Nope, clearly that does happen, but it's IMHO insane to seek for minute details caused by FIR filters and not fixing the EMI issues at hand. Are these not way more bothersome to you?
OK, you're right. Has nothing to with the original topic and I'll leave it at that.
I just couldn't understand how one would choose a non-filtered upsampling algorithm, get all the aliasing just because it's less stressful on the CPU. But as I said, that's obviously just me and I don't want to clutter this thread...
PGGB upsampled 88.2 experiment result interpretation
Hans, thank you for formatting the PGGB 88.2 upsampling results convienently in to an easy to read table. I hope you don't mind if I speak to some of the key indications, and don't hesitate to correct any misunderstanding I may have about the meaning of your table's data.
_______________________________________________________________________________
Perhaps, I should first clarify that files which are upsampled before being sent to a NOS DAC, are in fact, now Oversampled files. In this case, utilizing a NOS DAC avoids the data being passed through a second (internal) OS interpolation-filter, which OS DACs and OS disc players inherently feature. This was a primary point of the experiment.
The most obvious result is that half of us performed the experiment via a NOS DAC as intended, while the other half (which, includes myself) utilized an OS DAC or OS disc-player. That divergence is responsible for the most interesting result of the experiment, which is that 100% of NOS DAC users file comparisons resulted in no subjective preference for the 44.1 source files played NOS, over their upsampled 88.2 versions also played NOS. In other words, the data shows that they felt the PGGB 88.2 upsampling eliminated the characteristic subjective difference between NOS and OS playback. This is exactly what we were hoping to find, but it is lacking statistical confidence. It also indicates that, not-good-enough performance of typical integrated internal OS interpolation-filters is responsible for Oversampling's characteristic difference in sound from NOS playback. By extension, we can surmise that as the performance of an OS FIR interpolation-filter moves closer to a theoretically ideal windowed-SINC function, the more it sounds like NOS.
Those respondents utilizing either an OS DAC or an OS disc-player for the experiment, felt that they heard no preference in 42% of the file comparisons, but did hear a preference in the remaining 58% of comparisons. Further indicating that typical internal (chip based) OS interpolation-filters produce audible artifacts. Keeping in mind, as in all of our tests, that we simply are not a large enough group to place any statistical confidence in our results. These results are, however, starkly different between OS playback systems, and NOS playback systems. The 100% no preference heard result obtained by NOS DAC users just jumps out.
Hans, thank you for formatting the PGGB 88.2 upsampling results convienently in to an easy to read table. I hope you don't mind if I speak to some of the key indications, and don't hesitate to correct any misunderstanding I may have about the meaning of your table's data.
_______________________________________________________________________________
Perhaps, I should first clarify that files which are upsampled before being sent to a NOS DAC, are in fact, now Oversampled files. In this case, utilizing a NOS DAC avoids the data being passed through a second (internal) OS interpolation-filter, which OS DACs and OS disc players inherently feature. This was a primary point of the experiment.
The most obvious result is that half of us performed the experiment via a NOS DAC as intended, while the other half (which, includes myself) utilized an OS DAC or OS disc-player. That divergence is responsible for the most interesting result of the experiment, which is that 100% of NOS DAC users file comparisons resulted in no subjective preference for the 44.1 source files played NOS, over their upsampled 88.2 versions also played NOS. In other words, the data shows that they felt the PGGB 88.2 upsampling eliminated the characteristic subjective difference between NOS and OS playback. This is exactly what we were hoping to find, but it is lacking statistical confidence. It also indicates that, not-good-enough performance of typical integrated internal OS interpolation-filters is responsible for Oversampling's characteristic difference in sound from NOS playback. By extension, we can surmise that as the performance of an OS FIR interpolation-filter moves closer to a theoretically ideal windowed-SINC function, the more it sounds like NOS.
Those respondents utilizing either an OS DAC or an OS disc-player for the experiment, felt that they heard no preference in 42% of the file comparisons, but did hear a preference in the remaining 58% of comparisons. Further indicating that typical internal (chip based) OS interpolation-filters produce audible artifacts. Keeping in mind, as in all of our tests, that we simply are not a large enough group to place any statistical confidence in our results. These results are, however, starkly different between OS playback systems, and NOS playback systems. The 100% no preference heard result obtained by NOS DAC users just jumps out.
