A couple of riffs on the notion of "bit perfect"...

[spdif-usb]

Btw, is this named "decimation filtering" ? perhaps. But if so, it's quite confusing because "decimation" as such is downsampling.

Loosely speaking, "decimation" is the process of reducing the sampling rate. In practice, this usually implies lowpass-filtering a signal, then throwing away some of its samples. "Downsampling" is a more specific term which refers to just the process of throwing away samples, without the lowpass filtering operation.

In sigma delta A/D applications, the decimation filter serves a dual purpose. In the frequency domain, a coarsely-quantized signal (often just one bit) at a sample rate much greater than Nyquist (M*Fs, M some oversampling ratio) will have an even amount of noise across the spectrum from 0 to M*Fs/2, and there is a bunch of it because the resolution of the quantizer is coarse. The delta sigma modulator shapes this noise so that the total amount (over the entire bandwidth from 0 to M*Fs/2) stays the same (even increases), but the amount from 0 to some bandwidth of interest B gets much less. So, from designing the decimation filter in such way that it can, in the noise shaped environment, eliminate as much noise that's above the bandwidth of interest B as possible, comes greater SNR.

Please note that sigma delta modulators can be seen as state variable machines and that they will be characterized by undesired oscillation if the signal level is pushed too high, but that with current technology it has been made possible to minimize this important problem, i.e. by controlling the state variables as explained in the YouTube video that I previously linked.

I am not sure anymore whether someone said that "jitter precision" (or something like that) would be better on the higher sample rate, but altough mathematically this can be true, practically it works out differently. Think about DSD and its very high sampling rate, which now goes by accuracy of the bit (opposed to a 32 bit or so word). Not everybody will agree (maybe nobody) but I say that jitter has to be 30-36 dB better in order to represent the higher sample rate. This is not easy to see (and let's say I can be as wrong as that I can be right) because there's also the aspect of "time smearing". Okay, too difficult. But think about the high ferquency the "1 bit" square implies and how it needs to be filtered out. Well, in my view, when jitter anticipates the 32 bit audio word (x 2) then there will be 32 times more "infrequencies" at the 1 bit level implying *lower* frequencies. Now these have to be filtered out too.

Ah, ok, can't understand it myself anyway.

I agree that jitter rejection is of critical importance, but new ways have been found to tackle issues that typically arise when "phase-locking" mechanisms are applied. Please read the tech paper from ESS Technologies that I previously linked, it's the one about their SABRE line of A/D chips.

But isn't it all moot since no recording label of fame is providing Hires anyway (today) ?

"Slandering" Barry Diament again, are we? :grin:

Well, what about the size of the files ? One hour of 24/192 takes about 3GB. And believe it or not, but at least .WAV can't be stored larger than 4GB. So that's a sort of technical reason too.

Multichannel 24/192 FLAC files can be created from DTS-HD MA streams such as the ones commonly found on Blu Ray discs. Using eac3to to create these FLAC files, they can have filesize larger than 4 GB.

[spdif-usb]

P.S. - Some interesting comments on the 4GB filesize problem with .WAV files can be found here. ffmpeg->sox->wav wav/w64 length? - Doom9's Forum

[Jud]

PS: As you explained further down in the thread, this is obviously about recording in 352.8 (384) and when that is played back directly, not any filters have been around anywhere. Well, notice that this is the selling point of "Peter" at 2L. You may not know it, but they explicitly leave out ADC filtering. "It is just not necessary". FWIW, but it at least confirms what you're implying to be a good thing, Jud. One thing I can 100% guarantee you : what 2L does with this means of recording really works out. So, me as Anti-Hires #1 on the globe ... not so for 352.8 2L. It's audible in a fraction of a second and I never could do that at the length of full "Hires" normal albums.

So, very good subject ? Yes.

 

So Peter, have you tried listening to the 2L 352.8 recordings converted to 705.6 using Arc Prediction, and without that 2x conversion? Was anything lost or added, i.e., did one sound better to you than the other, and if so which?

[Miska]

The problem (IMO) with DSD is it causes non-linear distortions that cannot be fully dithered out, and it has to be converted to PCM if you want to use processing, i.e. a lossy conversion to PCM.

 

This is incorrect. All kinds of processing can be performed with DSD, it is just more expensive from processing power point of view. I have now all the same processing available for both PCM and DSD.

