Asynchronicity: A USB Audio Primer

Hi Juergen, thanks for all this. One small addition :<br />

<br />

<strong>For XP, I get only bit true, if a software offers ASIO out and I am using for this hardware written ASIO drivers or the ASIO4ALL wrapper. Without ASIO out (or KS out) I am getting no bit perfect out on native USB cards, and with no, I really mean no.</strong><br />

<br />

I know you said "USB cards", but since you're into RME ... RME's MME drivers are bit perfect for XP.<br />

But ... what about the Fireface USB version ? that would be an interesting twirl.<br />

Notice that MME does not exist for Vista.<br />

<br />

Peter<br />

<br />

<br />

@PeterSt: You are right, RME cards are bit true with MME and with ASIO drivers. I have three of them, in my older computers. The drivers work rock solid, this is a big plus for RME, but for the reason that the jitter is slightly too high for me and also the distortion on the analog out, I haven't bought any more in the last 3 years or so.<br />

<br />

@Audiozorro: I am sorry that I have to inform you that Brint (I do not know the correct spelling), the international sales manger of Ayre, has set back the Audio Midi Setup of the Mac Mini to 44.1 kHz. Is this now good to have Bit True or should he better stay with 88 in the display to have better luck in China?<br />

<br />

@riderforever: I have started with some measurements under Linux Ubuntu 9.04 and under Ubuntu Studio. The 16 Bit where fine, but for 24 Bit I haven't had enough time, so I stopped it here, because even only with XP, Vista, Tiger and Leopard, it takes a lot of time to verify all what is said in different post, if this is right or wrong.<br />

<br />

@Gordon: The 24 Bit walking zero signal is nearly perfect to measure Bit correctness, because the source signal is known and you are walking through the complete dynamic range of the digital system. The 16 Bit digital DC signal is perfect to look at what is happening with the bits does give only correct information if your system is 24 Bit (if you have a 16 Bit hardware BB PCM270x etc., they cut of the 17 – 24 Bit) so in this case even native Wave Out or native Direct Sound out looks correct in the 16 Bit world. But with a combined signal, as I described some pots earlier, I could check everything an one time, together with channel swapping. Concerning timing I have made also some other post, where I measured the jitter difference between native Wave Out (higher random jitter), native Direct Sound Out (higher discrete jitter) and “native” ASIO4ALL out or exclusively WASAPI out (as low as the hardware design can get), which is really clearly visible and I can tell you from the graph what native driver mode is used.<br />

<br />

So if someone set up a computer with a RME card and the DigiCheck software and use the digital in as a “slave” (not master mode), and play back the DC / Walking Zero signal from the host computer via digital out will be able to see whether the play back chain (software player and setting) is bit perfect or not, or use a dedicated measurement system like Audio Precision or similar.<br />

<br />

Juergen<br />

<br />

PS @Gordon: I visited Larry Key, a long time ago, at the Fi times, in Sausalito, and the funny thing was, that he lived next door to James Hetfield (Metallica), so was this the lucky number 8?<br />

<br />

PS @PeterSt: For Firewire I have only a KRK Ergo and this has only analog outputs, so I am sorry, that I can't make any digital Bit True tests.

Actually 4 is not from the Gang of four. It predates them. 4 in Chinese is the same sound as the word for "death." So 4 is avoided. 8 on the other hand is half of the character for "happiness". Therefore 88 is even better, the full character for happiness. In Hong Kong, people bid for auto license plates with lucky numbers. My father-in-law had AA 717, which was considered lucky, because 7+1=8 and 1+7=8, so he had a surrogate for 88. I suppose 176 would be even better, since it is 88 +88, double happiness. <br />

<br />

Larry

Jeez Chris,<br />

<br />

Things have got very technical on computeraudiophile in recent months!<br />

<br />

Seriously, there is a phD in probably all of these separate issues. Perhaps dozens of phD's actually. Is there a Berkley or Stanford or MIT department of electronic engineering looking at any of this stuff? What we need is academics with a non commercial interest to look at all this. Also throw in some people who have a sound background in audiology, statistical bias and how to design a proper double blind trial. Non commercial universities are the answer Chris.<br />

<br />

Aren't we getting a bit off the track...there are many more practicle issues in choice of DAC interface/software/computer chip etc...<br />

