz. I've made no edits.

This is during a loud part:
View attachment 489847

This is a quieter section. The ultrasonic noise is louder than all but the lowest frequencies:
View attachment 489848

Thanks for generating and posting these plots showing your 192khz PCM files coming from DSD. The ultrasonic noise increases above 25khz. Although I can't see the axis labels for the last two ticks (and the scale makes it hard to infer what they would be), because this is 192khz, I think that there should be a sharp decline around 96khz (1/2 the sampling rate) due to Nyquist as described by @Scytales, which I believe is apparent here. In the typical implementation of DSD64 playback shown by @NTTY (Denon DCD-SA1, Oppo 95, Yamaha CD-S2000, and Pioneer BDP-LX58) the ultrasonic noise peaks around 72khz or so and then has a slow decline, which is probably (if answer to my second question in post #22 above is 'yes') due to an analog FIR filter in the DAC chip. With transcoding DSD64 to higher rate PCM such as 176.4khz or 192khz, due to Nyquist, there is going to be a sharp decline in signal around 88.2khz or 96khz respectively, which is a much sharper decline than what the FIR filter does in DSD playback (NTTY post 113 in Oppo BDP-95 review goes all the way out to 360khz).

For direct DSD64 playback, it would seem that the audio engineers who designed the analog FIR filter implementation thought that the slow decline in noise after the ~72khz peak was sufficient to avoid problems.

I don't currently have the skills and means (would involve software tools) to remove ultrasonic noise from PCM files with DSD origin, although it is super respectable to have those skills and means. However if one is concerned about it, I know there is a straightforward way to avoid this ultrasonic noise with mainstream AVR. When playing DSD64 via HDMI using my Denon AVR-3808ci, I can avoid most of the ultrasonic noise (regardless of DSD or DSD transcoded to PCM) by using the Audyssey room correction (for example "Stereo" Mode), which should resample to 48khz PCM, so noise should decline rapidly around 24khz. If instead I do "DSD Direct" (Audyssey off) the FIR filter in the PCM1791a DAC chip will probably produce ultrasonic filtering similar to what NTTY saw for SACD players in DSD playback. If I use the Sony UBP-x800m2 playing DSD64 with transcoding to 176.4khz PCM into the AVR-3808ci in "Direct" Mode (Audyssey off), there should be a sharp decline in noise around 88.2khz, similar to what we see in MaxwellEq's plots where the sharp decline should be around 96khz.

 

[danadam]

Although I can't see the axis labels for the last two ticks (and the scale makes it hard to infer what they would be),

The existing ticks are:
100, 160, 250, 400, 630
then:
1000, 1600, 2500, 4000, 6300
then:
1, 1.6, 2.5, 4 (with e+04)

It's not that difficult to infer that the next tick is 6.3e+04 and the last one either 1e+05 or half the sampling rate, i.e., 9.6e+04.

Re. the noise, here's what Scarlett Book says:

E.2 Analog Post-filter
To protect analog amplifiers and loudspeakers, it is recommended that a Super Audio CD player
contain at its output an analog low pass filter with a cut-off frequency of maximum 50 kHz and a slope
of minimum 30 dB/Oct. For use with wide-band audio equipment, filters with a cut-off frequency of
over 50 kHz can be used.

 

[dualazmak]

The independent thread would be also of your reference and interest;
Does DSD recording benefit Japanese traditional instruments?

On that thread, I objectively and subjectively evaluated the DSD layer and CD layer of the specific hybrid SACD in my posts #42 and #56 (I also commented/noted in #59, #61 and #62).

 

[dualazmak]

You can find very nice example case of really low UHF noises in HiRes formats, I mean good-enough quality control (QC) in HiRes format release, in my post here #518 on my project thread.

 

[Sokel]

Other than theory, I did the simple thing.
I generated some DSD128 Pink Noise through Multitone Analyzer and threw it back in it to see how it looks like:

I then loaded it to foobar playing native DSD and captured the actual acoustic playback at 192kHz (the top of E-MU) at MLP at REW.

Can't see any problem there, drops like a rock after 20kHz and no rise after that whatsoever.

(the spikes you see at the right down corner are sometimes there with nothing playing, haven't searched what it is but have nothing to do with the the playback)

 

[Scytales]

@Sokel, what microphone did you use?

 

[Sokel]

@Sokel, what microphone did you use?

That's a good question I didn't think off.

For this one It's a Primo EM258, it's data sheet stops at 20kHz.

So, to test I just played back some PCM 192kHz Pink noise and the results are:

So probably a combination, mic and response, you're right you asked.

