Extreme filtering software upscaling

[rand129678]

1 hour ago, legend said:

all you have to do is now package it for sale so the rest of us can upsample other music!

 

I'm using HQPlayer's "sinc-M" filter, up-sampling to PCM705/768kHz, on the fly...

 

Sinc function filter, 1 million taps, > 240dB stop-band attenuation, on the fly.

 

There is already a 16 million tap filter in HQP for SDM up-sampling, so I'm sure the 16M tap filter will come to PCM up-sampling too.

 

Archimago recently included some measurements, using this Oppo 205 output feeding his A-to-D converter. Miska notes cleaner looking output from his Pro-Ject S2 DAC feeding his RME ADI-2 Pro...

 

https://archimago.blogspot.com/2019/02/measurements-look-at-hqplayer-325.html

 

I'm using a headless server. No interaction with a keyboard, monitor, mouse...

 

HQP is configured via web browser on iPad...

 

 

Edited February 27, 2019 by Music2496

[Ittaku]

1 hour ago, Music2496 said:

 

I'm using HQPlayer's "sinc-M" filter, up-sampling to PCM705/768kHz, on the fly...

 

Sinc function filter, 1 million taps, > 240dB stop-band attenuation, on the fly.

 

There is already a 16 million tap filter in HQP for SDM up-sampling, so I'm sure the 16M tap filter will come to PCM up-sampling too.

 

Archimago recently included some measurements, using this Oppo 205 output feeding his A-to-D converter. Miska notes cleaner looking output from his Pro-Ject S2 DAC feeding his RME ADI-2 Pro...

 

https://archimago.blogspot.com/2019/02/measurements-look-at-hqplayer-325.html

 

I'm using a headless server. No interaction with a keyboard, monitor, mouse...

 

HQP is configured via web browser on iPad..

Nice. It's only a matter of time then that high tap filters will become commonplace in more audiophile aimed software, rendering hardware solutions for this pointless.

[rand129678]

8 minutes ago, Ittaku said:

Nice. It's only a matter of time then that high tap filters will become commonplace in more audiophile aimed software, rendering hardware solutions for this pointless.

To be honest there are many many HQP users sharing observations across various forums... over many years now. And the high tap filters are not the overwhelming subjective favourite...

 

So I'm not sure about high tap filters becoming commonplace over time. 

 

But the beauty of HQPlayer is you have an abundance of choices...

 

You can choose extremely fast cut-off, linear phase high tap filters, >240db stop-band attenuation. Or the other end of the scale, like Ayre/Meridian/MQA, minimum phase and slower roll-off.

 

Ted Smith is a fan of linear phase and extremely slow roll-off and there is this option too. And many others in between the extremes.

 

All done on the fly and configured via any web browser (tablet, phone, whatever), in the case of HQPlayer Embedded.

Edited February 27, 2019 by Music2496

[Ittaku]

2 minutes ago, Music2496 said:

To be honest there are many many HQP users sharing observations across various forums... over many years now. And the high tap filters are not the overwhelming subjective favourite...

Admittedly the 1M tap filter still shouldn't be a match for the MScaler though because of their proprietary approximations towards an infinite sinc function. That's why brute forcing it to over 15M is where I found it had all its advantages in software rather than worrying about the WTA approximations.

[rand129678]

13 minutes ago, Ittaku said:

Admittedly the 1M tap filter still shouldn't be a match for the MScaler though because of their proprietary approximations towards an infinite sinc function. That's why brute forcing it to over 15M is where I found it had all its advantages in software rather than worrying about the WTA approximations.

Noted but I wouldn't say shouldn't be a match (knowingly?) to M-Scaler and also refer to proprietary in the same sentence.  We don't know fully what Bob is doing, just like we don't know fully what HQPlayer is doing...

 

Both only say approximate infinite sinc function, linear phase, 1 million taps, extremely fast roll-off, very very high stop-band attenuation....

 

Apart from that, both are proprietary... 

 

I heard M-Scaler and Dave myself and didn't think it made a significant enough difference to Dave alone. Dave was fantastic. I'm allowed to really like one Rob Watts product and not like another

 

Edited February 28, 2019 by Music2496

[Ittaku]

6 minutes ago, Music2496 said:

I heard M-Scaler and Dave myself and didn't think it made a significant enough difference to Dave alone. Dave was fantastic. I'm allowed to really like one Rob Watts product and not like another

 

I like the MScaler and hate the DAVE, so we're even.

