External clock for Analog to Digital Converter

[Eeyore]

For audio, I wouldn't think that even a microsecond of RMS jitter
would be audible.

1 part in 200 ?

Are you joking ?

Graham

 

[John Larkin]

1 part in 200 ?

Are you joking ?

Graham

I don't know if a microsecond would be audible, but I'm sure that
picoseconds can't be. Brownian motion pounding on a microphone
element, or on your eardrum, has got to wobble things more than that,
not to mention the s/n of the whole signal chain. Try quantizing the
effects of vibration of your walls as it affects echoes... you'll
easily get nanoseconds.

As I said, jitter doesn't raise the quiet noise floor; it just
extracts a bit of what signal energy is present, and spreads it around
the spectrum. You'd have to be able to hear that low-level grass
scores of dB below the instantaneous program material. The paper cited
here misses this entirely.

John

 

[Eeyore]

I don't know if a microsecond would be audible, but I'm sure that
picoseconds can't be. Brownian motion pounding on a microphone
element, or on your eardrum, has got to wobble things more than that,

It sets 0dB actually AIUI.

not to mention the s/n of the whole signal chain. Try quantizing the
effects of vibration of your walls as it affects echoes... you'll
easily get nanoseconds.

As I said, jitter doesn't raise the quiet noise floor; it just
extracts a bit of what signal energy is present, and spreads it around
the spectrum. You'd have to be able to hear that low-level grass
scores of dB below the instantaneous program material. The paper cited
here misses this entirely.

I'd be surprised that Julian missed anything.

Let's take your example of 1us. With 48kHz sampling that represents an error of 1 part
in 200.

Since the 'reconstructed audio' can be seen if you like as 'join the dots' puzzle and
those dots are placed according to voltage and time, it doesn't take too much to
realise an error in one can be seen as broadly similar to an error in the other. The
two effects can be transposed.

With a random distribution to the jitter, that would be like a s/n ratio of a mere 46dB
!

Jitter in the tens of nanoseconds range is considered quite poor btw. Hundreds of
picoseconds is considered good.

Graham

 

[John Larkin]

It sets 0dB actually AIUI.



I'd be surprised that Julian missed anything.

Let's take your example of 1us. With 48kHz sampling that represents an error of 1 part
in 200.

What's that word you like... eh? 1 usec is 1 part in 21 of the period
of 48K.

But that isn't the point. If the program material were 100 Hz, 1 usec
timing errors would result in minute amplitude errors. If it were 10
KHz, the errors would be a lot bigger. It's the signal frequency that
matters here, not the sampling frequency.

Since the 'reconstructed audio' can be seen if you like as 'join the dots' puzzle and
those dots are placed according to voltage and time, it doesn't take too much to
realise an error in one can be seen as broadly similar to an error in the other. The
two effects can be transposed.

With a random distribution to the jitter, that would be like a s/n ratio of a mere 46dB

Even if the s/n is 46 dB, could you hear -46 dB of random noise

*** while the signal level is 0 dB? ***

Jitter in the tens of nanoseconds range is considered quite poor btw. Hundreds of
picoseconds is considered good.

Considered by whom? Audiophiles? The same guys who argue about burning
in and cryotreating RCA cables?

John

 

[Eeyore]

What's that word you like... eh? 1 usec is 1 part in 21 of the period
of 48K.

Silly me. I slipped a zero. Not sure how I managed that.

But that isn't the point. If the program material were 100 Hz, 1 usec
timing errors would result in minute amplitude errors.

Who said it was anything to do with amplitude errors ? What's 'minute' btw ?

If it were 10 KHz, the errors would be a lot bigger. It's the signal frequency that
matters here, not the sampling frequency.

I don't see any supporting analysis.

Even if the s/n is 46 dB, could you hear -46 dB of random noise

*** while the signal level is 0 dB? ***

In this simple instance.

Considered by whom? Audiophiles? The same guys who argue about burning
in and cryotreating RCA cables?

Those are *AUDIOPHOOLS*.

Tell you what, why not try adding 1us jitter and listen to it ?

Graham

 

[John Larkin]

Silly me. I slipped a zero. Not sure how I managed that.



Who said it was anything to do with amplitude errors ? What's 'minute' btw ?

See below.

I don't see any supporting analysis.

Take a sample of a 1 volt peak, 100 Hz sine wave. Now booger the
sample time 1 ns early or late. The time error is 1e-7 * 360 degrees =
36 micro-degrees. If we sampled at the top of the sine wave, nothing
happens... slope is zero. At the steepest part, at the zero crossing,
the corresponding voltage error is 0.63 microvolts, just the sine of
the error angle.

Repeat for 10 KHz.

John