World’s First Valid Comparison of PCM versus DSD?

[Miska]

Why do you keep throwing that in ? That's comparing apples with oranges.

It certainly isn't a problem with material like the 24/192 PCM recordings from Barry.

SACDs ultrasonic artifacts are deliberately rolled off in commercial players, and perhaps that smoothing is preferred by many SACD devotees? i.e. it sounds less "digital".

 

24/192 still needs to be brickwall filtered, because otherwise you cannot reach -144 dBFS attenuation by 96 kHz (Nyquist frequency). While I have shown that for example Schiit Loki can still reproduce 250 kHz tone from DSD64, although attenuated.

[sandyk]

24/192 still needs to be brickwall filtered, because otherwise you cannot reach -144 dBFS attenuation

 

IIRC, the highest recorded frequency in Barry's recordings is around 57kHZ. Very few commercial recordings will have genuine musical content even that high .

What is the S/N of your amplifier ? Very few amplifiers even achieve 120dB unweighted S/N . Are you talking theoretically, not what is achievable in practice ? .

[Miska]

Sony had to introduce that HF rolloff due to happenings like amplifiers being destroyed originally, back in the days when Sony even sold tweeters with HF extension to 100kHz, especially for their new

(at the time) SACD format.

 

PCM needs HF filter and even a steeper one, because the content spectrum keeps repeating itself at every multiple of sampling rate. This is called "reconstruction filter" and you need it in all D/A converters regardless of type.

 

For example 19+20 kHz IMD test tone with PCM NOS DAC looks like this, even though it has fairly strong analog reconstruction filter. So think how it would look like in a filter-less DAC.

[Miska]

IIRC, the highest recorded frequency in Barry's recordings is around 57kHZ. Very few commercial recordings will have genuine musical content even that high .

 

But ADC chips usually go flat to around 80 kHz at 176.4/192 rates, which leaves less than 20 kHz to roll-off. I think the ADC chip Barry (Metric Halo) uses is AK5394A (delta-sigma ADC running at 6.144 MHz) which has digital filter with 87.072 kHz pass-band and 104.928 kHz stop-band at 192 kHz sampling rate and 120 dB attenuation. So the transition band is 17.856 kHz wide and it rolls off to -120 dB within that band.

 

What is the S/N of your amplifier ? Very few amplifiers even achieve 120dB unweighted S/N . Are you talking theoretically, not what is achievable in practice ? .

 

If we talk about 24-bit PCM we should talk about 144 dB. We can talk about 120 dB too, but then we should be talking about 20-bit PCM.

 

I like to talk about what I can computationally discover from digital data produced by ADC.

[sandyk]

I like to talk about what I can computationally discover from digital data produced by ADC.

 

You appear to be for the most part talking theoretically, not what is achievable in the real world due to Johnson noise and other considerations. That applies to both PCM and DSD PCM too.The valve amplifiers that many seem to favour would be lucky to achieve 100dB S/N, and with HF response typically well below that of a well designed SS amplifier, so it would all seem to be rather academic. .

[Miska]

You appear to be for the most part talking theoretically, not what is achievable in the real world due to Johnson noise and other considerations. That applies to both PCM and DSD PCM too.

 

Since I used to work on passive sonar systems, I use similar measures and methods to inspect quality too.

 

Even if the signal is noisy, there are good algorithms to dig out really quiet signals that have been buried in noise. And you can also easily hear -20 dB below the noise level (assuming white noise). How low you can actually go depends on how well you train your ears.

 

You can try yourself with this sample:

http://www.sonarnerd.net/noise/noise1k.flac

 

Here is Schiit Loki reproducing -120 dBFS 1 kHz tone:

[Hiro]

If we talk about 24-bit PCM we should talk about 144 dB. We can talk about 120 dB too, but then we should be talking about 20-bit PCM.

 

And if we look at 24bit PCM converters THD+N, we often can't even talk about 20-bit performance, but rather 17-bit....

[Jud]

If SACD/DSD were actually better, like Blue Ray, it would have become ubiquitous.

 

Yes, it's always the higher quality entertainment medium that wins out in the market, like Beta did over VHS...oh, wait.

