Optimising digital playback by tweaking transport: fact or fiction

[hochopeper]

One thing I've been thinking about lately ... often linear PSU's are based on regulator circuits that have a PSRR and noise bandwidth performance that drops away around audio frequency (often around 10kHz). Once into the MHz and even 200kHz or so it is likely that local capacitance and circuit layout far more impact on technical performance of the circuit. Some will use passive filters after the rectifier before the regulator, but I wonder if some of the newer regulator chips that are designed for wide bandwidth regulation might allow even better performance by reducing noise in these frequencies from 10kHz up to 200kHz or so. LM317, LT108x LM338 etc all present data for noise handling that falls off a cliff around 10kHz. Input filters on PSU may well deal with a lot of this, but I still wonder. Will be trying to get some measurements at some stage this year to explore that theory further.

[kdoot]

Any examples/good recommendations? And any tweaks you've used?

My preference is to start with a good overall design and leave tweaks for later. 

 

Things to look for:

- Asynchronous USB

- Lots of attention paid to clocks and clean power from the USB receiver onwards

- Dual clock crystals, one for 44.1/88.2/176.4 and one for 48/96/192

- Buffering can be helpful if well implemented

 

Things to try and avoid if you can:

- Relying on power from the computer's USB port

- Asynchronous Sample Rate Conversion (ASRC)

 

That last one is a key point with computer audio. We have these massively powerful CPUs in our computers and yet we often try and minimise their load while asking a dinky little embedded circuit worth a couple of bucks to do oversampling and digital filtering. The worst is when we rely on that as a jitter-reduction mechanism but even just as a general principle it's kind of backwards.

 

In my system the oversampling and filtering happens in the computer using Audirvana and the IZotope SRC. I only use 2x/4x oversampling because it avoids the error/noise involved in converting between the 44.1 and 48kHz frequency families. My DAC has no processing at all, so I can have full control over that process in software. But even on a DAC which has internal filtering, you can get some good results by pushing the incoming data stream up to a higher rate with quality filtering in your PC. 

 

176.4 or 192kHz audio data goes to my Audiophilleo which is an asynchronous USB device with two high quality clocks in it. Moreover, the clocks and the SPDIF coax output stage are both powered by a dedicated battery supply which gets isolated from the USB section during playback. The Audiophilleo SPDIF driver circuit is unique: it uses differential emitter-coupled logic circuits to produce a super clean, crisp and stable SPDIF signal. From there I could go to just about any DAC with a coax input and be confident I'm getting performance right at the top of that DAC's capability.

 

Other similar products include the OffRamp and new D-to-D converters from Belcanto and Wyred4Sound. I'm personally not a big fan of the HiFace products but lots of people use them to quite good effect, especially when they're internally-connected and using the I2S output instead of SPDIF.

 

Beyond that, there are DAC makers doing it all themselves. On the low end, Audio-GD (as always) rates a mention for their USB32 interface which supports 384kHz sample rates over USB. They've partnered that with the ESS SABRE 9018 (I think?) configured for tight tolerance on the jitter side, and of course they always emphasise their power supplies. Further up the chain there's Classe with their CP-800: everybody really should take a close look at their USB and clock implementation to get a feel for how this kind of stuff ought to be done. Sadly I've not had a chance to hear either of those products but I like they way they're approaching computer audio.

 

And ideally, if you get those fundamentals right, the need for tweaking is much reduced. The only thing I think about changing these days is some of the iZotope parameters. 

[davewantsmoore]

(6 or 8ch) i2s over LVDS with two RJ45

 

No doubt you're aware of the twistedpear "teleporter module" ?!

[davewantsmoore]

That last one is a key point with computer audio. We have these massively powerful CPUs in our computers and yet we often try and minimise their load while asking a dinky little embedded circuit worth a couple of bucks to do oversampling and digital filtering. The worst is when we rely on that as a jitter-reduction mechanism but even just as a general principle it's kind of backwards.

