Tackling USB / I2S interface jitter
Hi Ulas,
Thanks for your reply [post736]
> I am measuring jitter correctly, yes I trigger on the rising edge of BCK, but I also look at a number of following transients as well (timebase X10), this way yitter on both rising and falling edge of BCK can be examined. I added a oscillogram to proove this. The left oscillogram clearly shows that the reclock side effects are concentrated on the falling edge of BCK and the leading edge of BCK is relatively clean and contains very low jitter. The oscillogram on the right shows the reclocked BCK signal and the 48 MHz reclock signal. Now see what I mean? The next experiment is to trigger a digital one-shot circuit on the relatively clean BCK rising edge and overrule the two phase trailing edge, then both transients are clean. Then we have both, optimal sound quality and a fancy looking BCK oscillogram
> Yes you can laugh, but the circuit I build improved the USB / I2S interface in a way it had about the same sound quality as a direct I2S interface comming form my CD changer. I am pretty sure many USB / I2S interfaces don't even come close as sound quality is concerned. The big difference with standard asynchronous reclocking is that the leading edge of BCK has lower jitter and the two phase jitter is concentrated at the falling edge of BCK.
> I already explained that a high amplitude 2 phase jitter has far less negative effect on the audio signal than the 200ps continuous jitter from a CS8412. But you can also read Kusonoki's article about this issue. Now I further improved this by reducing leading edge jitter.
> Adding a asynchronous reclocker replases the continuous phase (phase noise) jitter with a consistent two phase jitter that in turn is locked to a stable crystal clock signal. If this type of jitter was worse than 200pS continuous jitter, sound quality had to be very bad, but at the contrary, sound quality improves.
> Personally, like you I preferably like to remove all jitter and obtain a perfect almost zero jitter signal, but as I said, the used USB reclocking setup is a good alternative that works, and I am even experimenting on improving this setup. I am just exploring solutions to this problem, that's all. I also tried various PLL setup's (slowly averaging systems) but basically adding a PLL makes things worse. If I am correct, PLL output frequency has to vary a bit in order to maintain phase locking, so basically it always jitters. If you got idea's to tackle jitter, given this setup I am very interested.
My approach to the USB interface is to accept the (large) amount of continuous jitter and add a reclock circuit that eliminates jitter. Each computer connected to the same interface can produce different results, even it's switched-mode power supply affects jitter. Ground loops can add huge amounts of jitter, so the first thing to start with is to use isolation transformers for BCK, WS and DATA. Then construct a circuit to tackle jitter, this way the USB / I2S output signal is consistent, regardless of USB cable length, USB power supply or the interference produced by the computers switched-mode power supply. Using a "better" USB chip brings only marginal improvement. If I am correct, jitter that emerges from a USB to I2S interface is a fundamental problem, not a problem that can simply be solved using "better" USB interface chip's.
> As you already noted, the 12 MHz clock is used for USB bidirectional data clocking, so it has to be related to packet timing. The PLL recovers BCK from this packet data (is related to the 12 MHz clock). Now by adding a separate 40 MHz crystal clock, a interference source is added (12 MHz and 40 MHz). Since the 12 MHz clock was very noisy, it also add's interference to the circuit. That's the reason I used a single 48MHz crystal instead, one stable clock source, less interference. A side effect of this setup is the relatively clean BCK rising edge.