<br />

...such as how does the human use the computer. Where is the audio gear positioned in the room? Are the DAC's available? How robust is the interface? How easy is it to access the digital files? Does it all crash etc etc........Nobody say's the ipod is the best electronics interface to play computer audio (far from it), but I bet it is the most popular device in the whole world right now for doing just that.... <br />

<br />

Why don't we leave the electronics to academics, and the practical stuff to the commercial marketplace?...which will soon sort out which interface/software/etc is better for everyone. When the universities produce the papers and they are scientifically reviewed, we will have our answers to the more complicated stuff! Forum members need to publish references from scientific journals if they wish to support their cases. It's getting that complicated.<br />

<br />

Just my thoughts Chris. please don't take offence....

From Computer Audiophile: "<cite>Hi Eric - Thanks for bringing up the fact that I spoke about Wavelength products and used Gordon Rankin as one source of data about USB Audio. It's critically important to keep this in the forefront. This is why I made it clear in the article who I used as sources of data during my research. Also, it's great that all the readers keep me honest by leaving comments on anything that may seem improper. It is almost impossible to research Asynchronous USB thoroughly without talking to Wavelength Audio. <br />

<br />

Again, I get your point 100% and it's always good to discuss it when reading any article.</cite>"<br />

<br />

I was also curious why the only 'expert references' you used for this article are companies that make asynchronous USB products. Perhaps you should have spoke with experts who implement the adaptive mode to balance the information you received to write this article.<br />

<br />

All the best,<br />

Elias

I've bee the owner of a Benchmark DAC1 Pre for about a year. I originally got it just as a DAC for CD. I soon discovered it was useful for playing ripped cd's from my XP computer and found that very beguiling. In your piece you have the following statement:<br />

<br />

"Another less common adaptive USB implementation is done using a TAS1020 chip. Manufacturers then have a choice of implementing the chip exactly like the PCM270x without additional programming or possibly using the example code provided by TI, or the manufacturer can purchase code from CEntrance, Inc. to use with the TAS1020. Popular devices using the CEntrance code are the Benchmark DAC1 variants, Bel Canto USB Link, and the PS Audio Perfect Wave DAC. Using the TAS1020 and CEntrance code greatly enhances the USB interface and allows native 24/96 playback without the need for additional device drivers or special software."<br />

<br />

The Benchmark documentation goes to great lengths to assert that their implimentation completely eliminates jitter. Your comments in the paragraphs following sugguest that this is not, or at least, may not be true, and that any non asyncronous design will have jitter. Could you comment on this please.<br />

Also, is the Benchmark design one of those that "Some listeners report as a Hi-Fi type of sound that is initially impressive, but long term listening may confirm otherwise." I've not noticed long term listening fatique, but I would like your opinion.<br />

<br />

paugust<br />

<br />

<i>"The Benchmark documentation goes to great lengths to assert that their implimentation completely eliminates jitter. Your comments in the paragraphs following sugguest that this is not, or at least, may not be true, and that any non asyncronous design will have jitter. Could you comment on this please."</i><br />

<br />

<br />

Hi paugust - I think even Benchmark would agree that it's impossible to completely eliminate jitter and the use of that phrase is used subjectively. Any non-asynchronous design will have jitter. So will any asynchronous design. Adaptive and asynchronous are two ways to do USB and each implementation varies greatly from manufacturer to manufacturer. <br />

<br />

<i>"Also, is the Benchmark design one of those that "Some listeners report as a Hi-Fi type of sound that is initially impressive, but long term listening may confirm otherwise." I've not noticed long term listening fatique, but I would like your opinion."</i><br />

<br />

I'm sure some listeners of the DAC1 agree with that statement and many others do not. If you haven't noticed any fatigue then you're in a good position.

<i><br />

"Any non-asynchronous design will have jitter. So will any asynchronous design."<br />

</i> <br />

<br />

Yes, but well implemented asynchronous can have lower jitter than async ever can and just makes better sense.