 

In my system the oversampling and filtering happens in the computer using Audirvana and the IZotope SRC. I only use 2x/4x oversampling because it avoids the error/noise involved in converting between the 44.1 and 48kHz frequency families. My DAC has no processing at all, so I can have full control over that process in software. But even on a DAC which has internal filtering, you can get some good results by pushing the incoming data stream up to a higher rate with quality filtering in your PC. 

 

Yes!

 

I would take this further, and say that this  (taking away oversampling, etc. from the DAC)  is one of the main reasons *** that higher sample rate audio can sound better.

 

It's not that higher sampling rates sound better themselves .... it's that the performance of oversampling/filtering in many DACs can be bettered by a transport which oversamples in a clock domain separate from the output  (ie.  not in 'realtime')

 

An extreme example of this is software audio filters built into players such as xxhighend or hqplayer ... which are much more advanced than what is (practical) possible with hardware solutions.

 

 

***  The other big reason being that some content recorded or distributed in higher sample rates had more "care" taken with it in recording/production.

[henry218]

Yes!

I would take this further, and say that this (taking away oversampling, etc. from the DAC) is one of the main reasons *** that higher sample rate audio can sound better.

It's not that higher sampling rates sound better themselves .... it's that the performance of oversampling/filtering in many DACs can be bettered by a transport which oversamples in a clock domain separate from the output (ie. not in 'realtime')

An extreme example of this is software audio filters built into players such as xxhighend or hqplayer ... which are much more advanced than what is (practical) possible with hardware solutions.

*** The other big reason being that some content recorded or distributed in higher sample rates had more "care" taken with it in recording/production.

thats interesting, i read from other threads that you set the ref5 with oversampling i forgot which but i think it was 2x. was it because u still using cdp? if you only use computer source would you prefer nos?

im using i2s now with cdp, its just better than AES/EBU and standard coax, although its a 1m cable (kdoot mentioned that it should be shorter ?) will i gain anything with shorter cable?

[hochopeper]

No doubt you're aware of the twistedpear "teleporter module" ?!

This is a new design I'm working on ...

Spaghetti of modules is not a proper implementation for high speed signals.

Edited January 28, 2014 by hochopeper

[kdoot]

thats interesting, i read from other threads that you set the ref5 with oversampling i forgot which but i think it was 2x. was it because u still using cdp? if you only use computer source would you prefer nos?

I've recently been experimenting with a DAC-19DSP. Didn't even bother trying the built-in USB for reasons as discussed earlier. The SPDIF input is limited to 96kHz because it's a DIR9001 chip. So I could only do 2x upsampling in my computer. I tried the DAC-19 in NOS mode with 88.2/96kHz input and it was OK. I find with my Metrum that you do get a better result at 176.4/192 though. Re-enabling oversampling on the DSP-1 sharpened things up but that DSP board combined with the DIR9001 seems to have a pretty strong effect on the sound and my computer-based efforts weren't improving things by much.

im using i2s now with cdp, its just better than AES/EBU and standard coax, although its a 1m cable (kdoot mentioned that it should be shorter ?) will i gain anything with shorter cable?

Shorter *could* be better, but as with so many things - "it depends". The ideal design has the clock in very close proximity to the DAC chip. But since you're sending the clock over a cable you have to watch for reflections and interference and attenuation and so on. Not my field of expertise though, maybe Chris or Dave will have something more useful to add.

[Green Wagon]

What dac? 

 

Sorry, my bad.

 

First dac I used was a beresford camain, on this I got the results of coax > optical > usb.

My current dac is a beresford bushmaster (v1).This doesn't have a usb input, only optical and coax.

With no other changes, the bushmaster is superior to the camain in all areas. (No usb, so cant test that.)

 

Yes both very 'budget' dac's, one day I'll be able to sample (ha, pun) something a little more, upmarket...

[hochopeper]

im using i2s now with cdp, its just better than AES/EBU and standard coax, although its a 1m cable (kdoot mentioned that it should be shorter ?) will i gain anything with shorter cable?

 

Need more info ... What is driving the i2s cable? What sort of cable? What is the DAC and how is it's i2s input configured? 

