What Does Jitter Sound Like? Let's Test

[alexwgoody]

jitter_1

 

I was sent this test by a buddy, and will likely do it tonight or this weekend. It is geared towards the pro audio market, and has higher levels of jitter than you'd get under most circumstances, but it should be interesting to test. To quote the instructions, "The reason I didn't ask "which has more jitter" is because it's irrelevant to the listening test - people need to use the subjective terminology as this is a non-technical test. There is no reason to be embarrassed; this is about learning what this sounds like - ear training."

 

Thoughts? Results?

[new_media]

Sounds like chicken.

[sandyk]

jitter_1

 

I was sent this test by a buddy, and will likely do it tonight or this weekend. It is geared towards the pro audio market, and has higher levels of jitter than you'd get under most circumstances, but it should be interesting to test. To quote the instructions, "The reason I didn't ask "which has more jitter" is because it's irrelevant to the listening test - people need to use the subjective terminology as this is a non-technical test. There is no reason to be embarrassed; this is about learning what this sounds like - ear training."

 

Thoughts? Results?

 

I would suggest that you contact "Barrows" who some time back posted a very informative post about the effects of "Jitter"

Both John Kenny and myself later expanded on that original post in another forum. I had a copy of his original post saved, but lost it when I upgraded my PC. BTW, some people even prefer the effects of a little added "Jitter"

Alex

[wgscott]

Sounds like chicken.

 

Tastes like phase error.

[Audio_ELF]

Sounds like chicken.

Yes but a Cookoo Maran hybrid sounds very different from a Rhode Island Red and that sounds very different to any bantam...

[sandyk]

WGS

Ask your delightful furry friend what it tastes like. My sincere apologies if the photo is of the much loved animal you lost a while back.

[christopher3393]

tastes like bait. remember eatin' bait? nightcrawlers, minnows, etc.?

[wgscott]

WGS

Ask your delightful furry friend what it tastes like. My sincere apologies if the photo is of the much loved animal you lost a while back.

 

It is. Her brother is still alive and well, however.

[bluesman]

BTW, some people even prefer the effects of a little added "Jitter"

Would they be jitterbugs?

[jhwalker]

Surprisingly, I was able to hear the differences. I correctly picked the original file, and ranked each of the others according to my preference. When I listened to the difference files, it was clear why I ranked them as I did.

 

Thanks for posting.

[JohnSwenson]

The pages on these files never actually states what was actually done to these files.

 

Since jitter as we normally use the term here is the timing variation of the clock at the DAC chip, how can you "add jitter" to the data in a file?

 

The best you can do is a simulation of what you think jitter does to the output signal. The description of the test files makes no attempt at explaining what was done. There are many different types of jitter, which was used in this jitter simulation test has not been specified.

 

Then in reality this simulation of what jitter does is played back on a system which has it's own builtin actual jitter which will cause a modulation of some form of the simulated jitter.

 

So while these sorts of tests are interesting, they do not necessarily tell you what real jitter (actually varying the timing of the clock at the DAC chip) actually sounds like.

 

John S.

[sandyk]

Surprisingly, I was able to hear the differences. I correctly picked the original file, and ranked each of the others according to my preference. When I listened to the difference files, it was clear why I ranked them as I did.

 

Thanks for posting.

 

John

Why should that surprise you when the levels of Jitter in the files are so high, that the binary contents are no longer the same ?

Alex

[Jud]

The pages on these files never actually states what was actually done to these files.

 

Since jitter as we normally use the term here is the timing variation of the clock at the DAC chip, how can you "add jitter" to the data in a file?

 

The best you can do is a simulation of what you think jitter does to the output signal. The description of the test files makes no attempt at explaining what was done. There are many different types of jitter, which was used in this jitter simulation test has not been specified.

 

Then in reality this simulation of what jitter does is played back on a system which has it's own builtin actual jitter which will cause a modulation of some form of the simulated jitter.

 

So while these sorts of tests are interesting, they do not necessarily tell you what real jitter (actually varying the timing of the clock at the DAC chip) actually sounds like.

 

John S.

 

I have a link somewhere around here to a program to add simulated jitter to the playback of a file in the quantity desired, in order to get some idea of whether you can hear various levels and how they affect the sound. Subject to some of the same problems you mention, John, though the developer is accessible and can explain what is being done.

