Why digital seems to be affected by power and cables

[petersv]

Add a reclocker to the output and you're golden, dac also needs to have an accurate clock to reconstruct the analogue wave form and preferably a decent class a output stage to minimize crossover / switching distortions

[rastus]

For an interview with (Prof) Malcolm Hawkesford with a lot about the audible effects of jitter, including how DAC power supplies and cables impact, have a read of http://linearaudio.nl/sites/linearaudio.net/files/Didden-Hawksford_Part1.pdf

 

Geeze, it's 21 pages long! Read all of it? Can't I just skip to the good part?

[davewantsmoore]

Human hearing is also extremely sensitive to timing errors, hence why master clocks and their accuracy are so vital to high end performance (in sub picoseconds - i.e. Femto clocks).

 

Ahh, this is a common subtle misunderstanding.... 

 

 

It isn't the "time difference" we hear.    Yes, we are sensitive to small timing differences....   but that sensitivity is NOT anything to do with what causes us to "hear jitter".

 

 

In a digital converter, the small timing differences result in added frequency compnennts to the signal.

This picture really makes it look like the "signal is badly timed":  http://www.apogeedigital.com/knowledgebase/wp-content/uploads/jitter1-266x300.png

 

 .....  but when you graph this same data differently you get this:   http://www.computeraudiophile.com/f8-general-forum/distortion-time-domain-digital-sound-signal-human-hearing-very-sensitive-jitter-vs-no-distortion-time-domain-digital-sound-signal-human-hearing-very-sensitive-jitter-26376/#post479921

 

See, that for jitter vs no-jitter.     Jitter is actually added frequency components to the signal.    In the chart Apogee posted ....  the slope of the line is changed from what it should be.   The changed slope, represents new frequency components (which is what we hear).

 

This might sound all technical and pedantic.....    to put it simply, it isn't the jerking in TIME that we hear  (like, as an extreme example -  if I spun a record at a varying speed) ...... it is the jerking in frequency (ie. additional unwated frequencies come out the other end).

 

 

PS - Just in case anyone is thinking it ......  I'm not saying here "Apogee are wrong".    Just that in trying to explain it in simple terms, they may inadvertently reinforce a misunderstanding.

 

Aside from the inherent linearity of the converter (with perfect input data) and the linearity analogue output circuit  ..... then jitter performance is the ONLY thing which matters in a converter.    So in a practical day-to-day perspecitve, jitter isn't "a" thing.   It is "the" thing.    :-)

[davewantsmoore]

OK Peter, I did read that link... so jitter is uneven distances between the transitions ("Notice that the distance between the transitions is uneven – this is jitter.")

 

Yes.

 

 

Obut with music 'words' there is no error checking? 

 

There is error checking, but no error correction.

 

 

So with music this causes "uneven intervals", is this correct? And this "uneven intervals" and "unevenness" causes "distorsion into the waveform". 

 

Yes.....     but it isn't the unevenness in time which is audible, like you would imagine.    It seems very intuitive that if I "made the signal jerky in time", that it would sound jerky like someone was speeding up and slowing down the music really quickly....  you can "imagine" what that would sound like.

 

 

When the red line in the apogee charts, doesn't go where it's supposed to ...... that represents added frequency components  (ie. what people common think of as "distortion").

 

 

How do we remove the jitter, if we haven't got a 'better' DAC?

 

By providing a new clock to the digital signal.

 

So the converter receives all the data ..... but instead of just using it right away, and suffering from the uneven arrival times .... it stores the data.      The it send the data into the DA converter, but it sends each pulse using a new clock reference, that has less uneven timing.

 

Reclocking the signal doesn't have to be necessarily done by another box inbetweeen your digital source, and digital destination...... many/most DACs already reclock the signal when it arrives  (otherwise they could not guarantee any level of performance - as it would depend on the incoming jitter)

 

Can't I just skip to the good part?