 

Furthermore, the ultrasonic noise of DSD still has to be rolled off in the playback chain in order not to overload the driver of a tweeter, and lossless data compression on DSD-encoded data is next to impossible.

 

That same ultrasonic noise exists in all modern SDM converter chips, even if they don't support DSD. SDM converters are much cheaper and have better linearity than multi-bit PCM chips and that's why practically all new converters are SDM and DSD is native language of these.

 

DST is the standard lossless DSD compression, but also plain standard data compressions like ZIP do fairly well with DSD. There's no reason why other methods couldn't be created, but I don't think those are needed since it already has ~30% better coding efficiency than PCM.

[Miska]

Much has been chewed upon this one. But when it's reasoned out for the whole chain (including the process of recording and mixing) it makes no sense to think it will ever (have) be(en) native DSD throughout.

And if it was, no mixing whatsoever can have been in order. The solution to this (at least that's what a few come up with) ? mix in analogue and make DSD of it. Well, maybe. But not of course.

 

You can mix in DSD, and do all kinds of other stuff like volume control, effects, etc.

 

I can do it in my software. And there's a DAW that can do it for you too:

Sonoma Multitrack DSD Recorder - DSD Editor

[Miska]

Do DSD recordings played back on DSD-capable DACs avoid the sample rate conversion problem, or are there sample rate (or other significant) conversions involved in this chain as well? More particularly, is the change of "1-bit" to, e.g., "6-bit" essentially a lossless change of digital "container" format, or are there potential audio problems that may apply to this stage?

 

Some do, some don't. It really depends on the particular DAC used and if it's a chip it depends on how it is configured. It is also possible to "upsample" and "downsample" DSD too, for example between DSD64 and DSD128.

 

Yes, you can play 1-bit SDM on 6-bit SDM without any fancy conversions just like you can play 16-bit PCM on 24-bit PCM DAC.

[Miska]

I do suggest that it is much more difficult for a DAC to take in a 24/192 or 24/176.4 signal and do a good job with it, than it is for a DAC to take in and competently process a 24/96 or 24/88.2 signal, which in turn is more difficult than processing a 16/44.1 signal.

 

It is actually vice versa in many cases. At 4x rates (192/176.4) you can get good results even with poor oversampling filters and less good analog filtering sections. With 2x rates you already need much better digital and analog filters to make the result good. And at 1x rates your aliasing bands are really close to audio band you need to pay extreme attention to detail to make it actually perform well...

[Miska]

I read this yesterday. It may be helpful.

 

Q&A with Charles Hansen of Ayre Acoustics | AudioStream

 

It was nice read indeed to see that someone agrees with me on so many points...

[Miska]

Hi, Robbie, been wanting to ask since I read this: Possible to apply a digital filter that neither interpolates nor decimates, but maintains the same sample rate, to counteract the phase shift introduced by an inexpensive analog filter?

 

I can answer this, yes you can. You can also have 1:1 apodizing resampling filter to modify ringing of the prior ADC decimation filter without actually changing the sampling rate (I offer also this possibility as a I didn't bother to prevent it either). Phase modifying non-rate-converting filters are also commonly part of digital room correction filters to counteract the phase anomalies created by loudspeaker cross-over filters and speaker element placements.

[Jud]

I can answer this, yes you can. You can also have 1:1 apodizing resampling filter to modify ringing of the prior ADC decimation filter without actually changing the sampling rate (I offer also this possibility as a I didn't bother to prevent it either). Phase modifying non-rate-converting filters are also commonly part of digital room correction filters to counteract the phase anomalies created by loudspeaker cross-over filters and speaker element placements.

 

Thanks, Miska, I really appreciate your answers to this and the other questions. I think there's a tendency sometimes for old information to be re-told as new, so it is nice to be brought up to date.

[spdif-usb]

This is incorrect. All kinds of processing can be performed with DSD, it is just more expensive from processing power point of view. I have now all the same processing available for both PCM and DSD.

Here is an excellent tech explanation on the topic of why non-linear distortions in DSD cannot be fully dithered out:

Nice Review of DSD by Andreas Koch

 

And here is how processing of DSD material is still done in practice:

PCM vs DSD. Why DSD and its variants?

That same ultrasonic noise exists in all modern SDM converter chips, even if they don't support DSD.

I never said it didn't; my point was to further illustrate the reason(s) why DSD does not, at least not in theory, offer any advantages over rate-converted PCM.

