Jump to content
IGNORED

Pro-Ject Pre Box S2 Digital: MQA HW decoding at reasonable cost


Recommended Posts

7 hours ago, Display said:

And are you sure Miska, that when playing DSD files, channels are not reverted?   (firmware 2.10

 

Yes, channel swap is certainly an issue, but other than that it works fine. I can swap the channels in the player, so it is possible to work around.

 

But patching macOS doesn't affect the channel swap in any case.

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
5 hours ago, Miska said:

 

Yes, channel swap is certainly an issue, but other than that it works fine. I can swap the channels in the player, so it is possible to work around.

 

But patching macOS doesn't affect the channel swap in any case.

 

The 2.11 firmware update resolves the DSD channel reversal, and a "pop" sound when the sample rate clock changes. PM me if you need the firmware file. You'll need Windows to install it though.

Everyone wants to date my avatar.

Link to comment
20 hours ago, left channel said:

The 2.11 firmware update resolves the DSD channel reversal, and a "pop" sound when the sample rate clock changes. PM me if you need the firmware file. You'll need Windows to install it though.

 

I already did... ;)

 

I have to admit I have not noticed any pops though. Maybe it is because I mostly keep my DACs running at constant rate though...

 

But my point was that this firmware update has nothing to do with macOS compatibility or the other related issues.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
2 hours ago, Miska said:

 

I already did... ;)

 

I have to admit I have not noticed any pops though. Maybe it is because I mostly keep my DACs running at constant rate though...

 

But my point was that this firmware update has nothing to do with macOS compatibility or the other related issues.

 

 

Oh that's right, sorry. I've gotten so many requests via PM on three forums, I've lost track. I don't mind helping while the designer is focused on finishing his new streamer, especially since it's not actually his job to distribute the updates. The people who are responsible for this support at Pro-Ject and the retailers have been unable to keep up with us early adopters so far, but I'm sure they'll be ready soon.

Everyone wants to date my avatar.

Link to comment

I have now tested the S2 dac with USB hooked to my main roon core (exteneded usb 2 cable "fished " down to main computer in my office which is my Roon core). 

 

Observations Using Roon via ipad and S2 connected to main living room audio system. 

 

- Tidal MQA works great. Roon passes all the data thu. I have not tried an actual MQA download yet. Of course, all DSP settings and any special volume features need to be off for the endpoint MQA will be passed thru. Plus, Exclusive mode needs to be chosen in roon. 

 

-I can only get DSD256 to work, no matter with set up (which is fine). If I pass thru the pure DSD signal (roon option), I get static. If I set roon to DSD over PCM, dsd 256 is max. Of course, I can have roon do a dsd conversion to pcm and the closest khz after conversion will desplay on S2. So with Roon passing thru DSD, I cannot get to DSD512. I believe DSD only works over PCM anyway with the S2.

 

-no clicks, pops, etc with latest upgrade. DSD L/R channels seem to be correct.

 

So now I am pretty happy. I just wish Roon supported Qobuz. But no biggie, I just change the s2 input to coax where I have my Bluesound Node2 connected. 

 

 

 

 

Link to comment
On 13/11/2017 at 6:13 AM, firedog said:

Don’t agree. Lots of DAC designers will tell you that the power supply and analog out section make more difference than the DAC chip. Lots of ESS based DACs sound different, even with the same chip.

 

Abso-friggen-lutely.

 

I know a few DAC designers and that's exactly what they concentrate on.

 

In fact the Grob DAC I posted the review of did exactly that - output - ultra high quality transformer - power supply deveped obver many years to find the best regulator and use them pretty much wherever you need power.

 

Digital power supplies are EVIL.

 

I have an Off-Ramp and it comes with a SMPS the maker claims is the bees knees.  I was with one of those DAC makers and just for the heck of it said lets put the SWMPS on my scope - horrid - that stuff going down the power cable and into the Off-Ramp then radiating about - yuck.   He built a linear supply for me and put that on the scope - dead flat even at maximum sensitivity.  And the difference in sound - easily better.

 

Thanks

Bill

Link to comment
On 29/10/2017 at 7:05 PM, Miska said:

You mean 2x unfolding followed by 2x upsampling with a poor filter...

 

Ah Miska 9_99_99_9

 

Yes - I see your point - but its a bit more nuanced than that.  They don't even reveal what order of spline they use for the downsampling and matching up-sampling - I think it varies and is encoded in metadata.

 

I have heard both the correct unfolding from a  Direct Stream and Explorer 2, and whatever up-sampling Auralic came up with - I cant really say which I prefer.