 

This stuff which has come up just now  (jitter).   Can be simple on the surface, but easy to misunderstand using "seems reasonable" type concepts.    The reason you will see it discussed (sometimes heatedly) online, is because many many people don't understand it well.    (like climate science, or economics, quantum physics, or....)

Edited May 6, 2016 by davewantsmoore

[rastus]

Add a reclocker to the output and you're golden, dac also needs to have an accurate clock to reconstruct the analogue wave form and preferably a decent class a output stage to minimize crossover / switching distortions

 

Like the "Remedy Reclocker" (https://wyred4sound.com/products/digital-converters/remedy-reclocker)? My wife will divorce me if I add another box thing, seriously... it would have to be a stealth operation. LoL! My PiFi DAC+ has RCA outputs, this Remedy Reclocker only has optical and coax inputs. And I looked at the whole range, https://wyred4sound.com/products/digital-converters, nothing for me. And man, the price.... geeze! I have read that these Remedy Reclocker sound much better with a LPSU. I had read up on the Remedy Reclocker before.

The "dac also needs to have an accurate clock to reconstruct the analogue wave form", so does the PiFi DAC+ have an accurate clock? I assume my Yammie RX-V2065 may have an accurate clock built in... in case I go with the DIGI hat (has optical and coax outputs, but is not a DAC in itself), and thus use the DAC built into my Yammie.

Who-ha, "a decent class a output stage", you got me there. Beg-ya-paron?

And, "minimize crossover / switching distortions", while interesting, does not make any sense to me what-so-ever.

Edited May 6, 2016 by rastus

[rastus]

Ahh, this is a common subtle misunderstanding.... 

 

I tend to agree with you there.... my brain 'compensates', least I thought it did.

[davewantsmoore]

Sorry, I probably don't do a fab job at explaining this..... it's a tough balance between over simplfying .... and talking about things which require too much detail to understand properly.

 

 

So much easier with a whiteboard and a 6 pack.

[rastus]

It isn't the "time difference" we hear.    Yes, we are sensitive to small timing differences....   but that sensitivity is NOT anything to do with what causes us to "hear jitter".

 

...it isn't the jerking in TIME that we hear  (like, as an extreme example -  if I spun a record at a varying speed) ...... it is the jerking in frequency (ie. additional unwated frequencies come out the other end).

 

So in a practical day-to-day perspecitve, jitter isn't "a" thing.   It is "the" thing.    :-)

 

So it's not a timing variation, but rather a frequency variation that causes some kind of reproduced sound distortion? Although the frequency variation is caused by a timing variation?

 

Oh OK, it is "THE" thing, meaning? The thing when considering a new purchase? When researching a device?

[rastus]

1. There is error checking, but no error correction.

 

2. When the red line in the apogee charts, doesn't go where it's supposed to ...... that represents added frequency components  (ie. what people common think of as "distortion").

 

3. By providing a new clock to the digital signal.

 

1. Because it's too late, the music data has already arrived and been delivered to the amp/speakers? So there is not any form of error compensation? Like adding a bit or a byte here and there to compensate? But, there is a 'loss' of sound signal? Not really, it's just a variation of the signal (when reproducing music, if the source is good). So is jitter peceivable? Or is jitter perceived as a 'messy' sound? Then we blame the amp, the speakers, the room, etc?

2. I would've assumed the result would be the removal of frequency components... or is this a result of an attenot by the DAC circuity to some how perform some type of error compensation? ie, adding to the signal.

3. Or a better clock, this is what I've read about the Raspberry Pi elsewhere. Damn subject keeps coming up, just like a flower.

Edited May 6, 2016 by rastus

[rastus]

So much easier with a whiteboard and a 6 pack.

 

I know where your coming from there brother, 7 years teaching Computer Network Engineering.... a whiteboard comes in handy! As for the 6pack, doubt that would help me understand, I go ga-ga!

