When is "High resolution" not quite that high?

[johnb]

Gordon, I have re-done the plots, this time with an FFT size of 8192 (I can go higher if required).



The difference in "Channel Silence" below 20k is mainly due to a difference in the exact clip of 'silence' I used. I've checked a different 'silence' elsewhere (from a different track on the recording) and the result was very similar.

[Gordon]

Thanks for #31 it is as I expected. Further increasing the FFT will only continue the trend of a spikier plot and lower spectral density, it won't alter the basic issue.

What is clear now is that from about 25kHz the DSD conversions have no S/N!! The yellow plot is an indication of what one might expect the other plots to do beyond 25kHz. IOW whatever signal may be present it's buried in noise! So whilst the theoretical bandwidth of DSD may well be increased cf PCM its S/N is wanting and so converting to 96 kHz is rather a waste of time and 48 will be adequate. One might even suggest that 44.1 would be better. At that rate at least 22 kHz audio would have an S/N of about 20dB and so eminently useful.

The root cause of this is insufficient shaping in the DSD system which could be improved by selecting a better coding order. Do we know whose equipment LSO live uses?

[johnb]

Gordon,

I can't find any details of the equipment used by LSO Live but the Channel Classics website gives the following information for the Fischer Mahler 4:

Microphones: Bruel & Kjaer 4006, Schoeps
Digital Converters: DSD Super Audio/GrimmAudio AD
Speakers: Audiolab, Holland
Software: Pyramix Editing, Merging Technologies
Mixing Board: Rens Heijnis, custom design
Mastering Room:B+W 803d series speakers, Classe 5200 Amplifier
Cables:Van den Hul

Incidentally, I've come across an interesting (accessible to laymen) article by Charles Hansen of Ayre Acoustics (the high-end consumer audio manufacturer) in which he outlines the development of, and wars between, the DVD-Audio/SACD formats and then outlines what he sees as the problems inherent in the DSD format.

The full article is here.

This is the section relevant to DSD noise:

1) Standard DSD has low noise levels in the audio band. But at 20 kHz the noise rises sharply, leading to one of Sony’s more obviously amusing marketing contradictions. On the one hand they say that its extended bandwidth leads to a more natural presentation of high frequencies in the music, but on the other hand, they say that the high amounts of high frequency noise doesn’t matter because it is inaudible. I suppose the answer to that question depends on which marketing person you are speaking to on any particular day...

However it does cause serious problems when recording with the techniques that have become commonplace over the last fifty years or more. Any time that the signal is manipulated, more and more noise is added to the signal. These high-frequency noises can lead to damage of downstream equipment including amplifiers and loudspeakers. Therefore the official Sony specification (the “Scarlet Book”) specifies a third-order low-pass filter starting at 50 kHz, and the actual usable frequency response of SACD doesn’t extend much beyond 30 kHz.

[Bryn]

Gordon,

I can't find any details of the equipment used by LSO Live but the Channel Classics website gives the following information for the Fischer Mahler 4:



Incidentally, I've come across an interesting (accessible to laymen) article by Charles Hansen of Ayre Acoustics (the high-end consumer audio manufacturer) in which he outlines the development of, and wars between, the DVD-Audio/SACD formats and then outlines what he sees as the problems inherent in the DSD format.

The full article is here.

This is the section relevant to DSD noise:

You gorra laff, innit?

[Dave2002]

You gorra laff, innit?

Yeah, guv!

Though I'm not sure that everyone will spot the flaw (s) ....

[Gordon]

...Incidentally, I've come across an interesting (accessible to laymen) article by Charles Hansen of Ayre Acoustics (the high-end consumer audio manufacturer) in which he outlines the development of, and wars between, the DVD-Audio/SACD formats and then outlines what he sees as the problems inherent in the DSD format.

Thanks jb!! That article is revealing and suggests our conclusions are not far off the mark!! A touch of Kings New Clothes perhaps.

However, going back to your #23 and #31 and the silence plots, they still puzzle me. In digital systems no signal means no quantising noise [check by taking a bit of silence from the Chandos file] so why is there such a lot in the DSD silence scan? It is reduced but by not a lot and it is nominally inaudible so no real problem to listeners. And where do those two notches come from; they were hidden until the finer resolution of the larger FFT dug them out. Are they inherent in the original sound or yet another artifact of conversion? They are not at any particularly significant frequency.