Last edited:
You're possibly late for the party. iZotope upsampling option is currently only available for the owners of previous versions. It was still available when I recommended to try Audirvana / Izotope no interpolation upsampling method a few years ago in the other NOS thread, but... Anyway, I'm not saying Izotope is proved to be perfect, but even if it's not perfect, it certainly increases the fidelity along with a good quality FIR filter here. I usually use Equilibrium @260000 step, regular DAC, not NOS. I suspect the difference between NOS and delta sigma would be minimal in my setup, and I guess the difference between particular converters is much more diverse than NOS vs delta sigma in general. Even comparing between delta sigma, we can hear so much difference, and NOS should be the same. If I find a NOS converter that sounds better than my current converter, Apogee Symphony, I would have no hesitation to switch (well, depends on the cost, of course).
I switched to Qobuz, because I can hear some difference between MQA and conventional compression. I was advocating MQA before, but no more. Qobuz sucks but that is another story...
Last edited:
Ken, Could we start a thesis, that
1 what we all describe as "NOS" sound is just the correct way of converting and reproducing digital sound.
2 Upsampling and interpolating filtering, when done good/better/correctly, is the only way to preserve this "correct way" - hence chip solutions (lacking power) are degarding the sound.
So NOS is NOT something special versus normal OS DACs and cd players. The latter are just not good enough
Conclusion:
The answer lies in finding the best possible resample function .
Not that I think you did not see that, on the contrary
But I just wanted to summarize it bit (probably mostly for myself... ).
Ok, in that case I give up...
Good luck with finding what you are looking for.
In this case it seems I have no idea what you are looking for.
Hi, Doede,
That seems to be the implication, doesn't it?
Is NOS the way that digital playback actually should sound? I think the PGGB experiment indicates that NOS is, at the least, more the way that digital playback should sound than is the sound of typical OS playback. However, I don't yet feel this is conclusive. Further testing, which we won't be doing, might reveal the Holy Grail that I've hoped to discover. Which is a merging of the best aspects of NOS sound, with the best aspects of OS sound. Such ideal merging may not exist, but I'm wondering whether this was sufficiently explored in our experiments.
I think that is essentially true. Either do 'proper' OS interpolation-filtering, without introducing audible artifacts, or don't do any at all (NOS). Doing OS less than adequately seems the worst choice of the three.
Yes, or lacking that, to go fully NOS. Optionally, including analog EQ of the Z.O.H. 20KHz response droop. Which I recommend to include, based on my own NOS experience.
I do think that you correctly understand what it seems we have determined.
Last edited:
mvs0, perhaps, slow down a bit. It's been a long, well discussed thread, and maintaining healthy communication can become challenging. Your input has been interesting, and your participation has been valuable. We are almost finished, and it would be unfortunate to see you depart the thread now.
There are a number of fashionable things pushing manufacturers in a certain direction, like 24 bits betters 20 bits and 192Khz/24 recordings are top of the bill.
But a few examples tend to point in a different direction.
1) Switching off the 8Fs upsampler and driving a PCM179X with 44.1Khz, will let a lot of SDM noise intrude into the Audio range.
Official magazine tests would probably make a big fuzz of this, but fact is that it can’t be perceived.
Marcel’s very sophisticated Valve Dac has some 30dB more noise as the 179X, but users are flabbergeisted with the sound. So how many bits do we really need?
2) Droop because of ZOH is regarded as something that’s unacceptable.
However, in the 88.2 test no reports were made of differences in the high audio range, so it’s safe to conclude that you can’t hear it.
3) Even stronger opposition is in general against supersonics starting directly above 22.05 Khz. Again within this limited group of contenders with NOS Dac’s nobody noticed a difference when playing the 88.2 version, despite the fact that the whole freq. range from 22.05 to 66.15 was emptied.
So to my feeling the yardsticks to measure performance are simply too high or at the wrong spots and don’t add to the perceived quality.
With the above one can say that when having a NOS Dac, the best filter is no filter, because filters cannot improve things, but a mediocre filter will damage the sound.
Hans
But a few examples tend to point in a different direction.
1) Switching off the 8Fs upsampler and driving a PCM179X with 44.1Khz, will let a lot of SDM noise intrude into the Audio range.
Official magazine tests would probably make a big fuzz of this, but fact is that it can’t be perceived.
Marcel’s very sophisticated Valve Dac has some 30dB more noise as the 179X, but users are flabbergeisted with the sound. So how many bits do we really need?
2) Droop because of ZOH is regarded as something that’s unacceptable.
However, in the 88.2 test no reports were made of differences in the high audio range, so it’s safe to conclude that you can’t hear it.
3) Even stronger opposition is in general against supersonics starting directly above 22.05 Khz. Again within this limited group of contenders with NOS Dac’s nobody noticed a difference when playing the 88.2 version, despite the fact that the whole freq. range from 22.05 to 66.15 was emptied.