[sandyk]

Miska

I wondered when you were going to bring the sonar bit into it. (grin) I reckon that Paul R would be past it for that kind of caper these days too, or even his old naval instructor. (LOL) Again, in the real world it simply doesn't matter at those levels. We aren't sitting there with headphones on and with ultra quiet electronics. For most people, the ambient noise level precludes things like that. From even my own experience, you even need to turn off any air conditioning, fans etc. to properly hear low level ambience with even a very good system. If you have a family, then forget really low level details unless they go out, and you live in a very quiet residential area. Even the sound of cicadas outside in summer, or rain in a tropical area makes this almost impossible for many.(Especially for those with ESL in their user names who reckon that the practical limit for most people's hearing is WAY below those limitations.) BTW, even a DSD1792 has a S/N and dynamic range of only 129dB .

 

As Hiro just pointed out.....

 

BTW, seems that Trendsetter Electronics has those lovely,low noise dual transistors at VERY good prices, and they were in stock a couple of days ago. They could be just the shot for some of an ex Sonar instructor's DIY designs !

 

Plug in LS352 or LS313. Trendsetter Electronics Online Store - Parts, Components and Electronics

Their present prices are VERY reasonable.

[Miska]

We aren't sitting there with headphones on and with ultra quiet electronics.

 

Hey, but that's how I test my things. In quiet room with headphones that have 30 dB passive ambient noise attenuation. You know, engineering needs to go have at least 10x safety factor. I prefer to design my digital filters with 180 dB or more stop-band attenuation.

 

But in any case, the particular ADC decimation filter in question had 120 dB in about ~18 kHz band, I would call that a brickwall filter. And it has time domain implications.

 

BTW, even a DSD1792 has a S/N and dynamic range of only 129dB .

 

So with enough gain applied, you could hear tone at -149 dB... (100+ dB of gain is not unusual for a passive sonar system)

 

Good thing underwater is that you don't have problem with cooling and you don't have any RF noise either, unless you bring it in...

 

BTW, seems that Trendsetter Electronics has those lovely,low noise dual transistors at VERY good prices, and they were in stock a couple of days ago. They could be just the shot for some of an ex Sonar instructor's DIY designs !

 

Plug in LS352 or LS313. Trendsetter Electronics Online Store - Parts, Components and Electronics

Their present prices are VERY reasonable.

 

Looks good, although they probably don't ship to Europe.

[sandyk]

Yes, it's always the higher quality entertainment medium that wins out in the market, like Beta did over VHS...oh, wait.

240 or so lines (if the machine was kept in good nick) , were high quality entertainment ? Hell, even S-VHS recorders had obvious limitations at around 400 lines. If we could record it, play it back later and recognise the faces we were for the most part happy. In my case, I owned 2 different S-VHS recorders over the years, a Blaupunkt and a JVC. They still had clear limitations compared with an off air ghost free signal .They remind me of the limitations of the vast majority of security cameras these days.

[Jud]

240 or so lines (if the machine was kept in good nick) , were high quality entertainment ? Hell, even S-VHS recorders had obvious limitations at around 400 lines. If we could record it, play it back later and recognise the faces we were for the most part happy. In my case, I owned 2 different S-VHS recorders over the years, a Blaupunkt and a JVC. They still had clear limitations compared with an off air ghost free signal .They remind me of the limitations of the vast majority of security cameras these days.

 

Notice I said higher, not that either was high quality. Though I did see some stuff from a friend with a Beta system that were the highest quality television pictures I'd encountered prior to the Faroudja line doublers.

 

Anyway, back to audio - anyone who thinks sound quality had anything at all to do with the failures of both SACD and DVD-A to become mass market media (though SACD in particular seems to have found a smallish comfortable niche for the time being) wasn't paying close attention to the uncanny ability of the music industry to find new feet to shoot itself in as regards their treatment of their most fanatical customers as thieves/pirates.

 

And by the way - IMHO, there's a lesson in the SACD/DVD-A "format wars" for today's manufacturers: You can spend time selling your stuff or dissing the other guy's stuff. Some people mistake the two for each other, but perhaps we can learn not to repeat that error.