 

[dualazmak]

@Sokel, what microphone did you use?

That's a good question I didn't think off.

For this one It's a Primo EM258, it's data sheet stops at 20kHz.

So, to test I just played back some PCM 192kHz Pink noise and the results are:



View attachment 489993


So probably a combination, mic and response, you're right you asked.

Yes, in measurement and analysis of actual room air sound, the microphone capability is one of the critical factors, always!

I too very carefully checked/confirmed my microphone (specially selected BEHRINGER ECM8000) capabilities as well as the "quality" of the test tone signal (in my case strictly QC-ed 20 Hz - 20 kHz flat white noise), as I shared in my posts #831 and #1,009 on my project thread.

Even when our microphones would not "hear" the room air sound above 25 kHz, our tweeters and/or super-tweeters can sing in the Fq zone of 25 kHz to say 50 kHz (Earthwork M50 microphone actually has flat response up to 50 kHz), we humans too cannot hear it, but our beloved pets (e.g. dogs, cats, birds) may hear it making them somewhat panic, as I wrote in my post #532.

 

[Scytales]

Yes, in measurement and analysis of actual room air sound, the microphone capability is one of the critical factors, always!

I too very carefully checked/confirmed my microphone (specially selected BEHRINGER ECM8000) capabilities as well as the "quality" of the test tone signal (in my case strictly QC-ed 20 Hz - 20 kHz flat white noise), as I shared in my posts #831 and #1,009 on my project thread.

Even when our microphones would not "hear" the room air sound above 25 kHz, our tweeters and/or super-tweeters can sing in the Fq zone of 25 kHz to say 50 kHz (Earthwork M50 microphone actually has flat response up to 50 kHz), we humans too cannot hear it, but our beloved pets (e.g. dogs, cats, birds) may hear it making them somewhat panic, as I wrote in my post #532.

It should be emphasized, though, that until now, the production of (significant) sound-pressure by tweeters above 20 kHz when DSD materials is played back remains speculative due to the lack of measured data.

 

[dualazmak]

It should be emphasized, though, that until now, the production of (significant) sound-pressure by tweeters above 20 kHz when DSD materials is played back remains speculative due to the lack of measured data.

I essentially agree with you since I do not have very expensive Earthwork M50 microphone (flat response up to 45 kHz) and I hence myself do not have experience of room air sound recording above 25 kHz up to 50 kHz using my own audio rig.

On the other hand, I could confirm the capabilities of Earthwork M50 (I mean flat response up to 45 kHz) in the occasion of cross-calibration of my BEHRINGER ECM8000 using well QC-ed white noise signal of 192-kHz/24-bit (up to 96 kHz)) and 96-kHz/24-bit (up to 48 kHz) (using Mr. TY's studio setup; ref. #831).

Consequently, since my super-tweeter FOSTEX T925A has good-enough Fq response up to 40 kHz (ref. #485 and #921), I do believe, if I have Earthwork M50, I will be able to properly record/measure the room air sound up to 40 kHz using well QC-ed 96-kHz/24-bit white noise signal (or using DSD track with much UHF noises) in the playing-recording setup (I mean even with my own audio rig) when no low-pass (LP) (high-cut) digital/passive filters/mechanisms are implemented in the signal path.

Please note that my OKTO DAC8PRO is perfectly transparent for even UHF sound/noises, having no low-pass (LP) (high-cut) digital/passive filters/mechanisms in it.

Therefore, in any way, let me emphasize again as I wrote in my post here #532 on my project thread;
- Summary of rationales for "on-the-fly (real-time)" conversion of all music tracks (including 1 bit DSD tracks) into 88.2 kHz or 96 kHz PCM format for DSP (XO/EQ) processing: #532

 

[MaxwellsEq]

It should be emphasized, though, that until now, the production of (significant) sound-pressure by tweeters above 20 kHz when DSD materials is played back remains speculative due to the lack of measured data.

I agree - I know I can't hear it and I suspect my tweeter and crossover network pretty much block this ultrasonic noise. There's a slight risk of intermodulation in the analogue domain back into the audible range.

The bigger concern to me is amplifier stability, especially with earlier Class D or some Class A/B amplifiers with insufficient open loop gain bandwidth product. As I have previously linked here in post #4, Bruno Putzeys thought it worth avoiding, since his comments on the Ncore are:..

My previous amp* has approximately a 1st order roll-off. Now, since the output filter naturally has a second order roll-off it means that this amp could be overdriven with out of band noise from e.g. DSD recordings (in fairness, only when you cranked a quiet recording high).

... and the Purifi design reduces that risk (my bold, BTW).

The link is here