> In my humble opinion, the following types of jitter are "preferred" starting with the best:
1) Zero jitter
2) Low jitter (synchronously reclocked)
3) Low jitter with some sporadic glitches (semi-asynchronous reclocking with digital one-shot)
4) Two phase jitter at trailing edge of BCK (semi-asynchronous reclocking)
5) Two phase jitter at both leading and trailing edge of BCK (asynchronously reclocked)
6) Continuous jitter (phase noise)
Hi Ulas,
Thanks for your reply [post736]
> I am measuring jitter correctly, yes I trigger on the rising edge of BCK, but I also look at a number of following transients as well (timebase X10), this way yitter on both rising and falling edge of BCK can be examined. I added a oscillogram to proove this. The left oscillogram clearly shows that the reclock side effects are concentrated on the falling edge of BCK and the leading edge of BCK is relatively clean and contains very low jitter. The oscillogram on the right shows the reclocked BCK signal and the 48 MHz reclock signal. Now see what I mean? The next experiment is to trigger a digital one-shot circuit on the relatively clean BCK rising edge and overrule the two phase trailing edge, then both transients are clean. Then we have both, optimal sound quality and a fancy looking BCK oscillogram
> Yes you can laugh, but the circuit I build improved the USB / I2S interface in a way it had about the same sound quality as a direct I2S interface comming form my CD changer. I am pretty sure many USB / I2S interfaces don't even come close as sound quality is concerned. The big difference with standard asynchronous reclocking is that the leading edge of BCK has lower jitter and the two phase jitter is concentrated at the falling edge of BCK.
> I already explained that a high amplitude 2 phase jitter has far less negative effect on the audio signal than the 200ps continuous jitter from a CS8412. But you can also read Kusonoki's article about this issue. Now I further improved this by reducing leading edge jitter.
> Adding a asynchronous reclocker replases the continuous phase (phase noise) jitter with a consistent two phase jitter that in turn is locked to a stable crystal clock signal. If this type of jitter was worse than 200pS continuous jitter, sound quality had to be very bad, but at the contrary, sound quality improves.
> Personally, like you I preferably like to remove all jitter and obtain a perfect almost zero jitter signal, but as I said, the used USB reclocking setup is a good alternative that works, and I am even experimenting on improving this setup. I am just exploring solutions to this problem, that's all. I also tried various PLL setup's (slowly averaging systems) but basically adding a PLL makes things worse. If I am correct, PLL output frequency has to vary a bit in order to maintain phase locking, so basically it always jitters. If you got idea's to tackle jitter, given this setup I am very interested.
My approach to the USB interface is to accept the (large) amount of continuous jitter and add a reclock circuit that eliminates jitter. Each computer connected to the same interface can produce different results, even it's switched-mode power supply affects jitter. Ground loops can add huge amounts of jitter, so the first thing to start with is to use isolation transformers for BCK, WS and DATA. Then construct a circuit to tackle jitter, this way the USB / I2S output signal is consistent, regardless of USB cable length, USB power supply or the interference produced by the computers switched-mode power supply. Using a "better" USB chip brings only marginal improvement. If I am correct, jitter that emerges from a USB to I2S interface is a fundamental problem, not a problem that can simply be solved using "better" USB interface chip's.
> As you already noted, the 12 MHz clock is used for USB bidirectional data clocking, so it has to be related to packet timing. The PLL recovers BCK from this packet data (is related to the 12 MHz clock). Now by adding a separate 40 MHz crystal clock, a interference source is added (12 MHz and 40 MHz). Since the 12 MHz clock was very noisy, it also add's interference to the circuit. That's the reason I used a single 48MHz crystal instead, one stable clock source, less interference. A side effect of this setup is the relatively clean BCK rising edge.
> In my humble opinion, the following types of jitter are "preferred" starting with the best:
1) Zero jitter
2) Low jitter (synchronously reclocked)
3) Low jitter with some sporadic glitches (semi-asynchronous reclocking with digital one-shot)
4) Two phase jitter at trailing edge of BCK (semi-asynchronous reclocking)
5) Two phase jitter at both leading and trailing edge of BCK (asynchronously reclocked)
6) Continuous jitter (phase noise)
Attachments
Ulas,
Most of us have been reading your posts in the background, and letting your smug comments slide by. But I think some of us, including myself, are getting tired and annoyed by your backhanded insults, especially to EC. I'm not denying you have extensive knowledge in digital audio, and may be you have something to contribute. I have yet to see a working design from you on this board, but you are a master at finding "faults" in others designs without the experience of actually hearing them. As far as can tell, EC has put more innovative work into a DIY DAC than any other DIYer I can remember, and he backs it up with actual data and listening. May be it's time to get off your high horse and join us on the ground. Your s**t doesn't smell any rosier than ours.