So to my feeling the yardsticks to measure performance are simply too high or at the wrong spots and don’t add to the perceived quality.
With the above one can say that when having a NOS Dac, the best filter is no filter, because filters cannot improve things, but a mediocre filter will damage the sound.
Hans
Last edited:
I upgraded Audirvana, hence iZotope is still selected in mine, though I am unsure if it is ever active. IMO newer recordings streamed at 44.1KHz/16bit (via Tidal with streaming set to exclude MQA) have dramatically improved over the past few years (certainly in my system). This begs the question if fully unfolded MQA files are any better than current NOS, OS or SRC implementations, including of the kind being tested here. Secondly, is a sonically corrupted MQA file (as the inference by MQA as in being limited to only the first unfold) better than NOS?
To be clear I have no device that can decode MQA. My impressions of MQA have always been that the first unfold of MQA files (in Tidal) were generally inferior to NOS, being never considered better than NOS. This is not to suggest that MQA files were not at times more palatable. However, in choosing MQA for reasons of palatability is considered a limiting factor to advancing audio reproduction. This is my primary reasons for not using it, notwithstanding that MQA seams more intended as ransomware than software.
In my limited understanding of MQA, and I've never auditioned it, it's primary claim to a subjective sound advantage is based on enabling, and most importantly, specifying the use of a relatively gentle anti-alias filter at recording, and an equally gentle anti-image reconstruction filter at playback. These filter specifications seem targeted at mitigating the 'undesired' artifacts introduced by OS (brickwall) SINC interpolation-filters.
As we saw in our experiment, whether an interpolation-filter is bad or not seems to depend on how closely it adheres to, or falls short of, the sampling theorem ideal of a true SINC function. Perhaps, the MQA solution indeed sounds better as claimed, I have no idea. It seems like it may mostly be a more complicated means of delivering the same sound already made possible by NOS, or by proper OS interpolation-filtering.
As we saw in our experiment, whether an interpolation-filter is bad or not seems to depend on how closely it adheres to, or falls short of, the sampling theorem ideal of a true SINC function. Perhaps, the MQA solution indeed sounds better as claimed, I have no idea. It seems like it may mostly be a more complicated means of delivering the same sound already made possible by NOS, or by proper OS interpolation-filtering.
Mqa is a marketing monster, look at Linn’s comment.
A supply chain monopoly
MQA is an attempt to not simply sell the same content again at a higher margin, or to maintain audio quality in streaming ecosystems: it is an outright land grab. It’s an attempt to control and extract revenue from every part of the supply chain, and not just over content that they hold the rights for. It really is quite extraordinary. Let’s break it down:
Manufacturers of recording equipment will have to license the technology and adapt their products
$$$ MQA Gets Paid $$$
Developers of recording software systems will require certified software plug-ins
$$$ MQA Gets Paid $$$
Recording and Mastering engineers must purchase and use certified equipment and software
$$$ MQA Gets Paid $$$
Artists must use studios and engineers utilising certified equipment and new workflows; or even pay to have their back catalogue ‘remastered’ in MQA. The costs, of course, are borne by the artist, either directly or recouped from royalties
Digital distributors will have to license MQA and purchase/lease a ‘Hyper-Security Module’ to encrypt/encode/watermark files ready for delivery to download services
$$$ MQA Gets Paid $$$
Download and Streaming service providers will have to agree to commercial terms and become partners from which…
$$$ MQA Gets Paid $$$
Physical media manufacturers can use MQA to author on to CD and DVD, presumably there will be licensing agreement required for this too
$$$ MQA Gets Paid $$$
Hi-fi manufacturers and software developers will have to adapt their products and license the technology
$$$ MQA Gets Paid $$$
End customers, having paid a premium for MQA music via licensed content providers, will also have to buy MQA certified players at increased cost, with a license paid for each unit shipped
$$$ MQA Gets Paid $$$
A supply chain monopoly
MQA is an attempt to not simply sell the same content again at a higher margin, or to maintain audio quality in streaming ecosystems: it is an outright land grab. It’s an attempt to control and extract revenue from every part of the supply chain, and not just over content that they hold the rights for. It really is quite extraordinary. Let’s break it down:
Manufacturers of recording equipment will have to license the technology and adapt their products
$$$ MQA Gets Paid $$$
Developers of recording software systems will require certified software plug-ins
$$$ MQA Gets Paid $$$
Recording and Mastering engineers must purchase and use certified equipment and software
$$$ MQA Gets Paid $$$
Artists must use studios and engineers utilising certified equipment and new workflows; or even pay to have their back catalogue ‘remastered’ in MQA. The costs, of course, are borne by the artist, either directly or recouped from royalties
Digital distributors will have to license MQA and purchase/lease a ‘Hyper-Security Module’ to encrypt/encode/watermark files ready for delivery to download services
$$$ MQA Gets Paid $$$
Download and Streaming service providers will have to agree to commercial terms and become partners from which…
$$$ MQA Gets Paid $$$
Physical media manufacturers can use MQA to author on to CD and DVD, presumably there will be licensing agreement required for this too
$$$ MQA Gets Paid $$$
Hi-fi manufacturers and software developers will have to adapt their products and license the technology
$$$ MQA Gets Paid $$$
End customers, having paid a premium for MQA music via licensed content providers, will also have to buy MQA certified players at increased cost, with a license paid for each unit shipped
$$$ MQA Gets Paid $$$
I found Linn's remarks more like the pot calling the kettle black. Notwithstanding the legitimacy of their arguments being accurate and genuinely concerning there was a time when nothing was any good unless authenticated by Ivor "the Master" Tiefenbrun. In other words, Linn's offence can seem to the extent offensive in that they hadn't thought of engaging in such manner themselves.
