When is "High resolution" not quite that high?

[Bryn]

When it's a Berliner Philharmoniker so-called 48/24, that's when.

I recently ordered the new Harnoncourt 'live' Schubert Symphonies, etc. set from Berliner Philharmoniker Recordings. What you get is 8 CDs, 1 Blu-ray, a 7 day 'voucher' for their concert streaming service, and download code for what they describe as "high resolution audio files of the entire album (24-bit / 48 kHz)". However, when you read the small print at the back of the handsomely presented hard-back booklet notes, you find this, "Recorded in 24-bit / 44.1kHz", so the downloads and Blu-ray audio are in fact up-sampled 44.1 kHz recordings, albeit with 24 bit quantization. Not quite so high a resolution, and necessarily replete with (hopefully inaudible) interpolation artifacts. Ho-hum.

https://shop.berliner-philharmoniker.de/harnoncourt-schubert-edition.html

Oh, and I also paid the extra for tracked and insured delivery. No tracking was provided and the parcel was left on my front doorstep by Parcel Farce for any opportunist thief to make off with. Fortunately I got there first, albeit at around 8pm when I returned from work.

[richardfinegold]

there ought to be a universal standard for High Resolution.
Another candidate for this topic would be the Alan Gilbert/NY Phil Nielsen SACDs (see my post elsewhere). it doesn't matter if you produce a recording in DSD if the origianl hall sonics are vague mush.

[Miles Coverdale]

The magazine Hi-Fi News provides an analysis of the high-res files it reviews which often shows that, for example, 96kHz files are, in fact, up-sampled 48kHz. Not everything is what it says it is on the packaging, unfortunately.

[mahlerei]

The magazine Hi-Fi News provides an analysis of the high-res files it reviews which often shows that, for example, 96kHz files are, in fact, up-sampled 48kHz. Not everything is what it says it is on the packaging, unfortunately.

Miles

Agreed. I think some transparency would help. For instance labels should specify the disc/download's native resolution. I've come across a number of downloads that have been compromised in the delivery chain. In the worst case two registered as 16/96 (!) on my M-DAC, so after heated discussions with the label - and two DSPs - I managed to get the offending files removed. it didn't end there, but that's a much longer story. And I've also encountered several 20-bit files masquerading as 24-bit ones....

[Jasmine Bassett]

Another one which is not what it seems is where the original is an old DSD file which is then offered as a 192 download.

I'm not sure I want all that noise pushed up into the inaudible parts of the spectrum by the noise shaping used in the original encoding to be faithfully reproduced up to 96 kHz.

[Dave2002]

Another one which is not what it seems is where the original is an old DSD file which is then offered as a 192 download.

I'm not sure I want all that noise pushed up into the inaudible parts of the spectrum by the noise shaping used in the original encoding to be faithfully reproduced up to 96 kHz.

I had to stop to think about that. If the "inaudible" parts were above 96 kHz, then downsampling to 96kHz could render them audible, or at least to have an audible effect. Otherwise they would surely still remain inaudible. However, that might be undesirable depending on how much energy had been moved to higher frequencies, and could cause problems for some electronics - though that does still seem unlikely. Some types of amplifier might also work badly with inputs with "unwanted" high frequencies - though again that does seem unlikely. If that were a potential problem most designers would filter the input to remove that risk.

On balance I doubt that there'd be any obvious effect for most people using most reasonable kit, and most amplifiers would not have any problems, nor pass any on to the speakers. Tweeters might just get some overload if the output were high enough, and eventually that could cause a problem - though again not very likely.

[Jasmine Bassett]

But why bother doing it in the first place, if it's not just playing the numbers game?

There is no useful information in that part of the spectrum.

[Dave2002]

But why bother doing it in the first place, if it's not just playing the numbers game?

There is no useful information in that part of the spectrum.

Some people might say exactly the same about anything above 20 kHz corresponding to 40 ksps, which does make the whole "point" of "hi-res" rather doubtful.

Maybe there are pet dogs and bats which can appreciate such recordings! Even dolphins!

[Jasmine Bassett]

Original high sample rate recording will have musical information above 20 kHz. The ability of any individual to perceive that extra information well depend on a lot of factors and cannot be dismissed.

High sample rate files derived from DSD originals will contain only quantization noise in the higher parts of the spectrum - that will not add anything positive to anyone's perception of the music.

[Dave2002]

Original high sample rate recording will have musical information above 20 kHz. The ability of any individual to perceive that extra information well depend on a lot of factors and cannot be dismissed.

Why not? If that is true, at what level would you set the cut off?

[Gordon]

Original high sample rate recording will have musical information above 20 kHz. The ability of any individual to perceive that extra information well depend on a lot of factors and cannot be dismissed.

High sample rate files derived from DSD originals will contain only quantization noise in the higher parts of the spectrum - that will not add anything positive to anyone's perception of the music.