Most of us have been reading your posts in the background, and letting your smug comments slide by. But I think some of us, including myself, are getting tired and annoyed by your backhanded insults, especially to EC. I'm not denying you have extensive knowledge in digital audio, and may be you have something to contribute. I have yet to see a working design from you on this board, but you are a master at finding "faults" in others designs without the experience of actually hearing them. As far as can tell, EC has put more innovative work into a DIY DAC than any other DIYer I can remember, and he backs it up with actual data and listening. May be it's time to get off your high horse and join us on the ground. Your s**t doesn't smell any rosier than ours.
USB / I2S interface optimized
Hi all,
As just noted, I was planning to trigger a digital one-shot circuit on the leading edge of the reclocked BCK signal in order to get rid of the nasty looking trailing edge, well it's already working. The result is a semi-asynchronous reclocked BCK signal that has very low jitter, looks fany on the oscillogram and produces exactly the same sound quality as with the direct I2S output from my CD-changer. Good enough?
The digital one-shot circuit consists of a D-flipflop that is triggered on the leading edge of the already reclocked BCK signal, then it is reset after 7 clock pulses (derived from a synchronous counter). Out comes the clean BCK signal as can be seen at the lower trace of the added oscillogram. The upper trace shows the reclocked signal before trailing edge correction.
A nice detail is that the USB cable is about 8 meters long and the whole USB/I2S interface is powered by USB (no adapter necessary), sound quality is the same when using different computers.
Hi all,
As just noted, I was planning to trigger a digital one-shot circuit on the leading edge of the reclocked BCK signal in order to get rid of the nasty looking trailing edge, well it's already working. The result is a semi-asynchronous reclocked BCK signal that has very low jitter, looks fany on the oscillogram and produces exactly the same sound quality as with the direct I2S output from my CD-changer. Good enough?
The digital one-shot circuit consists of a D-flipflop that is triggered on the leading edge of the already reclocked BCK signal, then it is reset after 7 clock pulses (derived from a synchronous counter). Out comes the clean BCK signal as can be seen at the lower trace of the added oscillogram. The upper trace shows the reclocked signal before trailing edge correction.
A nice detail is that the USB cable is about 8 meters long and the whole USB/I2S interface is powered by USB (no adapter necessary), sound quality is the same when using different computers.
Attachments
EC,
That is great news.
I was looking at having to purchase a CD2Pro, but I guess I can hold that purchase off for a while (to let my wallet recover) and use USB instead in the meantime!
So... for longer runs of wire, would one use a long USB cable? Or a long I2S cable? I wasn't too certain from reading your posts. It would be great if it were the former - that way everything can be kept to a single box, certainly a much neater and 'nicer' solution.
But still though, I wouldn't mind additional external boxes cluttering my place up - I mean, it's all about the sound right?
That is great news.
I was looking at having to purchase a CD2Pro, but I guess I can hold that purchase off for a while (to let my wallet recover) and use USB instead in the meantime!
So... for longer runs of wire, would one use a long USB cable? Or a long I2S cable? I wasn't too certain from reading your posts. It would be great if it were the former - that way everything can be kept to a single box, certainly a much neater and 'nicer' solution.
But still though, I wouldn't mind additional external boxes cluttering my place up - I mean, it's all about the sound right?
Where to put the USB interface?
Hi adhoc,
Thanks for your reply [post#749]
Yes the setup with the iMac is really something else , picking out a song from thousands of songs stored on a HD, just by typing a phrase is really impressive. The iMac with iTunes is going to stay for sure.
Now the sound quality of the USB to I2S interface matches my CD transport direct I2S output, it looks like all my CD players / changers will be catching dust on the attic soon.
