Consensus about upsampling to 512 DSD

[jabbr]

I upsample to DSD512 with SDM DAC or 768k with PCM DAC (max that the PCM1704 chip is recommended for)

[jabbr]

I’m coming to the impression that once upsampled to DSD512, that SQ entirely depends on the analog electronics ...

[jabbr]

4 hours ago, NOMBEDES said:

Worthless thread.  A DSD - (whatever) copy of a bad recording sounds like.....well,  a bad recording.

@jabbr has it right.

 

 

A worthless recording is always that ?'

 

What I am getting at, and I haven't worked out all the details yet on this, is that when upsample to DSD512, the quantization noise is shifted further out of the audible range and easier to filter away.

 

Lets also consider digital switching noise: at 44 kHz, this is close to audible (if not), but at 24 Mhz, this switching noise, ground plane bounce etc is further pushed out of the audible spectrum and more easily filtered away.

 

Same for Ethernet, 10Gbe switching noise? Not even @GUTB is able to hear that. USB, still 8kHz packets ... hmmm

 

But let's say there is noise on a 24 Mhz (DSD512) signal ... thats very high frequency.

The exception may be close-in phase noise -- is this noise centered around 24 Mhz (center), or is this at the phase offset (e.g. 1Hz,10Hz etc)? Hmm....

 

Anyways food for thought.

[jabbr]

14 hours ago, fas42 said:

Crucial grokking is the converse of what jabbr just stated, ie. " A worthless recording is always that " - if the content has value for you, then it ain't bad, or worthless, ever; in fact, it's tremendously valuable because it's highlighting with a sharp intensity where the playback chain is failing ...

 

Enough vague blather ... obviously if the content has value, then it’s not worthless. 

 

Regarding the actual topic of this thread: upsampling allows the recording to present itself in the best fidelity (which might be counterintuitive to some folks) — the reason being that the analog filter operates well outside the audible range. Noise shaping effectively pushes the noise into the high frequency range eg Mhz where it can be thus selectively filtered away.

[jabbr]

34 minutes ago, davide256 said:

Whereas I experience up-sampling as cleaning up digital filtering artifacts at the expense of randomizing signal at the lower limits of resolution.

 

If the correct upsampling algo is employed then the lowest level information is preserved 

[jabbr]

http://soundlab.cs.princeton.edu/learning/tutorials/sensors/node35.html

 

[jabbr]

http://soundlab.cs.princeton.edu/learning/tutorials/sensors/node36.html

[jabbr]

44 minutes ago, fas42 said:

 

In audio, nothing is simple: upsampling may help the analogue side, but there is the possibility that the circuitry doing the upsampling may degrade the SQ, because the digital chattering is impacting the analogue areas, negatively.

 

I have found that offline upsampling can be remarkably effective - create a huge file, of the upsampled track prior to playback - subjective presentation, especially in the treble, is greatly enhanced. So, it works - trouble is, massive data storage requirements ... ?

 

The circuitry doing the upsampling can be remote, across a network, in which case the noise from the upsampling server is isolated. I've used both fiberoptic and wireless ethernet, for example. Furthermore, "digital chatter" at DSD512 is in the 24Mhz range which is filtered out.

 

You can investigate offline vs online upsampling but such large files stress the NAS. Also if you are using something like Roon you may be slogging the files twice, first to the Roon server and then to the endpoint device. I have found in my own system that online upsampling works best but YMMV. The power requirements for upsampling PCM to 768k for both @PeterSt XXHE and @Miskas HQPlayer are miniscule compared to DSD. I @audiventory is excellent for offline conversion/upsampling. I particularly like it for SACD ISO to DSF conversion (tags). I use all of these products.

[jabbr]

1 hour ago, barrows said:

1. I do not believe such a "consensus" exits.

2. Oversampling to DSD is going to be entirely DAC dependent, with some DACs it will be an advantage, with others it will be of no advantage or even detrimental.  My experience with oversampling to DSD using ROON and Audirvana suggests that with DACs using the ESS 9018 or 9038 chips, that the sweet spot is DSD 256, I found no advantage going beyond DSD 256.

