Let's talk jitter.

[LHC]

On 31/10/2018 at 7:46 PM, LHC said:

https://www.jstage.jst.go.jp/article/ast/26/1/26_1_50/_pdf

 

Figure 1 in that Japanese research paper I posted earlier has an excellent diagram to show how jitter shifts the original waveform. When the shifted waveform is sampled it would have some extra sampled frequencies that was not in the original. It is a distortion alright. 

[davewantsmoore]

On 04/11/2018 at 3:23 AM, MLXXX said:

Plug-ins are available for the free audio editing software Audacity to create a constant smooth frequency deviation in the music (a plug-in called tremolo), or a random pitch deviation. These deviations are forms of jitter.

These error are NOT what happens to the output of a digital analogue converter due to jitter in the digital input signal.

On 04/11/2018 at 3:23 AM, MLXXX said:

Listening to the impact of these plug-ins on the quality of the sound could be interesting for experimenters!

If you are suggesting that listening to these, is what jitter will sound like ..... then you are very mistaken about what jitter is (in a digital to analogue converter).

[Ittaku]

4 minutes ago, LHC said:

 

Figure 1 in that Japanese research paper I posted earlier has an excellent diagram to show how jitter shifts the original waveform. When the shifted waveform is sampled it would have some extra sampled frequencies that was not in the original. It is a distortion alright. 

Oh there was no argument about whether it is distortion or not, simply that the scale used for describing said distortion shouldn't be compared to say THD.

[MLXXX]

13 hours ago, rmpfyf said:

 

I would tend to think that jitter changes the shape of a waveform.

To some extent. But then a slightly non-linear amplification characteristic introduces harmonic distortion and intermodulation distortion. And these change the shape of the waveform also, to some extent.

I have found it interesting to examine the effects on waveform shape of a psychoacoustic codec operating at a high enough bitrate to sound practically transparent  (e.g. the Advanced Audio Codec). If you try to align the analogue version of the music after it has been compressed and decompressed using the codec, with the original waveform, you may have difficulty!  There are differences all over the place!  Of course a design goal of a psychoacoustic codec operating at a high bitrate is to omit waveform information that is very unlikely to be perceived by human ears. 

[zenelectro]

1 hour ago, LHC said:

 

Given this thread is a spin-off from another fiercely argued debate, I got to ask you how did you managed your expectation bias when testing?

Give me a break.... LoL 

 

I have been doing this for years and have learned the hard way not to have any expectations.

This damn pursuit has slapped me in the face with so many failures that were hopeful improvements the improvements are often completely unexpected. 

I don't know what else to tell you.

[davewantsmoore]

On 03/11/2018 at 12:36 AM, LHC said:

where are the published DBT results to what forms of jitter are audible

There is a problem with this ..... the resulting distortion from different types or amounts of jitter isn't easy to predict.    So if we said what does "blah" jitter sound like .... it isn't necessarily a question which will have a consistent result.

 

In general avoiding jitter is a good thing... less jitter is better .... but "what does is sound like" can only be really analysed by focusing on the effect (the non-linear distortion) rather than the cause (some amount of inconsistency in the digital input signal)

[davewantsmoore]

14 hours ago, rmpfyf said:

I would tend to think that jitter changes the shape of a waveform.

It 100% does.   That's what we've been saying all along.   Jitter is non-linear distortion.   It adds new frequency components to the output signal  (in the same way that "THD" and "IMD" do) .... ie.  that the amplitude vs time view (ie. the "waveform") is a different shape.

 

As mentioned - it does not (as MLXXX implies) create a "timing error" in the analogue output signal.    The original frequency components occur at the correct time.   The waveform is a different shape because there are also new frequencies.

 

14 hours ago, rmpfyf said:

I like my music NOS. Does this increase a propensity to jitter having an audible effect? Probably! 

Why would it?

 

[LHC]

3 minutes ago, zenelectro said:

Give me a break.... LoL 

 

I have been doing this for years and have learned the hard way not to have any expectations.

This damn pursuit has slapped me in the face with so many failures that were hopeful improvements the improvements are often completely unexpected. 

I don't know what else to tell you.

 

That is ok, I fully accept your explanation.

 

It was just that this thread originated from another debate about DBT and the like, and you know how these things go. It felt like a duty to pose the question (someone has to do it I suppose), but it wasn't meant to be a personal challenge. 

[davewantsmoore]

44 minutes ago, davewantsmoore said:

These error are NOT what happens to the output of a digital analogue converter due to jitter in the digital input signal.
 

If you are suggesting that listening to these, is what jitter will sound like ..... then you are very mistaken about what jitter is (in a digital to analogue converter).

 

My comments might benefit from some clarification.