The best part is, the USB interface can be integrated in the octal D-I DAC, skipping the differential I2S interlink all together, obtaining optimal sound quality. So I am already thinking of putting only 3 I2S input's on the octal D-I DAC and add a USB input.
To answer your question, with this particular USB to I2S interface, use a long USB cable, skip the I2S interlink and put the USB interface in the box.
Hi adhoc,
Thanks for your reply [post#749]
Yes the setup with the iMac is really something else
, picking out a song from thousands of songs stored on a HD, just by typing a phrase is really impressive. The iMac with iTunes is going to stay for sure.Now the sound quality of the USB to I2S interface matches my CD transport direct I2S output, it looks like all my CD players / changers will be catching dust on the attic soon.
The best part is, the USB interface can be integrated in the octal D-I DAC, skipping the differential I2S interlink all together, obtaining optimal sound quality. So I am already thinking of putting only 3 I2S input's on the octal D-I DAC and add a USB input.
To answer your question, with this particular USB to I2S interface, use a long USB cable, skip the I2S interlink and put the USB interface in the box.
Hello EC,
Thanks for that reply. However, your reply has just triggered something in my mind:
Are you sure that the output from your iMac is even bit-perfect? In other words, is your iMac/iTunes setup covertly resampling your stored music?
I ask this because in Windows based machines at least, (badly done) internal resampling is by default done covertly through Windows kmixer (48kHz), and can only be bypassed via enabling kernal-streaming, directsound (not sure about this one) or ASIO.
AFAIK, in Windows at least, it is impossible to get a bit-perfect stream out of iTunes via the standard user interface.
I'm not sure how you would test your setup for 'bit-perfect-ness', but 2 possibilities come to mind. If you could somehow send a digital stream of a rip of a HDCD-encoded CD to an external HDCD compatible DAC (as is present on many redbook players) and check whether the HDCD light to light up, you would be able to definitively say that 'yes, this is a bit-perfect stream'.
Alternatively, you could digitally stream a DTS-encoded surround signal to a DTS receiver and listen for music (as opposed to white noise if resampling took place).
Or perhaps after streaming a digital-out back into your iMac, and after adjusting for delay and what-have-you, a bit-for-bit comparison could be made?
Just my 0.02c.
Thanks for that reply. However, your reply has just triggered something in my mind:
Are you sure that the output from your iMac is even bit-perfect? In other words, is your iMac/iTunes setup covertly resampling your stored music?
I ask this because in Windows based machines at least, (badly done) internal resampling is by default done covertly through Windows kmixer (48kHz), and can only be bypassed via enabling kernal-streaming, directsound (not sure about this one) or ASIO.
AFAIK, in Windows at least, it is impossible to get a bit-perfect stream out of iTunes via the standard user interface.
I'm not sure how you would test your setup for 'bit-perfect-ness', but 2 possibilities come to mind. If you could somehow send a digital stream of a rip of a HDCD-encoded CD to an external HDCD compatible DAC (as is present on many redbook players) and check whether the HDCD light to light up, you would be able to definitively say that 'yes, this is a bit-perfect stream'.
Alternatively, you could digitally stream a DTS-encoded surround signal to a DTS receiver and listen for music (as opposed to white noise if resampling took place).
Or perhaps after streaming a digital-out back into your iMac, and after adjusting for delay and what-have-you, a bit-for-bit comparison could be made?
Just my 0.02c.
Hi smms73,
Thanks for your reply [post#731],
The main difference with the Cambridge CD player is that the octal D-I DAC uses no oversampling and increases resolution 8 times, this is a different approach. The Cambridge DAC increases resolution After 4 X oversampling by sequencially firing the 4 DAC chip's at 8 BCK intervals, resulting in 16 times more samples.
The problems associated with digital interpolation filters are still present here.