 

Edit: HAY! I caught up to yas! So this is about the clocks at both ends of the signal, the sending device AND the receiving device? Both clocks must be accurate? And, the question begs an answer.... if I can't perceive jitter then I can ignore it, but if I can perceive jitter, then it needs to be delt with.

 

Maybe this is the gnawing 'feeling' I have inside about the sound quality of my system... maybe it's my brain on some deep psychological level trying to tell me, "It don't sound 'right'! Fix it!" I really thought that was just my OCD, really.

Edited May 6, 2016 by rastus

[davewantsmoore]

Although the frequency variation is caused by a timing variation?

 

Yes.   Frequency = cycles / second

[davewantsmoore]

Oh OK, it is "THE" thing, meaning? The thing when considering a new purchase? When researching a device?

 

Yes.     We can characterise the performance of a DAC in two very broad ways.

 

1) It's performance when it has "perfect" input data

2) It's performance when it has "worse" input data

 

 

ie.   #2 is the presence of jitter in the data .....  OR the presence of electrical noise / interference (like what could be caused by less than adequate cables or power supplies) ... which  manifests itself as jitter.

 

ie.  some unwanted electrical noise or pulse gets into the DA converter....   this pulse gets mixed with the input signal somewhere  (the square wave in the Apogee pictures) ..... this distortion in the input data square wave, looks like sudden "jitter" to the DA converter.    The jitter is usually very large, very irregular,  and on the time-scales of a digital signal it usually "comes and goes"....  ie.  there'll be a big spike, then nothing for many nanoseconds.... then another big spike.    Many DA converters do not deal with this well.    Converters used for professional applications usually pay more attention to this, as they need to keep their level of performance constant, even when faced with a "dirty environment".

[rastus]

Yes, the frequency of a CPU is measured in cycles per millisecond. Oh, I see, hang on.... I got to read that again, let it 'sink in'.

 

 

Edit-1:

 

So it's 'garbage' in the signal that results in a type of distortion. And it's periodic. Why? Is there some kind of 'build up' and then a type of release? Or is the DA converter off-loading what it can no longer deal with, or could never deal with?

"like what could be caused by less than adequate cables or power supplies" <---- This all adds up to money, you know that right? Including a high performing DAC, more money.

The cause is "the presence of electrical noise / interference", the outputted result is jitter in the form of distortion (or garbage)?

 

 

Edit-2:

 

So the frequency variation is a result of a timing inaccuracy, and the timing inaccuracy would be caused by either badly produced signal quality in the first place, or a cheap and inadequate DA converter?

Edited May 6, 2016 by rastus

[davewantsmoore]

1. Because it's too late, the music data has already arrived and been delivered to the amp/speakers?

 

In short yes.    The analogue output is a constant signal .... so there is "no time" available to make a correction.

 

UNLESS... the digital device stores the incoming data .... and then "later" .... sends the data on.   (which is in short how reclocking works) .... although it quickly becomes complex to discuss  

 

What many people seem to not appreciate (because the evidence is complicated, I guess) is how robust digital transmission is.    It's a very good question to ask  "so what about all the data which gets transferred wrong?"  (if it's not corrected then that doesn't seem very good).      The reality is that the data is all transferred correctly, even when things get quite bad.

 

The data is just a 1 or a 0, and so it's very easy to get right  (it will still work out that 0.6 / 0.2 / 0.7 / 0.01 / 0.9 ..... is 1/0/1/0/1) .... but the timing between each syllable has infinite possibilities of precision  (and so it is never perfect)

 

 

1. So is jitter peceivable? Or is jitter perceived as a 'messy' sound?

 

Yes.   

 

Jitter (see the charts I posted earlier jitter vs no jitter) is ends up as non-distortion  (added frequencies which aren't supposed to be there).

