I'm surprised it worked

[Jim_F]

I've been playing around with online and offline upsampling for years now (including with XXHE), most recently offline using iZotope's 64-bit SRC (selected as a DSP effect in a Sony SoundForge batch) trying out settings from posts over at Computer Audiophile (primarily from folks who use the real-time iZotope SRC available in Audirvana+ on the Mac).  I even recently coughed up the bucks for AuI Converter producer edition, which can off-line upsample to DSD512 (the resulting .dsf files are **huge**, but they sound good, and I can play them back in HQPlayer or foobar over USB to an iFi iDSD Micro).

I've been a little frustrated with the sound of PCM upsampling recently (variously, to 96k, 176.4k, or 192k, played back over various DACs,
NOS and otherwise).

So on an impulse I conceived an experiment, which I frankly didn't expect to work.  I have a dCS Purcell hardware (synchronous) sample-rate converter that I decided to take out of the closet and try again, just for kicks.  Using the Purcell to go from 44.1 to 192 is problematic unless you have a dCS DAC, because when the Purcell is outputting 192k, it's doing so over dual-wire AES.  There are a number of discussions on the Web with folks lamenting the difficulty of getting from dual-wire to single-wire, but I discovered a way to do it using **two** Apogee Big Bens: dual-wire from the Purcell into the first Big Ben (input set to AES "DOUBLE"), then ADAT S/MUX4 from the first Big Ben's optical output to the second Big Ben's optical input (with the second BB's input set to S/MUX 4, of course), and then from the second Big Ben you can take quad-speed single wire out via AES or even coaxial S/PDIF.  There's no way to do this with a single Big Ben (at least not with any firmware revisions I have access to).  Nor can I do it with one or even two RME ADI-192DD digital-to-digital converters.

So the first DAC I tried with this setup was a Wyred4Sound DAC2 (with the ES9018 "Sabre32" chip), which I certainly thought sounded better fed by the Purcell at 192k than the last time I'd heard it.

But then I thought -- I wonder if there's any way I can get to hear the Purcell through my Phasure NOS-1a.  The Phasure is a USB-only DAC (unless there's an S/PDIF input I haven't found ;-> ).  There've been naive queries on the Web from people wanting to know if there's any way to run USB-only DACs from S/PDIF inputs (like this one: http://www.head-fi.org/t/381903/spdif-usb ) and the answer to such queries has always been a brusque NO!

To cut to the chase, here's how I got it working.  I connected an RME Babyface USB interface to a Toshiba Windows 7 laptop with the Babyface driver installed on it, and connected the ADAT S/MUX4 optical at 192k from one of those Big Bens to the Babyface.  I also connected the Phasure NOS-1a to another USB port on the laptop (the laptop also has the Phasure's driver installed).

The trick to tying them together was the discovery that the freeware DAW Audacity (I'm using 2.0.6, the latest) has a feature called "software playthrough" that lets you turn on monitoring and feed audio from your selected recording device (in my case, the RME Babyface) to your selected playback device (the NOS-1a).  This playthrough monitoring doesn't work with all drivers -- WASAPI doesn't work, and when I tried an Audacity with ASIO support compiled in, the ASIO drivers didn't work either.  It has to be either MME or DirectSound.  Either of those work.

So I've got an elaborate S/PDIF (well, ADAT S/MUX4 to be precise, but it could just as easily be S/PDIF) to USB "converter" going, to be able to listen to the Purcell doing 44.1->192 to the Phasure NOS-1a.  It sounds pretty good!  (On the Purcell, I've got 24-bit output, "Filter 1", and 9th-order noise shaping selected.)

Sometimes, things that "shouldn't" work, do.

[Jim_F]

I just discovered, quite by accident, that Windows 7 will do "playthrough" from one sound device to another automatically, without the need of other software.  In Control Panel -> Hardware and Sound -> Sound -> Manage Audio Devices, under the Recording tab, if you pick a device and then click on the Properties button, there's a "Listen" tab on the next screen.  If you check the box "Listen to this device" you can then select any other device in the "Playback through this device" drop-down.

