Some of HiRez remastered songs are nothing but upsampled 44.1K files, but most of newly mastered recordings are done by newer generation of ADC directly from master tapes, so they should arguably and technically be better. The reason why I said 'arguably' is some of them, such as BlueNote HiRez versions are much less compressed now, and they sound totally different from what we're accustomed to. I'm not sure if that's a part of anti-loudness war syndrome, buy in my opinion, tastefully well compressed music sounds better than tasteless uncompressed version of the same song. Anyway, I mean it's not always 1:1 comparison is possible between 44.1K and HiRez, because the mastering process is different.
Doede, Marcel,
As I recall, the DF bypass mode of the PCM1794A also sets the DAC in to mono, so the multiplexed I2S input data stream must be demultiplexed to mono left and right streams before arrival to the mono DACs. That would seem to enable the simple synchronous integer multiplication of WDCK and BCLK to make the DACs read the same input sample data multiple times in succession. Effectively, producing an upsampled rate increase, but without filtering of the samples by the OS interpolator.
For example, if you were to multiply a CD rate clocks of, WDCK = 44.1KHz and BCLK = 2.8224MHz by two, then multiplying those clock signals now become, WDCK = 88.2KHz, and BCLK = 5.6448MHz. Which should force each DAC to read each sample of it's own mono data stream, twice. Voila', upsampling by sample repetition, and without interpolation-filtering. Providing the higher frequency clock signal to drive the SDM noise-shaping process more effectively.
Perhaps, the most reliable means to generate the required multiplied clock rates is to simply use the recovered SCLK to drive a synchronous-counter IC based clock-divider chain. The derived clocks can be utilized to drive a Input Receiver set to, slaved I2S output mode. Then tap whatever are the desired clock signal frequencies from the counter outputs, which will always be synchronous to the input sample data. Although, the maximum obtainable BCLK frequency taken from the divider-chain would be SCLK/2. I've successfully utilized such a synch. counter based divider-chain in an experiment where I was blanking every other sample to approximate return-to-zero DAC operation. The divider-chain worked well.
Last edited:
Multi-purpose Hans, hence I was thinking of a device like the portable TASCAM DR-100 MK3 or the cheaper DR-40. These both have microphone input and balanced XLR input.
I stopped using the turntable for some time. This was for reasons that digital reproduction was becoming significantly better than analog on my turntable setup (as commented on by others who apparently don't seem to care about hurting my feelings). This was in comparison to many digital recordings produced/reproduced over the past few years, and the motivation for work on the phono section. I am currently listening to the turntable on one breadboarded channel single ended (half stereo), although the phono-stage has balanced outputs that might be an improvement in CMRR, hence the desire for a balanced XLR input recorder. In any event the turntable is sounding significantly better.
It is expected that current digital recordings of the kind found especially good were never recorded at 44.1KHz. Hence however the Tidal/Audirvana/whatever was altering the original recordings the outcome is good. It was interesting that from my perspective the file comparisons you provided the upsampled files were both inferior to the NOS files, notwithstanding one set being subsequently downsampled. To me this suggests greater problems in mechanisms of upsampling, seemingly being verified by my experiences with listening to Tidal files. This suggests that recordings have advantage in being done at higher sampling rates/ resolution and that playback files can prove most effective at 44.1KHz.
Gerrit
Only in the sense of whether the architecture includes noise-shaping to move the in-band quantization noise out-of-band. Doede has a great spectrum analyzer image, back in post #947 (https://www.diyaudio.com/forums/digital-line-level/371931-makes-nos-sound-95.html#post6718498), showing how the quantization noise of the PCM1794A moves undesirably in-band as the noise-shaping process is run at a lower than recommended frequency.
On second thought, my suggestion in post #1023 would only work if the sample data could be read by the DAC in parallel, which it can't be for the PCM1794A, nor for most any other audio DAC. So, no go. Many DIR chips will, however, output the current sample twice in succession, should the next sample not yet be ready to send to the DAC. Which, would be the case should the I2S clock signals be run at a rate greater than that required to extract the input samples using a clock divider chain to produce the I2S clocks. As described in post #1023 above.
Doede, If you read this, I'm wondering whether you could simply drive the PCM1794A SCLK pin at some synchronous multiple of BCLK, to get the quantization noise moved fully out-of-band, instead of driving it at the BCLK rate?
Doede, If you read this, I'm wondering whether you could simply drive the PCM1794A SCLK pin at some synchronous multiple of BCLK, to get the quantization noise moved fully out-of-band, instead of driving it at the BCLK rate?
Last edited:
Final Tests Coming...
FYI - Watch for two new listening experiments. One will test a possible root cause explanation for why typical OS interpolation-filters introduce audible artifacts. The other will listen for whether 176.4KHz PGGB upsampled files sound superior to their 44.1K source files. Both played via a NOS DAC. The final hope being, to discover playback which is subjectively superior to NOS. More details on both experiments soon.
