Sign in to follow this  
Followers 0
tones

Sampling frequencies and their relevance to the sound

45 posts in this topic

'Allo, 'allo, 'allo, wot's all this then? I have a few days off (as opposed to the usual few off-days) and a new bit springs up like a weed. Good luck to it.

As something worthy of technical note, I am proposing a link to good ol' (or bad ol') HFC:

http://forum.hifichoice.co.uk/viewtopic.php?t=24946

which I hope James will allow, because, later in the discussion of vinyl enters the HFC resident technical whiz adb (qualified in electronics). The bit of interest is p.5, where he does a great exposition on sampling rates and why people's intuitive ideas on the subject are completely up the creek. It's not a complete explanation, but it's the best we'll get without resorting to the mathematics (in which I for one would be lost by the first few lines).

0

Share this post


Link to post
Share on other sites

tones wrote:

'Allo, 'allo, 'allo, wot's all this then? I have a few days off (as opposed to the usual few off-days) and a new bit springs up like a weed. Good luck to it.

As something worthy of technical note, I am proposing a link to good ol' (or bad ol') HFC:

http://forum.hifichoice.co.uk/viewtopic.php?t=24946

which I hope James will allow, because, later in the discussion of vinyl enters the HFC resident technical whiz adb (qualified in electronics). The bit of interest is p.5, where he does a great exposition on sampling rates and why people's intuitive ideas on the subject are completely up the creek. It's not a complete explanation, but it's the best we'll get without resorting to the mathematics (in which I for one would be lost by the first few lines).

Funnily enough, I was reading that yesterday.

I think the difficulty comes when you try and bring this to the real world and design and build a practical implementation. This is where the shortcomings of digtal come in.

One thing he doesn't cover which I think is extremely important is hoe to preserve linearity at low levels and minimise quantisation errors of low level signals.

0

Share this post


Link to post
Share on other sites

As I understand it (and I readily admit that my knowledge of digital electronics amounts to spelling the words correctly (most of the time)), this is the real-world situation - this is what actually happens. Like adb, I believe that digital has no shortcomings in comparison with vinyl (neither of course will ever be - can ever be - perfect).

As to linearity andquantisation levels, I don't know, but I suspectthe effect of these are much exaggerated (ditto "jitter"). I suspect that it's another case of the well-known audiophile disease of taking a known physical phenomenon and either exaggerating its significance out of all proportion or completely misattributing it.

I have both vinyl and CD to a reasonably high quality (Meridian 588, Linn LP12/Ekos/Klyde) and I find it easy to hear the differences - the record has occasional surface noise and the CD doesn't. So, it's CD for me. Other folk (such as yourself obviously) think differently. Thats' fine. We only have our own ear/brain combinations to satisfy.

0

Share this post


Link to post
Share on other sites

tones wrote:

As I understand it (and I readily admit that my knowledge of digital electronics amounts to spelling the words correctly (most of the time)), this is the real-world situation - this is what actually happens. Like adb, I believe that digital has no shortcomings in comparison with vinyl (neither of course will ever be - can ever be - perfect).

Yes, it is what actually happens and it's possibly to correctly re-construct all freqencies up to fs/2. When people say that you don't have enough points to follow the curve, they are mis-understanding. Any curvature is a sign of higher frequencies in the signal. As we are accepting band limiting, we can safely ignore this.

As to linearity and quantisation levels, I don't know, but I suspect the effect of these are much exaggerated (ditto "jitter"). I suspect that it's another case of the well-known audiophile disease of taking a known physical phenomenon and either exaggerating its significance out of all proportion or completely misattributing it.

Not at all exaggerated- in fact the linerarity of raw CD is only good at high levels. Only at 0dBFS levels do you have the full 65536 steps to quantise a waveform. At lower levels, quantisation errors become large. Unfortunately, quantisation noise is correlated to the signal and the evidence indicates that correlated noise is subjectively more audible and deleterious. However, this can be ameliorated almost totally with the addition of dither, which is carefully shaped noise and effectively turns the quantisation error signal into random un-correlated noise at the expense of dynamic range.

Additional filters are required to remove aliasing images (Nyqusit theory says that these will appear at multiples of the sampling frequencies). In the case of CD, this requires an aggressive filter to sharply delineate the stop-band from the pass-band. These "brick wall" filters can cause frequency response aberrations (rippled response in HF) and/or time domain anomalies at HF, hence some people's preference for non-oversampling DACs without digital filters.