 

The original frequency components are not "shifted in time".    New frequency components are added (just like in THD/IMD for example) ... and that make the waveform a different shape.

 

Using a filter to move the timing of the original frequency components as suggested, will not at all produce the same effect as jitter.

 

 

 

 

[Guest rmpfyf]

45 minutes ago, davewantsmoore said:

It 100% does.   That's what we've been saying all along.   Jitter is non-linear distortion.   It adds new frequency components to the output signal  (in the same way that "THD" and "IMD" do) .... ie.  that the amplitude vs time view (ie. the "waveform") is a different shape.

 

As mentioned - it does not (as MLXXX implies) create a "timing error" in the analogue output signal.    The original frequency components occur at the correct time.   The waveform is a different shape because there are also new frequencies.

 

Jitter is not additive - it’s a small but important difference.

 

There may be slight changes in frequency as waveform shapes are changed, but there are no new frequencies added; there may be new spectral energy of low amplitude though that’s a signal processing artefact, not an evidence of a new noise source.

[MLXXX]

 

1 hour ago, davewantsmoore said:

These error are NOT what happens to the output of a digital analogue converter due to jitter in the digital input signal.

 

The plug-ins obviously can only go so far in modulating the signal in a way that will mimic the effects of different types of real world jitter.

 

I see on review that the built-in tremolo plug-in I had referred to does not offer the frequency/phase modulation I was looking for. I have now located a vibrato plug-in, and have amended my earlier post.  (I've found it's easier to use a recent version of Audacity, and its built-in effects manager functionality at the effects tab.)

 

The Vibrato plug-in could be set to mimic the effect of mains ripple (50Hz, or more likely 100Hz) modulating the clock speed of a clock used to control an ADC.   Below is a screen capture of Audacity version 2.30 (on a Windows 10 platform) with the available parameters for this effect displayed:

 

 

[MLXXX]

56 minutes ago, davewantsmoore said:

As mentioned - it does not (as MLXXX implies) create a "timing error" in the analogue output signal.    The original frequency components occur at the correct time.   The waveform is a different shape because there are also new frequencies.

It appears you may not have considered a wobble that occurs slowly, over many clock cycles, such as a clock speed variation that varies in sympathy with the peaks and troughs of 100Hz mains ripple.

 

If the ADC clock speed wobbled at a slow rate during the recording process then playing back the digital recording will yield an analogue waveform that slips backwards in forwards in time  relative to the original analogue waveform.

 

Conversely a DAC using a clock whose speed varied in accordance with 100Hz mains ripple would subtly shift the timing of the entire reconstructed analogue waveform backwards and forwards in time, during the playback of a recording made at a constant clock speed. 

[Ittaku]

1 hour ago, LHC said:

It was just that this thread originated from another debate about DBT and the like, and you know how these things go. It felt like a duty to pose the question (someone has to do it I suppose), but it wasn't meant to be a personal challenge. 

In my experience, I would trust a non DBT sighted listener who has described incidences of changes in their system that they said made no audible differences...

[MLXXX]

2 hours ago, davewantsmoore said:

This is not the same as jitter in a digital to analogue converter.

 

In your example, the "jitter" is in the analogue system  (ie. the timing of the movement) .... in a DAC the jitter is the timing of the digital signal....  but the result is not an error in the timing of the analogue output.   The result is that there are additional frequency components added to the analogue output.

 

 

I'm sorry if people already realised that .... but if they didn't, then it is very important to note/appreciate.

 

I've only just seen this post of yours, @davewantsmoore. Please see my post immediately above that explains how a slow wobble in clock rate of either the ADC or the DAC will shift the timing of the reconstructed analogue waveform back and forth when the digital file is played, compared with the timing in the original analogue waveform captured with the ADC.
 

[davewantsmoore]

53 minutes ago, rmpfyf said:

There may be slight changes in frequency as waveform shapes are changed, but there are no new frequencies added

Perhaps I have misunderstood what you mean ...... Let's be very clear.

 

If the output signal has a changed shape from the input signal ... this means that there ARE new frequency components present in the output which were not present in the input.

 

You cannot have a waveform which changes shape which does not have new frequency components present in it.

 

59 minutes ago, rmpfyf said:

There may be slight changes in frequency as waveform shapes are changed, but there are no new frequencies added; there may be new spectral energy of low amplitude though that’s a signal processing artefact, not an evidence of a new noise source.

You may need to expand on this....

 

If I compare the output signal from a converter which has some jitter ... and the same converter with more jitter .... the non-linear distortion components (ie. the new frequencies, aka "new spectral energy") are louder in the later version.    When the jitter is extreme, these distortion components could be quite loud.

 

You appear to be saying that these new frequency components don't exist ... and that they are a "signal processing artefact".   This doesn't seem at all right.   What do you mean?!