 

However having your HQPlayer I suspect you might do a better job than the Auralic mob.

 

Thanks

Bill

Link to comment
3 hours ago, bhobba said:

Yes - I see your point - but its a bit more nuanced than that.  They don't even reveal what order of spline they use for the downsampling and matching up-sampling - I think it varies and is encoded in metadata.

 

I have heard both the correct unfolding from a  Direct Stream and Explorer 2, and whatever up-sampling Auralic came up with - I cant really say which I prefer.

 

However having your HQPlayer I suspect you might do a better job than the Auralic mob.

 

Yes, but that doesn't matter at all. Do not try to copy MQA, at least not the filters, you'll just degrade quality. Certainly you don't want to use spline for downsampling. And I don't think they actually use either. Their "splines" at upsampling (what they call rendering) phase remove most of their decimation filter. But there's no magic there.

 

If you want to save bandwidth/file size to equal amount or more than MQA while preserving better quality, use good filters and cut out the stuff above music content. This won't affect the music content's transient response as long as your filter transition band doesn't hit the music content. And you don't want that if you have DXD to begin with. You'll keep the transient response intact!

 

You can get more than "benefits" of MQA by not trying to do the same, but doing it different and better.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
2 hours ago, Miska said:

Yes, but that doesn't matter at all. Do not try to copy MQA, at least not the filters, you'll just degrade quality. Certainly you don't want to use spline for downsampling. And I don't think they actually use either. Their "splines" at upsampling (what they call rendering) phase remove most of their decimation filter. But there's no magic there.

MQA "rendering" is a perfectly normal FIR filter (with bizarre coefficients). No splines involved.

Link to comment
49 minutes ago, mansr said:

MQA "rendering" is a perfectly normal FIR filter (with bizarre coefficients). No splines involved.

 

Yes, I've seen it. Of course you can also turn IIR filter into FIR under certain conditions, but not vice versa. Not claiming anything about splines here, just continued to use the same terminology not to confuse more (that's the reason for double quotes).

 

I guess the spline part could theoretically be involved in creating those bizarre coeffs... Or someone just pulling those out of their sleeve, that is also possible (except for 14), given the very low attenuation and number of taps. For some reason 14 looks like normal equiripple FIR, 1 and 3 being sort of low order IIR looking. Others look like EQ'd version of 1 and 3.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
39 minutes ago, Miska said:

I guess the spline part could theoretically be involved in creating those bizarre coeffs... Or someone just pulling those out of their sleeve, that is also possible (except for 14), given the very low attenuation and number of taps. For some reason 14 looks like normal equiripple FIR, 1 and 3 being sort of low order IIR looking. Others look like EQ'd version of 1 and 3.

Yes, 14 stands out as fairly normal-looking (has anyone seen it used?). All of them seem to have fairly deliberate frequency responses, so I doubt someone was just drawing shapes.

Link to comment
28 minutes ago, mansr said:

Yes, 14 stands out as fairly normal-looking (has anyone seen it used?). All of them seem to have fairly deliberate frequency responses, so I doubt someone was just drawing shapes.

 

Btw, what factors did you use to draw the plots? Because the plots I get for the x8 rates look different...

 

Looking at the plot for 3 I feel like I have idea what they did.

 

mqa-filt3.thumb.png.0f62bb3fe14761b43587a87f683a7f9c.png

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
20 hours ago, Knight Rider said:

Have you guys tested Audioquest Jitterbug with Pre box s2 digital? Does it makes any audible difference?

 

We've tested other anti-jitter products on it, and they did make a difference. Using the external power adapter (instead of just USB power) improves the sound even more, and some report that upgrading the power adapter may help as well.

Everyone wants to date my avatar.

Link to comment
5 hours ago, left channel said:

We've tested other anti-jitter products on it, and they did make a difference. Using the external power adapter (instead of just USB power) improves the sound even more, and some report that upgrading the power adapter may help as well.

 

Do you happen to have measurement figures for that? :)

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
10 hours ago, Miska said:

Btw, what factors did you use to draw the plots? Because the plots I get for the x8 rates look different...

 

Looking at the plot for 3 I feel like I have idea what they did.

 

mqa-filt3.thumb.png.0f62bb3fe14761b43587a87f683a7f9c.png

Factors? I used the freqz function in Matlab/Octave. This is filter 3 at x8:

mqa-f3-x8.thumb.png.7652ff6622b6df9fdd59878d448f310d.png

Looks a lot like your plot but with higher resolution.