SDM converters are much cheaper and have better linearity than multi-bit PCM chips and that's why practically all new converters are SDM and DSD is native language of these.

As I already explained earlier in the thread, the YouTube video I linked shows the HyperStream modulator was designed to deal with the problem of unwanted oscillation occurring in sigma delta modulators whenever the signal level is pushed too high. Yes, the internal data stream is also 1-bit using a very high clock frequency, so there are indeed huge similarities with DSD here, but for the sake of correctness (although I could be wrong ) the "native language" of HyperStream is actually PWM (Pulse Width Modulation), i.e. not DSD, the "native language" of which is PDM (Pulse Density Modulation).

DST is the standard lossless DSD compression, but also plain standard data compressions like ZIP do fairly well with DSD. There's no reason why other methods couldn't be created, but I don't think those are needed since it already has ~30% better coding efficiency than PCM.

Fair enough, but the data compression ratio is still nowhere nearly as strong as that of FLAC.

[PeterSt]

P.S. - Some interesting comments on the 4GB filesize problem with .WAV files can be found here. ffmpeg->sox->wav wav/w64 length? - Doom9's Forum

 

Well, I too can tweak everything when I'd like to. Similarly I would be able to tweak the 2GB of WAV playback limit, but who is waiting for that (going outside of standards).

 

So, outside tweaks, 24/352.8 would limit to 15 minutes and 32 seconds of track length.

[PeterSt]

So Peter, have you tried listening to the 2L 352.8 recordings converted to 705.6 using Arc Prediction, and without that 2x conversion? Was anything lost or added, i.e., did one sound better to you than the other, and if so which?

 

Jud, proceeding with my previous post (which is by pure coincidence), I just tried and ...

the xth "tweak" to let improperly organized WAV headers work - which in this case is about a change last September to let certain HDTracks files work, now does not work for 2L files. So, the last few versions of XXHighEnd won't play them as it appears now ("Wrong Header" message).

 

So, must solve this first ...

Peter

[spdif-usb]

Well, I too can tweak everything when I'd like to.

The W64 format can hardly be called a "tweak", though (IMO). With new requirements come new standards, as usual. According to Wikipedia: "The RF64 format specified by the European Broadcasting Union has also been created to solve this problem." WAV - Wikipedia, the free encyclopedia

[Jud]

The W64 format can hardly be called a "tweak", though (IMO). With new requirements come new standards, as usual. According to Wikipedia: "The RF64 format specified by the European Broadcasting Union has also been created to solve this problem." WAV - Wikipedia, the free encyclopedia

 

The lack of a widely used standard poses problems for everyone along the chain, though.

[spdif-usb]

The lack of a widely used standard poses problems for everyone along the chain, though.

That's why the word "standard" always reminds me of what Bill Gates once wrote for the New York Times.

https://groups.google.com/forum/?fromgroups#!msg/alt.folklore.computers/mpjS-h4jpD8/9DW_VQVLzpkJ

[Jud]

Here is an excellent tech explanation on the topic of why non-linear distortions in DSD cannot be fully dithered out:

Nice Review of DSD by Andreas Koch

 

 

The paper cited for that proposition at the link you give was authored in 2001. 1-bit encoding and decoding appears to be capable of better sound than it was in 2001. The industry has long since decided against the direction recommended in that paper (i.e., 1-bit encoding/decoding happens internally in virtually all A/D and D/A chips). Since 1-bit format is used internally in virtually all A/D and D/A processing, use of PCM almost inevitably means additional conversions, giving rise to the question of whether such conversions are audible.

 

Another issue is whether PCM conversions themselves are inevitable, i.e., necessary for editing. I've read your citation (where Bruce Brown in a couple of short paragraphs says he uses the analog domain for editing, then talks about his understanding of what's happening internally when he edits DSD, while giving the caveat that his understanding may be incorrect), as well as other material, and don't know a simple conclusive answer.

[Miska]

Here is an excellent tech explanation on the topic of why non-linear distortions in DSD cannot be fully dithered out:

 

Nice Review of DSD by Andreas Koch

 

That paper is just outdated and incomplete/incorrect. SDM signal processing has proceeded quite a lot since.

 

And here is how processing of DSD material is still done in practice:

 

PCM vs DSD. Why DSD and its variants?

 

And I know how I'm doing it in practice... That's the way Sonoma does it, and my way is different.

