[James Tanner]

^
Hey Gordon - What can I say - great piece!

thanks,
james

[drubin]

Great stuff, Gordon.

[Sasha]

James and the rest of the gang,

Bob emailed me yesterday so I figure I would clear up a few things about USB.

First if your willing to write drivers for all os's then USB is wide open. I did this for years and would rather not do it again. Therefore I like to use what is called the Native drivers.

Now there is USB 1.1 and USB 2.0 but really when we talk audio it is Class 1 or Class 2 that really makes a difference. Class 1 was the basic audio specification which includes all rates up to 102,3K sampling rate. Why? Because the Class 1 spec says that an audio frame will happen ounce ever 1ms or greater. The specification says only 1023 bytes per frame is allowed thereby setting the max rate to 102,3K.

First the Class 2 Audio spec is only supported under MAC OSX 10.5.x and greater. There is no present plans to do Class 2 on Windows or Linux, though I am betting within a year from now that this will happen.

The big problem is that to get to xx/192 (I say xx as I am working on the ESS part and have 32/96 working and should have 32/192 soon) you need both High Speed USB 2.0 and Class 2. In the Class 2 audio spec you can now use what are called sub frames. This allows you to basically send a sub frame every 146uS which means you can pass some serious data.

The big difference and something that I have been working on for like the last 5 years is the jitter is not really understood that much. I have been saying the following and testing seems to back it up that the use of upsamplers or FIFO reclocking devices only act as like a low pass filter to jitter. You can pour though JA's testing and see that this is the case in many of the USB input tests the jitter is significantly higher than the SPDIF inputs on the same dac. I believe that the damage is done before the upsamplers and FIFO reclocking systems and that not all of this is removed or can be removed.

So we basically have two types of USB Interface for audio ASYNC and Adaptive. With Async you also have PLL generated clocks (not good) or low jitter clocks that are connected at the dac and feed back to the USB controller (TAS1020 or others) and that clock is used to generate the Bit Clock and the Word Clock as well as clock out the data.

As an example here is some jitter numbers I did using the Wavecrest jitter measurement system on Word Clock. I used the Faber Acoustics Oscillator (Fs=44.1K) as the basis for the stream sending a 1KHz Sine wave to the unit under test. FYI Word Clock is always used for jitter measurements as it will always be the worst case and this does not reflect the overall jitter of the system as most current chips do have a fair amount of jitter rejection. But this is measured at the output of the Word Clock for the following chips and or software:

TAS1020 standard Adaptive Word Clock Jitter 2838ps
TAS1020 using my slow Adaptive code 1131ps
TI/BB PCM2706/07 I2S output Word Clock 3743ps
TAS1020 using PLL derived asynchronous output I2S 433ps
TAS1020 using low jitter MCLK input at 11.2896MHz 73.2ps

The main problem with Adaptive is that in a sense you are changing the clock in the USB controller every 1ms. In the TI reference code for the TAS1020 this is set to 4ms which makes it better and why you see and improvement from 3743ps to 2838ps but if the Soft PLL is made to work a little smarter you can see it can be improved even more to 1131ps. But you can really see that removing the 1ms change required to meet the Adaptive USB and going to Asynchronous USB that you really start to get into acceptable territory.

I am not really a fan of upsampler chips because the math units in these are always compromised. The good thing is that the computer upsamplers use floating point and have a ton more processing power and to me always sound better. But as you can see from above, reduction of the jitter going to the dac chip would be a requirement for any of the adaptive USB setups to meet High End standards.

James if you have any questions on USB don't hesitate to ask. I have been pretty open with my information. I think USB is really the only interface capable of High End Audio. Sure Ethernet or any connection will work but then you have to write drivers. Firewire really could have been it but they sat on their laurels for years and now even Apple is taking it off their product line.

Well back to work, thanks
Gordon
Wavelength Audio

Gordon,

Could you please elaborate more on “the use of upsamplers or FIFO reclocking devices only act as like a low pass filter to jitter”?
Does this mean that jitter attenuation with upsampling/reclocking is effective in higher audio frequencies only, while it does almost nothing in lower audio frequencies?
Reason I am asking is that my experience has been that it is nearly impossible to get good bass reproduction with multi-box solutions (source of any kind > DAC, any transport), source must have extremely low jitter, and it seems there is nothing that beats megabuck single box solution in this aspect of performance (player where there is no clock on optical transport).
In other words, relatively low jitter will give you decent higher frequency performance, but you must send extremely low jitter signal to DAC to get good bass performance.

[wavelength]

Sasha,

By low pass I mean that it acts more like a filter, not like a real low pass filter.

