• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Does DSD sound better than PCM?

Status
Not open for further replies.

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO

JohnYang1997

Master Contributor
Technical Expert
Audio Company
Joined
Dec 28, 2018
Messages
7,175
Likes
18,292
Location
China
I do feel there is difference. I hear more dynamic from pcm and smoother on dsd. I think it's just that pcm being superior.
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
The impressive DSD jitter results are mostly by Jussi (Miska) the author of HQPlayer.

He’s also the one showing charts up into the MHz range. His claim is that noise there might become audible with some equipment due to intermodulation into the audible band. I’m skeptical that this is the case, but don’t have the equipment to measure it.

If there is IM in the normal audio band we can measure it. Without commenting on the likely hood of this, even though its technically feasible, do we have any measurements that correlate and support the proposition?

Another view is don't use an amp that is wide open to RF frequencies.
 

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,630
Likes
10,205
Location
North-East
If there is IM in the normal audio band we can measure it. Without commenting on the likely hood of this, even though its technically feasible, do we have any measurements that correlate and support the proposition?

Another view is don't use an amp that is wide open to RF frequencies.

I think @Miska had some examples. Not sure I was convinced IM was the issue, but maybe he can explain.
 

Roen

Addicted to Fun and Learning
Joined
Oct 14, 2018
Messages
681
Likes
247
By measuring both audio band and out of band behavior. Traditional measurements on audio band (20 kHz and 100 kHz). And checking that there are no correlated components or discrete spurious tones above audio band, such as images for example.
Do you have any preferred settings for 16/44.1, 16/48, 24/88.2, 24/96, 24/192, DSD64 and DSD128?
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Do you have any preferred settings for 16/44.1, 16/48, 24/88.2, 24/96, 24/192, DSD64 and DSD128?

Nowadays I do most measurements and listening with "poly-sinc-ext2" filter. For DSD, depending DAC's DSD filters, either ASDM7 or ASDM5 modulator. For PCM, if DAC can do 20 bits or more resolution, TPDF or Gauss1 is fine for dither. For 16-bit and high rates noise shaping is good. For PCM inputs, linearity sweep is good starting point to see how much accuracy DAC actually has to set exact output dithering resolution. As PCM (R2R) example, Holo Spring (2) gives optimal results with 20 bits, up to 1.536 MHz. However, for example on macOS one is limited to 16-bit output at those > 1MHz rates, but using NS5/NS9 noise shaper and 16-bit at 1.4112/1.536 MHz rate gives practically same audio band SNR as 20/24-bit at same or lower rates. For Spring1, 20-bit TPDF ditther at 352.8/384k is fine for the R2R section. From audio-band performance perspective Holo Spring's optimal point is DSD256. If you look from wide-band perspective it is at DSD512. So in the end between the two it is largely system-dependent.

ASDM7 is more aggressive and complex so it puts more demand on ultrasonic filtering capabilities, but is otherwise cleaner and pushing the noise down more in audio band. ASDM5 is more gentle. But at higher DSD rates usually DAC's analog noise floor dominates and the digital noise floor of DSD is way below that. I've done quite a bit of measurements where in < 100 kHz band both ASDM5 and ASDM7 looks exactly the same. Only above that you can see a difference if analog filter doesn't cut all the noise away.

For DSD128 noise corner is around 50 - 60 kHz (like pretty much for most ADC's too). So 100 kHz band measurement still shows some difference between the two at that rate.

I think suitable way to look at bandwidths is like:
DSD64: flat noise floor bandwidth equivalent of 44.1/48k PCM, but without AA-filter effects
DSD128: flat noise floor bandwidth equivalent of 88.2/96k PCM, -"-
DSD256: flat noise floor bandwidth equivalent of 176.4/192k PCM, -"-
DSD512: flat noise floor bandwidth equivalent of 352.8/384k PCM, -"-

Now most "PCM" ADC's based on SDM look like DSD128 from noise profile point of view. Very few ones looks like DSD256. I have not seen many that would do PCM at 352.8/384k with flat noise floor and flat frequency response up to 176.4/192k... For example RME ADI-2 Pro's noise floor looks pretty much exactly the same regardless if you run it at 705.6/768k PCM or DSD256 (with compliant noise filter).
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
If there is IM in the normal audio band we can measure it. Without commenting on the likely hood of this, even though its technically feasible, do we have any measurements that correlate and support the proposition?