Seems to work fine.

[Nick]

I just discovered, quite by accident, that Windows 7 will do "playthrough" from one sound device to another automatically, without the need of other software.  In Control Panel -> Hardware and Sound -> Sound -> Manage Audio Devices, under the Recording tab, if you pick a device and then click on the Properties button, there's a "Listen" tab on the next screen.  If you check the box "Listen to this device" you can then select any other device in the "Playback through this device" drop-down.

Seems to work fine.

Jim_F hi,

Great post and a result getting a recombined 192k stream into your NOS1a. I would be interested to hear what the resulting sound is like compared to playing at 192k directly to the NOS1a. Thinking about the end setup I have a hunch that the change might be due to both dCS up-sampling and PC. I am working on what might be  the related area in the PC at the moment so this is very interesting.

Is there any chance you could post thoughts on what changes in sound reproduction ?

Many thanks,

Nick.

[PeterSt]

Hey Jim,

Nice ! Really nice.

A few remarks, for whatever it may be worth :

The NOS1(a) can have SPDIF input but it must be supported in hardware and by the driver. Neither are, just because it seems useless. However, from of the NOS1-USB it has been an option, today not being a topic any more. So it got forgotten. Btw Nick, aren't you one of the few who has this SPDIF input ? Still, the driver support for it never got official (because not needed).
Side note : SPDIF originally could take 24/96 only, although later things changed in the USB firmware and nobody knows (me neither because not tried) whether now it can take 24/192.

Then ASIO;
I can not be sure at this time whether you ran into "ASIO not supported", Jim. So in the NOS1 driver I mean. But in the end it just is, but here too, the support of it has been left out by me, because - at least back at the time - this was useless because it was regarded that XXHighEnd would be the only software used for the NOS1. And XXHighEnd itself does not support ASIO.

The PassThrough thing from W7 (and above);
Yes, that works. I'd say also with WASAPI but now I am not sure again. So what I recall is that when W7 was around XXHighEnd did not support KS yet, and one of the first things I tried was that PassThrough. So must have been WASAPI. However, it did not work well enough, so indeed it would be my idea that WASAPI does not work (out).

Coincidentally, these days I am virtually working on these things. So about the PassThrough and ASIO as well. Not to let dCS upsamplers work, but to let work external sources of music. Say, additional options to be "general" enough so that all will be accepted and played through the NOS1a (think Internet and streaming). Here the "a" matters, because of the expectation that such stuff isn't much detrimental to SQ now (normal NOS1 would not be the best). Ultimately though all should run through "upsampling/filtering" as such, so think "through XXHighEnd". This is where it gets more complicated in the "PassThrough" realm, and which is why I said "virtually".

Hope this made some sense and thank you for sharing this, Jim.
Peter

[Nick]

Btw Nick, aren't you one of the few who has this SPDIF input ? Still, the driver support for it never got official (because not needed)

Peter hi,
Yes I still have the physical connection for SPDIF, but of course not a live connection live now days.
Regards,

Nick.

[Jim_F]

Peter,

You wrote:

> What I recall is that when W7 was around XXHighEnd did not support KS
> yet, and one of the first things I tried was that PassThrough. So must
> have been WASAPI. However, it did not work well enough. . .

One thing that occurs to me is that with the setup I'm using (with Windows 7 playthrough), there are two independent clocks -- the one in the S/PDIF or ADAT coming in, and the local clock in the Phasure NOS-1.  So, sooner or later, depending on how closely the clocks are matched (and the master clock in my system is an Antelope Isochrone, so I imagine the two clocks are pretty close, but still not absolutely identical), there will be a playback glitch eventually, because a buffer in the computer will either overflow or run dry.

I know there are DACs out there that are deliberately designed to work this way (Doede Douma's DDDAC comes to mind, and there are others) -- "exotic" D/A boxes that clock their DAC chip locally rather with the extracted clock from incoming S/PDIF, and that let the two clocks freewheel against each other, glitches be damned (for the sake of the sound quality in between the glitches ;-> ).