[Miska]

Really curious about the results of this comparison (limitations of the analog filtering notwithstanding), and even more so about this one, 32/176.4 PCM vs DSD128, as the 16/352.8 pcm thing doesn't exist in nature

 

OK, since it's holiday time I had some more time to take measurement plots...

 

UD-501 at 176.4k PCM with "sharp" filter:

 

Since we have possibility to keep the same analog filters, but switch off the oversampling filters we can play with things.

 

So same, but in NOS mode at 176.4k:

whoops, quite a lot of timing spread.

 

Since 44.1/88.2/176.4/352.8 (and standard DSD rates) are not in sync with 1 kHz signal (unlike 48/96/192/384), what happens if we use 352.8 instead and zoom-in to see more (notice new horizontal scale):

this is a bit better already.

 

How does this compare to DSD? Let's see

ok this is still better than 352.8k PCM, although we use same base clock rate. But pretty close.

 

Now let's see how the 20 kHz sine works out in this case, since it's not in sync with sampling rate either.

 

For reference, let's take DSD again.

 

Let's see how it looks like at 352.8k sampling rate

looks decent. Again close match with the DSD.

 

Now let's drop to 176.4k rate, the oversampling rate used for TDA1541A and many other older R2R ladder DACs.

OK, this ain't pretty anymore.

 

Let's see how 88.2k fares, it's still 2x to RedBook.

Oh dear...

 

And let's see how the NOS RedBook PCM manages with our 20k sine.

This was supposed to look like sine, wasn't it?

[Miska]

I also did some plots with one Sabre DAC.

 

1k square at 176.4 PCM:

 

1k square at DSD128:

[magnum innominandum]

Miska,

 

I am only interested on how to make current DAC technologies sound great, that means delta-sigma.

 

Seems many DAC makers did not get that Memo. Pure multibit ADC's and DAC's remain in heavy use in fields outside audio, where what counts is hard performance massively averaged FFT's and "personal opinion on sound".

 

Leaky oversampling filters, or even worse (shrug) NOS DACs, tend to have much worse IMD because of intermodulation with the ultrasonic images and those are heavily correlated unlike delta-sigma noise.

 

If you declare ultrasonic images to be "IMD" (which they are not) - yes they measure much poorer IMD. But that is because you are not measuring correctly. If I declared the rising noisefloor above 15K of DSD to be "IMD" you would take exception I presume?

 

If not I can take a simple old fashioned analogue audio analyser and declare that DSD at -20dBFS has around 1% IMD, depending on the precise filtering used...

 

So, if we analyse IMD we either include noise, in which case a standard PCM DAC looks great, a PCM DAC at high sample rates with a sensible analogue filter looks very good and filtered DSD and unfiltered PCM look awful, or we include the harmonic products, in which case non of these have much of anything in terms of IMD, competent design given.

 

The ultrasonic images of (unfiltered) PCM and the ultrasonic noise of DSD are not IMD, but we may view both as "non-harmonic" distortion. Which is the lesser evil (or is it filtered PCM) is down to listening and I suspect to personal preference.

 

In all honesty, my Pass D1 maybe an antique, but it still sounds by far more "analogue" to these ears than any Delta Sigma system I heard in recent times (e.g. Oracle).

 

So now give me a PCM DAC that has both clean time domain response and clean frequency domain response at the same time. What I usually see is this kind of time domain response:

[ATTACH=CONFIG]9653[/ATTACH]

...or this kind of frequency domain response:

[ATTACH=CONFIG]9654[/ATTACH]

 

Funny,you claim to be showing a "PCM DAC", but what you are showing based on the file namesis actually the Teac UD-501 which uses a Delta Sigma DAC. You are quite living the dream, aren't you?

 

PCM has been already spoiled at the ADC side with brickwall filters.

 

Delta Sigma derived recordings have already been spoiled at the ADC side with massive amounts of noise.

 

So, neither is particularly good, but the problems differ.

 

So some people prefer to live with the obvious sonic flaws of delta sigma systems, while others prefer to live with the flaws of PCM. I suspect a system less imperfect than DSD and PCM would be instantly recognised as superior by both.