The SAA7220 output's is 32 bit wide. When 4 interpolated samples should be placed at equal distance the delay's are BCK, BCK8, BCK16 and BCK24. So the Cambridge DAC also uses equal interpolated sample distance.
> If I am correct, the SAA7210 output is 64 bit wide / 2.8224 MHz
> Why is my design 64 bit wide?
The octal D-I DAC is a NOS DAC primarily intended as a external DAC that has a integrated source selector, since I decided to use multiple differential I2S interfaces instead of SPDIF, the octal D-I DAC should be able to accept both the philips format (64 bit wide) and the sony format (48 bit wide). It switches automatically between them.
At the moment I am thinking of using 3 differential I2S inputs and one USB input.
> When you bypass the SAA7220, you end up with 4 times resolution enhancement in NOS mode, so resolution is probably not high enough. The corrective filter (if present) might affect sound quality in NOS mode as well.
Thanks for your reply [post#731],
The main difference with the Cambridge CD player is that the octal D-I DAC uses no oversampling and increases resolution 8 times, this is a different approach. The Cambridge DAC increases resolution After 4 X oversampling by sequencially firing the 4 DAC chip's at 8 BCK intervals, resulting in 16 times more samples.
The problems associated with digital interpolation filters are still present here.
The SAA7220 output's is 32 bit wide. When 4 interpolated samples should be placed at equal distance the delay's are BCK, BCK8, BCK16 and BCK24. So the Cambridge DAC also uses equal interpolated sample distance.
> If I am correct, the SAA7210 output is 64 bit wide / 2.8224 MHz
> Why is my design 64 bit wide?
The octal D-I DAC is a NOS DAC primarily intended as a external DAC that has a integrated source selector, since I decided to use multiple differential I2S interfaces instead of SPDIF, the octal D-I DAC should be able to accept both the philips format (64 bit wide) and the sony format (48 bit wide). It switches automatically between them.
At the moment I am thinking of using 3 differential I2S inputs and one USB input.
> When you bypass the SAA7220, you end up with 4 times resolution enhancement in NOS mode, so resolution is probably not high enough. The corrective filter (if present) might affect sound quality in NOS mode as well.
hi maxlorenz,
the address still not good.
for your order with 20pcs TDA1541a
date code was HSH8944-2
s/n 11145.
all fit with IC sockets to U.
But I had a little problem, The address still not clear, some words looks the hotmail not support. can U send to my personal address.
aasiu@netvigator.com
regarding the groups buy for more 1543 & other chips such as the AD1865 or PCM56.
I will offer to personal match the same date code to diyers.
thx
thomas
the address still not good.
for your order with 20pcs TDA1541a
date code was HSH8944-2
s/n 11145.
all fit with IC sockets to U.
But I had a little problem, The address still not clear, some words looks the hotmail not support. can U send to my personal address.
aasiu@netvigator.com
regarding the groups buy for more 1543 & other chips such as the AD1865 or PCM56.
I will offer to personal match the same date code to diyers.
thx
thomas
hi adhoc,
trust me, pls did not buy the pro2. I had 2 & all sold already.
since pro2 looks need calibrate for te accurancy of the reading.
Diyers not easy to do this. If U like to get a good transport.
The first suggest was CD-M9 series, M1,M4 will better but now is too rare.
still had some M9 transport in markets.
forget 12.1 series.
Another choice was choose Plextor Cd-rom with any types of cd-rom transport control kit.
this is my comment.
thx
thomas
trust me, pls did not buy the pro2. I had 2 & all sold already.
since pro2 looks need calibrate for te accurancy of the reading.
Diyers not easy to do this. If U like to get a good transport.
The first suggest was CD-M9 series, M1,M4 will better but now is too rare.
still had some M9 transport in markets.
forget 12.1 series.