 

but for the sake of correctness (although I could be wrong ) the "native language" of HyperStream is actually PWM (Pulse Width Modulation), i.e. not DSD, the "native language" of which is PDM (Pulse Density Modulation).

 

That's the way probably all DSD DACs work, from practical point of view you can consider PDM being equal to PWM. The difference is only theoretical. I would say probably all SDM and PCM DACs are sample-and-hold type. For ADCs it may actually vary up to being 50/50 or something.

 

Fair enough, but the data compression ratio is still nowhere nearly as strong as that of FLAC.

 

When the data is 30% less to begin with, the remaining compression ratio brings the total bandwidth savings to roughly equal of FLAC. Not that it would matter in practice, audio is so low bandwidth anyway that dealing with it as uncompressed is not a problem.

 

(good quality full-HD video with H.264/AVC compression is 24-30 Mbps at least peaking around 50 Mbps with pans)

[Miska]

Jud, proceeding with my previous post (which is by pure coincidence), I just tried and ...

the xth "tweak" to let improperly organized WAV headers work - which in this case is about a change last September to let certain HDTracks files work, now does not work for 2L files. So, the last few versions of XXHighEnd won't play them as it appears now ("Wrong Header" message).

 

Peter, there's nothing in WAV about header ordering, so you shouldn't be expecting anything about the order. Preferably headers should precede the actual audio data, but for me even that doesn't matter too much. I don't actually recall making any hacks ever to my WAV parser to deal with any of the variants.

[sandyk]

Miska

Not related to the present exchange between Peter and yourself, but reminded by what Peter said.

Have you any suggestions as to why some software, such as cPlay and SeeDeClip Duo Pro will not open ripped DVD-A .wav files, or Barry Diament's 24/192 .wav files from his DVDs ? I found that if you chop off around 50mS from the start of the .wav files in Sound Forge 9, and presumably other Audio Editors,that they can then be opened and processed with SeeDeClip Duo Pro, and played using cPlay.

Alex

[Jud]

Miska

Not related to the present exchange between Peter and yourself, but reminded by what Peter said.

Have you any suggestions as to why some software, such as cPlay and SeeDeClip Duo Pro will not open ripped DVD-A .wav files, or Barry Diament's 24/192 .wav files from his DVDs ? I found that if you chop off around 50mS from the start of the .wav files in Sound Forge 9, and presumably other Audio Editors,that they can then be opened and processed with SeeDeClip Duo Pro, and played using cPlay.

Alex

 

Hi guys - Interesting topic, but can I ask that you make it a new thread if you want it to go further? Would probably be more helpful to anyone searching the forum for answers on that subject if it were its own thread.

 

Thanks. :-)

[Paul R]

I have a friend who is convinced that playing a WMA file through Windows Media Player is "bit perfect." i do not believe that is the case,and am looking for an easy, simple, and cheap way to prove or disprove bit perfect playback under Windows. Any ideas?

 

-Paul

[mav52]

I have a friend who is convinced that playing a WMA file through Windows Media Player is "bit perfect." i do not believe that is the case,and am looking for an easy, simple, and cheap way to prove or disprove bit perfect playback under Windows. Any ideas?

 

-Paul

are these files your friend has copied using Media Player with the error correction box checked under 'more options', just wondering

 

here's is a old thread

http://www.computeraudiophile.com/f8-general-forum/anything-wrong-windows-media-player-and-wma-lossless-format-995/

[Paul R]

I was aware that there is a WMA lossless format, though admittedly, I have never used it. I guess the question is does Media player play it back bit perfect with any settings? Not sure how his files were copied or RIPed Mav- though I believe he used dbPowerAmp to RIP them. -Paul

are these files your friend has copied using Media Player with the error correction box checked under 'more options', just wonderinghere's is a old thread http://www.computeraudiophile.com/f8-general-forum/anything-wrong-windows-media-player-and-wma-lossless-format-995/

[Miska]

I have a friend who is convinced that playing a WMA file through Windows Media Player is "bit perfect." i do not believe that is the case,and am looking for an easy, simple, and cheap way to prove or disprove bit perfect playback under Windows. Any ideas?

 

It may be as much bit perfect as iTunes on OSX, so after choosing correct output format and setting volume to max it may or may not be.

 

Easiest way to confirm bit perfect playback is to use a DoP WAV-file (or loss-less WMA) and play it to a DSD DAC and listen...