Are 24bit/192KHz music files really better than the CD standard?

I am clearly not a DSP guy so I believe almost anything concerning such.
Having now DACs that can handle 24bit/192KHz, I thought of updating a bit of my music collection to 24bit/192KHz level. Looking for a suitable site to buy music in that quality, I noticed this article: 24/192 Music Downloads are Very Silly Indeed . The article concludes in large that 24bit/192KHz music serves no purpose. The article appears very serious and the argumentation is based on sampling according to the Nyquist-Shannon Theorem.
If I had to describe a time-varying signal through sampling, my intuition would urge me to use as many equidistant samplings as possible with as fine a resolution as possible. Without doing any calculations, I would be able to reproduce the original signal quite closely by just repeating the sampled values. The article states that I am not the only person being foolish (luckily there are co-foolish so I do not feel singled out) and that I am wrong.
The way I understand the article, sampling according to the Theorem should ensure full information about the signal up to the limit (half the sampling rate) given by the Theorem, eventually through Fourier calculations.

From search on this forum I am left with the impression that 24bit/192KHz music is a clear improvement. Perhaps one improvement is that Fourier calculations become obsolete, at least in part.

My questions are, do I have a benefit from 24bit/192KHz music compared to my present CDs?
If not, are the many disclosures about new high quality music standards really just a commercial scam? Is a difference the amount of calculations I need to do on samplings according to the Theorem in order to reproduce the signal?

Thanks for any reply. Sorry if I have overlooked existing replies to my questions.
 
Last edited:

ubergeeknz

Member
2018-08-31 12:32 pm
Just my understanding and observations, I'm not an expert in the field by any means...

From what I understand the main improvement from using high sample rates is that the recording of higher frequency content is possible.

Due to the above, filtering can happen well outside the audible range (instead of very close to it, affecting some of the audible frequencies). Some kinds of filtering can cause distortion artifacts back into the audible range as well. These issues are mitigated with a higher than needed sample rate.

The other concepts center around our ability to process phase information beyond the frequency to which we can hear, also that we perhaps sense in some other way high frequency content which we cannot, strictly speaking, "hear".

More bits per sample give a higher possible dynamic range. But it seems like the 106dB possible with a 16 bit sample is already overkill for musical content.

Add to this - many "high res" recordings were not recorded or mastered for high res formats. So whatever you might gain in fidelity as already been lost.
 
Do you have a cat who likes music? If so, 192 kHz is definitely better than 44.1 kHz.

If not, and no other pets either, then it all depends on two things:

1. Whether ultrasonic signals are really always inaudible, also when heard in combination with other signals.
2. Whether the recordings are peak sample normalized and whether your playback chain can handle it when they are.

1) The only experiment I know of that indicates that music may sound better with ultrasonics included is a 1990's Japanese experiment involving EEG equipment. Japanese gamelan players listening to Balinese gamelan recordings had different brain waves when a super tweeter reproducing signals above 26 kHz was turned on than when it was off, even though they could not consciously hear the sound of the super tweeter on its own. From the description of the experiment, it wasn't clear to me whether it was double- or only single-blind, single-blind tests are known to be rather unreliable.

2) Many if not most DACs and digital signal processing chips clip on peak sample normalized music, the Benchmark Media site has some good explanations why. This effect gets less severe with increasing sample rate, but you can solve it completely by digitally reducing the volume before any type of filtering is applied to the signal.
 

Galu

Member
2018-04-17 6:50 pm
It depends on which era of music you listen to.

If we look at the history of recording, PCM digital first displaced analogue tape around 1970.

The recordings in the early 70s were made at 13 bit resolution, increasing to 14 bits by the end of the decade - and no amount of fiddling with bit rates today will restore the missing information.
 