For example you can look at most of the USB dacs that have been tested in Stereophile. Most of these have SPDIF inputs also which tested allot better than the USB input. All of the companies that use reclocking FIFO's and upsamplers believe that this approach will rid the jitter from the receiver (be it SPDIF, USB or others) and that the only remaining jitter will be that of the intrinsic jitter of the dac chip. If this were true then why is it that with each of these Adaptive USB/SPDIF dacs is the SPDIF jitter significantly less.

Example... well known dac with SPDIF/USB TAS1020 Adaptive code for USB 24/96 (@44.1) WCLK jitter again 2838ps, SPDIF WCLK jitter 673ps, system jitter for USB was 250ps and SPDIF was 150ps.

If it were true that the upsampler would rid itself of the jitter from the receiver then the system jitter should have been the same, or at least close.

Now the reason I think the damage is done between the receiver and the upsampler is that the WCLK jitter out of the upsampler was almost the same for both SPDIF and USB. But with this drastic of a difference the only place the damage could have been done was between the receiver and the upsampler. That is why I think it is critical for applications to think of jitter at the source were it begins. Not to try and fix it later.

Thanks
Gordon

[ian.ameline]

Gordon,

Excellent posts.

I'm interested when you say that the math units in the resamplers are always compromised. This is likely true, but the compromises are likely not terribly meaningful for this particular application (mainly applying a FIR filter to fixed point data). For software resamplers and audio processing, I usually see them using single precision float for the computation -- this is adequate for 16 and 20 bit samples, but not quite what you need for 24 bit. (remember that float has 23 bits of fraction precision, and while IEEE754 claims no more than 0.5 ulp error, floating point allows for all sorts of problems with numerical stability -- Have a good read of the following http://docs.sun.com/source/806-3568/ncg_goldberg.html -- "What Every Computer Scientist Should Know About Floating-Point Arithmetic" -- and I would add "And what almost None of them Actually Do"). Given the published specs for the Burr Brown resampler that bryston is using, and looking at their patent covering the technique (7262716), it looks like that math is being done in fixed point with at least 28 bits of precision (at one point in their math (sample rate estimation) they're holding 36 bits of precision as the result of the only divide operation in their pipeline). From the math they're doing -- mainly just multiplies, adds and subtracts, this should guarantee no more than 0.5ulp error in a 24 bit integer result, and probably less. This is significantly better than what single precision float would produce. (I'm not such a great believer in using the FP units when you're dealing with fixed point data -- the pitfalls of floating point can be *very* easy for the unwary to fall victim to, and are still all too easy for those knowlegable about FP math to get caught up by as well. So implementing an algorithm like this using double precision floats can still cause problems -- it's best to remain in fixed point for this sort of thing.)

The quality of resampling has been highly variable -- recent versions of windows and OS/X are quite good these days, but older versions (Windows XP in partucular) have been pretty bad. It's a hard problem to get right, and very easy to screw up.

But it looks like TI got it right with the SRC4392. And from listening to my BDA-1, it sure sounds like it as well.

-- Ian.

[bob stern]

Here's a mathematical analysis that supports Gordon's statement that sample rate conversion greatly reduces, but does not eliminate, jitter.  Hence, it is best to minimize jitter at the interface between the source (transport or computer) and DAC. 

Jitter filtering is less effective at low frequencies.  See the graph in Figure 9 and the conclusions on the last two pages, under the heading "AASRCs and Jitter".

The section beginning on page 7 addresses Ian's point regarding distortion caused by interpolation and arithmetic imprecision.  I don't understand the math, but I believe the distortion component due to interpolation is eliminated in Bryston's synchronous SRC.

http://www.iet.ntnu.no/courses/fe8114/slides/upsanddownsofasrc.pdf

[alexone]

James and the rest of the gang,

Bob emailed me yesterday so I figure I would clear up a few things about USB.

First if your willing to write drivers for all os's then USB is wide open. I did this for years and would rather not do it again. Therefore I like to use what is called the Native drivers.

Now there is USB 1.1 and USB 2.0 but really when we talk audio it is Class 1 or Class 2 that really makes a difference. Class 1 was the basic audio specification which includes all rates up to 102,3K sampling rate. Why? Because the Class 1 spec says that an audio frame will happen ounce ever 1ms or greater. The specification says only 1023 bytes per frame is allowed thereby setting the max rate to 102,3K.

First the Class 2 Audio spec is only supported under MAC OSX 10.5.x and greater. There is no present plans to do Class 2 on Windows or Linux, though I am betting within a year from now that this will happen.

The big problem is that to get to xx/192 (I say xx as I am working on the ESS part and have 32/96 working and should have 32/192 soon) you need both High Speed USB 2.0 and Class 2. In the Class 2 audio spec you can now use what are called sub frames. This allows you to basically send a sub frame every 146uS which means you can pass some serious data.