These will be the final listening experiments of our investigation. After which, I will write and post an investigation summary, and conclusion.
FYI - Watch for two new listening experiments. One will test a possible root cause explanation for why typical OS interpolation-filters introduce audible artifacts. The other will listen for whether 176.4KHz PGGB upsampled files sound superior to their 44.1K source files. Both played via a NOS DAC. The final hope being, to discover playback which is subjectively superior to NOS. More details on both experiments soon.
These will be the final listening experiments of our investigation. After which, I will write and post an investigation summary, and conclusion.
Last edited:
I'm coming in WAY late to this but I can relate what I've heard. I've listed to LOTS of DACs over the years, including delta sigma dacs with selectable filter roll off and numerous non-oversampling (NOS) ones. Delta sigma is problematic in itself for some reason, plus the lack of digital filter is the main reason why NOS sounds better. NOS still has idiosyncrasies sonically but I think it's superior to anything delta sigma. Traditional ladder dacs with oversampling are good to0, hence my conclusion that it's BOTH delta sigma and digital filtering, probably mostly the former.
But I am not sure the DAC produces lowest distortion at these higher modulator rates. The data sheet figures seem to indicate the contrary.
Distortion aside, 12.288MHz appears to be the highest modulator speed available on 1792 being 96000x128, ref top of P32 data sheet.
TCD
Better late, than never. Welcome.
Perhaps, you would participate in our final two listening experiments when they are released?
Thanks. 104 pages of discussion already??? I worry that we engineers get way into the weeds and overanalyze things. Occam's razor based on listening I think is the tell here. As I said previously, delta sigma for all it's measurable goodness was created as a cheaper substitute for ladder dacs. I believe there is a flaw in it that manifests itself as a problem psychoacoustically. I say that because I've listened to at least a dozen, all sorts of DAC chips and filter roll offs. None were keepers. Second, some digital filter roll-offs exhibit the same psychoacoustic problem. I recall listening to a Primare CD player once. It used the vaunted 1704 ladder dac chip. Made my ears ring, literally. Unlistenable. Several Naim cd players I've owned used the same 1704 and sounded fine ...
Yes, but the noise is correlated with the audio signal, not purely random. So, it may be audible to some people as excessive brightness, or something like that.
Yes, I would love to participate. Sorry I missed your question previously
Possibly. Spectrum analyzers show, however, that the quantization noise can be spectrally relocated completely to the ultrasonic (see post #947), and so, can be rendered totally inaudible itself. Maybe, assuming that I correctly interpret what you are suggesting, should the relocated quantization noise not be sufficiently de-correlated from the signal, it could produce an audible modulation of the signal. I don't know.
ESS says they trained all of their executive team how to hear signal correlated noise from dacs (except Martin Mallinson who was unable to learn how). ESS says the human ear is exquisitely sensitive to very low levels of such noise. Some people can hear it without any training. Doesn't sound like resistor noise, it tends to sound like a change in the sound of music; sort of a bright sounding distortion, perceptually speaking.
The noise can be measured by FFT: Noise floor is measured with and without audio present, or else it is measured at various DC offsets when the dac is fed a PCM signal representing the particular offset. Any change in the measured noise floor indicates signal correlated noise.
The noise can be measured by FFT: Noise floor is measured with and without audio present, or else it is measured at various DC offsets when the dac is fed a PCM signal representing the particular offset. Any change in the measured noise floor indicates signal correlated noise.
Last edited:
You certainly can get weird artefacts out of an improperly dithered sigma-delta DAC or ADC, and the single- bit versions can't be dithered in accordance with dither theory. Multibit sigma-delta DACs can be dithered properly, but they require DEM algorithms, some of which produce artefacts of their own.
The most well-known issues are idle tones that get modulated by the signal during soft passages. In the chaotic mode of my valve DAC, I've used chaos to get rid of those - a specific type of chaos that Lars Risbo recommends in his PhD thesis: one open-loop pole just above +1 and one just below -1 in the z domain. It indeed suppresses the tones, but instead you get nonstationary noise (woosh-woosh-woosh) when playing silence. Fortunately the other modes (quasi-multibit with embedded pulse width modulator) have no such issues.
The most well-known issues are idle tones that get modulated by the signal during soft passages. In the chaotic mode of my valve DAC, I've used chaos to get rid of those - a specific type of chaos that Lars Risbo recommends in his PhD thesis: one open-loop pole just above +1 and one just below -1 in the z domain. It indeed suppresses the tones, but instead you get nonstationary noise (woosh-woosh-woosh) when playing silence. Fortunately the other modes (quasi-multibit with embedded pulse width modulator) have no such issues.