[johnb]

Gordon, from what I can see on the web (yes, a goodly number of pinches of salt are required) the two notches seem to be characteristic of the DSD noise.

Not too dissimilar DSD noise profiles are shown in websites here and here both of which appear to show the noise without any low pass filter having been applied. The two graphs have a logarithmic frequency axis so have have tweaked my previous (very) rough and ready effort to be comparable.

I think the difference between my first and second attempts to plot the 'silence' is mainly due to the clip that I selected - I think my first selection must have had some low level audio signal present.

As for why there is noise present in silence - pass.

[IMG][/IMG]

[Gordon]

Thanks jb for #37 and comment. I agree that the internet if full of strange things some of it abject prejudiced nonsense and some of value. If the acknowledged experts disagree [despite certain potential commercial biasses] then people like me who have been out of the business too long must stand and watch with interest and try and pick some understanding from the pieces. This thread has certainly exposed some issues!!

Going back to Bryn's original query though - there does seem to be a need for labelling on downloaded files including something about the meaning of translation between formats. A health warning? "The content in this download file may have been derived from other formats and may not necessarily achieve the full potential of the file format presented." Having said that the way that downloads are marketed and sold there is virtually no chance of achieving such labelling. At least Chandos have put their hands up and dealt with their position pretty well.

[Dave2002]

What amused me about the article (msg 33 - http://www.ayre.com/insights_dsdvspcm.htm) and the comparison files supposedly accessible from it is that they are "needle drops" from LPs, played back through a moving coil cartridge. Some are definitely quite old recordings, too.

I recently read an article about John Crabbe in which it revealed that at one time he investigated the tracking behaviour and interactions between cartidges and the vinyl being played, and he discovered that significant vibrations were being put back into the vinyl. Because of the mechanics of moving coil cartridges he realised that this would be more significant for moving coils, and then he suggested that perhaps the supposed "better" sound quality of MC cartridges was actually due to the additional - to some extent programme related - distortion/noise being fed back into the cartridge.

http://www.stereophile.com/interviews/john_crabbe_firebrand/index.html

I have no way of playing the .dsf (DSD) files, and the PCM files failed to download.

[Gordon]

What amused me about the article (msg 33 - http://www.ayre.com/insights_dsdvspcm.htm) and the comparison files supposedly accessible from it is that they are "needle drops" from LPs, played back through a moving coil cartridge. Some are definitely quite old recordings, too.

Well, yes, a few questions to beg there! If he wants to demonstrate his point he needs good wideband audio that hasn't been anywhere near a digital system. To do that and have a fighting chance of capturing whatever content might exist beyond 20kHz he needs state of the art analogue recordings, pre 1980.

Why chose LP? Convenience perhaps having no access to studio quality analogue. I assume his LPs were also mastered pre 1980 so they they also are free of any digital footprints. The vinyl artifacts should not however negate the experiment, even if they might cloud the digital ones, they are common to both digital codings, and will exercise those codings in the critical mid range where DSD should theoretically have an advantage over PCM due to the noise shaping.

This is visible in jb's plots in #31 and 37. The noise floor reported by Audacity for 96/24 PCM is around -118dB spectral density which we can reasonably assume underlies the rest of the spectrum ie the mid band. This level isn't necessarily the theoretical floor of 24 bits; it is probably well above that because of other noise sources in the system - microphones for example [but then that applies to DSD too, noise shaping only suppresses its own noise not that in the input content] - and the possibility that dither is used in the ADC which may lose about 2 bits. It is also greater because of Audacity's resolution bandwidth which we don't know - it's implied in the FFT size. So we should not take too much notice of the absolute values of the left hand scale but we can make use of the reported differences between the systems.

In that same area, assuming that the silence noise level of DSD remains shaped as we see it when the system is exercised by signal [the floor rises by about 10dB], the density is around -130dB at least as far as 22kHz where it starts to rise quickly. The mid band difference of 10dB or so might be significant in what is heard. Then again it is so low down that one is asking a listener to hear that 10dB in a dynamic range of 80+dB.