Another view is don't use an amp that is wide open to RF frequencies.

How I look at things:
- Correlated/discrete HF -> bad, because it potentially generates discrete IM tones
- Noise HF -> less bad, because potential IM tones are also noise (random) -> less audible and if audible just background hiss (like tape or radio noise)
But overall, for proper reconstruction, both audio band and wide band output must be clean. Which, of course, is only part of the story skipping over all the reconstruction filter details and such. But if/when something happens with any of the higher frequency output we'd want that to be something that disturbs the listening experience the least? In any case there shouldn't be any images that would indicate incomplete/imprecise reconstruction.

We cannot know what kind of amps there are after the DAC, so we need to account for all kinds of possibilities, such as analog class-D amps, etc. One dominating factor is how the amp's THD/IMD vs frequency profile looks like.

Then again, whenever you limit bandwidth in analog domain using traditional analog filters, in an amp or such, you also tend to exhibit phase shift at top audio octaves... And in addition this can create notable issue on transient (step) response too... I've somewhat studied this using band-limited 7 kHz square wave. Some examples...

iFi micro iDSD at 705.6k PCM (this is the same for DSD512 too because it is defined by the analog filter):
iDSDmicro-square7k-7056.png


Marantz HD-DAC1 at DSD128 (CS4398 in Direct DSD mode):
Marantz-square7k-dsd128.png


For my Open Hardware DSC1 project, I tried to design transient optimized 4th order analog reconstruction filter, so here's how that one does at DSD512:
square7k-dsd512.png


Partially that is due to how the conversion stage analog FIR works, 4th order filter alone wouldn't be enough anyway.

Complex topic? Oh yeah...
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO
Mixing an RF'ish signal to baseband is possible -- that is an EMI/RFI problem.

IMD creates tones in a couple of places, loosely stated... Even-order IMD produces tones near DC and near multiples (2x, 4x, etc) of the input signals. Odd-order IMD products land very near the original signals and at odd multiples (3x, 5x, etc.) of the input signals. Odd-order IMD is perhaps the most audible because it is near the input tones and yet not harmonically related.

Filter theory is a huge subject so I'll leave that aside with the comment that achieving steep rolloff to suppress images without compromising either the time- or frequency-domain response (or both) is challenging. Analog filters with the best pulse (time-domain) integrity tend to have the slowest roll-off in the frequency domain. Pick your poison (design trade).

FWIWFM - Don
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Yeah, I think both iFi and Marantz in question above have 2nd order analog filters. The third (DSC1) has 4th order analog filter.

In addition, all these have D/A conversion stage filter, iFi (TI/BB chip) has analog FIR, Marantz has switched capacitor filter and DSC1 has analog FIR too.
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
Hi Miska

It's an interesting debate and it would be great to have a (non argumentative good natured :) ) discussion on it.

Ok, first thing I would like to discuss is the use of square waves to demonstrate a point.

Digital audio systems are band limited by definition. They will be limited at the front end whether it is due to the microphone, it's amp, the mixer etc or the ADC it's bandwidth setting and anti alias filter.

To reproduce a square wave perfectly you require infinite bandwidth as it has harmonics that stretch to infinity. The "distorted" part of a square wave are by definition harmonics outside of the bandwidth of the system. In audio system that should be outside of the normal audio band.

You ear doesn't hear those ultrasonic harmonics so therefore effectively hears the "distorted" square wave regardless of how well you reproduce it, even if you have an infinite bandwidth system. Let's ignore what speakers and their inherent crossover filters do to the signal.

So what I am saying here is the damage is already done in this respect. You can't recover out of band signals in the replay chain. Your square wave will never be square and it doesn't need to be as its signals you can't hear anyway.


In band phase shift seems intuitively like something you want to avoid. A general question, does anyone have any links to research on the audibility of a simple minimum phase change at the limits of the audible band?


Yes, as acknowledged previously IM is a possibility. However it is neatly avoided by having appropriate amplifier input filtering. IMO we shouldn't confuse a failing in an amplifier design with this issue. Basically I don't want an amp that picks up my local AM radio station ;)

I would temper that comment with the fact that there is a grey area of just how much bandwidth an amp should have and the how low frequency any ultrasonic output from any particular dac might go, so it's not entirely black and white.