 

I'll poke around and have a look for it if anyone is interested in trying it.

[jhwalker]

John

Why should that surprise you when the levels of Jitter in the files are so high, that the binary contents are no longer the same ?

Alex

 

Because my equipment is not very revealing

[sandyk]

Because my equipment is not very revealing

 

Your equipment, or your hearing ? (grin)

[Miska]

Some time ago I created couple of test files with systematic and random jitter. Audibility of jitter vs level largely depends on type of the jitter (it's frequency characteristics).

 

And also on sampling rate, so putting some amount of jitter into 44.1k PCM file is different than putting same amount of jitter into 5.6 MHz SDM file (typical DAC chip runs at).

[Don Hills]

... Since jitter as we normally use the term here is the timing variation of the clock at the DAC chip, how can you "add jitter" to the data in a file? The best you can do is a simulation of what you think jitter does to the output signal. The description of the test files makes no attempt at explaining what was done. There are many different types of jitter, which was used in this jitter simulation test has not been specified.

 

Page 2 states the composition of the jitter signals. I accept that the author assumes a working knowledge of how jitter affects the audio signal. (Classic textbook phase modulation.) "The best you can do" is pretty good. The output waveform from the DAC reproducing the test files is identical to that produced by the equvalent amount of actual clock jitter. (But see my last paragraph below.)

 

Then in reality this simulation of what jitter does is played back on a system which has it's own builtin actual jitter which will cause a modulation of some form of the simulated jitter.

 

The assumption is that the system will have levels of jitter well below the amount added to the test files. In my opinion, someone who cares enough to do the test is likely to have a sufficiently competent DAC.

 

So while these sorts of tests are interesting, they do not necessarily tell you what real jitter (actually varying the timing of the clock at the DAC chip) actually sounds like. ...

 

The test demonstrates the effect of jitter on the sample clock for a classic NOS R2R type DAC. I agree that the same jitter for an oversampling D-S DAC *may* result in a different output.

[Miska]

The test demonstrates the effect of jitter on the sample clock for a classic NOS R2R type DAC. I agree that the same jitter for an oversampling D-S DAC *may* result in a different output.

 

You could also generate ones with 50/100 or 60/120 Hz waveform as PSU related modulation would create. And in addition combine it with white noise jitter.

 

Higher the sampling rate, more same level pulls the frequency.

[Don Hills]

You could also generate ones with 50/100 or 60/120 Hz waveform as PSU related modulation would create. And in addition combine it with white noise jitter.

 

That would be a common type of jitter. So might signal-correlated jitter. As John said, there are many sources of jitter, each with a different audible effect (if it's bad enough to be audible).

 

Higher the sampling rate, more same level pulls the frequency.

 

I'll have to think about that. My current understanding is that the same amount of jitter results in the same audio output spectrum regardless of the sample rate.

[JohnSwenson]

Page 2 states the composition of the jitter signals

 

Aha, I missed the test setup file. I like that way of doing it, I think that is a far better test than many of the others I've seen. It's still not varying the clock at the DAC, but varying the clock of the ADC is still a very interesting way to do it.

 

John S.

[Don Hills]

Aha, I missed the test setup file. I like that way of doing it, I think that is a far better test than many of the others I've seen. It's still not varying the clock at the DAC, but varying the clock of the ADC is still a very interesting way to do it.

 

John S.

 

And I, in turn, missed that setup PDF. Clever of him. I had assumed (you know what they say about "assume") that he'd done the jitter injection in software. The resulting analogue output will (theoretically) be the same regardless of where you inject the jitter - ADC clock, in software or DAC clock. I agree that the actual topology of the ADC or DAC can cause it to behave differently than theory predicts. This actually makes software injection the most accurate method if you're trying to determine thresholds of audibilty, because it avoids the implementation-dependant variables.

 

Some ten years ago, Arnie Kreuger did a test where he added jitter to an S/PDIF datastream and fed it to a couple of DACS. One was severely affected, one wasn't.

(Not a) good explanation of jitter in TAS - Hydrogenaudio Forums

 

I seem to recall seeing somewhere a thread where someone wrote a software jitter injector that allowed supplying a user-defined jitter signal. I'm still looking to see if I can find it again. Even if I can't, it'll be a useful coding project. Different jitter signals (noise, deterministic, signal-correlated) generate different artifacts, which become audible at different thresholds. It would make a useful training tool for people interested in identifying the audible effect of various jitter signals.