Jitter is another issue, and due to clock errors when rebuilding the analogue signal. This has the effect of generating spurious signals, which may or may not be correlated to the audio.

It seems we have learnt a lot about digital, especially in the last 5 years or so, and it is maturing as a technology, but I think there is still a lot to learn.

As far as analogue goes, the ultimate resolution, I would hazard is actually higher than CD, and is basically limited by the size of the vinyl molecules (it can be calculated based on groove size, etc.). It is however still "quantised" after a fashion. The absolute dynamic range is limited by the noise floor, rather than resolution but this noise is truly random and therefore uncorrelated, making it subjectively less objectionable, though the dynamic range is significantly poorer than CD. Additionally, the bandwidth limiting is more graceful meaning fewer problems with filtering. Thus a less theoretically perfect system can sound very good indeed, because it has to do less processing and thus is easier to implement in practice.

Interestingly, I was thinking about photography the other day and the luminance dynamic range of the best D-SLRs is about 10 stops, or 10-bits. Seems our ears might be better than our eyes. ;)

I have both vinyl and CD to a reasonably high quality (Meridian 588, Linn LP12/Ekos/Klyde) and I find it easy to hear the differences - the record has occasional surface noise and the CD doesn't. So, it's CD for me. Other folk (such as yourself obviously) think differently. Thats' fine. We only have our own ear/brain combinations to satisfy.

You're perfectly entitled to your own opinion, I don't have any problem with that! Currently, I prefer my vinyl setup - it has much more colour and better rendering of timbre. The CD player is no slouch, it's just that my vinyl seems to give a more natural rendering of the performance. This is of course a subjective opinion only.

Just my 2p on the subject.

Edit: typos.

0

Share this post


Link to post
Share on other sites

i_should_coco wrote:

Interestingly, I was thinking about photography the other day and the luminance dynamic range of the best D-SLRs is about 10 stops, or 10-bits. Seems our ears might be better than our eyes. ;)

Aye.. I remember a demo years ago where the effect of different bit rates on video was shown - anything up to 10 bit was largely rubbish (at least on a grade 1 monitor), 10 and higher was largely indistinguishable (we're still talking 625 line grade 1).

Minimum "acceptable" for audio was 12 bit compressed - IIRC NICAM...

0

Share this post


Link to post
Share on other sites

Leonard Smalls wrote:

i_should_coco wrote:

Interestingly, I was thinking about photography the other day and the luminance dynamic range of the best D-SLRs is about 10 stops, or 10-bits. Seems our ears might be better than our eyes. ;)

Aye.. I remember a demo years ago where the effect of different bit rates on video was shown - anything up to 10 bit was largely rubbish (at least on a grade 1 monitor), 10 and higher was largely indistinguishable (we're still talking 625 line grade 1).

Minimum "acceptable" for audio was 12 bit compressed - IIRC NICAM...

Not sure if I was clear, I didn't mean 10-bit colour (most computers these days use 24-bit, but that's 8 bits per colour). I mean the absolute luminance dynamic range, from pure black to pure white. Up until recently cameras have used 12-bit processing and I think the latest models use 14-bits.

0

Share this post


Link to post
Share on other sites

i_should_coco wrote:

Not sure if I was clear, I didn't mean 10-bit colour (most computers these days use 24-bit, but that's 8 bits per colour). I mean the absolute luminance dynamic range, from pure black to pure white. Up until recently cameras have used 12-bit processing and I think the latest models use 14-bits.

IIRC, which I possibly don't as it was a long time ago, 10 bits was the minimum just for Y to get good quality... Fair enough I spose as MPEG1 used 12 for Y,U and V!

The bit rate for a proper component system, such as the early SonyD1, was 10 each. Which got expensive for early HDTV (at 1250 lines) as they used one D1 machine for each component, at getting on for £100k per machine!

If you looked at Panasonic's (I'm still only talking video here, know bugger all about still cameras!) D3 that used only 8 bits per sample for uncompressed composite video.. Still £30k per machine though - now you can pick one up for pennies.

0

Share this post


Link to post
Share on other sites

Uncle Ants wrote:

Fascinating stuff. So if I understand this right the sampling frequency just limits the maximum frequency recorded and has no effect on resolution within that limit. Is that right?

Yes, that's it.

0

Share this post


Link to post
Share on other sites

i_should_coco wrote:

Uncle Ants wrote:
Fascinating stuff. So if I understand this right the sampling frequency just limits the maximum frequency recorded and has no effect on resolution within that limit. Is that right?