Link to comment
8 hours ago, Miska said:

 

Do you happen to have measurement figures for that? :)

 

 

Subjective tests only. But the designer did measure the power results, and that's why they decided to add an auto-switching feature: when you plug in an external adapter, it switches away from USB power. (The manual was apparently written before they did that; a corrected version of that section is available in my firmware folder.)

 

Everyone wants to date my avatar.

Link to comment
10 hours ago, mansr said:

Factors? I used the freqz function in Matlab/Octave. This is filter 3 at x8:

mqa-f3-x8.thumb.png.7652ff6622b6df9fdd59878d448f310d.png

Looks a lot like your plot but with higher resolution.

 

I mean the plots you shared earlier where there were all the filters collected in single picture. Those didn't look like these, those were for the 2x filters?

 

I used my own little Octave script that calculates various things about filters. But it doesn't interpolate the responses like that, so it is more like bare data. Same result anyway.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
11 minutes ago, Miska said:

I mean the plots you shared earlier where there were all the filters collected in single picture. Those didn't look like these, those were for the 2x filters?

Those were probably only for 2x and/or 4x. The 8x filters were added in a later Bluesound firmware version.

 

11 minutes ago, Miska said:

I used my own little Octave script that calculates various things about filters. But it doesn't interpolate the responses like that, so it is more like bare data. Same result anyway.

Interpolate? The frequency response exists for all frequencies. I don't see why you think calculating it at only a small number of points is more accurate.

Link to comment
4 minutes ago, mansr said:

Interpolate? The frequency response exists for all frequencies. I don't see why you think calculating it at only a small number of points is more accurate.

 

Where did I say it is more accurate? :D

 

My plot is just from plain 2x padded complex representation of the coeffs. For me it is easier if I can see actual data points in the plots.

 

But I don't know how relevant the plotting details are for this thread anyway. I already regret posting anything about the topic.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
On 11/29/2017 at 8:36 AM, mansr said:

MQA "rendering" is a perfectly normal FIR filter (with bizarre coefficients). No splines involved.

 

I think you may know more in the signal processing area than I do.  I am but a humble applied mathematician with an interest in Audio, amongst other things.

 

A spline of order 0 is a rectangle, of order 1 a triangle, and weirder shapes as you go higher - the detail can be found it many books on applied math - its used in many areas - not just signal processing - where curve fitting of some sort is required.

 

Whenever MQA talks about how it does what it does (or others try to explain what they have been told) it talks about using triangle sampling ie a sampling with an order 1 spline:

https://www.soundonsound.com/techniques/mqa-time-domain-accuracy-digital-audio-quality  

 

However as the above says:

Even better results are possible using higher-order ‘B-spline’ kernels, which allow both the position and intensity to be identified of two or more separate pulses occurring within the same sampling period!

 

It is thought, from what others have posted on various forums, that's what they use in practice rather than the triangle and encode in metadata somewhere the right way to decode ie get the best up-sampled approximation from whatever they used to encode it with.  For a spline of order 1 ie a triangle - its dead simple - just linear interpolation.   Higher order splines have other ones.  Naturally its still just a Finite Impulse Responce (ie FIR) filter eg even linear interpretation is just a FIR filter.  It must be FIR because they claim to keep impulse width to a minimum.

 

It's all detailed in their patent:

https://patentscope.wipo.int/search/en/detail.jsf;jsessionid=C54A108611AAB6A34392194DDAEFDDC7.wapp2nA?docId=WO2014108677

 

Why do they do it?  They claim it leads to better impulse responses - the response to a Dirac pulse being only 10us wide using their methods while others ring like a bell.   You of course can use an apodising filter that puts all the ringing after the pulse - that is much better since we are used to hearing ringing after pulses - but before - our ears aren't quite used to that - it never occurs in nature.  Also Dirac pulses do not actually occur in nature (they aren't even actual functions - just some mathematical trickery applied math guys like me use) and if you use a regular linear phase brick-wall filter beyond the highest frequency of actual musical content - nothing happens.  Experiments I have done show in actual material its rare to find stuff above 30k - and even rarer above 50k.  Basically simple 96/24 with a brick-wall filter is all you really need - but that may change in the future as even higher res material becomes available.  I have checked so called 192/24 material from HD downloads - its usually HD ie better than 44.1/16 but 192/24 it mostly isn't - although in a few cases it stands up to scrutiny.  I have one DXD master that goes all the way to 176k- but its the only one I have been able to find.  So IMHO MQA is fixing a problem that doesn't really exist.  What is a problem is the band-witch/size requirements for high res material.   A better way than FLAC to losslessly reduce the size/bandwidth of material is needed.  MQA is one attempt - but IMHO way sub-optimal - better ways exist without detailing them.  