[rand129678]

dCS released their first up-sampling box (separate to DAC box) in the late 90's...

 

I recently read that HQPlayer is nearly 10 years old...

 

So up-sampling upstream of the DAC box is obviously not a new concept...

 

Rob has said he really wants 1 million WTA taps in his pocket for when he travels, so no doubt a future Chord Hugo will be able to do it all, directly driving a pair of headphones (or DAC). He has said the challenge is isolating the RF the M-Scaler FPGA generates...

 

I'm sure he'll figure it all out and there won't be a need for this 2-box solution in future.

Edited February 28, 2019 by Music2496

[Ittaku]

Just now, Music2496 said:

dCS released their first up-sampler in the late 90's...

 

I recently read that HQPlayer is nearly 10 years old...

 

So up-sampling upstream of the DAC both is obviously not a new concept...

 

Rob has said he really wants 1 million WTA taps in his pocket for when he travels, so no doubt a future Chord Hugo will be able to do it all, directly driving a pair of headphones (or DAC). He has said the challenge is isolating the RF the M-Scaler FPGA generates...

 

I'm sure he'll figure it all out and there won't be a need for this 2-box solution in future.

If he wasn't so hung up about using an fpga, and did it using commodity PC components he could easily do it... Upsampling and oversampling has been an integral part of most DACs since time immemorial. Massive tap filters has only been a recent thing thanks to modern PC and FPGA performance.

[rand129678]

5 minutes ago, Ittaku said:

Upsampling and oversampling has been an integral part of most DACs since time immemorial.

 

Noted. I referred to up-sampling before the DAC, both using higher quality filters and also reducing DSP (RF noise) inside the DAC box (away from analogue electronics...)

Edited February 28, 2019 by Music2496

[Ittaku]

Just now, Music2496 said:

 

Noted. I referred to up-sampling before the DAC, both using higher quality filters and also reducing DSP inside the DAC box (away from analogue electronics...)

Gotcha, like the deconstructed smashed avo digital I was talking about earlier.

[rand129678]

17 minutes ago, Ittaku said:

If he wasn't so hung up about using an fpga, and did it using commodity PC components he could easily do it... Upsampling and oversampling has been an integral part of most DACs since time immemorial. Massive tap filters has only been a recent thing thanks to modern PC and FPGA performance.

If you chat with him on Head-Fi enough, you will know he really loves his Hugo2 when travelling (when he can't carry his Dave/TT2/M-Scaler around, although he does that too).

 

This is why he's said he's trying to figure out how to get 1 million taps into a portable DAC/amp but he says the RF isolation from the analogue components is a nightmare.

 

 

Can you achieve the same processing power with commodity PC components, whilst simultaneously having the same footprint as the FPGA he uses in Hugo2, for example? The former is not a problem of course (there are hundreds, maybe thousands of HQPlayer + foobar + sox users around the world). The latter (keeping footprint tiny) may be the challenge? I don't know.

 

 

 

Edited February 28, 2019 by Music2496

[Ittaku]

Just now, Music2496 said:

Can you achieve the same processing power with commodity PC components, whilst simultaneously having the same footprint as the FPGA he uses in Hugo2, for example? The former is not a problem of course (there are hundreds, maybe thousands of HQPlayer + foobar + sox users around the world). The latter (keeping footprint tiny) may be an issue? I don't know.

I'd say yes. NUCs are pretty small and allow for some very powerful CPUs and tons of memory these days, and they're off the shelf equipment, with 10x10cm boards. Custom made ones could be even smaller. I guarantee it's possible. However, when you have a hammer, everything looks like a nail; and the FPGA is Chord's hammer.

[rand129678]

7 minutes ago, Ittaku said:

I'd say yes. NUCs are pretty small and allow for some very powerful CPUs and tons of memory these days, and they're off the shelf equipment, with 10x10cm boards. Custom made ones could be even smaller. I guarantee it's possible. However, when you have a hammer, everything looks like a nail; and the FPGA is Chord's hammer.

Noted. I think people think assume the FPGA is only doing up-sampling and noise shaping... and forget it’s also doing the incoming jitter rejection and a couple other key functions (have to search Head-Fi but it’s all there...)

