Ethernet Renderers jitter measurements

0 Members and 1 Guest are viewing this topic. Read 19429 times.

audioengr

Ethernet Renderers jitter measurements
« on: 9 Dec 2017, 01:09 am »
I have my DLNA renderer, the Interchange and I have a Raspbery Pi SBC and a Digi HAT board with BNC S/PDIF output that can do DLNA as well as Roon. I set both of these up for DLNA so I could play tracks from Jriver.

This particular Digi board, which is a HAT board for the Raspberry Pi SBC, is claimed to have around 4psec of jitter in the S/PDIF output.  I believe this is typically measured using a DAC and spectrum analyzer.  I guess they put a single frequency to it and measure deviation.  This is the way of measuring jitter for Stereophile and others.

I don't feel that this is an accurate method of measuring jitter, I never have.  Because I'm lucky enough to have spent the big bux on a programmable scope (originally $130K) with jitter software, I can make direct jitter measurements.  I purchased a 75 ohm terminated input for the scope.  That tiny plug-in was over $1K alone.  Brother.  At least I can plug the S/PDIF output from any device directly into the scope and make all kinds of jitter and signal integrity measurements.

So, I selected a music track that has a lot of dynamics in it and I took samples during the duration of that track.  Then I stopped the sampling and generated the plots.  This is a distribution of real time intervals, around 42K samples over 5 minutes.

Digi HAT jitter distribution, 20psec per division:



Interchange jitter distribution, 10psec per division:



The Interchange has about 10psec of jitter and the Digi Hat board 60-70psec of jitter.  This proves to me that the current method of measuring jitter is insufficient and misleading.  The sound quality of the Digi HAT is great for a $150 system, but their jitter claims are somehow wrong.  I don't believe it's their fault.  They are using a flawed measurement method that everyone else uses. 

My Interchange is more musical to me, and I think I know why from these plots.  This is the correlation between jitter and sound quality that I have strived for.  I think I have something here.  I may have to share this method with John Atkinson.  It's time we had a decent correlation.

Here are the S/PDIF signals, first the Digi HAT:



And the Interchange:



The rise and fall times for the two devices are almost identical, around 1.11nsec, but my Interchange has the same risetime with less voltage swing.  The Interchange has a swing of .5V P-P like the spec requires and the Digi Hat board is more like .75V P-P.

Steve N.
« Last Edit: 2 Nov 2018, 08:23 pm by audioengr »

ketcham

  • Industry Participant
  • Posts: 285
Re: Renderer jitter measurements
« Reply #1 on: 9 Dec 2017, 01:29 am »
Steve,

enlighten me regarding the deviation for each unit.  Seems one order of magnitude less for your renderer. 

Can you light smoke under ABC to get their firmware updated for ROON...they're killing me.

audioengr

Re: Renderer jitter measurements
« Reply #2 on: 9 Dec 2017, 02:00 am »
Steve,

enlighten me regarding the deviation for each unit.  Seems one order of magnitude less for your renderer. 

Can you light smoke under ABC to get their firmware updated for ROON...they're killing me.

The standard deviation for the Interchange is about 20psec.  The Digi HAT is difficult to measure.  If you include all of the humps, it is 100-120 psec.

I think ABC should do Roon, and I don't really know why they are dragging feet.  They can sell a lot more SBC's to me.

BTW, I have new mods for your Overdrive SX to get to these low jitter numbers.

Steve N.
« Last Edit: 14 Dec 2017, 06:10 pm by audioengr »

carlito1

  • Industry Participant
  • Posts: 8
Re: Ethernet Renderers jitter measurements
« Reply #3 on: 14 Dec 2017, 01:13 pm »
Hello Steve

  ok I see your measurements with your specific tool. When was your programmable scope last calibrated ?

 Its hard to take your measurements at face values , especially since you are selling a competitor product , but we will try to see if we can get one .

  At last , I would like to refute 2 things. BNC is 75Ohm and yes voltage swing is higher because we did take in consideration the cable loss...

audioengr

Re: Ethernet Renderers jitter measurements
« Reply #4 on: 14 Dec 2017, 06:01 pm »
Hello Steve

  ok I see your measurements with your specific tool. When was your programmable scope last calibrated ?

 Its hard to take your measurements at face values , especially since you are selling a competitor product , but we will try to see if we can get one .

  At last , I would like to refute 2 things. BNC is 75Ohm and yes voltage swing is higher because we did take in consideration the cable loss...