The big difference and something that I have been working on for like the last 5 years is the jitter is not really understood that much. I have been saying the following and testing seems to back it up that the use of upsamplers or FIFO reclocking devices only act as like a low pass filter to jitter. You can pour though JA's testing and see that this is the case in many of the USB input tests the jitter is significantly higher than the SPDIF inputs on the same dac. I believe that the damage is done before the upsamplers and FIFO reclocking systems and that not all of this is removed or can be removed.

So we basically have two types of USB Interface for audio ASYNC and Adaptive. With Async you also have PLL generated clocks (not good) or low jitter clocks that are connected at the dac and feed back to the USB controller (TAS1020 or others) and that clock is used to generate the Bit Clock and the Word Clock as well as clock out the data.

As an example here is some jitter numbers I did using the Wavecrest jitter measurement system on Word Clock. I used the Faber Acoustics Oscillator (Fs=44.1K) as the basis for the stream sending a 1KHz Sine wave to the unit under test. FYI Word Clock is always used for jitter measurements as it will always be the worst case and this does not reflect the overall jitter of the system as most current chips do have a fair amount of jitter rejection. But this is measured at the output of the Word Clock for the following chips and or software:

TAS1020 standard Adaptive Word Clock Jitter 2838ps
TAS1020 using my slow Adaptive code 1131ps
TI/BB PCM2706/07 I2S output Word Clock 3743ps
TAS1020 using PLL derived asynchronous output I2S 433ps
TAS1020 using low jitter MCLK input at 11.2896MHz 73.2ps

The main problem with Adaptive is that in a sense you are changing the clock in the USB controller every 1ms. In the TI reference code for the TAS1020 this is set to 4ms which makes it better and why you see and improvement from 3743ps to 2838ps but if the Soft PLL is made to work a little smarter you can see it can be improved even more to 1131ps. But you can really see that removing the 1ms change required to meet the Adaptive USB and going to Asynchronous USB that you really start to get into acceptable territory.

I am not really a fan of upsampler chips because the math units in these are always compromised. The good thing is that the computer upsamplers use floating point and have a ton more processing power and to me always sound better. But as you can see from above, reduction of the jitter going to the dac chip would be a requirement for any of the adaptive USB setups to meet High End standards.

James if you have any questions on USB don't hesitate to ask. I have been pretty open with my information. I think USB is really the only interface capable of High End Audio. Sure Ethernet or any connection will work but then you have to write drivers. Firewire really could have been it but they sat on their laurels for years and now even Apple is taking it off their product line.

Well back to work, thanks
Gordon
Wavelength Audio

great input. cool forum. best results.

wow.

al.

[ian.ameline]

Here's a mathematical analysis that supports Gordon's statement that sample rate conversion greatly reduces, but does not eliminate, jitter.  Hence, it is best to minimize jitter at the interface between the source (transport or computer) and DAC. 

Jitter filtering is less effective at low frequencies.  See the graph in Figure 9 and the conclusions on the last two pages, under the heading "AASRCs and Jitter".

The section beginning on page 7 addresses Ian's point regarding distortion caused by interpolation and arithmetic imprecision.  I don't understand the math, but I believe the distortion component due to interpolation is eliminated in Bryston's synchronous SRC.

http://www.iet.ntnu.no/courses/fe8114/slides/upsanddownsofasrc.pdf

That's a really good paper covering the basics of sample rate conversion. Anyone who thinks upsampling or downsampling is simple and/or easy should read this -- it should open your eyes. As talking Barbie says; "Math is hard!"

My argument in my previous post was to show that the BB SRC in the BDA-1 has sufficient bits of precision to avoid numerical instabilities affecting the correctness of its results (ie the math units are *not* compromised). But or course,  that correctness also depends on the correct algorithms being used and the appropriate selection of filter characteristics and the generation and precision of the filter coefficients.

As to the distortion caused by resampling -- I believe it will be there whether you have the upsample button on or off on the BDA-1, as the SRC is being used to reclock and attenuate the jitter on the input regardless. But according the the spec sheet for the SRC, it will be at or below -140db -- That's amazingly good.

-- Ian.

[ian.ameline]

Speaking of jitter attenuation, Stereophile has reviewed and measured another external dac that uses the BB SRC4392, and found 76.1ps jitter with 24 bit data, and 149ps with 16 bit data. Again, extraordinary numbers...

I see no reason to believe that the BDA-1 will produce significantly different measurements for jitter. (Other measurements will be different due to different dacs, output stages and power supplies)

http://www.stereophile.com/digitalprocessors/208mfx/index4.html

I'm looking forward to read what John Atkinson has to say about the BDA-1. (James, are you guys going to get Stereophile to review the BDA-1?)

-- Ian.