It's considered bad engineering practice, but so was NOS once upon a time (and still is, at least at 44.1k).  "Good" engineering practice would be to put an asynchronous sample-rate converter (like a CS8420, AD1896, or SRC4192) in between the two clocks.

There was an amusing thread on Computer Audiophile a while back in which "Miska" (of Signalyst/HQPlayer fame) debated with a non-technical audiophile about this sort of thing.
http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/mytek-stereo-192-a-5555/index113.html
The audiophile was using a setup with an independent clock in the DAC, and raving about the sound quality, and Miska was trying to explain why this inevitably leads to glitches and is considered bad engineering practice.  They never did reach a mutual understanding.

The answer to this conundrum is that a "civilian" audiophile is perfectly entitled to put up with glitches at home for the sake of the sound in between, but in a studio or other professional setting this would be considered completely unacceptable.  But even the "civilian" audiophile is not entitled to complain about the glitches, under these circumstances.

[Nick]

Jim_f

For sure two word clocks will run out of synchronisation and buffer glitches must happen as a result at some point. I have plenty of experimental evidance of the impact on sound quality that relative clock speed in the replay chain (5 clocks looked at so far more on the way) has on sound quality. This is both within the PC in what would have been considered the data domain and withing the DAC itself.

My guess however is that the word clocks running at different speeds is a secondary effect here (unless the two word clocks are grossly out that is, think 100s to 1000s of beats per second out).

There is another significant difference between the two setups you have tried that is likley to be the prime reason that the sound is changed. It would be helpful if you can you post your impression of how the sound has been changed ?

Hash improved bass, clearer highs, more detail, dynamics ?

Thanks,

Nick.

[PeterSt]

Jim,

So, sooner or later, depending on how closely the clocks are matched (and the master clock in my system is an Antelope Isochrone, so I imagine the two clocks are pretty close, but still not absolutely identical), there will be a playback glitch eventually, because a buffer in the computer will either overflow or run dry.

This description is pretty much what I recall. Could work for a while and then disotortion started or whatever. But the "while" is key.

Btw wanted to look into it again today but I forgot.

Thanks again,
Peter

[PeterSt]

Hash improved bass,

Was this intentional, Nick ?
Seems like a nice phenomenon.

[Nick]

Hash improved bass,

Was this intentional, Nick ?
Seems like a nice phenomenon.

No not intensional  , just what changed with the sound ?

There is another large difference between the two test setups that is not being considered. Intermodulation of the two word clocks is going to have an impact sure but there is something else. Think where the data originates from in the two setups. Its very different in the two cases, there is a whole subsystem of the PC not being used when the pass through setup is being used. This really really matters :-).

Regards,

Nick.

[Jim_F]

Nick,

You wrote:

> Great post and a result getting a recombined 192k stream into your NOS1a.
> I would be interested to hear what the resulting sound is like compared
> to playing at 192k directly to the NOS1a. . .

Well, I'm not actually seriously recommending a setup like this to anybody.  My intention was more along the lines of announcing "Look at the flying pig!" and suggesting a way in which somebody with a lot of electronics at their disposal (USB interfaces, digital format converters, what-have-you) can play with their toys.

But as to sound quality -- well, this might be rather system-dependent in my case.  The system I'm experimenting with is rather a "dog's dinner", as the British say.  My Quad ESL-63 electrostatics started showing their age a while ago (I've had them for more than 20 years, and I bought them used) -- they started arcing-over occasionally, while turned on but not playing, in response (it seemed to me) to the downstairs neighbors' cooking smells (like an audio smoke detector, if I'm not totally imagining the correlation).  So it's time to have them cleaned and/or serviced and/or refurbished (and/or recycled -- but I don't want to think about that ;-> ).