I quite agree JB, to imagine that some musical instruments don't emit harmonics above 20 kHz goes against logic and physics. 20 kHz is a limiting feature of the average [young] human auditory system and, as you say, is variable such that some individuals may be able to perceive acoustic energy above it. My experience of these things is that these people are rare - standards are not designed for the rarities in this world but for the mass.

The question is how much energy is present up there - not a lot compared with the whole spectrum? If it is missing does it matter to the bulk of the listening population? It's not impossible to regenerate it in playback anyway - a whiff of distortion will do the trick. If it does need to be captured at source then it follows that a large bit depth may be needed. Modern mastering machines should be able to capture low level spectral components well above 20kHz and so mastering formats should use high sampling frequencies and large bit depth if only to allow for processing in post.

But returning to the original post:

This problem is an unwanted side effect of digital media. I'd agree that there needs to be some clarity about source vs distribution coding parameters, preferably with a verified and defined file history. Fairly easy to do for a physical medium like a disc but rather hard in a file except by embedding it in the file format where the average consumer can't necessarily find it prior to purchase. Where it involves PCM sample rate and bit depth it should not be too much of a problem especially if the sample rate ratio is a simple one like 2:1 but if DSD gets involved a great deal of care is needed.

Labels like 48/24 are not guarantors of quality; they are format frameworks into which signals like audio can be fitted and preserved, providing that the required engineering precision implied by the sampling and quantising processes is achieved. For example, 24 bits is just a measure of the instrumental capacity of a hardware/software storage scheme. Garbage in Garbage out is a good principle familiar to engineers; just because a file’s sample rate is 96 kHz doesn’t mean its spectral content extends to 48 kHz. Similarly a 24 bit sample doesn’t necessarily mean that the LSBs carry anything useful.

Part of the problem is not necessarily theoretical but practical. We had a longish discussion about this sort of thing some months back [in April 2013] in a wide ranging thread about Vinyl conversion in which we got moved on to sampling formats:

http://www.for3.org/forums/showthread.php?7926-Vinyl-to-CD-again/page1&highlight=interpolation

If you are going to change source standard eg 96/24 to 44.1/16 then defining the two ratios isn't enough, it's how the filtering [ie computations] is done that matters and that filtering can involve noise shaping as well requiring bit depths greater than 24.

Any fool can define a filter to change a sample rate but to do it correctly and with the required precision takes some thought and will invariably involve a compromise in the size of the filter ie how many taps with a law of diminishing returns. I dread to think what happens in some cheap software format conversion packages. So in addition to the rate/bit depth change history of an audio file there should be a filter spec as well. If you are not doing it in real time then some speed constraints can be removed but the required precision remains. I have suspicion that many of the “high resolution” files offered are derived from CD rips. Exceptions may be those files from the original record company who have access to the original masters.

Clearly a 28 kHz sinusoid can be captured with 96 kHz sampling but not with 48 [it would translated to an alias at 20kHz]; downsampling has to remove anything above 24 kHz because it has no place in the baseband of a 48 kHz Nyquist system. However an 8 kHz sinusoid can exist in both and should in principle be represented equally accurately in each and that is the task of the down sampler filter. If the bit depth in the downsampled world is less than that in the original then provided the downsampled LSB is meaningful the system has delivered its potential and can do no more and so there is a theoretical loss of precision some of which is mitigatable by noise shaping. What that loss means subjectively is anyone guess. Some of the differences that sharp eared audiophiles seem to hear between formats may be nothing to do with the sampling or bit depth but the filters used.

[Jasmine Bassett]

I think we are in total agreement Gordon.

It's been a long time since I read some of the AES papers by Lipschitz and J. Vanderkooy from about 15 years ago which first illustrated some of DSDs less well publicised characteristics but this quote (from Wikipedia) neatly sums up the problem I have with old DSD recordings being "converted" to 192 /24 files:

<<Since around 1989, 1 bit delta-sigma modulators have been used in analog to digital converters. This involves sampling the audio at a very high rate (2.8224 million samples per second, for example) but only using a single bit. Because only 1 bit is used, this converter only has 6.02 dB of dynamic range. The noise floor, however, is spread throughout the entire "legal" frequency range below the Nyquist frequency of 1.4112 MHz. Noise shaping is used to lower the noise present in the audible range (20 Hz to 20 kHz) and increase the noise above the audible range. This results in a broadband dynamic range of only 7.78 dB, but it is not consistent among frequency bands, and in the lowest frequencies (the audible range) the dynamic range is much greater — over 100 dB. Noise Shaping is inherently built into the delta-sigma modulators.>>

I don't know if there are any digital to digital algorithms to go from DSD to Hi Res PCM but as you suggest above these would need to be exceptionally well designed to achieve anything worthwhile. If the conversion was done via the analogue domain then I would hope that the playback device would have a suitable LPF to remove all the high frequency noise but then there's nothing left to suggest 192 sampling would be of any use.