 

1 hour ago, MLXXX said:

The plug-ins obviously can only go so far in modulating the signal in a way that will mimic the effects of different types of real world jitter.

No.   The plug-ins do a completely different thing.   It's not like they "kinda go close".   They're completely different.

 

Let's say I had a signal which is a 5khz sine.     Your plug-in will change the 5khz so it is a different frequency.    The 5khz is gone, and in it's place is 5004Hz (or whatever) sine.

 

That's not what jitter does!   For our 5khz sine wave... when the digital signal is jittered ... and converted back to analogue.... what we get is the exact same 5khz sine .... and we get some new waves which are (usually much) lower in level at other frequencies.

 

 

The view you are putting forward of jitter being like "doppler distortion" is correct.... but it is the digital signal which is distorted like that.     The analogue signal which comes out of the DA converter has no such "doppler like distortion".    It has new frequency components.    The original frequency components are correctly "in time".

 

 

 

 

 

 

 

 

 

[davewantsmoore]

53 minutes ago, MLXXX said:

If the ADC clock speed wobbled at a slow rate during the recording process then playing back the digital recording will yield an analogue waveform that slips backwards in forwards in time  relative to the original analogue waveform.

Sure.

 

Sorry - I was only focusing on the performance of a DAC .... ie.   assuming that what was encoded into the digital signal is exactly what we wanted.

[LHC]

48 minutes ago, Ittaku said:

In my experience, I would trust a non DBT sighted listener who has described incidences of changes in their system that they said made no audible differences...

 

Sure, that is fair and reasonable. Again, I was just posting the question; I have no expectation of how people may reply. 

[Ittaku]

5 minutes ago, LHC said:

Sure, that is fair and reasonable. Again, I was just posting the question; I have no expectation of how people may reply. 

I understand, I was just pointing out the opposite indirectly - that I don't trust those who hear a difference no matter what they change in their system.

[Guest Eggcup The Daft]

3 minutes ago, Ittaku said:

I understand, I was just pointing out the opposite indirectly - that I don't trust those who hear a difference no matter what they change in their system.

It depends on the meaning of "trust". For me, if someone hears a difference in their system after any change, I believe they hear that change. It's real to them. I've been there.

 

However, I don't immediately believe that the output of their system has changed, and that makes many of the changes reported by people unlikely to be heard by all listeners who make the same change. And if the output of their system has changed, even then the change may not be applicable to all systems or even all similar systems.

 

I don't believe that if someone doesn't hear a change, either, that no change has occurred in the system that others may hear. I've been there as well.

[MLXXX]

2 minutes ago, davewantsmoore said:

No.   The plug-ins do a completely different thing.   It's not like they "kinda go close".   They're completely different.

 

Let's say I had a signal which is a 5khz sine.     Your plug-in will change the 5khz so it is a different frequency.    The 5khz is gone, and in it's place is 5004Hz (or whatever) sine.

 

That's not what jitter does!   For our 5khz sine wave... when the digital signal is jittered ... and converted back to analogue.... what we get is the exact same 5khz sine .... and we get some new waves which are (usually much) lower in level at other frequencies.

 

We appear to be discussing the nature of frequency modulation.

 

If you apply a 100Hz sine wave voltage to a variable capacitance diode that is part of an RC [Resistor Capacitor] network determining the frequency of a free-running sine wave oscillator, with centre frequency 5kHz, the oscillator frequency will always be on the move, never settling on a steady state rate of oscillation. In a certain sense it is never at any particular frequency! (Perhaps it is closest to achieving a steady frequency at the positive and negative peaks [turning points] of the modulating sine wave.)

 

Similarly, if you sample a steady analogue 5kHz sine wave at 96kHz but wobble the sampling frequency at 100Hz and you later on play back the 96KHz file, the reconstructed analogue waveform you get will be a sinewave  that wobbles in frequency, never settling on any particular value, although centred on 5KHz. On a spectrum analyser (that relies on a Fourier analysis) you will see various components displayed, despite the fact that the reconstructed waveform never settles on any particular value, and certainly doesn't "settle" on 5kHz. [It passes through 5kHz very rapidly!]

 

I might leave it that.

[Ittaku]

2 minutes ago, Eggcup The Daft said:

It depends on the meaning of "trust". For me, if someone hears a difference in their system after any change, I believe they hear that change. It's real to them. I've been there.

 

However, I don't immediately believe that the output of their system has changed, and that makes many of the changes reported by people unlikely to be heard by all listeners who make the same change. And if the output of their system has changed, even then the change may not be applicable to all systems or even all similar systems.

 

I don't believe that if someone doesn't hear a change, either, that no change has occurred in the system that others may hear. I've been there as well.