 

The challenge would be doing all of that plus input interfaces plus headamp section plus power supply electronics plus analogue electronics, into something with overall footprint of Hugo2 to fit in your pocket... with an FPGA he can do it all (save for 1M WTA tap filtering, for now...).

 

Before anyone asks why Hugo2 is important, you need to spend some time on Head-Fi and talk to Rob himself about why Hugo2 is important for him.

 

Gotta be careful playing the armchair DAC designer game.... we all do it of course but there’s always more to it than meets the eye.

 

 

 

[Ittaku]

1 minute ago, Music2496 said:

Noted. I think people think assume the FPGA is only doing up-sampling and noise shaping... and forget it’s also doing the incoming jitter rejection and a couple other key functions (have to search Head-Fi but it’s all there...)

 

The challenge would be doing all of that plus input interfaces plus headamp section plus power supply electronics plus analogue electronics, into something with overall footprint of Hugo2 to fit in your pocket... with an FPGA he can do it all (save for 1M WTA tap filtering, for now...).

 

Before anyone asks why Hugo2 is important, you need to spend some time on Head-Fi and talk to Rob himself about why Hugo2 is important for him.

 

Gotta be careful playing the armchair DAC designer game.... we all do it of course but there’s always more to it than meets the eye.

Oh I was simply talking about the upsampling component, I know nothing about implementing a DAC. I know heaps about PC hardware and tons more about software.

[rand129678]

5 minutes ago, Ittaku said:

Oh I was simply talking about the upsampling component, I know nothing about implementing a DAC. I know heaps about PC hardware and tons more about software.

That’s where the bigger picture comes into it, when saying “If he wasn't so hung up about using an fpga”

 

His end game is having 1M taps (or more later) inside a DAC, that fits in pocket that can carry when travelling.

Edited February 28, 2019 by Music2496

[Ittaku]

1 minute ago, Music2496 said:

That’s where the bigger picture comes into it, when saying “If he wasn't so hung up about using an fpga”

 

His end game is having 1M taps inside a DAC, that fits in pocket that can carry when travelling.

Roger. Though he said himself the RF noise is the issue when close to the DAC. Me, I already have something like 4 components making up my digital chain (after the PC) so another would be no big deal to me, it's just there are limits to what sample rates I can use with my chosen combination of DSP and DAC. That's why I'm getting others to test the higher sample rates. Again, the DAC itself is far more important than the upscaling here, I'm just looking for solutions to optimise what I can. An MSB reference DAC without any upscaling sounds light years better than a Qutest with an MScaler.

[rand129678]

 

3 minutes ago, Ittaku said:

Roger. Though he said himself the RF noise is the issue when close to the DAC.

Yes, I’m the one that shared that quote. Also shared the quote where he states this is what he’s working on solving, longer term...

 

He’ll work it out. Dave 2 is apparently a long way away (not even on his mind he says) but I’m sure he’ll have it sorted by then. Then that will filter down to the portables eventually.

 

 

[legend]

3 hours ago, Music2496 said:

 

I'm using HQPlayer's "sinc-M" filter, up-sampling to PCM705/768kHz, on the fly...

 

Sinc function filter, 1 million taps, > 240dB stop-band attenuation, on the fly.

 

There is already a 16 million tap filter in HQP for SDM up-sampling, so I'm sure the 16M tap filter will come to PCM up-sampling too.

 

Archimago recently included some measurements, using this Oppo 205 output feeding his A-to-D converter. Miska notes cleaner looking output from his Pro-Ject S2 DAC feeding his RME ADI-2 Pro...

 

https://archimago.blogspot.com/2019/02/measurements-look-at-hqplayer-325.html

 

I'm using a headless server. No interaction with a keyboard, monitor, mouse...

 

HQP is configured via web browser on iPad...

 

 

One of the reasons for me not going down the HQplayer path is that I found it quite difficult to discover anything useful about the number of taps it uses and could they be varied etc - for example by putting 'HQplayer' and "taps' into Google.  I guess I could troll through a user group but usually find the signal to noise ratio often so low it not worthwhile eg the Official M-Scaler forum on Head Hi-Fi.

 

I could also not find a user manual on a reasonably recent version (only one from 2016) and gather you have to download HQplayer before you can access one.  I respect the right of companies to protect their IP but in this case it out me off!