Edited January 29, 2014 by hochopeper

[hochopeper]

Having just re-read the OP I have some stuff to add from a technical perspective, will take some time to write up though to make it coherent ... will be back later with some more relevant post(s) from a technical perspective.

Edited January 29, 2014 by hochopeper

[Guest myrantz]

- Asynchronous USB

What's the advantage of async over sync in the context of audio? Support higher bitrates? Does async give better sound? If possible, are you able to describe it.

 

Don't think computer can oversample (that's a hardware thing) but computers can upsample. And on that part you may have something there - coz the higher the sample rate, the smaller oversample mode you need (at least with the STX card). Curiously this option of changing the OS mode was exposed previously, but is now hard coded dependant on the sample rate (IIRC, been a while).

 

I can only assume using a different mode will affect the performance somehow (perhaps the filter of the sound card expect the same number of bits everytime?)..

 

Thanks for the detailed post, it's very helpful.

 

Sorry, my bad.

 

First dac I used was a beresford camain, on this I got the results of coax > optical > usb.

My current dac is a beresford bushmaster (v1).This doesn't have a usb input, only optical and coax.

With no other changes, the bushmaster is superior to the camain in all areas. (No usb, so cant test that.)

 

Yes both very 'budget' dac's, one day I'll be able to sample (ha, pun) something a little more, upmarket...

Hey my Asus STX with Burson OpAmps is cheaper than yours, and yet here I am mixing with you guys ...

 

I did consider more upmarket solutions, e.g. the Marantz NA11S1. So it's 1792A vs DSD1792, Burson vs HDAM SA2, computer SMPS vs hifi PS. But seems the STX is still better to me.

 

I never have much success with USB myself, but that's because I have a ground loop (and somehow USB connection has a noise that's worse than a coax one).

[henry218]

Need more info ... What is driving the i2s cable? What sort of cable? What is the DAC and how is it's i2s input configured?

its audio gd master7, the cable is blue jeans's rca to bnc to cec tl51xr.

not sure how the i2s configured, but i can hear a significant improvements over coax and aes/ebu, especially on musical instrument

[hochopeper]

its audio gd master7, the cable is blue jeans's rca to bnc to cec tl51xr.

not sure how the i2s configured, but i can hear a significant improvements over coax and aes/ebu, especially on musical instrument

 

how is the cectl51xr's i2s connection made?

[kdoot]

What's the advantage of async over sync in the context of audio? Support higher bitrates? Does async give better sound? If possible, are you able to describe it.

It's about which end of the chain is in control of the rate of data transmission. With isochronous or synchronous USB, the audio device has to adapt its timing to keep pace with the rate of data transmission as determined by the computer. This is bad, because the USB controller in a computer is built around a 12MHz crystal and 1ms timeslicing which have no relation whatsoever to 44.1 or 48kHz audio sampling frequency groups. The result is lots of jitter.

With asynchronous USB audio, the audio device has a small buffer and it signals to the computer when it's ready to receive another burst of data to keep the buffer from being emptied. The rate of emptying the buffer can be controlled by an audio-optimised clock.

So the difference is all about the clocking. With async, a stable and audio-friendly clock down at the DAC end is in control. Otherwise your audio devices are flapping about all over the place trying to restore a smooth, regular flow to data tossed out by the PC at highly irregular intervals. They can do a remarkable job, but it's better not to have to.

 

Don't think computer can oversample (that's a hardware thing) but computers can upsample.

From my reading, I don't think that's a meaningful distinction. But specifically, I mean take the common CD audio at 44.1kHz and mathematically process it to produce a new data stream at a higher rate (176.4kHz in my case) with digital filtering shifting any noise above 22.05kHz up into the truly ultrasonic ranges. Also, I prefer a minimum-phase filter with no pre-ringing for use with my particular DAC, and software gives me that kind of flexibility.

 

And on that part you may have something there - coz the higher the sample rate, the smaller oversample mode you need (at least with the STX card). Curiously this option of changing the OS mode was exposed previously, but is now hard coded dependant on the sample rate (IIRC, been a while).