[James Tanner]

Speaking of jitter attenuation, Stereophile has reviewed and measured another external dac that uses the BB SRC4392, and found 76.1ps jitter with 24 bit data, and 149ps with 16 bit data. Again, extraordinary numbers...

I see no reason to believe that the BDA-1 will produce significantly different measurements for jitter. (Other measurements will be different due to different dacs, output stages and power supplies)

http://www.stereophile.com/digitalprocessors/208mfx/index4.html

I'm looking forward to read what John Atkinson has to say about the BDA-1. (James, are you guys going to get Stereophile to review the BDA-1?)

-- Ian.

Hi Ian,

Yes Stereophile has had one now for about a month.

james

[drubin]

I read Charles Hansen's wonderful white paper last night, in which he explains how one-box players generally avoid jitter as an issue.  (I knew that, but he made the case so clearly and convincingly that it got me thinking.)  From a jitter perspective, at least, the BCD-1 should therefore sound better than the BDA-1.  (I'm assuming the DAC and analog output stages of the two are fairly similar, but perhaps that's not the case.)  Those who have heard them both -- what do you think?

[miatadan]

Hi

I work for audio shop that sells Bryston BDA-1. As expected build quality and sound first class.
I noticed many different types of inputs for this dac. Is there a reason ethernet connection not available such as used in Linn Akurate DS, Majik DS or Logitech Transporter Dac?
I like to leave laptop at desk other end of room connected to dac. As far as I know usb cables not good very 30 ft length but ethernet connection not affected by distance.
Any thoughts on this?

Thanks

Dan

[EDS_]

Speaking of jitter attenuation, Stereophile has reviewed and measured another external dac that uses the BB SRC4392, and found 76.1ps jitter with 24 bit data, and 149ps with 16 bit data. Again, extraordinary numbers...

I see no reason to believe that the BDA-1 will produce significantly different measurements for jitter. (Other measurements will be different due to different dacs, output stages and power supplies)

http://www.stereophile.com/digitalprocessors/208mfx/index4.html

I'm looking forward to read what John Atkinson has to say about the BDA-1. (James, are you guys going to get Stereophile to review the BDA-1?)

-- Ian.

Hey Ian,

I have a Musical Fidelity XDAC V-8.  And love it - espcecially via USB.  For reasons James may or may not want to opine about the BDA - 1 is just a better sounding machine.  I've got one for a few more days.  As soon as I feel more confident I will talk about my time with a  BDA - 1 in more detail.

[phurbag]

Hi James,

Are there any plan for an upgrade of the BDA-1 USB interface in the next months (to handle files of res higher than 48/16)  ?

Phurbag

[James Tanner]

Hi James,

Are there any plan for an upgrade of the BDA-1 USB interface in the next months (to handle files of res higher than 48/16)  ?

Phurbag

Hi Phurbag,

No immediate plans.  We do use the USB currently with the I2S input but we are going to hold off until the USB chips support the 192/24 that the rest of the BDA-1 inputs support.

james

[95Dyna]

Hi James,

Are there any plan for an upgrade of the BDA-1 USB interface in the next months (to handle files of res higher than 48/16)  ?

Phurbag

Hi Phurbag,

No immediate plans.  We do use the USB currently with the I2S input but we are going to hold off until the USB chips support the 192/24 that the rest of the BDA-1 inputs support.

james

Hi James,

Will it be possible to upgrade the USB connection on existing BDA-1's when the 24/192 chips are available?

Bill

[James Tanner]

Hi James,

Are there any plan for an upgrade of the BDA-1 USB interface in the next months (to handle files of res higher than 48/16)  ?

Phurbag

Hi Phurbag,

No immediate plans.  We do use the USB currently with the I2S input but we are going to hold off until the USB chips support the 192/24 that the rest of the BDA-1 inputs support.

james

Hi James,

Will it be possible to upgrade the USB connection on existing BDA-1's when the 24/192 chips are available?

Bill

Hi Bill,

That's the plan.

james

[95Dyna]

Thanks James.  I couldn't tell if phurbag was referring to upgrades to going forward BDA-1's or to retrofits as well.  That's great news for those of us who are in the decision process for new digital sources.

[Wig]

 Guys,

 Just received my BDA-1 and would like to replace fuse with a hi-fi fuse but can't seem to find a allen or hex that fits. Anyone know what tool will work?

 Thanks,
 Wig

[bob stern]

Just received my BDA-1 and would like to replace fuse with a hi-fi fuse but can't seem to find a allen or hex that fits. Anyone know what tool will work?

Do you mean the top plate screws?  Looks like you need a Torx driver.  But how do you know there's a standard fuse cartridge inside?  Could be an unusual shape, or even soldered.  You'd better wait for James' advice.