 

Hard to say ... lots of folks are getting the best results with DSD512, is this a limitation of the ESS 9018 or 9038?

 

1 hour ago, barrows said:

3. So every DAC is likely to be different, and one has to try with their own DAC, in there own system, to see what sounds best.

4. The higher the rate of DSD, the more important a really good master clock becomes, so if one is looking to oversample to high rates of DSD, pay attention to master clock quality.

 

Huh? On what basis and what is your definition of "a really good master clock"?

 

On my initial analysis, I'd say the opposite is true. Suppose, for example, that a 1Ghz clock has a 10 Hz phase offset error of -60 dBc/Hz ... is that good? average? ... Now realize how many samples are being averaged in the audio band between 20 and 20kHz.... the potential accuracy of that signal is off the chart.

 

1 hour ago, barrows said:

 

Luckily, as this oversampling is software based, it is relatively easy for anyone to try oversampling to DSD at different rates to see what works best for them.

 

I dunno, sounds like you are spreading uncertainty about upsampling?

 

[jabbr]

1 hour ago, barrows said:

I would suggest that the above is not entirely true, here's why:  Let's take for example the PS Audio DS, and let's say we oversample everything to DSD 128 first.  Now if we do that, yes the DS will still oversample again up to its very high running rate, but the fact is that the filters necessary for going from DSD 128 up should not produce any audible artifacts, as all of their artifacts will be so far out of band to be entirely inaudible.  

It is also very possible that one can run a much better filter algorithm and modulator in software than what the DS can do in its fairly limited FPGA (in comparison to even an I5 based computer).  Personally, if I used a DS DAC at home, I would experiment with oversampling to everything to DSD 128 in Audirvana (and I would not be surprised at all if this made for a nice sonic improvement).

 

In my own experience upconverting to DSD128 and then making the jump to DSD512 *** ... or even trying to upsample DSD256 recordings to DSD512 ... it takes much much more resources than simply going from PCM44 or PCM192 to DSD512 using HQPlayer. I'm sure @Miska can confirm this but I seriously doubt a Xilinx 6 series FPGA can do this better/more efficiently. Audirvana hadn't done anything like NAA the last time I checked so this has ruled out that for me. But yeah I can see why if you want to offline convert to DSD128 and then stream the files to a DAC unchanged, that you'd have a vested interest in the position that DSD128 is "good enough".

 

In any case, I'll repeat, when you upsample in to the 22-24 Mhz range, you are also "pushing" any waveform related noise into that range and further away from the audible spectrum. Thus better.

 

*** I use @audiventory AuI for offline upsampling.

[jabbr]

6 minutes ago, Em2016 said:

Not so fast! It turns out there are physical limitations such as electronic component speeds, finite clock slopes, etc., that limit the amount of performance gain we could expect from raising the sample rate above a certain threshold. Such limitations are indeed starting to affect performance with quad DSD in D/A converters.

 

Of course, thankfully there are transistors, logic families, clocks, etc that are capable of handling increased frequencies without such physical limitations. Specific ICs will of course be constrained. FYI just to discuss what the real physical limitations currently are, Xilinx is incorporating ~5 Ghz (yes Ghz) 14-16 bit ADC/DACs onto their new series of "RF SoC/FPGA"s intended for things like phased array radar etc ... so with multibit there is even more room for upsampling (not that this is needed for audio, hardly, but just to discuss what the current physical limitations are)

[jabbr]

http://www.analog.com/en/analog-dialogue/articles/improved-dac-phase-noise-measurements-enable-ultra-low-phase-noise-dds-applications.html

 

Discusses phase error effects in the *100Mhz-Ghz* sampling range (again phased array radar applications) ... we've got the technology to do audio at DSD512 ?