 

OK, so you have a NOS R2R DAC and you claim it measures great? Quite contradictory unless it's something like Phasure NOS1 that is intended to be run with off-board oversampling. I'm even more eagerly waiting to see some measurement results and hear why such great device would be discontinued.

 

My DAC has a HDCD Digital filter, set to minimal filtering. It is not NOS (it is old, not new old stock), but as the HDCD part works it must be bitperfect.

 

Why Pass Labs discontinued it? Ask Nelson Pass, I think the PCM-63 became unavailable, it may be their amplifiers sold better, who knows. It is still the best DAC I have heard with any kind of recording etc.

[magnum innominandum]

Miska,

 

OK, since it's holiday time I had some more time to take measurement plots...

 

Could it be that your problem is the fact that you use a Delta Sigma DAC to demonstrate the behaviour of a PCM System?

 

That is like using a tractor with heavily loaded trailer to demonstrate how a Camaro corners?

[Hiro]

I also did some plots with one Sabre DAC.

 

1k square at 176.4 PCM:

[ATTACH=CONFIG]9739[/ATTACH]

 

1k square at DSD128:

[ATTACH=CONFIG]9741[/ATTACH]

 

 

Many thanks for posting all these additional measurements ..... most interesting indeed!

[Miska]

Could it be that your problem is the fact that you use a Delta Sigma DAC to demonstrate the behaviour of a PCM System?

 

That's 99+% of the market today so it is most valid. But I can go and dig out the old BB DAC from storage too.

[Miska]

If you declare ultrasonic images to be "IMD" (which they are not)

 

No, I was talking about intermodulation between ultrasonic images and between ultrasonic images and in-band tones.

 

So, if we analyse IMD we either include noise, in which case a standard PCM DAC looks great

 

No, most of the time it doesn't because the ultrasonic image level is higher than level of ultrasonic noise from DSD. And in addition intermodulation products of/with ultrasonic images generate discrete frequencies while DSD noise just lifts the noise floor. I have not heard complaints about disturbing background hiss with DSD.

 

In all honesty, my Pass D1 maybe an antique, but it still sounds by far more "analogue" to these ears than any Delta Sigma system I heard in recent times (e.g. Oracle).

 

Try out $150 Schiit Loki and then try to make some PCM thingie sound equally good at the same price. Or go with dCS Vivaldi and upsample to DSD on the other edge.

 

Funny,you claim to be showing a "PCM DAC", but what you are showing based on the file namesis actually the Teac UD-501 which uses a Delta Sigma DAC. You are quite living the dream, aren't you?

 

Well, I didn't start the reference to a particular DAC, but since other people started posting figures of this particular DAC and you started questioning my measurements based on those I gave my set of results with that particular DAC. So other people can take & verify the results.

 

Delta Sigma derived recordings have already been spoiled at the ADC side with massive amounts of noise.

 

Well, again, that's 99+% of records currently on market. But disregarding that noise part, in addition when delivered as PCM it also has time domain performance spoiled...

 

So some people prefer to live with the obvious sonic flaws of delta sigma systems, while others prefer to live with the flaws of PCM.

 

I don't mind, I support both at equal level. Although I provide more oversampling filters for PCM than for SDM. So in a way I still have better support for PCM than SDM.

 

My DAC has a HDCD Digital filter, set to minimal filtering. It is not NOS (it is old, not new old stock), but as the HDCD part works it must be bitperfect.

 

As far as I know, HDCD decoder does contain oversampling filter. So the exact same samples seen by the decoder input don't end up at the analog conversion stage driving the ladder.

 

Why Pass Labs discontinued it? Ask Nelson Pass, I think the PCM-63 became unavailable, it may be their amplifiers sold better, who knows. It is still the best DAC I have heard with any kind of recording etc.

 

How does this qualify as objectively better? I'm completely fine everybody having their own preferences when it comes to sound, but that doesn't qualify as objective technology evaluation.

[Miska]

OK, measurements galore!