Not sure what you're talking about here. My DAC is a NOS DAC. There is no further processing performed beyond what I'm doing in software. No "sound card" involved either, unless you want to think of my Audiophilleo that way.

 

I can only assume using a different mode will affect the performance somehow (perhaps the filter of the sound card expect the same number of bits everytime?)..

Oversampling, as I understand it, is a mathematical process. My suggestion is that a modern PC's CPU can probably do a better job of it than the tiny embedded circuit in most DAC silicon. Even if your DAC is the type that upsamples everything to a really high rate like 384 or more internally, having performed the critical filtering in software first might deliver a better result than trusting the DAC to do everything.

 

Thanks for the detailed post, it's very helpful.

Well, you're welcome

[hochopeper]

From my reading, I don't think that's a meaningful distinction. But specifically, I mean take the common CD audio at 44.1kHz and mathematically process it to produce a new data stream at a higher rate (176.4kHz in my case) with digital filtering shifting any noise above 22.05kHz up into the truly ultrasonic ranges. Also, I prefer a minimum-phase filter with no pre-ringing for use with my particular DAC, and software gives me that kind of flexibility.

Oversampling, as I understand it, is a mathematical process. My suggestion is that a modern PC's CPU can probably do a better job of it than the tiny embedded circuit in most DAC silicon. Even if your DAC is the type that upsamples everything to a really high rate like 384 or more internally, having performed the critical filtering in software first might deliver a better result than trusting the DAC to do everything.

 

 

My understanding agrees with yours, they're analogous processes. However it seems to be a convention that the up/over sampling filter inside the DAC chip is called oversampling filter and the up/over sampling filter in the computer is called upsampling filter. In reality they are both a DSP filter and are comprised of an IIR and/or FIR filter. One option that has been less well explored is that some DAC chips actually allow you to program your own filter coefficients as well as swapping between some pre-defined filter designs, this hasn't been exposed to the consumer on many (any?) products but is something I'd like to see - along with a more robust/objective method for comparing filters within the context of the consumer's setup and allow them a simpler way to compare the filters and select their preference without needing to organise a friend to come around or a partner to sit patiently through your comparison(s).

[davewantsmoore]

This is a new design I'm working on

 

10-4 ... I did not necessarily mean to use, as much as to 'copy'

[hochopeper]

10-4 ... I did not necessarily mean to use, as much as to 'copy'

From the looks it is a simple circuit they've implemented there, not much more than what would be in the IC datasheet ... 

[Guest myrantz]

I'm more confused about everything now.. (Esp the sync/async USB part)...  Learning process.. Cheers guys and thanks for the info.

[henry218]

how is the cectl51xr's i2s connection made?

thats the thing, from CEC is SPDIF coax then go to M7 i2s, many on the headfi forum experienced improvements over coax, and i can hear the improvement easily although not day and night.

i wonder why...

[davewantsmoore]

thats interesting, i read from other threads that you set the ref5 with oversampling i forgot which but i think it was 2x. was it because u still using cdp?

 

Yes, that's right.    I use TV and BDP as other sources than the computer.   Also the AudioGD DSP1 unit in that DAC uses a reasonably sophisticated DSP  (although 'realtime')  so it is on par with better transports/DACs in that regard.... and internally it over samples to higher rates which I can feed it (even from the computer), so 2x works best.

 

In my main system right now, I use a MiniDSP which works internally at 24/96, and sends all inputs to ASRC if not 96khz.    I use DSP (same as kdoot) in the computer to convert all to 24/96, so to bypass (or at least not tax) the ASRC inside the minidsp.   It does sound noticeably better.

 

 

 

im using i2s now with cdp, its just better than AES/EBU and standard coax, although its a 1m cable (kdoot mentioned that it should be shorter ?) will i gain anything with shorter cable? 

 

It is difficult to generalise about what you'll find better cos there are so many variables.