Another choice was choose Plextor Cd-rom with any types of cd-rom transport control kit.
this is my comment.
thx
thomas
hi maxlorenz,
do u means where is the location of my chips come from.
some was china, mainly was taiwan.
mainly my 1541a chips was from philips hong kong. around 1998, philips hong kong will close some of dept & move to china. repair stock parts sold to public in good price.
my 1541a was taken that time. some of other parts such as BC caps & other parts were sold to two old couple. THis two old couple always go to large company to use good price to buy the remainstock & resell. This recent days I met them & ask for the old stock of philips parts. They told me that still had some, so I went back to their store 7 check. The most suitable for me was BC caps 40V 470uf. it means every my coming dac need 32pcs per dac. But will really very good for this caps. All caps was made in Holland.
my uncle was one of brand name computor monther board producer. His purchasing dept help me lot for chasing for chips & price not bad.
thx
thomas
do u means where is the location of my chips come from.
some was china, mainly was taiwan.
mainly my 1541a chips was from philips hong kong. around 1998, philips hong kong will close some of dept & move to china. repair stock parts sold to public in good price.
my 1541a was taken that time. some of other parts such as BC caps & other parts were sold to two old couple. THis two old couple always go to large company to use good price to buy the remainstock & resell. This recent days I met them & ask for the old stock of philips parts. They told me that still had some, so I went back to their store 7 check. The most suitable for me was BC caps 40V 470uf. it means every my coming dac need 32pcs per dac. But will really very good for this caps. All caps was made in Holland.
my uncle was one of brand name computor monther board producer. His purchasing dept help me lot for chasing for chips & price not bad.
thx
thomas
Attachments
Dear Thomas,
You wrote: (post#724)
From that post I supposed you preferred to open a new thread for us to put orders.
Do you prefer direct mail instead? Or maybe post at one of your existing threads?
I think a new thread will be better for you: less confusion.
8 chips is enough for me!
All I (we) want is a secure place where to put the orders. I think email will be conflicting.
Ecdesigns wrote:
Sorry to insist about CDPRO transport
I mailed Nico Thevissen, from http://www.daisy-laser.nl/
and he told me that it outputs I2S in Philips format:
"The output is Philips format no oversampling.
See IC spec. "
Regards,
Nico
-------------------------------------------------------------------------------
By "IC specs", does he meant SAA7324 M2 (CD10)
servo-decoder???
http://www.daisy-laser.com/products/CD/modules/CDPro2/SAA7324_2.pdf
Thanks for any help.
M
You wrote: (post#724)
From that post I supposed you preferred to open a new thread for us to put orders.
Do you prefer direct mail instead? Or maybe post at one of your existing threads?
I think a new thread will be better for you: less confusion.
8 chips is enough for me!All I (we) want is a secure place where to put the orders. I think email will be conflicting.
Ecdesigns wrote:
Sorry to insist about CDPRO transport
I mailed Nico Thevissen, from http://www.daisy-laser.nl/
and he told me that it outputs I2S in Philips format:
"The output is Philips format no oversampling.
See IC spec. "
Regards,
Nico
-------------------------------------------------------------------------------
By "IC specs", does he meant SAA7324 M2 (CD10)
servo-decoder???
http://www.daisy-laser.com/products/CD/modules/CDPro2/SAA7324_2.pdf
Thanks for any help.
M
Hi ecdesigns
Thanks very much for you reply (post752)
-You are correct, the saa7210 output is 64bit wide.
-I will change the delay line of the Cambridge for 64 bit wide, (bck, bck16, bck32, bck 48), and bypass the SAA7220 again.
the output only have one ne5534(i/v converter) for each dac, and a simple rc filter in the end.
-This modifications will make the dac work like your design, but only 4 tda1541a.
Thanks very much for you reply (post752)
-You are correct, the saa7210 output is 64bit wide.
-I will change the delay line of the Cambridge for 64 bit wide, (bck, bck16, bck32, bck 48), and bypass the SAA7220 again.
the output only have one ne5534(i/v converter) for each dac, and a simple rc filter in the end.
-This modifications will make the dac work like your design, but only 4 tda1541a.