[jabbr]

7 hours ago, barrows said:

@Miska, just listening tests with ESS 9018 and 9038, and also not via HQP, so perhaps those results could be different.  But, in subjective terms I could hear no advantage for DSD 512 vs. DSD 256.  I do hear significant improvement going for DSD 128 to DSD 256 with the ESS chips.

 

In my listening experience, the audible improvements higher than DSD256 are subtle and wouldn’t be surprised if DSD512 sounds identical at least with many implementations. 

Quote

I have a DSC "2" here (from Pavel, you are probably aware of this implementation?) but I have yet to get it up and running.  Too many projects, but I am looking forward to giving it a try with DSD 256 and 512. 

 

I think the major differences are with the analog stages. Being a fan of balancing, which could  add to accuracy, the I-V stage and quality of the analog filters do more to determine the sound (in my experience)

 

Lets just say that the iFi Micro BL sets a difficult bar for OPA output in the <$1000 DSD512 capable market ??‍♂️

[jabbr]

1 hour ago, Ron Scubadiver said:

 

It did not work for me.  That does not mean it will not work for anyone else.  Grace M9xx DAC on Win & Mac.

Looks like it’s only DSD128 at best. Hard to say but doesn’t really allow too much upsampling ??‍♂️

[jabbr]

1 hour ago, mansr said:

If that's not happening, the only thing one of these coolers will do soundwise is add fan noise.

To combine two threads, it’s possible that the vibration from the cooling pad sitting on the lap creates a more pleasing listening environment ?

[jabbr]

Hey, different setups will give different results and different folks will have different preferences. An N3700 is not made for DSD upsampling!!!

 

FWIW, I use these type of low powered machines as NAA to receive a DSD stream and pass along to the DAC. PCM upsampling uses vastly less CPU power.

 

 

[jabbr]

54 minutes ago, GUTB said:

Public service announcement: digital volume control sucks, use a real discrete analogue  pre-amp.

 

Public service announcement: learn some math and explain to yourself why a “digital volume control” applied to a DSD512 stream has less noise / distortion than the best potentiometer you know ?

 

Or do we need DSD1024 just for your ears?

[jabbr]

25 minutes ago, GUTB said:

Digital attenuation sucks because the result isn't hi-fi. Altough an active preamp stage adds distortion a well-matched preamp sounds much better. Fact.

 

Let’s take 120 dB SNR preamp which is a 1000000 ratio.  So an analog preamp with a 5V full range would have a noise floor of 5 microVolts. Got that?

 

DSD512 is what? 200 dB? So you’ve got about 80 dB of “volume control” before you impact on the -120dB preamp?

 

(that’s a very rough call I’m doing on my phone — @mansr or @Miska can be more precise) 

 

In any case do you get it? 

[jabbr]

7 hours ago, GUTB said:

 

In theory with enough bandwidth digital attenuation shouldn’t impact the sound. However in reality a discrete analogue line stage is needed to get the best results.

 

The advantage of a line stage is not with volume control, rather matching impedance and level.

 

Quote

 

Matched to the amp and lesser extent the source. Make sure impedance ratios are appropriate and don’t use a pre with a lot of gain with a high-gain amp, and as with everything else go by ear.

 

A pre can be as simple as a current amplifier / buffer eg the well known pass “b1”

Quote

if anyone wants to test this fact out for themselves just cut out the preamp and use either your DAC or something like HQPlayer to control volume in a high quality fashion. The SQ will suffer 10 out of 10 times.

If you have a pre you like with you amp and use HQPlayer as a volume control the SQ won’t suffer.

[jabbr]

26 minutes ago, GUTB said:

 

Alright a unicorn DAC which obviously disqualifies itself from discussion.

 

Care to name an example of a DAC and amp that won’t benifit from a pre?

 

Phasure NOS1a with new G3 output stage into  FirstWatt J2 as one not unicorn example. 

 

The iFi Micro BL as another example where you can either send the output direct or via the amplification stage (it has a headphone amp) — better direct.