 

This can sound like

 

Raised noise floor

• Generally, a dead or flat sound
• Muddy or even missing bass (as you start with much less signal to noise ratio in the bass)

 

Actual audible distortion

• If the added frequencies are so high in level that they are themselves specifically audible, it can be a harsh, dissonant tone to the audio

 

People should be careful hearing these qualities and concluding they have jitter  (you may, but be careful).    I can go into the studio and record something, and the recording actually sounds dead / flat / harsh / muddy, etc..     It's a trap sometimes to think that you want your favourite music to "sound better", when really it might actually be supposed to sound the way it does    ..... and yes, that can even apply to a single virtuoso violinist with a million dollar instrument recorded direct to DSD in an 400 year old abandoned church.    

 

 

 

2. I would've assumed the result would be the removal of frequency components... or is this a result of an attenot by the DAC circuity to some how perform some type of error compensation? ie, adding to the signal.

 

Nope.   It is just simply the fact that in this picture:   http://www.apogeedigital.com/knowledgebase/wp-content/uploads/jitter1-266x300.png

 

That when the slope of the line changes (from where is was supposed to be) .... what that represents is that the signal contains a different blend of frequency components to what it was supposed to.

 

 

A simplified example of that is these two pictures here:    http://www.computeraudiophile.com/f8-general-forum/distortion-time-domain-digital-sound-signal-human-hearing-very-sensitive-jitter-vs-no-distortion-time-domain-digital-sound-signal-human-hearing-very-sensitive-jitter-26376/#post479921

 

The signal is supposed to be just one frequency  (the chart with just the single big spike - ie. one frequency) ..... but the result is that there are many frequencies  (we have extra unwanted spikes)

 

 

3. Or a better clock, this is what I've read about the Raspberry Pi elsewhere. Damn subject keeps coming up, just like a flower.

 

Same concept.

 

If the DAC were going to rely on the timing embedded into the data .... then we would want the source clock to be good  (and the transmission medium, eg. the cables, etc.....  to be non-distorting)

If the DAC were going to provide a new clock signal of it's own .... then we would also want that clock to be good.

 

 

The clock timing is important .... but potentially not as important as external interference like from poor power supplies and cables.    It's a little like saying that:   my car vibrates a little bit all the time when I drive it .... but if there is an earthquake when I'm driving.... then?  

 

teaching

 

Yep.  Someone might understand 90% of the concept, and then just need one bit spun back over 10 times before it clicks.    Especially with a complex topic like this.

 

 

2. I would've assumed the result would be the removal of frequency components... or is this a result of an attenot by the DAC circuity to some how perform some type of error compensation? ie, adding to the signal.

3. Or a better clock, this is what I've read about the Raspberry Pi elsewhere. Damn subject keeps coming up, just like a flower.

 

[petersv]

@@rastus , indeed; ignorance is bliss. Essentially jitter inhibits an emotional connection and engagement to the music (which is why we're all in this right?) it doesn't immediately present like obvious frequency distortions etc

[rastus]

"UNLESS... the digital device stores the incoming data .... and then "later" .... sends the data on.   (which is in short how reclocking works) .... although it quickly becomes complex to discuss"

Buffering. So reclockers slow down the signal, buffers, and then spits it out at the correct rate/frequency/timing. Believe me, I have no wish to discuss it right now, thanks, maybe later.

I understand about data, and data transfer (network engineer), but with networks timing doesn't really become an issue (for the most part) because buffering is used, and checking and resending. Personal note: there was someone on SNA talking about FIFO buffering that may be an answer to the jitter and the Pi DACs. Must find his post.

Revisited the graphs: it's sporadic and right across the frequency range. ahhh, "added frequencies which aren't supposed to be there".

OK, raised noise floor and/or actual audible distortion.... requiring turning up the center speaker gain to hear it, not bright and alive sound, muddy bass resulting in turning the bass up and then affecting the center speaker even more, harsh, dissonant tone.... so many other factors can cause/affect the sound of a system resulting in these issues...

The chart shows more than one frequency being affected. I see now, it was only one frequency supposed to be displayed.