 

This all takes me back many years... I've been in telecommunications for over 40 years, and I well remember early jitter problems and the solutions to them - for example, modems with DCE clock, where the modem clocked the data out of the computer as it required it. Going further back, jitter on telegraph (teleprinter) circuits was handled by "asynchronous" transmission. There's nothing new under the sun.

[sandyk]

The resulting analogue output will (theoretically) be the same regardless of where you inject the jitter

 

However, in the posted examples the checksums are different.

[Don Hills]

However, in the posted examples the checksums are different.

 

Yes, and this is expected and valid.

Here's a greatly simplified example:

 

Take a smoothly rising signal (the slope of a triangle wave, for example.)

1--2--3--4--5--6--7--8--

 

For a perfect ADC, 3 sequential samples might be: 2----4----6

 

Now imagine jitter causes the second sample to be taken late, the 3 samples might be: 2------5--6

When reproducing them through a perfect DAC, the result is: 2----5----6

This puts a "kink" in the slope.

 

For a DAC with jitter causing the second sample to be output late:

Input 2 4 6

Output 2------4--6

Again, a kink in the slope.

 

Finally, you can simulate jitter by manipulating the samples.

Assuming 3 samples 2 4 6:

Add "jitter": you get 2 5 6 (or 2 3 6 etc).

Output this through a perfect DAC: you get 2----5----6 (or 2----3----6).

 

Hopefully this shows why the checksums are different.

 

But regardless of whether you apply the jitter at the ADC, in the digital samples or at the DAC, you get the same result when you examine the analogue output with a spectrum analyser (or your ears).

[sandyk]

In the real world the checksums are still the same with typical amounts of Jitter , although I know that you will refuse to accept that they can sound different while still having identical checksums. Perhaps in these samples they felt the need to greatly exaggerate the amount of Jitter so that the average person could hear it ? Has anybody tried something like say a PSU with a superimposed 100kHZ sawtooth on top of the DC output voltage, or even a SMPS with accentuated HF ripple ?

[Don Hills]

In the real world the checksums are still the same with typical amounts of Jitter , although I know that you will refuse to accept that they can sound different while still having identical checksums. Perhaps in these samples they felt the need to greatly exaggerate the amount of Jitter so that the average person could hear it ? Has anybody tried something like say a PSU with a superimposed 100kHZ sawtooth on top of the DC output voltage, or even a SMPS with accentuated HF ripple ?

 

You're getting mixed up between jitter on the bitstream into the DAC and jitter on the DAC clock. Jitter on the bitstream has no effect unless it's so bad that the receiver loses lock. It only matters if the jitter on the bitstream causes jitter on the DAC clock(*), as can happen with poor DAC design or implementation.

 

Since it's not practical for the average listener to introduce a known amount of jitter to their DAC clock to see what it sounds like, the jitter has to be introduced earlier - by jittering the ADC clock (as in the Crane Song examples) or by adding jitter mathematically to the digital samples. This will, of course, result in non-identical files. This does not invalidate the audible results.

 

Put simply:

As applied to digital audio, jitter is a variation in the time that an ADC samples an analogue input, or a DAC outputs an analogue value.

That jitter can be applied to the ADC clock, or the DAC clock, or mathematically added to the digital samples.

The analogue output from the DAC will be identical in all 3 cases.

 

Instead of wondering about the level of jitter in the Crane Song samples, perhaps you could read the first paragraph from the web page:

" ... Depending on the playback system and the level of ear training it may or may not be possible to hear the differences in the files. The level of jitter has been pushed a bit to help this experiment. ... "

 

As for the effect of PSU voltage variations, it will be highly implementation dependant, even with DACs using the same base chipset. A better PSU may greatly improve one DAC but have no effect on another. So while Alex and John's PSU may well improve a Mac Mini, it's not a universal panacea.

 

Note that "a superimposed 100kHZ sawtooth on top of the DC output voltage" is simply an instance of "accentuated HF ripple".

 

 

(*) As John Swenson has pointed out, in a poor design jitter or other noise on the input can also result in variations of the supply or reference voltages of the DAC chip, but we're getting well down the rabbit hole now.