As already mentioned, it all depends on the masters. Getting 48k copies of 48k masters might be worth it, or hi-res transfers from analogue, though 40 year old tapes have there own problems. The other worry, are they remastered, as in over compressed? My guess is most are not worth it, they may even be worse than the CD.
 

oscroft

Member
2019-09-07 1:29 pm
As I understand it, higher sampling rates are nothing to do with producing very high frequencies for cats (or anyone/anything else) to hear. The Nyquist theorem is sound and can reproduce the human hearing range at the CD sampling frequency of 44.1hKz just fine, but that's not what higher sampling rates are all about.

As ubergeeknz says, the difference it makes is in the filtering in the analog conversion, when the sampling frequency needs to be filtered out. If the file is sampled at 44.1kHz, that's only a little over an octave above 20khz, so the filtering needs to be very steep. But very steep and very clean filters don't exist, so the filter produces distortion that reaches down into the audible range.

With a higher sampling rate, the filter can be less steep - and the higher the sampling rate, the less steep the filter. And the less steep the filter, the less distortion there is extending down into the audible range.

It's the same reason we have upsampling DACs, which resample a 44.1/48kHz source in the digital domain before analog conversion - upsample, then use a gentler filter that produces less distortion.

I'm using a DacMagic Plus upsampling DAC which upsamples to 384kHz before analog conversion. It's a relatively modestly priced DAC, but I can hear a definite improvement over a couple of non-upsampling DACs I've compared it to.

Now, it might be simply that it's a better DAC than the comparisons, but the DacMagic Plus has three different filter settings to choose from, and they definely do sound different - I'm not sure which I prefer, but to my ears two of them are defintely better than the third.

Whether there's any difference between a source sampled at 192kHz and a 44.1kHz source upsampled, I don't know - but I am sceptical.

As for 24 bit samples, from what I understand, the maximum dynamic range that's practically possible extends to only 19 or 20 bits, with any greater resolution being below the thermal noise floor. So 24-bit should be better than 16-bit, but not to the full 24 bits. Whether it really is better? I've only heard a few 24-bit tracks, and I can't hear a difference on my system (which is not high end, so I can't say more than that).
 
Last edited:
Oscroft's response reminds me that I forgot a third issue:

3. Whether your DAC uses linear-phase interpolation filters with too much passband ripple

3) If the DAC uses a linear-phase interpolation filter with too much passband ripple, this will cause a pre-echo. Unlike the ultrasonic sin(x)/x-shaped pre-ringing that all linear-phase interpolation filters have, this pre-echo is in the audible frequency range. The higher the original sample rate and the less steep the filter, the shorter the time between the pre-echo and the main signal and the better it gets masked.
 

enryfox

Member
2019-02-06 8:08 pm
As already mentioned, most of musical recordings hardly use 14 bit, the remaining 10 bit are used just to more accurately capture the master tape hiss.

192 kHz might make sense just to reproduce all the frequency content of the recorded music (provided the source is not filtered during mastering) but whether or not they can be heard is still a matter of dispute. Beside that, you need a Hi-Res certified audio chain (amplifier and loudspeaker) otherwise ultrasonic content might add distortion in the audible region.

24/192 is mostly marketing, the industry cannot tolerate that a standard defined in the late '70s (16/44.1) is still optimal for music listening.
 
Many thanks for all the competent replies.
In general, the replies are much closer to the article I cited than I initially expected. I did not find the identity of the author but with what is presented in the article and compared to your comments, it is someone competent.

To summarize:
16bit is generally enough for music with 19-20 bits marking the thermal noise floor such that 24 bits may be used to ensure a dynamic margin for post treatment.
44.1K/48K sampling gives a sufficient bandwidth for normal audio use while ultrasonic frequencies may have some influence on the sound perception and higher sampling rates will counter some clipping and pre-echo effects.
The advantage of higher frequency sampling mainly relates to the need for less steep filters which is particularly important with analog filters and for low distortion.

I found this page: 2L High Resolution Music .:. free TEST BENCH , offering recordings according to various standards. With those recordings I can make a comparison myself as suggested.

