384kHz/192kHz Audio Foolness

Status
Not open for further replies.
Jump to Latest
diyralf

diyralf

Member
Joined 2016
#1
Question: I would be interested in whether the difference between 44.1 kHz and 48 kHz / 96 kHz / 192 kHz in the audio recording can be heard.
Answer: No. A higher sampling rate than 44.1 kHz is meaningless since frequencies above 20 kHz can not be heard by anyone. The DIN standard has even set a limit of only 16 kHz.

Question: Why do manufacturers offer these formats?
Answer: Business. Marketing.

Question: What sampling frequency is necessary to accurately reproduce an Hi-Fi audio signal.
Answer: 44.1 kHz
 
Last edited:
#4
While I see your point, I'm not sure I agree.

44kHz is the absolute minimum birate needed to describe a 22kHz frequency, so long as they're in-phase. If you get a 22kHz sine wave, phase shift it by 90 degrees and then take 44kHz samples, every sample will read zero since they coincide exactly with the zero points of the 22kHz sine wave.

I would therefore argue that you need samples at around 4-8x your maximum audible frequency, so 96kHz sampling seems like a good idea. Not for the extra bandwidth, but in order to retain high-frequency phase information.

Chris
 
#9
Without their context these figures are saying nothing at all !

Higher sampling rates make it easier to implement the necessary anti aliasing filters.

And because it is theoretically possible to sample a steady-state input signal (i.e. a sinusoidal) whose frequency is half that of the sampling frequency doesn't mean that it is wise to do so with music in practice.
There is room for discussion however which choice of sampling-rate would be the best.

Regards

Charles
 
#10
chris661 said:
44kHz is the absolute minimum birate needed to describe a 22kHz frequency, so long as they're in-phase. If you get a 22kHz sine wave, phase shift it by 90 degrees and then take 44kHz samples, every sample will read zero since they coincide exactly with the zero points of the 22kHz sine wave.
Click to expand...
True (but irrelevant).

I would therefore argue that you need samples at around 4-8x your maximum audible frequency, so 96kHz sampling seems like a good idea. Not for the extra bandwidth, but in order to retain high-frequency phase information.
Click to expand...
False deduction from an irrelevant statement.

To fully describe a 22kHz sinewave using digital sampling you need any sampling frequency greater than 44kHz, so 44.1kHz is fine. You just need a good anti-aliasing filter before the ADC, and a good image filter after the DAC.

This issue has been gone into in thread after thread, so no need to start a new one. Find a good book or website on digital audio, then read it and try to understand it. If you think it is wrong, then this is probably because you don't understand it because you haven't thought enough about it or your mathematical background is insufficient.
 
#12
I would put the idea that in case of audio reproduction at home you don't really need much or any AAF if your amp is good enough.
NOS ppl are happy with little AAF and class D amps also put out a lots of out of band noise.
AAF is important if the fed unit like (pre) amp is sensible for out of band signals.
Of course other industries like RF needs proper AAF.
 
#14
The terminology is that ADCs use AAFs (anti-aliasing filters) and DACs need AIFs (anti-imaging filters). If your DAC is putting out significant ultrasonics you may well notice a loss in dynamics without bandlimiting the DAC's output to the audio band. At least that's been my experience with NOS DACs but its quite possible my amps sucked.

ClassD amps do put out significant out of band noise but they don't much like it on their inputs, again in my experience.
 
#17
DF96 said:
True (but irrelevant).


False deduction from an irrelevant statement.

To fully describe a 22kHz sinewave using digital sampling you need any sampling frequency greater than 44kHz, so 44.1kHz is fine. You just need a good anti-aliasing filter before the ADC, and a good image filter after the DAC.

This issue has been gone into in thread after thread, so no need to start a new one. Find a good book or website on digital audio, then read it and try to understand it. If you think it is wrong, then this is probably because you don't understand it because you haven't thought enough about it or your mathematical background is insufficient.
Click to expand...

The extra 100Hz sampling rate is sufficient, because then even if a 22kHz wave does come along, it'll be picked up within milliseconds. Though it'll also drop out every so often, giving a tremolo effect. I guess if you're only going up to 20kHz it's not a problem.

So I fired up Audacity, made two different projects (one with a 44.1kHz sample rate, and one with a 192kHz sample rate), and generated a 22kHz sine tone in each.

An externally hosted image should be here but it was not working when we last tested it.


I guess the anti-aliasing would straighten out the top one, right?

FWIW I have a reasonable mathematical background, having recently completed a Physics degree. There was one lecture on digital audio, and the Nyquist frequency bit always bugged me.

Chris
 
#18
chris661 said:
I guess the anti-aliasing would straighten out the top one, right
Click to expand...
No. The anti-image filter does that. As you were starting from a pure sine wave you didn't need an anti-aliasing filter because you guaranteed that the input was already bandlimited.

If you generate a 22kHz sine wave in a 44.1kHz system then the raw data is a 22kHz sine plus a 22.1kHz sine. The latter is the image which the filter will then remove.

Congratulations on completing a physics degree. That means that you should have enough maths to understand sampling, although it may take a while. I too have a physics degree but it took me a while to understand how Nyquist got it right. Most people look at the raw data (e.g. on a sampling digital oscilloscope) and conclude that much faster sampling is needed - but that is without an anti-image filter.
 
Last edited:
#19
Is it possible to take into account the phase response of the anti-aliasing filter, when designing an anti-imaging filter? In other words, could the combined phase response made flat within the passband? Probably it could be done with a DSP, question if does it make any sense.
 
#20
DF96 said:
(1) No. The anti-image filter does that. As you were starting from a pure sine wave you didn't need an anti-aliasing filter because you guaranteed that the input was already bandlimited.

(2) If you generate a 22kHz sine wave in a 44.1kHz system then the raw data is a 22kHz sine plus a 22.1kHz sine. The latter is the image which the filter will then remove.

(3) Congratulations on completing a physics degree. That means that you should have enough maths to understand sampling, although it may take a while. I too have a physics degree but it took me a while to understand how Nyquist got it right. Most people look at the raw data (e.g. on a sampling digital oscilloscope) and conclude that much faster sampling is needed - but that is without an anti-image filter.
Click to expand...

(Numbers added by me)
1 - Makes sense.
2 - It does look suspiciously like a beat frequency, which would be 100Hz. So the image filter looks at what's sampled, compares it to the sample rate, figures out what's an image (ie, what's above the Nyquist limit?) and what's actual signal, and filters accordingly. Neat.
3 - Thanks!

So Nyquist was right, so long as you've got the ability to do some checks on the sampled signal. With no filtering on the input or output, you would need a very high sample rate but the filters we can apply mean the extra bandwidth isn't necessary, so you can save space on your harddrive.

The programming behind it is beyond me, I suspect. I think I'm happier messing with the bits of cardboard and magnets at the other end of the signal chain.

Thanks for the lesson!

Chris
 
Status
Not open for further replies.
Top Bottom