Needless to say there is no way I will purchase any of these files.

[Dave2002]

I do broadly agree with msgs 11 and 12. My point of msg 10 is that there are probably very few people who can reliably claim to hear a difference with any signals over 20kHz in a recording, and whether such components are desirable or useful is an open question which very few will be able to answer. I believe that it is possible to record to up to 100 kHz, and indeed experiments on animals would appear to show this. Of course we can't ask the animals whether they appreciate the quality of reproduction, but one interesting story concerns studies of dolphins. One experimenter set up microphones near a pool, and a dolphin was put in the pool. The experimenter realised that the dolphin was making sounds to which he (and perhaps others, reacted), and at different pitches. The recordings actually showed that the dophin had effectively run a hearing test on the humans nearby, and had emitted a lot of different noises to which the humans had not reacted. Once the dolphin had checked out that humans can't hear above 20kHz (or perhaps lower) it never bothered with high frequency utterances after that.

I don't know enough about DSD and noise shaping to be convinced of problems which can arise, though I do believe it is possible for such noise, possibly at high levels, to be mapped down into audible ranges by various forms of signal processing. Re garbage in - garbage out, some recordings are just not right even before any attempt is made to transmit or store them, as rfg has pointed out. Put the microphones in the wrong place, and it usually won't matter too whether the sampling rates or bit depths are high res or not. A slight caveat there though, it may sometimes possible to rescue recordings, particularly if they are of historical significance. This is particularly true with image processing, where sometimes appropriate filters can make a big difference to the appearance of an image. It is possible that some aspects of audio may be similar.

Regarding resampling, at least with audio the signals are approximately continuous functions, so with care resampling can be done with minimal damage. In video work there are significant problems in converting between frame rates for different systems, such as US TV, European TV, film etc., and some of the conversions might be quite ugly with the technologies we currently use.

[Gordon]

Thanks JB. That Wiki article is reasonably good. I hesitate to step into DSD file processing, it's complex [or easy when you know how!]. I know of no proven DSD-PCM tools but they must exist [a number of forums talk of proprietary solutions], after all it is "only" a mapping process from a string of DSD single bits [in DSD1] that describe a local variation in the audio amplitude but with no absolute anchor value [as in PCM] other than what has passed before. DSD is a kind of waveform following system based on a form of ∆-∑ modulation, a derivative of Differential PCM invented in the 1950s, where differences in successive audio amplitudes are capable of being coded in a stream of single bits, not several as in PCM proper.

The single bit represents a small amplitude change that one could compare [carefully] with the quantising step size of PCM - each expresses the resolving power of their respective systems. The maximum rate of change of the waveform determines the frequency at which that waveform needs to be sampled given the step size such that all amplitude changes that the waveform can experience can be captured accurately. It's a kind of "up a bit"/"down a bit" signalling to a decoder. The trick is to convert these anchorless differences into anchored PCM sample values. The dynamic range of DSD, claimed to be about 120 dB [assumed ref FS Unweighted], will determine the PCM bit depth which will need to be about 18 not allowing for computation round off, so 20 seems fine and 24 seems excessive.

The nominal claimed audio bandwidth in DSD is said to be around 50kHz or even 100 kHz [this looks arbitrary, having no obvious connection to anything else so is undefined] which means that to preserve it [why you may ask] a sample rate at 96 kHz is only just about acceptable so possibly that's why 192 is offered. Like you, I don't see what 192/24 has to offer. But, as Andy Grove said, Only the Paranoid survive!".

http://www.amazon.co.uk/Only-Paranoid-Survive-Andrew-Grove/dp/1861975139

[Gordon]

........Regarding resampling, at least with audio the signals are approximately continuous functions, so with care resampling can be done with minimal damage. In video work there are significant problems in converting between frame rates for different systems, such as US TV, European TV, film etc., and some of the conversions might be quite ugly with the technologies we currently use.

I worked in digital TV for many years and these are significant problems in the digital world. The problem is that whilst the spatial dimensions of a picture are manageable - more pixels ie HDTV - the temporal sampling - ie the frame rates of 25 or 30 Hz - ire far too low to match the human visual system but for historical and practical reasons it is not easy to increase. The bit rate needed and therefore the transmission bandwidths are directly a function of frame rate and so a doubling doubles everything or forces a bit rate reduction technology like MPEG4 or HEVC or whatever to work a lot harder and also the hardware has to work at a very high speed. Current HDTV [1920 x 1080 pixels at 25 Hz frame rate with 24 bits per pixel] gives an uncompressed bit rate of about 1.2 GigaBits/second.

Temporal Up sampling in TVs is commonplace now but not very well done. To do it properly you have to use motion compensation and that means lots of MIPS and memory in a consumer device where every penny counts.