You can get a used 2.5GHz Tek scope with JIT3 here:

https://www.ebay.com/itm/Tektronix-TDS7254-Digital-Phosphor-Oscilloscope-Opt-1M-SM-ST-jitter3/151576491510?epid=1700314859&hash=item234aa9b5f6:g:yKEAAOSwk5FU0I7L

Steve N.
« Last Edit: 14 Dec 2017, 08:20 pm by audioengr »

carlito1

  • Industry Participant
  • Posts: 8
Re: Ethernet Renderers jitter measurements
« Reply #5 on: 14 Dec 2017, 07:28 pm »
Our datasheet on the BNC connector says its 75Ohm.

  Can you check the voltage output at the end of the cable with the proper termination ? You will have a small drop.

Also can you explain why you assume that jitter is adding on the "peaks" that you found ? How exactly are they adding up..

At last , if you renderer has 16ps and ours 60ps but its about 1/20 of the price... I would say its a pretty good result for our device .

carlito1

  • Industry Participant
  • Posts: 8
Re: Ethernet Renderers jitter measurements
« Reply #6 on: 14 Dec 2017, 07:39 pm »
One last thing... BNC signal must be 1.0 Vp-p and not 0.5

http://www.rane.com/note149.html

audioengr

Re: Ethernet Renderers jitter measurements
« Reply #7 on: 14 Dec 2017, 07:50 pm »
Our datasheet on the BNC connector says its 75Ohm.

I guess some of these have a smaller collet in them.  I stand corrected.

Quote
Can you check the voltage output at the end of the cable with the proper termination ? You will have a small drop.

The plots are terminated into 75 ohms.  The scope has a special plug-in that is 75 ohms.  The output impedance of the driver should be 75 ohms and the termination is 75 ohms, so the drop should be 1/2 compared to open termination.

Quote
Also can you explain why you assume that jitter is adding on the "peaks" that you found ? How exactly are they adding up..

Jitter is the deviation from the nominal period.  Each of the four peaks is a deviation, although the outer ones are less frequent.  I usually use the standard deviation measured by the scope, but it's more complicated when there are several peaks.

Quote
At last , if you renderer has 16ps and ours 60ps but its about 1/20 of the price... I would say its a pretty good result for our device .

Remarkably good.  You have done a great job with this design.  Actually about 1/30 of the price.  If you could fit some other through-hole decoupling caps, I think it might even be better.  I have found that ceramic caps achieve good functionality, but film and electrolytic caps reduce jitter even more.

Steve N.


audioengr

Re: Ethernet Renderers jitter measurements
« Reply #8 on: 14 Dec 2017, 07:53 pm »
One last thing... BNC signal must be 1.0 Vp-p and not 0.5

http://www.rane.com/note149.html

That is the AES3 interface.  I don't think I can drive long cables like that, and I would not want to try anyway.  Terrible jitter would result.

Steve N.

carlito1

  • Industry Participant
  • Posts: 8
Re: Ethernet Renderers jitter measurements
« Reply #9 on: 14 Dec 2017, 09:18 pm »
I guess some of these have a smaller collet in them.  I stand corrected.

The plots are terminated into 75 ohms.  The scope has a special plug-in that is 75 ohms.  The output impedance of the driver should be 75 ohms and the termination is 75 ohms, so the drop should be 1/2 compared to open termination.

Jitter is the deviation from the nominal period.  Each of the four peaks is a deviation, although the outer ones are less frequent.  I usually use the standard deviation measured by the scope, but it's more complicated when there are several peaks.

Remarkably good.  You have done a great job with this design.  Actually about 1/30 of the price.  If you could fit some other through-hole decoupling caps, I think it might even be better.  I have found that ceramic caps achieve good functionality, but film and electrolytic caps reduce jitter even more.

Steve N.

  Thx you Steve , unfortunately because of the small PCB size there is not much space for extra capacitors. Anyway we are developing a new DigiOne (signature) that runs on pure battery power with lots of film capacitors ..

 Good luck with your project .

audioengr

Re: Ethernet Renderers jitter measurements
« Reply #10 on: 14 Dec 2017, 10:03 pm »
  Thx you Steve , unfortunately because of the small PCB size there is not much space for extra capacitors. Anyway we are developing a new DigiOne (signature) that runs on pure battery power with lots of film capacitors ..

 Good luck with your project .

I would recommend that your battery supply also have Ultracaps.  Even the best LI batteries don't seem to have the capability to supply the di/dt needed for digital.

I used to offer such a supply, one that actually charged the LI batteries and then disconnected from the AC and ground.  It was very good, but ultimately my new AC supplies are better.

Steve N.

Johnny2Bad

Re: Ethernet Renderers jitter measurements
« Reply #11 on: 14 Dec 2017, 11:59 pm »
{snip} .... 

My Interchange is more musical to me, and I think I know why from these plots.  This is the correlation between jitter and sound quality that I have strived for.  I think I have something here.  I may have to share this method with John Atkinson.  It's time we had a decent correlation.

... {snip}.

Steve N.