 

But go ahead try to twist & turn and redefine your argument to try and maintain your self imposed “fact”ual accuracy. 

[jabbr]

4 hours ago, GUTB said:

 

Which preamps did you try which you found inferior to direct?

Not necessarily inferior but not necessarily better would be NP's own recommended B1: https://www.passdiy.com/project/preamplifiers/b1-buffer-preamp. This basic circuit also forms the basis of active filters/crossover such as the B4.

 

But again, I prefer DACs with excellent analog output stages --- the quality of the output stage makes the biggest difference in SQ in my experience.

[jabbr]

4 hours ago, Ralf11 said:

But, mea culpa, I have an ARC LS25 Mk II which gives me more functionality and euphonicizes the sound with its tubey goodness.  I happen to like the way it sounds with a simple outboard DAC. 

 

Nothing at all wrong with the classic tube sound -- this should be about having fun. A few years ago I was listening to a bottlehead crack/speedball that I was endlessly rolling caps and tubes. Ultimately I chose to increase accuracy, bass control etc but different folks have different tastes. 

[jabbr]

34 minutes ago, Audiophile Neuroscience said:

Still, I just don't like the idea of mangling with the signal even if its in the digital domain. So, why an analogue pre-amp to mangle the signal then?

 

How a “volume control” works on a DSD stream may not be intuitive to you. The DSD stream literally has a component in the digital domain — the carrier at 22-25 MHz — and a component in the analog domain — the audio. The DAC can simply remove the digital carrier and leave the analog signal alone. 

 

The signal hasn’t been “mangled” rather the upsampling is done to a different level. 

 

Quote

 

What has been traditionally offered is impedence matching, source selection, Voltage gain - line level pre-amp (excluding phono) to supposedly feed to a power amp to amplify current in its most linear section of its output. It is claimed that if you feed line level output into a power amp, unless the input sensitivity is sufficiently high, you will lose dynamics.

 

 

If your amp likes a preamp then by all means use it, rest assured, however, that what the preamp is has doing has nothing to do with volume control, rather amplitude and impedance matching. It will work just as well as when you “volume control” your DSD512 stream — to keep this focused on the topic of this thread...

 

Quote

All that theory aside, an analogue pre-amp just sounds better than using a digital volume control. I have auditioned this extensively with various high end products that offer digital volume control eg a Bricasti DAC which I owned at one time but other products and in different systems. A pre-amp just sounds better for me.

Again, use your preamp along with volume control in your DSD upsampling software! It will work just as well if not better. 

 

BTW: the Bricasti DAC only accepts up to DSD128, so the conclusions you’ve drawn from this example may not apply to the topic at hand. also really hard to know what any one DAC is doing under the hood. 

[jabbr]

6 minutes ago, fas42 said:

Are you saying that years of experience trying things, and coming up with the same answers every time, which always confirm an underlying concept, count for nought?

 

No, unless you can articulate your “same answers every time” such that a reasonably competent person can replicate, or else provide an objective, independently verifiable measurement confirming,  no it’s not worth anything. 

 

What I hear is that you supply the same explanation in every single thread — prove  to me that you aren’t a bot ?

[jabbr]

10 hours ago, fas42 said:

 

There are a lot of medical people floating around here - I've got the impression over the years that a rather important first step is to make a correct diagnosis of a patient's complaint, rather just than trying every possible treatment that may vaguely relate to the symptoms ...

 

Oooph — the Flexner report was published in 1910 https://en.m.wikipedia.org/wiki/Flexner_Report

 

It derided the then current practice of medical training under apprenticeship, where a patient would see any of a number of traveling salesmen “snake oil” purveyors who would listen to their complaints, “diagnose” and treat with a concoction of hopefully harmless herbs or potions or worse. 

 

It called for a more standardized and science based training. There are text books. There are well articulated methods. Treatments need to be proven. (Nothing is perfect however!)

 

Diagnosis: there is a specific teachable diagnostic method — you have nothing specific. 

 

Not that you are killing people with bad audio but you make my point.