Re: poor power supplies: That's the only reason I signed up to this forum, to get a good quality low noise local made power supply for my Pi (Geiseler). Now I find myself 'roped' into this discussion, and other discussions on SNA, and I'm really enjoying that... but still haven't pressed the button on the power supply! Geeze, sorry Clayton! Get to it soon brother.

 

PS: @@davewantsmoore > Well your explanations were rather well explained and explanatory. Your ability to answer the many questions I've asked has been most helpful (depending how one looks at it).

 

PSS: @@petersv > "indeed; ignorance is bliss." <----- Never a truer word, but then I' had an inkling something might be amiss, I just put it down to OCD, now I have my 'justification'... won't help with the missus though. Exactly... we are all hear for very similar reasons, if not the same, a love of fine tunes (and how to reproduce them fine tunes to a finer level). I trust my intuition: if it don't feel right, then something may be amiss. Very helpful response bro!

Edited May 6, 2016 by rastus

[davewantsmoore]

I understand about data, and data transfer (network engineer), but with networks timing doesn't really become an issue (for the most part) because buffering is used, and checking and resending.

 

Yes.    In a computer network (my profession also), the system is digital to digital.   As long as the correct 1s and 0s come through, it doesn't matter what the timing between each bit is.....  and of course, if the 1 or 0 was wrong somehow.  The error detection methods know, and ask for the data to be sent again.

 

 

In a DAC ....  it's digital to analogue (and the timing is critical).      Or it's digital to digital, then pause for a sec inside the DAC and organise a new clock, and then digital to analogue (using that new clock)

 

In DAC .....  if the 0 or 1 was wrong somehow, then the error detection system knows .....   but there is no mechanism to ask the source to resend (no bidirectional communication).   All the DAC can do, is mute the output  (which is exactly what the error detect is for - to stop bad noises coming out the speakers if the input signal was corrupt).

 

 

 

 

Revisited the graphs: it's sporadic and right across the frequency range. ahhh, "added frequencies which aren't supposed to be there".....   The chart shows more than one frequency being affected. I see now, it was only one frequency supposed to be displayed.

 

 

 

Remember what you are looking at is a super simple example.   One single frequency  (and what the jitter looked like) ..... if it was music, there's many more frequencies.....  and so for the jittered result, there would be WAY MORE extra spikes ...... so much more that it just looks like "the noise floor".    When is why jitter typically sounds like a compromised signal to noise ratio.

 

See how in this picture, the noise floor of the Rpi, is 100 times higher than the noise floor of the CD player.

https://hifiduino.files.wordpress.com/2014/11/i2sphasenoise-2.png?w=595&h=451

 

 

 (depending how one looks at it).

 

Heh  

[CraigS]

Geeze, it's 21 pages long! Read all of it? Can't I just skip to the good part?

 

Well yeah - do what I did and search for the term "jitter" and just read those bits.  Although there is quite a lot on the subject!

[rastus]

owha, wait up a bit... so... I did have my PiFi DIGI+ hat connected directly to my Raspberry Pi (40 pin connector), the DIGI has both coax or optical (I tried both), then through the cables (coax or optical) to my Yammie AVR (RX-V2065) with built in DAC chip (TI Burr-Brown PCM1680DBQR) out to my speakers. Then, some time back, I read about jitter (didn't really get it, so ignored it).... and then I got cold feet and decided to replace the DIGI with a PiFi DAC+ hat (TI Burr-Brown PCM5122) because it was a later DAC chip, sounded better than the Yammie's built in chip too (to me).

So in the first setup, if there was any jitter at all, it would have been between the DIGI and the Yammie AVR, via coax or optical (both sounded the same to me)... or would it have been introduced before, say bu the Pi itself?

But in the second setup, the Digital to Analogue conversion is taking place in the PiFi DAC+ hat, that is inserted firmly on the Pi itself, then that is sent out via RCA connectors to the AVR, bypassing the DAC chip in the Yammie. So, any introduced jitter would be taking place between the Pi and the PiFi DAC+ hat via the I2S connector.