Having experienced the period back in the 70-80’ties with the attempts to make us all change for quadraphonic surround systems (it failed solidly), I particularly understand the statement that the music and HiFi industry has difficulties accepting such an old standard as used for the CDs still being the de facto standard in use. They have to try to offer us something new but with your technical expertise and the article I cite, 24bit/192KHz may have a rather disappointing future.

Once more thanks
 
Last edited:
Since music is mainly non-sinusoidal you need a lot more bandwidth then overly simplified Nyquist-Shannon Theorem dictates.

A hard strike on a kick drum generates content well into the MHz range due to the steep non sinusoidal shock wave generated.

The theorem also leads to a formula for perfectly reconstructing the original continuous-time function from the samples.

From: Nyquist–Shannon sampling theorem - Wikipedia

There is nothing "continuous" with the leading edge shock wave from a drum or other transient sounds (which makes up most of the content in music).

There is a huge difference between 16/44,1 and 24/192 when compared with well recorded and very little processed source material. If you take a nice BIS recording of a grand piano and a/b compare between high rez and normal CD quality the difference is quite startling.
 

enryfox

Member
2019-02-06 8:08 pm
Since music is mainly non-sinusoidal you need a lot more bandwidth then overly simplified Nyquist-Shannon Theorem dictates.

A hard strike on a kick drum generates content well into the MHz range due to the steep non sinusoidal shock wave generated.


No matter what is the frequency contents of a specific sound, our hear cannot detect anything beyond 18/20 kHz and that is true at best up to to 25 years of age. Why bother reproducing something we cannot possibly hear ?
Using an analogy, would you buy a TV that emits ultraviolet and X rays for the sake or "accurately reproducing the image" ?
 
Sign up Qobuz for free (1 month), and compare 44.1K and 96K or 192K mastered versions of the same song. I recommend Audirvana as a player (also 30 days free), and you should clearly hear the difference. Many older songs are through different mastering process (= different sound), but newly mastered songs are mostly hires mastered and down sampled to 44.1K.
 
Oscroft's response reminds me that I forgot a third issue:

3. Whether your DAC uses linear-phase interpolation filters with too much passband ripple

3) If the DAC uses a linear-phase interpolation filter with too much passband ripple, this will cause a pre-echo. Unlike the ultrasonic sin(x)/x-shaped pre-ringing that all linear-phase interpolation filters have, this pre-echo is in the audible frequency range. The higher the original sample rate and the less steep the filter, the shorter the time between the pre-echo and the main signal and the better it gets masked.

Actually, it does not matter if DAC's "internal" FIR filter setting is Linear or not. Even the filter on ESS chip does sound bad . We can upsample 44.1K with or without interpolation in real time (Audirvana or some other desktop software) and can move DAC's internal cheap FIR filter to 2x or 4x frequency range, which should make any DAC (arguably except Chord DAC) sounds better. FIR is sn extremely heavy process and should not be handled by tiny DAC chip.
 
Actually, it does not matter if DAC's "internal" FIR filter setting is Linear or not. Even the filter on ESS chip does sound bad . We can upsample 44.1K with or without interpolation in real time (Audirvana or some other desktop software) and can move DAC's internal cheap FIR filter to 2x or 4x frequency range, which should make any DAC (arguably except Chord DAC) sounds better. FIR is sn extremely heavy process and should not be handled by tiny DAC chip.

For the record, I'm not against linear-phase filters at all. It is easy to find a reasonably short linear-phase FIR filter coefficient set that reduces pre-echoes to well below -120 dB, a 1970's Fortran program written by James McClellan will readily do that for you when you enter the right parameters. However, filters with larger passband ripples and larger pre-echoes are somewhat cheaper and are therefore used in many DACs. I haven't a clue how good, bad or ugly the filters in ESS chips or in Audirvana are.