I recently read an article about John Crabbe in which it revealed that at one time he investigated the tracking behaviour and interactions between cartidges and the vinyl being played, and he discovered that significant vibrations were being put back into the vinyl. Because of the mechanics of moving coil cartridges he realised that this would be more significant for moving coils, and then he suggested that perhaps the supposed "better" sound quality of MC cartridges was actually due to the additional - to some extent programme related - distortion/noise being fed back into the cartridge.

http://www.stereophile.com/interview...and/index.html

Yes I had seen something about that article. I don't want to get too distracted into vinyl matters here!! But the clear implication of his remarks is that the vinyl disc is alive with vibrations of its own brought about by the groove forcing the stylus to follow it - Newton's Third Law. That implies that the groove walls are deforming slightly in order to generate the force needed to cause the stylus to move and remain in the groove - as per Newton, the stylus inertia is applying a deforming force. Given that the stylus tip is small [ca 1 milligram] that force doesn't sound large until one realises that the accelerations involved can be in the thousands of g and so the instantaneous stylus force looks more like several grams. Given that the contact area is very tiny [ca a square micron or less?] the stress on the vinyl can be quite high and so its Young's Modulus will determine the size of any deflections. The forces will be dependent on the groove modulation and so will vary dynamically. Remember also that already the playing weight [1-2 grams] will be acting on that same small contact area and will produce its own relatively constant bias deflection. Given these deflections what chance that the stylus, even if its maintains full groove contact at all times which is debatable] is producing an accurate version of what the cutter set in the groove geometry? Provided that the Young's Modulus is high enough the deflections are small - but energetic enough to stimulate the disc vibrations - and provided also that the range of forces applied does not exceed the vinyl elastic limit then no serious harm comes to the vinyl. However we do know that excessive playing weight and poor mounting compliance which adds effectively to the stylus mass produces plastic deformation in the grooves. LP and vinyl are endlessly fascinating; one wonders that it works at all - paranoia heaven.

EDIT: For those with an interest, if you can find this old JAES paper it is worth a read I've just come across it in an old file!:

Barlow D A, "Limiting Factors in Gramophone Reproduction" JAES, Vol5 No2 April 1957, pp109-117. Reprint of Wireless World Vol 63 1957 pp228-230 and 290-294.

It is listed on the AES web site and s is available for free to members. I am no longer a member so can't get at it without paying $20!!

[Dave2002]

I don't want to get too distracted into vinyl matters here!! But the clear implication of his remarks is that the vinyl disc is alive with vibrations of its own brought about by the groove forcing the stylus to follow it - Newton's Third Law. That implies that the groove walls are deforming slightly in order to generate the force needed to cause the stylus to move and remain in the groove - as per Newton, the stylus inertia is applying a deforming force. Given that the stylus tip is small [ca 1 milligram] that force doesn't sound large until one realises that the accelerations involved can be in the thousands of g and so the instantaneous stylus force looks more like several grams. Given that the contact area is very tiny [ca a square micron or less?] the stress on the vinyl can be quite high and so its Young's Modulus will determine the size of any deflections. The forces will be dependent on the groove modulation and so will vary dynamically. Remember also that already the playing weight [1-2 grams] will be acting on that same small contact area and will produce its own relatively constant bias deflection. Given these deflections what chance that the stylus, even if its maintains full groove contact at all times which is debatable] is producing an accurate version of what the cutter set in the groove geometry? Provided that the Young's Modulus is high enough the deflections are small - but energetic enough to stimulate the disc vibrations - and provided also that the range of forces applied does not exceed the vinyl elastic limit then no serious harm comes to the vinyl. However we do know that excessive playing weight and poor mounting compliance which adds effectively to the stylus mass produces plastic deformation in the grooves. LP and vinyl are endlessly fascinating; one wonders that it works at all - paranoia heaven.

I don't want to distract you with vinyl matters either, but I just don't see how we can judge systems based on inputs which are as inherently limited as most LPs. There are a lot of issues there. Many LPs will have been made from masters produced from analogue tape systems, and these will also have limitations.