As previously mentioned, do we have any measurements of in band IM being correlated to a DAC US output so we can see what level and nature they are? If its happening at an audible level it will be measurable.
 
Last edited:

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Digital audio systems are band limited by definition. They will be so at the front end whether it is due to the microphone, it's amp, the mixer etc or the ADC it's bandwidth setting and anti alias filter.

To reproduce a square wave perfectly you require infinite bandwidth as it has harmonics that stretch to infinity. The distorted part of a square wave are harmonics outside of the bandwidth of the system. In audio system that should be outside of the normal audio band.

I know, as noted above this is bandlimited square wave (same source file in all cases)... Especially for DSD, the bandwidth limits are in MHz range anyway.

From the response you can see the shape is not limited by linear-phase digital domain bandwidth, rather than analog domain response of the reconstruction filter. And further on transient response of the analog reconstruction filter (as you know, different types of analog filters have different type of response). I don't know exact parameters for the others, but DSC1 filter is said 4th order with fc=100kHz. Probably not hugely different from other DAC filters that aim for hires (up to 192 kHz sampling rate). One of the aspects I looked at when deciding filter parameters was amount of phase shift at 20 kHz.

Square waves are good for inspecting step-response (directly connected with impulse response) of the analog filter. How well the system can reconstruct perfect step response is one of the interesting factors (in my opinion). You don't need to have anything digital to test this, just analog square wave generator connected to input of the analog reconstruction filter is enough.
 
Last edited:

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
Yeah, I think both iFi and Marantz in question above have 2nd order analog filters.
The iFi Nano post-DAC filter is two RC stages. I've never seen a Micro, so I can't say if it differs.
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
Aahh, didn't see the bit about bw limited square wave. DSD may have BW limits in the MHz but our amplifiers, speakers and ears do not.

So we are talking about the effects of any phase shift (in the audible band).

Do you have any research links that show the subjective audibility of this?

Secondly what about the band limiting that I mentioned already exists in any recording?
 
Last edited:

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Aahh, didn't see the bit about bw limited square wave.

So we are talking about the effects of any phase shift.

Do you have any research links that show the subjective audibility?

Secondly what about the band limiting that I mentioned already exists in any recording?

I don't start with research about audibility, but about absolute perfectness. Mathematically we are supposed to be able to reconstruct the signal perfectly, so that's what I'm looking for. I don't take stance on how many out of billions of people are able to hear what.

I'm talking about reconstruction in it's all aspects, for which one part is phase response. Others are clean step response, etc. These are all related but different views of the same thing.

P.S. Reason why I chose 7 kHz square is that is not in sync with 44.1k sampling rate and it still has fundamental and first harmonic within the RedBook Nyquist band. But the test is really about hires (PCM or DSD, doesn't matter), at or above 352.8k sampling rate the analog filter usually begins to dominate anyway. 3 kHz would probably work fine too.
 

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
Accidentally building such a thing was a useful lesson for my 12-year-old self.

I had some challenges building first version of 1 MHz bandwidth current-feedback hybrid BJT-MOSFET power amp. The first set of power MOSFETs literally exploded because the thing started to oscillate. :D
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
I don't start with research about audibility, but about absolute perfectness. Mathematically we are supposed to be able to reconstruct the signal perfectly, so that's what I'm looking for. I don't take stance on how many out of billions of people are able to hear what.

I'm talking about reconstruction in it's all aspects, for which one part is phase response. Others are clean step response, etc. These are all related but different views of the same thing.

P.S. Reason why I chose 7 kHz square is that is not in sync with 44.1k sampling rate and it still has fundamental and first harmonic within the RedBook Nyquist band. But the test is really about hires (PCM or DSD, doesn't matter), at or above 352.8k sampling rate the analog filter usually begins to dominate anyway. 3 kHz would probably work fine too.

I applaud your quest but searching for perfection IMO is a fruitless search. You will never get there. :)

Surely the quest has to be relevant to what people can actually hear? You must have a view on audibility because you wouldnt waste your time on the project if you didnt think it was beneficial (well not unless this was a cynical exercise in extracting cash out of naive audiophiles, which Im sure its not). As a (ridiculous) example, what would be the point of producing a speaker that can reproduce signals up to 1MHz?