Anyway, in the meantime I looked around for a pair of relatively modest speakers that wouldn't be too embarrassed to stand in for Quads, and I bought (used, unheard) a pair of Paradign Signature 2 v2 (with the beryllium tweeters).  These are being driven at the moment by an unmodified Carver Pro ZR1600 (a Tripath-based amp -- full-range Class D -- that some folks were raving about 11 years ago) whose balanced inputs are being fed by an Audio Experience (a Chinese direct-sale brand) "Balanced A2" tube balanced line stage.  (I've basically retired my tube amps.  Unless I experience a change of heart, which I suppose is always possible, I'll be using full-range Class D [Tripath is defunct, but ICEpower, Hypex UcD et al. are still around] for power amplification, and tubes only in line-level preamp stages, for the foreseeable).  So anyway, like I said, a bit of a dog's dinner.

As I mentioned, I've been experimenting with software upsampling ever since I stumbled across a computer program called Eximius DVD2One more than a decade ago.  Most recently, I've been playing with realtime upsampling in XXHE, HQPlayer, and foobar2000 (with the SoX plugin).  And even more recently I've been playing around with an iFi iDSD Micro, to see what upsampling to DSD sounds like (in HQPlayer, foobar2000 with Maxim Anisiutkin's ASIO Proxy driver, or offline with Yuri Korzunov's -- Audio Inventory's -- "AuI ConverteR 48x44 PRODuce-RD").

In fact, it was after hearing the DSD results that I got really annoyed with my collection of upsampled PCM files, so just on a lark I decided to try out the dCS Purcell again -- to go back to the unit (or at any rate the consumer successor to the dCS 972 pro unit that started it all) that created the "upsampling" juggernaut back in '99 (Jonathan Scull's review of that in _Stereophile_ remains to this day the most affecting piece of audio porn I've ever read, http://www.stereophile.com/digitalprocessors/260/ ).

So what's the difference with the Purcell (in comparison to a batch of files I upsampled with iZotope's 64-bit SRC to, variously, 192k, 96k, and 176.4k -- [settings: Steepness 4; Max filter length 2,000,000; cutoff scaling 1.28; Alias suppression 200.00; Prering 0.00%]).  This is totally subjective, and may have more to do with my current mood than anything else, but -- a bit more midrange (not getting lost between the bass and the treble), plenty of ambience, but with a bit less "hollowness" than before, and most importantly, a bit less sharpness in the upper midrange (especially on piano tone -- it's important to be able to listen to Alfred Brendel play Mozart piano concertos on Philips without getting a headache ;-> ).

It's perfectly possible one might be able to match the sound of the Purcell with a different software upsampler, or with different settings.  There's been a lot of discussion on Computer Audiophile (and even Hydrogen Audio) about trying to match the characteristics of Ayre's "apodizing" filter by playing with settings in SoX or Izotope to minimize pre-ringing, use minimum-phase filtering, etc., and I was going along with all of that.  Maybe I was barking up the wrong tree (I doubt if the Purcell follows that mantra).  On the other hand, dCS's filter algorithm (**whatever** it is) **was** the one that got everybody's juices flowing in the first place, 15 years ago (assuming it wasn't just a marketing stunt that all the reviewers got suckered in by).  On the other hand, I've seen an article on line from International Audio Review ( http://www.iar-80.com/page21.html ), presumably from back around the turn of the century, that claims that the Purcell's "magic" had nothing to do with upsampling per se, but everything to do with the noise-shaping that it applies (which is actually a user-selectable option) when converting back to fixed-point output after doing its internal arithmetic).  I have no idea if the author of that article knows what he's talking about (I fear the worst ;-> ).  So who knows?

[PeterSt]

Ok. I wouldn't know how "hash improved" bass sounds.

And to be really honest, what description is that ?
Of course it is not important at all, but hash improved highs I would understand right away.

Peter

[Nick]

Jim hi,

Love the story about the ESL63s I certainly understand why you have had them so long. I also had a used pair form many years, I was so attached to them I could not bring myself to part with them for 5 years after I had bought a set of horn speakers. They do things others just cannot manage :-).

Thanks for the description of sound. Now I read your first post again I think the laptop playing music in the pass through USB mode maybe using its disk sub system after all to read the music data.