I agree that it might be possible to compare the DSD vs PCM whatever the input, but surely we have to try to get something which at least approximates to what most of us might want to hear. Years ago I tried to make some very low bit rate encodings of some audio, including bird song. I was interested to hear that to me the bird song sounded OK. The results on other inputs were less good, so I gave up, though I did find out some quite interesting things (to me) about mapping the signals up to very high frequencies and then mapping then back down into the audible range. One of my friends commented that although I couldn't hear what was probably horrible distortion in the bird song, many birds might not have liked it.

I have never been able to confirm whether they were upset, or would have been, for many reasons.

There are merits in trying to do theoretical calculations, and also taking measurements, but if the end point is to try to produce something which humans can listen to, then basing one's analysis on theory and measurements alone is not completely sound. OTOH, the views of some that one can only tackle these issues by listening are also suspect, even if attempts are maded to get consensus by listening panels, and controlled conditions, and statistical averaging there are many flaws in some of the subjective tests and test procedures.

There are several problems with using humans in tests. One is listening fatigue, which I really believe is a factor. Another is that human hearing and attention is adaptive, so that one may either become tolerant of small distortion and ignore it, or may become sensitive to it, so that it becomes a very significant issue. Yet another problem is that the choice of material to play for the tests is often not going to show up the difficulties, if any, in recording then reproducing a faithful rendition.

Re vinyl again - very briefly - there were papers about vertical tracking angle, and discussions about what vertical tracking angle to use for cutters and replay cartridges in the 1970s. The assumption was that to minimise distortion in the difference channel (vertical) that the geometry of the cutting equipment and the replay equipment should be very similar. Different companies, both in recording and cartridge/pickup manufacture had slightly different views on this. Eventually there was standardisation of the vertical tracking angle, but some experiments and measurements - I think at CBS by Ben Bauer and others, showed eventually that elastic and also plastic deformation in different stages of production and replay materials gave a measurable discrepancy in the effective vertical tracking angle at the cutter and replay cartridge.

[richardfinegold]

I own many SACDs, which use DSD, and fwiw, they (mostly) sound much more natural, or "organic" than standard red book
CD. Some people have described them as "analog like" although I am not sure that I ever had analog sound so good.
So whatever theoretical trickery is used in DSD, it does the job, quite well.

[Gordon]

#41 - bring back Dynagroove!

[Dave2002]

#41 - bring back Dynagroove!

For a few seconds I thought you were refering to that quad system which used high frequency sub-carrier modulation. I can't find a current web page about that, but I wonder what the grooves looked like after a few playings. Amazingly there was also a video disc system which gave a few minutes of video. Again, I wonder what the grooves looked like - perhaps they self destructed as in a well known TV series - though I think the TV show used tapes.

Dynagroove - https://en.wikipedia.org/wiki/Dynagroove

[Gordon]

Perhaps you're thinking of CBS' SQ or Sansui's QS perhaps? RCA's CD-4 system IIRC I think they had supersonic carriers around 30kHz using FM in the L and R on which the rear channels were placed, L and R remaining in their usual places for compatibility. BAnd width on disc was 45 kHz!

Google "quadraphony discs" and a lot will come from a wiki. Look at its references too this one for example:

http://midimagic.sgc-hosting.com/quadrafon.htm

Before discrete methods became possible for multichannel audio, broadcasters in particular [and record companies] were looking for a backward compatible was of adding surround to existing stereo without too much sacrifice. Matrix systems blossomed forth in due abundance, many of them slight variations of another. Some are very well known. I remember colleagues in audio around 1980 [I was then working on digital video recording and audio was the usual poor cousin!] doing some experiments with quad with Capital Radio in London. There was a multitrack tape recording of a Karajan concert that Capital [Yes Capital] had negotiated at the RFH about that time for use in quad experiments. I believe it was actually broadcast by Capital. If only I had found that tape before I left!!

The video disc saga also spawned a number of alternative formats one being Philips' optical disc technology [google "Philips LaserDisc"] which moved on to CD of course. Didn't take off until DVD came long. I have in front of me the proceedings of an IEE colloquium back in November 1981 complete with glossy promotion brochures describing them all. One of the others was from RCA [google "RCA Selectavision disc" and the third was a JVC system "VHD" [google "video high density disc"] that was promoted by Thorn EMI in the UK and was a rather neat technology that is not easy to describe briefly.

But we digress!