 

The argument over MQA is whether all this 'tom foolery' sounds better.

 

Opinions vary - I like it - others don't.

 

However they make the claim the system with the smallest time domain smear is always preferred - that claim IMHO is utter BS - as I explained with the frequency of actual material.   Also better in audio is a very subjective thing.  I have seen claims like that before, sat with some fellow audiophiles and the opinions usually vary all over the place.  This should be known to Bob Stuart, and I think he does know it, but puts all sorts of marketing spin around it, for reasons that are obvious to anyone with even an inkling of knowledge of how business works.  I am very much a babe in the woods regarding marketing and that other business BS - but even I see whats going on.   Will it succeed - who knows - we will see.

 

At the moment all I can say is I, and pretty much everyone I have listened with, prefer Tidal MQA to Tidal 44.1.  It seems one of those rare things most people I know agree on - like I say its usually all over the place.   But I have seen posts here where others have a different experience - so really its just the people I know - in reality like nearly every other thing in audio its all over the place.

 

What is MQA's advantage then?  Deciding will need much more material with actual content over 50k - and listening tests using that.  Only with such material can the claim of MQA that the recording with the better lime smear is always preferred be put to the test.   My opinion is it will be the same - all over the place - some will say - wow - others blah.

 

Thanks

Bill

Link to comment
1 hour ago, bhobba said:

You of course can use an apodising filter that puts all the ringing after the pulse - that is much better since we are used to hearing ringing after pulses - but before - our ears aren't quite used to that - it never occurs in nature.

 

Apodizing filters don't need to be minimum phase, they can be also linear phase or something between the two... Or a filter like MQA rendering filters. The properties just make it replace side effects of the original "filter kernel" with the other one.

 

1 hour ago, bhobba said:

At the moment all I can say is I, and pretty much everyone I have listened with, prefer Tidal MQA to Tidal 44.1.  It seems one of those rare things most people I know agree on - like I say its usually all over the place.   But I have seen posts here where others have a different experience - so really its just the people I know - in reality like nearly every other thing in audio its all over the place.

 

What is MQA's advantage then?  Deciding will need much more material with actual content over 50k - and listening tests using that.  Only with such material can the claim of MQA that the recording with the better lime smear is always preferred be put to the test.   My opinion is it will be the same - all over the place - some will say - wow - others blah.

 

The problem I have is the technical approach that is trying to enforce certain types of filters on everyone. While also hampering possibilities to use things like digital room correction DSP that certainly have much bigger effect on sound quality. And as we know, it doesn't really save bandwidth either compared to properly encoded FLAC.

 

I don't mind so much Tidal using it, since that is only a single service and their application can decode it. And Tidal content is not preserved for long time. But I would mind if digital downloads I purchase would suddenly be available only as MQA. I don't want to risk that I would need to purchase same content again at later time. Plus I'm not too fancy paying MQA-tax on everything from content to DACs.

 

Signalyst - Developer of HQPlayer

Pulse & Fidelity - Software Defined Amplifiers

Link to comment
5 hours ago, bhobba said:

I think you may know more in the signal processing area than I do.  I am but a humble applied mathematician with an interest in Audio, amongst other things.

 

A spline of order 0 is a rectangle, of order 1 a triangle, and weirder shapes as you go higher - the detail can be found it many books on applied math - its used in many areas - not just signal processing - where curve fitting of some sort is required.

 

Whenever MQA talks about how it does what it does (or others try to explain what they have been told) it talks about using triangle sampling ie a sampling with an order 1 spline:

https://www.soundonsound.com/techniques/mqa-time-domain-accuracy-digital-audio-quality  

 

However as the above says:

Even better results are possible using higher-order ‘B-spline’ kernels, which allow both the position and intensity to be identified of two or more separate pulses occurring within the same sampling period!

 

It is thought, from what others have posted on various forums, that's what they use in practice rather than the triangle and encode in metadata somewhere the right way to decode ie get the best up-sampled approximation from whatever they used to encode it with.  For a spline of order 1 ie a triangle - its dead simple - just linear interpolation.   Higher order splines have other ones.  Naturally its still just a Finite Impulse Responce (ie FIR) filter eg even linear interpretation is just a FIR filter.  It must be FIR because they claim to keep impulse width to a minimum.

A spline is a piecewise polynomial interpolation. A FIR filter is something different. Whereas splines of order 0 or 1 are trivially expressed as FIR filters, higher order splines are not.

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...