Just eMail JA. He reads and responds. His address is on their site.

audioengr

Re: Ethernet Renderers jitter measurements
« Reply #12 on: 15 Dec 2017, 12:16 am »
Just eMail JA. He reads and responds. His address is on their site.

I emailed JA.  He knows me and usually looks me up at shows, although I have not attended too many lately.

Steve N.

carlito1

  • Industry Participant
  • Posts: 8
Re: Ethernet Renderers jitter measurements
« Reply #13 on: 15 Dec 2017, 09:06 am »
I would recommend that your battery supply also have Ultracaps.  Even the best LI batteries don't seem to have the capability to supply the di/dt needed for digital.

I used to offer such a supply, one that actually charged the LI batteries and then disconnected from the AC and ground.  It was very good, but ultimately my new AC supplies are better.

Steve N.

 Yeap of course ultracaps are included. Multiple.

jespera

  • Jr. Member
  • Posts: 7
Re: Ethernet Renderers jitter measurements
« Reply #14 on: 22 Oct 2018, 09:18 pm »
Steve,

in the measurements of the spdif output squares,
the flat pieces appear to be overlayed by noise — not ringing.
This is the case both for the digione and the interchange
measurements.

What is thw source of that noise?

Can you elaborate a bit on how your jitter measurement works?

Have you tried measuring the jitter from combining the digione
with the synchro mesh?

Thx

Jesper



audioengr

Re: Ethernet Renderers jitter measurements
« Reply #15 on: 23 Oct 2018, 01:21 am »
Steve,

in the measurements of the spdif output squares,
the flat pieces appear to be overlayed by noise — not ringing.

This is the case both for the digione and the interchange
measurements.

What is thw source of that noise?

Are you referring to the spectral or distribution plots?  What do mean by "squares"?

Quote
Can you elaborate a bit on how your jitter measurement works?

It measures the period of a repetitive signal tens of thousands of times.  With S/PDIF it locks onto the shorter of the periods since the signal has a changing period.

Quote
Have you tried measuring the jitter from combining the digione
with the synchro mesh?

No, but I suspect that it will be similar to all other sources driving the SM.  Very low jitter on the output.

Steve N.

jespera

  • Jr. Member
  • Posts: 7
Re: Ethernet Renderers jitter measurements
« Reply #16 on: 23 Oct 2018, 11:31 am »

The opening post of this thread shows four screen shots. I'm referring to screen shots 3 and 4
that show square waves in them. What is causing the noise on the flat part of the squares?

Have you observed any impact on the jitter measurements from using linear power supplies?

thx

Jesper
 

audioengr

Re: Ethernet Renderers jitter measurements
« Reply #17 on: 23 Oct 2018, 05:16 pm »
The opening post of this thread shows four screen shots. I'm referring to screen shots 3 and 4
that show square waves in them. What is causing the noise on the flat part of the squares?

Okay, this small overshoot and ringing on the square-waves is because of the ultra-fast rise-times.  It settles out quickly, so not a problem.  This small ringing has no effect on the jitter.  In past lives, I usually optimized transmission-lines by allowing a small amount of overshoot and ring.  This produces better performance.

The small noise riding on the rest of the waveform is typical in digital systems.  It's thermal noise, power supply noise and ground-plane noise.  The reason you are even seeing it is that the B/W of the scope is 7GHz.  Most scopes will show a flat-line, but it's not real.  There is always a little noise riding on the waveform.

Quote
Have you observed any impact on the jitter measurements from using linear power supplies?

My fast Hynes-based LPS, the Dynamo reduces the jitter a little.  It's audible.  Next time I set-up measurements, I will capture some comparisons of this.

Steve N.

jespera

  • Jr. Member
  • Posts: 7
Re: Ethernet Renderers jitter measurements
« Reply #18 on: 23 Oct 2018, 10:05 pm »

Sorry for pestering you with stupid questions. Here are some more:

What does b/w of 7GHz mean? What is the sampling freq of the scope?

Is there any way of reducing the noise on the squares? Would the use of
discrete components (transistors, tubes, chokes, etc) rather than chips help?

Do you think it would be audible?

thx

Jesper
 
 

brj

Re: Ethernet Renderers jitter measurements
« Reply #19 on: 24 Oct 2018, 03:12 am »
Steve can answer far better than I, but while some of the 'noise' on the square wave trace might actually be noise, most of what you're seeing is just the nature of a band-limited signal.  Rather than trying to explain via a forum post, I'd recommend the following video as background:

https://www.youtube.com/watch?feature=player_embedded&v=cIQ9IXSUzuM

I'd encourage interested parties to watch the entire video, but the square wave relevant info and how all DAC reconstruction filters will truncate the higher order harmonics needed to reduce the 'ripple' you're seeing starts around 17:30 or so.