The TAS1020B implementation, as in the Benchmark, does have some severe jitter at the TAS output. This you can measure at the chip output itself, or you can read the CEntrance implementation.<br />

<br />

But at the TAS output, the Benchmark uses an ASRC (asynchronous sample rate converter) with a fixed crystal oscillator on the output to re calculate the digital data at a different and fixed sample rate.<br />

<br />

With this process, you do really remove the jitter, because you create a totally new digital data stream, but the drawback is that this output stream is far from bit true for ever.<br />

<br />

So some people like this method of ASRC because they get very low jitter and are highly immune against input jitter, but they are also some people, that couldn’t live with the sort of clean, tending to sterile sound of ASRCs devices. So it is up to everyone’s preferences.<br />

<br />

I hope this helps a little bit for clarification and I am sure, if you want to need more detailed information, the Benchmark team will help you out.<br />

<br />

Juergen<br />

Thanks to both Chris and Juergen for your quick replies to my Newbe question. As I look through your CA Academy, Chris, I am begining to undersand more and more, but, in the process come up with more and more questions, as I try to decide on how best to move into a CA system. I hope to get some advice on that topic, but I should probably post such questions on a different part of the forum.<br />

<br />

Juergen, in your reply, you state: "With this process, [using a asyncrhonous sample rate converter], you do really remove the jitter, because you create a totally new digital data stream, but the drawback is that this output stream is far from bit true for ever."<br />

<br />

Are you saying that any time you convert to a different sample rate, (even if your upconverting), that you loose the status of bit perfect? If so, this makes sense to me, but I would think the converted bit stream would be as good a representation of the music as the original, (again as long as your upconverting). In other words, no information would be lost so why wouldn't the result be just as good? Or, am I missing something? Also, do you have a theory as to why ASRC devices should sound clean and sterile to some while Asynchronous USB converters, presumably, would not? <br />

<br />

This is a great forum. Thanks in advance for your reply<br />

<br />

Paugust<br />

<br />

<br />

<br />

<br />

<br />

I will try to make it very simple and easy, to explain the difference between oversampling (integer number) and up sampling (fixed output rate). I know this explanation will be easy, so I hope that not too many will chime in and try to explain in a more detailed way, what was wrong with my very simple post.<br />

<br />

A lot of, or nearly every DA converter, does oversample the signal, in order to get rid of the alias frequency and have to use only a very soft analog output filter. With this method, in the first digital filter stage, the original data still remains, and there is “only” added some calculated signals between the original data, depending on the used digital filter. So still the original data are mostly there as an “anchor”.<br />

<br />

With an up sampling, (ASRC) to a fixed output rate, you take the original data and calculate a complete new set of data. With this method you loose your “anchor” in the audio band. Sure there are big differences between different ASRCs and the newer once do measure and sound better than the older ones, but the point that you do calculate all data totally new, and have no original anchor points at the output, does makes the difference.<br />

<br />

Juergen<br />

<br />

PS: All DSP guys, please be patient with my answer given to a “Newbe”. I do not want to write an DSP compendium. Thank you.<br />

Juergen,<br />

<br />

Another thing to note with an ARSC is that these devices are usually between the system receiver (be it SPDIF, USB, Firewire whatever) and the DAC chip. Many of these do complex fixed math with only a 24 bit result. Meaning a lot of the low level information is getting thrown out.<br />

<br />

With the oversampling filters inside the dac, which most of the math is the same it is a little different as the dac processing can use a wider word to output the data without losing the information in the math.<br />

<br />

Most of this math is pretty simple table driven multiplication, summation stuff but you have to remember that it has to be done very quickly so a lot of the time the precision is dumped for faster processing.<br />

<br />

Thanks<br />

Gordon

Gordon, are you saying dithered 24-bit DSP will throw away audio information that was originally present in a dithered 16 or 24-bit recording? <br />

<br />

Can you name a single A/D or D/A that gives true 24-bit performance? The highest performing (32-bit) chip that I know of will actually only acheive about 21 bits of performance (dynamic range). So, with a properly dithered 24-bit ASRC, I can't understand how "a lot of the low level information is getting thrown out." Even if it was 32-bit ASRC....or 64-bits for that matter, no D/A chip on the planet will have sufficient dynamic range to maintain that low level information. Also, the best dynamic range from an A/D will max at 21-bits...so if the original recording didn't have 24-bit performance, why would a 24-bit ASRC throw out "a lot of the low level information?" <br />

<br />

All the best,<br />

Elias<br />

<br />

From Computer Audiophile: "I think even Benchmark would agree that it's impossible to completely eliminate jitter and the use of that phrase is used subjectively. "<br />