We have a test on this forum where member @Blumlein 88 has performed an 8x loopback recording of some tracks and so far everybody is really struggling to correctly identify the 8th generation against the original.

It would be interesting to see how you perform, its just good fun dont take it seriously, but I think it does put things into perspective.

https://www.audiosciencereview.com/...-choose-the-8th-generation-digital-copy.6827/



OK, lets presume that your work can have an audible positive impact. What do we do about the fact that the recording has all the limitations that you are trying to eliminate?

A second question, OK lets say you get a perfect signal out of your reconstruction filter. Would I be correct in saying that you now need an infinite bandwidth amplifier (and speaker) with no phase shift to reproduce it?
 
Last edited:

Miska

Addicted to Fun and Learning
Audio Company
Joined
Feb 20, 2019
Messages
615
Likes
448
I applaud your quest but searching for perfection IMO is a fruitless search. You will never get there.

I know, and I don't have a problem with that. Keeps me busy. I have long ago accepted that we mere humans will never reach perfectness on anything.

We have a test on this forum where member @Blumlein 88 has performed an 8x loopback recording of some tracks and so far everybody is really struggling to correctly identify the 8th generation against the original.

Are they? Elsewhere I explained why in my opinion the DAC in question (yours!) is actually partially fixing faults of the original. ;) Opening the files in audio editor the difference is immediately obvious.

But that loop is heavily bandlimited, so one needs to understand what it can represent and what it cannot. It is more bandlimited than for example output of your DAC. (further band-limiting is done by the ADC's anti-alias and decimation filters that also have other effects)

It would be interesting to see how you perform, its just good fun dont take it seriously, but I think it does put things into perspective.

I don't generally use myself as a yardstick. I have my interest areas and I work on those and I hear differences on those areas. For hard numbers I put my measurement rigs at work. My quest is "what can I do to with DSP to make this overall system perform better". Since I'm limited by time and resources I just hope hardware manufacturers come up with better DACs and don't settle in sentiment of "good enough". Imagine if people would have done that when CD was first rolled out! If I could put 72 hours in a day I could do more hardware stuff. Now I just applaud every time I see new DACs coming out from someone!

I have to admit I'm sometimes tired of listening the same test material 100th time during a day! :D
 

March Audio

Master Contributor
Audio Company
Joined
Mar 1, 2016
Messages
6,378
Likes
9,317
Location
Albany Western Australia
I know, and I don't have a problem with that. Keeps me busy. I have long ago accepted that we mere humans will never reach perfectness on anything.

Are they? Elsewhere I explained why in my opinion the DAC in question (yours!) is actually partially fixing faults of the original. ;) Opening the files in audio editor the difference is immediately obvious.

But that loop is heavily bandlimited, so one needs to understand what it can represent and what it cannot. It is more bandlimited than for example output of your DAC. (further band-limiting is done by the ADC's anti-alias and decimation filters that also have other effects)

I don't generally use myself as a yardstick. I have my interest areas and I work on those and I hear differences on those areas. For hard numbers I put my measurement rigs at work. My quest is "what can I do to with DSP to make this overall system perform better". Since I'm limited by time and resources I just hope hardware manufacturers come up with better DACs and don't settle in sentiment of "good enough". Imagine if people would have done that when CD was first rolled out! If I could put 72 hours in a day I could do more hardware stuff. Now I just applaud every time I see new DACs coming out from someone!

I have to admit I'm sometimes tired of listening the same test material 100th time during a day! :D

:) Im making no claims about my dac, it is an ESS9038 implementation so very representative as to whats currently out there from other manufacturers.

...and yes people are very much struggling to identify which is which, I have seen the results. I think it highly unlikely that my dac is fixing any issues of the ADC or the original recordings. Thats really not plausible proposition that it provides an inverse to those issues.
In any case surely it will still have the DAC filter issues you allude to? Plus you have already stated that the differences are readily visible in an audio editor.

So if we further band limit and filter (repeatedly) do we not have worse problems with phase?

Perhaps we could ask @Blumlein 88 to repeat the test with a square wave?

I think the point I would make is that I am most certainly not denigrating your work, however I think there are bigger fish to fry - it doesnt address the issue anywhere else in the chain beyond the DAC, which means it can only have very limited impact, but also that we need to establish correlation to audibility to say if there is actually any impact.
 
Last edited:
Status
Not open for further replies.
Top Bottom