Sure. What I mean is that if a person has things they don't hear a change with some things, then I'm more likely to trust them when they say they actually do hear a change. It's not an absolute, but more a guiding principle for how much faith I trust in peoples' own impressions without me listening, or without a DBT.

[MLXXX]

31 minutes ago, davewantsmoore said:

Sure.

 

Sorry - I was only focusing on the performance of a DAC .... ie.   assuming that what was encoded into the digital signal is exactly what we wanted.

But the converse does apply, as I see it. It doesn't  matter whether the clockrate slow wobble is in the ADC clockrate at the time of the recording of the digital file, or in the DAC clockrate at the time of playback of the digital file.  In both cases the reconstructed analogue waveform gets shoved backwards and forwards in time during playback, compared with the original analogue waveform.

A lot of this discussion of (extremely?) minor effects is hypothetical of course unless we can establish that human listeners can actually detect an audible difference if the particular type of effect is present above a certain level, and that that effect -- at that level or above -- might reasonably be expected to be encountered in a real life audio device.

[Guest Eggcup The Daft]

11 minutes ago, MLXXX said:

Similarly, if you sample a steady analogue 5kHz sine wave at 96kHz but wobble the sampling frequency at 100Hz and you later on play back the 96KHz file, the reconstructed analogue waveform you get will be a sinewave  that wobbles in frequency, never settling on any particular value, although centred on 5KHz. On a spectrum analyser (that relies on a Fourier analysis) you will see various components displayed, despite the fact that the reconstructed waveform never settles on any particular value, and certainly doesn't "settle" on 5kHz. [It passes through 5kHz very rapidly!]

Sorry to be dumb here, but  surely, if you wobble the sampled frequency (as it's sampling a point in time)  is still at 5kHz? You said, the wave is steady. I can't see how you then sample a different frequency?

 

I'd expect the playback to deviate from a pure sine wave, as the wobble will cause the ADC to sample the waveform at a slightly different point in time from the intended one.

[MLXXX]

2 hours ago, Eggcup The Daft said:

Sorry to be dumb here, but  surely, if you wobble the sampled frequency (as it's sampling a point in time)  is still at 5kHz? You said, the wave is steady. I can't see how you then sample a different frequency?

 

I'd expect the playback to deviate from a pure sine wave, as the wobble will cause the ADC to sample the waveform at a slightly different point in time from the intended one.

I'll use an example.

 

Assume a fragment of a linear PCM audio file (say, part an original file with the extension ".wav") is found that has no header information to tell you whether it was recorded at a sample rate of 44.1KHz or a sample rate of 48.0kHz. In repairing the file in an audio editor program you might decide to add a header that indicates a sample rate of 44.1Khz, and then settle back in your armchair and listen to the audio played back on your media player.  If the playback seems to be at the right speed and pitch you may feel pleased that you have not only recovered part of a recording but you have correctly guessed the sample rate used to record it! On the other hand if it sounds a bit slow, you might go back and change the header to 48KHz. And then try that on your media player. Voilà! It now sounds right!

 

A media player will play back a PCM digital recording using the specified sample rate. If the specified sample rate is wrong, the playback will either be too fast and too high pitched, or too slow and too low pitched.

 

So the digital sample rates of the recording and the playback need to match each other for accurate reproduction. A slow wobble in either of them will result in speed and pitch anomalies in the playback.

Edited November 5, 2018 by MLXXX

[Guest Eggcup The Daft]

20 minutes ago, MLXXX said:

I'll use an example.

 

Assume a fragment of a PCM audio file is found that has no header information to tell you whether it was recorded at a sample rate 44.1KHz or a sample rate of 48.0kHz. In repairing the file in an audio editor program you might decide to add a header that indicates a sample rate of 44.1Khz, and then settle back in your armchair and listen to the audio played back on your media player.  If the playback seems to be at the right speed and pitch you may feel pleased that you have not only recovered part of a recording but you have correctly guessed the sample rate used to record it! On the other hand if it sounds a bit slow, you might go back and change the header to 48KHz. And then try that on your media player. Voilà! It now sound right!

 

A media player will play back a PCM digital recording using the specified sample rate. If the specified sample rate is wrong, the playback will either be too fast and too high pitched, or too slow and too low pitched.

 

So the digital sample rates of the recording and the playback need to match each other for accurate reproduction. A slow wobble in either of them will result in speed and pitch anomalies in the playback.

I don't believe this example is anything like the same as shifting individual samples of a fixed frequency. We're dealing with jitter in the ADC case described before - so "noise and artefacts" (or what looks like noise and artefacts, anyway) alongside a distorted 5kHz waveform is what is expected. Just because the shift is periodic doesn't change that.

Edited November 5, 2018 by Eggcup The Daft