<br />

I'm not sure I understand what you are saying here. Why is it impossible to completely eliminate jitter? At what stage? At the USB receiver? At the DAC chip? <br />

<br />

Best,<br />

Elias

Elias,<br />

<br />

No 24 or even 32 bit ADC or DAC get's close to the signal to noise ratio of the data. But many agree that making sure the data is true is a step closer to truer output.<br />

<br />

Think of it this way... many phono systems have less signal to noise ratio than a 16 bit dac does. But of course we know that analog smashes most 24 bit system in their reproduction.<br />

<br />

But let's look at how an ARSC works and how the low level detail is thrown out. All digital filters work on a circular table system. There are two tables the first are the samples at time t, t-1, t-2, t-3 and so forth for the length of the table. This is usually a circular buffer and the pointer advances instead of moving the data and therefore t-X where X is the size becomes t for the next sample. The other table is the coefficients. Each one of the coefficients are less than one or a fraction. Since the beginning when Fourier and Laplace (early 1800's math, wonder what they were thinking???) the coefficients resolution will determine the true output of the filter, in this case the upsampler.<br />

<br />

BUT wait we are dealing with fixed math, in most cases the ALU is only 24 bits so how does this work and retain the information that we want to keep? Well it doesn't... it basically uses fixed coefficients and when they are multiplied by the sample the remainder is thrown away and then accumulated with the other sample times their correlated coefficient.<br />

<br />

In regards to jitter rejection, yes jitter cannot never be fully gotten rid of. Most companies assume that jitter is only effected at the dac chip. But what happens if the jitter is so high going into the ARSC that the error happens here. Because what is the difference between the protocol on the input to the ARSC and the DAC?<br />

<br />

Also companies really don't understand how jitter elimination works. It's funny but they all seem to think that the jitter output is a function of only the new asynchronous master clock. But this is not correct... Jitter reduction is like a filter, only some of the jitter is removed the more you have on the input side of the ARSC the more that comes out.<br />

<br />

This is easily seen Elias even in the measurements Stereophile has made on the DAC1. If truly the ARSC got rid of all the jitter and only the intrinsic jitter of the ARSC/DAC and new asynchronous master clock. Then the jitter measurements for the SPDIF and USB would be the same. But they are not....<br />

<br />

The ARSC in your unit does a great job of lowering the overall jitter and many I am sure find the output exceptional. I just personally do not like the sound of any of the ARSC that I have ever listened to. This is what set me on my journey to do the asynchronous USB (not to be confused with ARSC) as I did not feel that Adaptive gave me low enough jitter to make a reference system from.<br />

<br />

But these are just concepts of what companies do. This is why Benchmark sounds different than Wavelength. A consumer really should listen to any device before making a decision as to which they should buy. As ARSC and Asynchronous USB are only a few things that make up the sound of a unit.<br />

<br />

You as the consumer will only know if you listen which is right for you.<br />

<br />

Thanks<br />

Gordon

The CS8421 ASRC is pretty good. It is working internally with 32 Bit, can dithering down at the end to 24 Bit. When using a good crystal oscillator and a very power supply, the jitter at the output is extremely low, but still some sonic effects, some more positive, some more negative.<br />

<br />

When measuring this device, I have no distortion component within the original audio band, measuring down to – 172 dBFS. Out of the original audio band, between 22 k and 48 k I have some uncorrelated distortion lines, 5 in the area of – 140 dBFS and 2 tiny in the area of – 160 dBFS.<br />

<br />

But still, what I have said above about calculation a totally new set data set is still true.<br />

<br />

The different working modes are audible, even with this extremely small uncorrelated out of band distortion. Comparing Bit True mode (no SRC, no digital filter, no jitter reduction), slave through mode (no SRC, but digital filter, no jitter reduction) and ASRC mode (ASRC with digital filter and jitter reduction).<br />

<br />

Juergen<br />

Rather than diss the level of discussion, revel in being able to listen in. I am a PhD and an expert in my area. So listening in to experts in another area is wonderful and a great learning experience.<br />

<br />

G

What about us po' folks? A V-Link is as far as I can go with all this. My DAC's got a TI PCM2902 to change USB to SPDIF at 16/44 only. No high rez through there! Coaxial and TosLink 24/96. V-Link should work, right?