I'm not saying this is a distance thing, that because the I2S bus is closer than a long cable that it won't have jitter issue, I'm saying I know coax and optical can have jitter, so what about I2S? I suspect that RCA cables don't experience jitter either, am I correct there? Electrical impulses rather than data packets (words).

 

Edit: Hang on, I think you mention some of this already.... reading....

Edited May 6, 2016 by rastus

[CraigS]

I know the head engineer at Naim very well (I recruited him from University 25 years ago into Wharfedale, when I was CTO there).  His signature design is the Naim Statement (google it), for which you need house mortgage deep pockets to buy.

 

Last time we talked was at the Bristol Audio show, and he was describing the internal reclocking in the DAC he had designed and had just launched.  Basically there were several very accurate, low jitter clocks, closely spaced in frequency, and a FIFO.  The input data had its clock roughly restored with a PLL, and that was used to clock the data into the FIFO.  Then the accurate clock was chosen which kept the stack pointer of the FIFO at about half full.  If it was emptying, a slower accurate clock was chosen, and if it was getting too full a faster accurate clock was chosen.  The switching between accurate clocks was inaudible - it was the ultra-low jitter of these that was the important thing. I forget what the type number for the product was, but most reviewers did not really want to send the review sample back.

[petersv]

I2S isn't affected by jitter like SPDIF is.

[davewantsmoore]

So in the first setup, if there was any jitter at all, it would have been between the DIGI and the Yammie AVR, via coax or optical (both sounded the same to me)... or would it have been introduced before, say bu the Pi itself?

 

• The PI created the digital signal, which contained some jitter (due to the clock in the PI and other things in the PI)
• ... the cable, distorts the signal  (that distortion looks like jitter the receiver)
• The AVR receives the signal  (it's circuitry might also introduce some jitter here) .....
• Then the AVR has a choice (depends how it's designed).   It can either just accept the digital data (and the timing), and put that straight into its DA converter.   That's not usually common, as a very jittery intput would make it perform badly.
• What it typically does, is apply a new clock to the data ..... this reduces the unknown level of incoming jitter, to a known level.

 

 

 

Simple?!   .....   but not all "reclocking" is created equal.   Simply applying a new clock to the data is trivial ....   but it is difficult to make a system where the new timing is truly and completely decoupled from the original timing.

[CraigS]

I2S isn't affected by jitter like SPDIF is.

In that it has a clock and data as separate signals, yes.  But it is a standard that is applied to short connections inside a piece of equipment, not as an external standard to connect two pieces of equipment.  There is no defined characteristic impedance, because it is designed for transmission over a few centimetres.

 

Now it *can* be designed to work as an external connection, but like all these things the devil is in the detail. (i) it needs a defined characteristic impedance and a cable to suit, (ii) it needs galvanic isolation of both clock and data and (iii) it needs differential line drivers (with enough oomph; an engineering term, you understand) and receivers.

[rastus]

Remember what you are looking at is a super simple example.   One single frequency  (and what the jitter looked like) ..... if it was music, there's many more frequencies.....  and so for the jittered result, there would be WAY MORE extra spikes ...... so much more that it just looks like "the noise floor".    When is why jitter typically sounds like a compromised signal to noise ratio.

 

OK, "no bidirectional communication".... and all "the DAC can do, is mute the output", so do some DACs do this better than others? Or, is it some DACs don't do this at all?

"Remember what you are looking at is a super simple example." <---- Yes, I do get the relevence of this... music + jitter = berry bad.

Yes, I have seen that picture before, and did read the article it was attached to, and didn't really understand much, other than something might be amiss.

[petersv]

Yes other dacs do it better, ess saber have a time domain jitter patent implementation that rebuilds lost samples by over sampling and rebuilding in an intelligent way. Those chips are expensive though, 9018 is ~x20 the cost of regular dacs and have nightmare power supply requirements so are rare to see in under $1k dacs etc