 

Suffice to say that I2S wasn't designed to be used in the way you are.   You cannot be sure of the termination requirements (excuse any potential oversimplification there Chris) for either end unless you're able to measure the digital signals  (ie.  you know how to comprehend what a 'scope shows you).

 

Making the cable short wouldn't hurt.

 

It's like Chris mentioned before.   The audiophile idea that "I2S is superior" has a lot of ifs, buts, and technical requirements, packed away inside it.    I'm not saying it isn't working better for you.... but isn't anywhere near as straight forward as most people discuss it.

[davewantsmoore]

My suggestion is that a modern PC's CPU can probably do a better job of it than the tiny embedded circuit in most DAC silicon

 

Most definitely.  

 

To elaborate for those watching on ---- An 'embedded circuit'  (unless it is also performing reclocking) needs to operate within the clock cycle of the DAC / transport ... so can only perform a finite amount of processing, and hence include a limited amount of logic.

 

A computer (or other transport) with the ability to operate on the digital audio before it has become exposed to the clock cycle of the DAC, can perform an unlimited amount of processing / logic during the re-sampling, because it does not matter how long the processing takes.

 

It is not impossible to do this 'advanced processing' with 'embedded circuits' ....  but it's much easier on a computer .... and regardless, when the processing power of some 'embedded circuit'  (say within a DAC or transport) begins to approach that of a PC, then I'd say the lines are blurring as to whether you ARE using a "computer for your over-sampling".

 

 

 

I'm disappointed at how wordy and incoherent that got.

[davewantsmoore]

thats the thing, from CEC is SPDIF coax then go to M7 i2s, many on the headfi forum experienced improvements over coax, and i can hear the improvement easily although not day and night.

i wonder why...

 

It's still unclear how you are doing this.

 

 

It sounds like you are using 4x SPDIF->BNC cables.... which would mean that the M7 has it's I2S connections exposed via 4x BNC or RCA connectors.   I haven't seen one like that, but it is surely possible.

 

 

The CEC has I2S output.   How is this exposed?  (it will be 4 wires or cables,  Eg.  4x BNC, 4x RCA, 1x RJ45) .... AFAIK the CEC transports usually use a DB9 which has I2S on it   (they call it supralink or something)

 

The M7 has I2S input.    How is this exposed?   (They normally do it with RJ45)

[davewantsmoore]

i wonder why...

 

I2S is the internal communication protocol used within devices.      Using I2S between two boxes, will bypass the circuits which are normally used to turn I2S into SPDIF and back again.

 

I2S was not designed to cover distances or cope with interference, or generally be suitable for human use.

[hochopeper]

It's still unclear how you are doing this.

 

 

It sounds like you are using 4x SPDIF->BNC cables.... which would mean that the M7 has it's I2S connections exposed via 4x BNC or RCA connectors.   I haven't seen one like that, but it is surely possible.

 

 

The CEC has I2S output.   How is this exposed?  (it will be 4 wires or cables,  Eg.  4x BNC, 4x RCA, 1x RJ45) .... AFAIK the CEC transports usually use a DB9 which has I2S on it   (they call it supralink or something)

 

The M7 has I2S input.    How is this exposed?   (They normally do it with RJ45)

 

That's what I was trying to ask, thanks Dave!

 

Wait till you see how much I can ramble when I try to reply to the OP later tonight ... had long meeting just now to daydream about 1) how to explain this in some sort of vaguely concise/balanced manner (ie less than 10,000 words) and 2) graph theory and why the hell I didn't do computer science as well as engineering like I had originally planned but instead chose to teach this stuff to myself 'in my spare time' as a way to solve data questions at work ... heh first world problems.

Edited January 29, 2014 by hochopeper

[davewantsmoore]

From my reading, I don't think that's a meaningful distinction. 

 

My understanding agrees with yours, they're analogous processes. However it seems to be a convention that the up/over sampling filter inside the DAC chip is called oversampling filter and the up/over sampling filter in the computer is called upsampling filter. 

 

Yes.

 

This is probably where a lot of the ambiguity comes from when talking about digital audio.