Does anyone know what kind of passband ripple classical oversampling filters such as the SAA7220 have, if any?
The datasheet only says it has a length of 120 taps and interpolates by a factor of four. A CD player I used to own with four times oversampling and TDA1541 DACs suppressed the 43.1 kHz image by about 62.2 dB when playing a 1 kHz sine wave (71.6 dB at 34.1 kHz, 68.16 dB at 54.1 kHz when playing 10 kHz). This includes analogue filtering, so I guess the stopband rejection is of the order of 55 dB.
Guessing that the transition band is from 20 kHz to 24.1 kHz and forgetting about droop compensation and coefficient rounding, that could mean that its response is something like the first two attached figures (vertical scale in dB, horizontal frequency divided by the 176.4 kHz sample rate). If its transition band stops at 22.05 kHz, it is more like the third picture, which doesn't even reach 55 dB of suppression and which has far more passband ripple. All in all, my best guess are the first two pictures (and the text file with the impulse response), but there is a lot of guessing involved.
The datasheet only says it has a length of 120 taps and interpolates by a factor of four. A CD player I used to own with four times oversampling and TDA1541 DACs suppressed the 43.1 kHz image by about 62.2 dB when playing a 1 kHz sine wave (71.6 dB at 34.1 kHz, 68.16 dB at 54.1 kHz when playing 10 kHz). This includes analogue filtering, so I guess the stopband rejection is of the order of 55 dB.
Guessing that the transition band is from 20 kHz to 24.1 kHz and forgetting about droop compensation and coefficient rounding, that could mean that its response is something like the first two attached figures (vertical scale in dB, horizontal frequency divided by the 176.4 kHz sample rate). If its transition band stops at 22.05 kHz, it is more like the third picture, which doesn't even reach 55 dB of suppression and which has far more passband ripple. All in all, my best guess are the first two pictures (and the text file with the impulse response), but there is a lot of guessing involved.
Attachments
Thx, Interesting to see.
We still had a lot to learn in those days about the damaging effect of ripple in the passband.
Hans
I extracted the coefficients from my SAA7220 and put them into LTSpice based on Kendall Castor-Perry's blog posts about FIR simulations.
The ripple all occurs on the part of the response which is rising to offset the 'droop' of the following Bessel filter so its difficult to get a precise reading. But it looks to be in the range 0.02 to 0.03dB peak to trough.
My sim shows in the region of 50dB but its not even, I guess due to only having 12bit coefficients.
Happy to share the LTSpice file if you'd like to have a play.
I ran a plot and posted it here : Alternative to the SAA7220 for TDA1541 dac's
Their arguments were nicely articulated Hans. By the way I am still using an LP12 and currently designing a moving coil phono stage around it. The LP12 has advantages along with limitations that appear correctable to some degree. I am also looking for a digital recorder to generate some digital files from the turntable as you have done. Any recommendations?
Getting back to the topic of NOS it is wondered if analog "recorded" files at 44.1KHz being played back at 44.1KHz files are better/worse than analog recorded files at 88.2KHz being played back at 88.2KHz, hence both NOS. It isn't clear to me that 88.2KHz NOS is better than 44.1KHz as one would think this would have become a new standard at some point in the 4 decades since 44.1KHz and 48KHz became standards.
Gerrit