[rand129678]

16 minutes ago, Ittaku said:

That's why I'm getting others to test the higher sample rates.

 

While I’m a long time and happy HQPlayer user, doing everything on the fly (no patience for offline upsampling) I’d love to have a listen to your files if possible?

 

Can you PM me a link to an original file plus your best mega up-sampled version?

 

A PCM705/768k version with highest/best combination of taps and stop-band attenuation?

Edited February 28, 2019 by Music2496

[rand129678]

6 minutes ago, legend said:

One of the reasons for me not going down the HQplayer path is that I found it quite difficult to discover anything useful about the number of taps it uses and could they be varied etc - for example by putting 'HQplayer' and "taps' into Google.  I guess I could troll through a user group but usually find the signal to noise ratio often so low it not worthwhile eg the Official M-Scaler forum on Head Hi-Fi.

 

I could also not find a user manual on a reasonably recent version (only one from 2016) and gather you have to download HQplayer before you can access one.  I respect the right of companies to protect their IP but in this case it out me off!

It’s easy! No PhD required

 

All these questions can be directed to Jussi btw. He doesn’t spend time talking about taps much because it’s only one of the key parameters... but you can ask him anything.

 

And the ‘M’ in his sinc-M filter stands for million... as in 1 million taps...

 

[Ittaku]

Uploading a 705.6kHz sample as we speak. Given the dodgy upload speed of (me still being stuck on) ADSL, it'll be another hour and a half before it's available. Will PM the offending parties when it's time.

[rand129678]

8 minutes ago, Ittaku said:

Uploading a 705.6kHz sample as we speak. Given the dodgy upload speed of (me still being stuck on) ADSL, it'll be another hour and a half before it's available. Will PM the offending parties when it's time.

Awesome. How many taps and what's the stop-band attenuation you were able to get? Extremely steep?

 

 

Edited February 28, 2019 by Music2496

[Ittaku]

18 minutes ago, Music2496 said:

Awesome. How many taps and what's the stop-band attenuation you were able to get? Extremely steep?

30 million is more than enough, but I used 360 million. I usually limit attenuation to 170dB since it's so far below the bit depth that even dithering or other noise can no longer bring it out in the signal, so that's the attenuation. The 0dB bandwidth is 99.9999%.

Edited February 28, 2019 by Ittaku

[rand129678]

7 minutes ago, Ittaku said:

30 million is more than enough, but I used 360 million. I usually limit attenuation to 170dB since it's so far below the bit depth that even dithering or other noise can no longer bring it out in the signal, so that's the attenuation. The 0dB bandwidth is 99.9999%.

This is lower than Chord though.

 

Rob Watts: I am actually very pleased by the sound of the new filter; but to do this I needed over 200 dB of attenuation, to ensure zero aliasing.

 

And HQPlayer has >240dB stop-band attenuation with the sinc-M filter, as I previously mentioned.

 

I know talking about higher than 170 dB attenuation sounds  ridiculous but Rob talks about perception of depth and transients improving... And HQPlayer believes this is more important than just focusing on number of taps alone (which is why the dev has never talked about tap numbers much before)...

 

How many taps can you do at PCM705k with 192dB stop-band attenuation (32-bits) and same extreme steepness?

 

Edited February 28, 2019 by Music2496

[Ittaku]

3 minutes ago, Music2496 said:

This is lower than Chord though.

 

Rob Watts: I am actually very pleased by the sound of the new filter; but to do this I needed over 200 dB of attenuation, to ensure zero aliasing.

 

And HQPlayer has >240dB stop-band attenuation with the sinc-M filter, as I previously mentioned.

 

I know talking about higher than 170 dB attenuation sounds  ridiculous but Rob talks about perception of depth and transients improving... And HQPlayer believes this is more important than just focusing on number of taps alone (which is why the dev has never talked about tap numbers much before)...

 

How many taps can you do at PCM705k with 192dB stop-band attenuation (32-bits) and same extreme steepness?

 

There's not much difference in number of taps 360->430, it only uses more ram. I think that's marketing now. I get 204dB attenuation if I just choose 32 bits. There is no mechanism for aliasing to occur at -170dB with 24 bits of resolution...

Edited February 28, 2019 by Ittaku