 

Now BB DF1700 8x oversampling digital filter + PCM1700 R2R ladder DAC (I think the analog reconstruction filter was maybe 3rd order 30 kHz corner):

 

1 kHz square rise time with same time scale as earlier:

 

I had to change the time scale to make it look closer like the earlier ones:

 

1 kHz square scope plot zoom-out:

Notice the irregularity of the second high level period due to not-in-sync wave vs sampling frequency.

 

20 kHz sine looks decent given the 352.8 kHz conversion rate and low reconstruction filter corner frequency:

 

Output spectrum of the 20 kHz sine doesn't look very pretty though, notice the image frequencies repeating at both side of the 352.8k conversion frequency:

 

Analyzer IMD reading is 0.025% for 19+20 kHz test tone.

[sandyk]

And by the way - IMHO, there's a lesson in the SACD/DVD-A "format wars" for today's manufacturers: You can spend time selling your stuff or dissing the other guy's stuff. Some people mistake the two for each other, but perhaps we can learn not to repeat that error.

 

Hi Jud

 

Judging by this thread, hell will freeze over first. Most people hate ads on TV that specifically denigrate another manufacturer's product. The fact is, that the once mighty Sony is a mere shadow of what it once was, and as Barry would say about the majors of the recording industry, they keep shooting themselves in both feet.

I hope that you are enjoying quality time with your family, and the weather is pleasant ?

Best Wishes

Alex

[Allan F]

Hi Jud

 

Judging by this thread, hell will freeze over first. Most people hate ads on TV that specifically denigrate another manufacturer's product. The fact is, that the once mighty Sony is a mere shadow of what it once was, and as Barry would say about the majors of the recording industry, they keep shooting themselves in both feet.

I hope that you are enjoying quality time with your family, and the weather is pleasant ?

Best Wishes

Alex

 

The majors in the recording industry are all mere shells of there former selves. Self inflicted foot injuries no doubt played a major part, as evidenced by the industry's reluctance or inability to change and adapt in the face of piracy.

[Nes]

I'm kinda put off by all this. I have great respect for Charles Hansen and Ayre and other audio entrepreneurs. I am in the market for a DAC and the new Ayre QB-9 DSD was high on my list. Now I'm not so sure. Why implement DSD playback at all on the QB-9? It drives down the value proposition of the PCM 'side' of the DAC. Why not stick to your guns and invest the time and resources to even further optimize PCM playback? Even if only in my head, I now would feel 'cheated' listening to DSD on a QB-9.

 

For the record, my current source is a Theta Compli and I mostly listen to my (growing) collection of ˜250 SACDs. Reason is SACD playback clearly outperforms CD playback with the Compli in my system, so much so that I only have very few CDs that I still play as I can't listen to the rest. I also recently added a Schiit Bifrost (Uber), but have been underwhelmed by it (20 hours on it so far).

[robocop]

Nes you shouldn't be put off. By adding DSD capability you at least have the opportunity to play back DSD. You never know someone will crack SACD's one day and you can play them back from your computer. I bought the Benchmark DAC2 with DSD for future use. In the mean time I play all my PCM files. These latest editions of DAC's have not only DSD capability but take PCM listening to new heights. PCM playback is being optimised all the time.

[astrotoy]

Nes you shouldn't be put off. By adding DSD capability you at least have the opportunity to play back DSD. You never know someone will crack SACD's one day and you can play them back from your computer. I bought the Benchmark DAC2 with DSD for future use. In the mean time I play all my PCM files. These latest editions of DAC's have not only DSD capability but take PCM listening to new heights. PCM playback is being optimised all the time.

 

You should read the lengthy threads about ripping SACD's - Sony was the one who allowed it, building their first Play Station 3 models with the ability to rip SACD's that they so desperately wanted to be copied protected. Ironic that the left hand didn't know what the right hand was doing. I'm having a great time playing both two channel and especially multichannel rips of my 400+ SACD's on my E28. I went the other way, with both a PCM DAC (BADA2) and a DSD DAC (E28). When and if one of the top guys makes a better multichannel DSD DAC than the E28, I probably will go for that. Don't think anyone has so far.

 

Larry