Yes, that's it.

Excellent I learned something today. I was under the impression that the higher the sampling rate the higher the resolution at all frequencies ... so what does increased bit depth do for us?

0

Share this post


Link to post
Share on other sites

Uncle Ants wrote:

i_should_coco wrote:
Uncle Ants wrote:
Fascinating stuff. So if I understand this right the sampling frequency just limits the maximum frequency recorded and has no effect on resolution within that limit. Is that right?

Yes, that's it.

Excellent I learned something today. I was under the impression that the higher the sampling rate the higher the resolution at all frequencies ... so what does increased bit depth do for us?

It allows more discrete steps to "measure" the amplitude of the waveform at that point in time. As we only have 65536 steps with 16 bits, 65536 gives a dynamic range of 96dB. We can only represent the amplitude as one of these discrete values. If the actual value differs sligthly, we round up or down. This is called "quantisation error" and becomes worse when trying to capture low level signals.

If we have a waveform with a 2V peak to peak amplitude, the smallest step is 30.15mV and half of this is the maximum error. That's tiny compared to a full level signal. At lower levels, e.g. -40dB down, the signal is 0.02V, i.e. 20mV, so as you can see the smallesy step will be much greater than the signal leading to a huge error.

Increasing bit depth to 24 bits, gives possibility of 16777216 steps, and using the same 2V reference, gives a smalles step of 0.000119mV. About a ten-thousandfold increase.

0

Share this post


Link to post
Share on other sites

i_should_coco wrote:

Uncle Ants wrote:
i_should_coco wrote:
Uncle Ants wrote:
Fascinating stuff. So if I understand this right the sampling frequency just limits the maximum frequency recorded and has no effect on resolution within that limit. Is that right?

Yes, that's it.

Excellent I learned something today. I was under the impression that the higher the sampling rate the higher the resolution at all frequencies ... so what does increased bit depth do for us?

It allows more discrete steps to "measure" the amplitude of the waveform at that point in time. As we only have 65536 steps with 16 bits, 65536 gives a dynamic range of 96dB. We can only represent the amplitude as one of these discrete values. If the actual value differs sligthly, we round up or down. This is called "quantisation error" and becomes worse when trying to capture low level signals.

If we have a waveform with a 2V peak to peak amplitude, the smallest step is 30.15mV and half of this is the maximum error. That's tiny compared to a full level signal. At lower levels, e.g. -40dB down, the signal is 0.02V, i.e. 20mV, so as you can see the smallesy step will be much greater than the signal leading to a huge error.

I see, so its something which is going to matter on highly dynamic material but matter not a jot on say modern rock and pop stuff where it's all within 15db of the end stops.

0

Share this post


Link to post
Share on other sites

Uncle Ants wrote:

i_should_coco wrote:
Uncle Ants wrote:
i_should_coco wrote:
Uncle Ants wrote:
Fascinating stuff. So if I understand this right the sampling frequency just limits the maximum frequency recorded and has no effect on resolution within that limit. Is that right?

Yes, that's it.

Excellent I learned something today. I was under the impression that the higher the sampling rate the higher the resolution at all frequencies ... so what does increased bit depth do for us?

It allows more discrete steps to "measure" the amplitude of the waveform at that point in time. As we only have 65536 steps with 16 bits, 65536 gives a dynamic range of 96dB. We can only represent the amplitude as one of these discrete values. If the actual value differs sligthly, we round up or down. This is called "quantisation error" and becomes worse when trying to capture low level signals.

If we have a waveform with a 2V peak to peak amplitude, the smallest step is 30.15mV and half of this is the maximum error. That's tiny compared to a full level signal. At lower levels, e.g. -40dB down, the signal is 0.02V, i.e. 20mV, so as you can see the smallesy step will be much greater than the signal leading to a huge error.

I see, so its something which is going to matter on highly dynamic material but matter not a jot on say modern rock and pop stuff where it's all within 15db of the end stops.

Not exactly, there can still be low level harmonic content riding on the large waveform, and you stil need to capture that.

0

Share this post


Link to post
Share on other sites

Interesting thread. So what's the standard sample depth or rate used by recording studios? Is most stuff recorded at CD level, or is it recorded at a higher level and down sampled to fit the relevant formats being marketed?

0

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!


Register a new account

Sign in

Already have an account? Sign in here.


Sign In Now
Sign in to follow this  
Followers 0

  • Recently Browsing   0 members

    No registered users viewing this page.