The playback setup is sort of smart with the dCS and word clocks, really understand where your coming from with the flying pig comment, you just would not think it would work  There is certainly a lot that could change the sound, word clock quality, over sampling algorithm or other things. Your post caught my eye because I'm looking at the PC's storage sub system at the moment. What you have described is the sound is sort of consistent with the PC's disk subsystem being improved or in this case perhaps removed from the replay chain. But again re reading your first post I think the laptop is still reading its disk but there are newly introduced SPDIF and upsampleing steps in the data path.


Kind regards,

Nick.

[Nick]

Ok. I wouldn't know how "hash improved" bass sounds.

And to be really honest, what description is that ?
Of course it is not important at all, but hash improved highs I would understand right away.

Peter

Peter hi I agree, more likely to hear reduced hash is the highs not in the bass. I was typing a little too fast on a tablet  . Sorry for the confusion.

Regards,

Nick.

[Jim_F]

Nick,

You wrote:

> But again re reading your first post I think the laptop is still
> reading its disk but there are newly introduced SPDIF and upsampleing
> steps in the data path.

I would suspect that in fact it's not using the disk at all (that is, on the laptop that's "bridging" the RME Babyface USB input to the Phasure NOS-1a USB output via the built-in Windows 7 playthrough capability).

Remember, there are actually **two** computers in the system I'm describing.  The computer that has the music files and the music player is the one I usually use for audio playback -- an HP minitower (quad-core i7, with Windows 7 64-bit).

The (current) architecture of the whole system is not unlike that of a dual-PC system using HQPlayer on a PC connected over a LAN to HQPlayer's NAA (Network Audio Adapter) on a second PC that's actually performing the job of feeding the audio to a USB DAC (or the similar arrangement using JRiver Media Center on one PC connected over a LAN to a second PC running Jplay and feeding audio to a USB DAC).  Except that instead of using Ethernet to connect the "control PC" to the "audio PC", I'm just sending the data via S/PDIF (or ADAT).  And it's the RME Babyface USB interface that's getting the ADAT (or S/PDIF) -- the laptop only sees USB data coming in and going out; it's certainly not doing any sample-rate conversion, and it's not reading or processing audio files.

I've actually got a sort of "optical bus" running around the house.  The source PC (running foobar2000, usually) is connected over Firewire to an RME Fireface 400 that's clocked by an Antelope Isochrone.  The Fireface sends optical output to the first audio-system "station" on the bus -- the "station" is either an Apogee Big Ben or an RME ADI-192DD that takes in the optical and then passes it along to the next "station" (the optical links are up to 50' Hosa glass cables).  The Big Ben or ADI-192DD can then feed a DAC, or a signal processor (like the Purcell), etc., etc.

I also installed that "Fidelizer 6.5" program on the laptop (the Toshiba laptop's also a quad-core i7 machine running Windows 7 64-bit; both machines have plenty of memory -- 6 or 8 GB, I can't remember exactly), and put Fidelizer into "Extremist" mode, so the laptop should be pretty quiet --
I don't think Windows would have any need to page to the hard drive, and I can't see why it would be putting any audio on the hard drive -- I would think the buffering would all be done in memory.

Oh, speaking of the Purcell -- I just noticed that in fact when you're doing upsampling from 44.1k to either 176.4k or 192k, with 24-bit output word length, both dither and noise shaping are automatically turned off, and cannot be turned on, at least with the firmware in the unit I'm using (these things go through multiple generations). I hadn't noticed this before (I'm used to using the Purcell in other modes, such as doing 44.1k->96k and then
reducing output word length to 18 bits to feed an Audio Note NOS DAC capable of taking 96kHz). So much for the theory in that IAR article about noise shaping!

And as far as playback glitches are concerned -- they do happen from time to time (every 15 minutes or so maybe?) and are quite audible, even with Windows 7 playthrough and Fidelizer.  They sound like buffer underruns to me -- the audio drops out, comes back, drops out for about a second or two altogther, then goes on for a long while before the next one.  There were a lot more of them when I was using Audacity to do the playthrough (and they were different -- raspberry-sounding things: brrrrrr for a second or so.  A buffer overflow in that case?).  So again, I'm not really recommending this to anybody.