HQ Player

[Miska]

6 hours ago, copy_of_a said:

 

A question re the gauss-hb and gauss-hb-s filters.

The manual says they are "(slightly) leaky around Nyquist but extremely high attenuation".

May I ask how much attentuation they provide?

 

Actually the attenuation comment in manual regarding gauss-halfband-s is a copy-paste error I have already fixed for next release.

 

Both have very similar transition, but halfband-s trades attenuation to do same in shorter length. They reach full attenuation by 24.1 kHz which is where image of 20 kHz tone would be.

 

poly-sinc-gauss-halfband (44.1k source):

poly-sinc-gauss-halfband-s (44.1k source):

Thus, -s is tuned to reach 25-bit accuracy while former plot is mostly limited by the plotting tool.

 

5 hours ago, Schafheide said:

I agree - gauss-hb-s is, indeed great - maybe an apodizing version would be useful - if it is possible to produce such a version?

 

It is not possible. By definition, halfband filters pass the original data through as-is and only add new samples between. This also means it cannot fix any errors in the source data and will thus faithfully reproduce all the errors as well. This is why the manual also says "Only suitable for highest technical quality source materials". Which means it is not suitable for 95+% of RedBook content.

 

[Miska]

6 hours ago, audiofool said:

Would there be any advantage or disadvantage to re-dither 44/16 cd source before processing with hqplayer to sdm output?

My thinking is similar to using apodizing filters to fix issues, maybe adding dither is good in some cases? 

 

Bad idea... In particular with 16-bit sources.

 

54 minutes ago, bogi said:

Dithering is not applicable to DSD data, since randomizing values of lowest bit(s) makes sense with multibit binary weighted data presentation like PCM encoded audio is. An equivalent thing to dithering for SDM (DSD) data is noise shaping.

 

It is also applicable to SDM, but in a different way. HQPlayer modulators are dithered, which also means that if you run same source data through the processing multiple times, you will never get same output bitstream twice.

 

[audiofool]

17 minutes ago, Miska said:

 

Bad idea... In particular with 16-bit sources.

 

 

It is also applicable to SDM, but in a different way. HQPlayer modulators are dithered, which also means that if you run same source data through the processing multiple times, you will never get same output bitstream twice.

 

I was thinking specifically for 16-bit sources.  In the case where a cd master is truncated to 16 bits instead of dithered - I guess once the damage is done...

[copy_of_a]

2 hours ago, Miska said:

Actually the attenuation comment in manual regarding gauss-halfband-s is a copy-paste error I have already fixed for next release.

 

Both have very similar transition, but halfband-s trades attenuation to do same in shorter length. They reach full attenuation by 24.1 kHz which is where image of 20 kHz tone would be.

Many thanks, Miska - much appreciated!!

 

2 hours ago, Miska said:

"Only suitable for highest technical quality source materials". Which means it is not suitable for 95+% of RedBook content.

This morning I‘ve recorded the output of my Desktop-DAC with gauss-hb and gauss-hb-s.

With a Tascam DR100 🙄 - but, hey, that‘s all I have to digitally record analog line level input. Anyway.

I have a 44.1kHz/16 album that doesn‘t trigger the Apodizing Counter in HQP due to a steep high quality LPF but the actual musical content goes up to 22K (with quite some energy).

While I would find the images produced by the hb filter at least acceptable (images peak below -90dB) the hb-s filter shows pretty strong images. It‘s fine, though, with recordings that roll off gently and don‘t contain high levels above 20kHz.

So I would say the Apodizing Counter is a great guide if you use for instance ps-gauss-short but the use of gauss-hb-s sure requires further attention. With an appropriate recording it‘s really great!

[Miska]

24 minutes ago, copy_of_a said:

While I would find the images produced by the hb filter at least acceptable (images peak below -90dB) the hb-s filter shows pretty strong images. It‘s fine, though, with recordings that roll off gently and don‘t contain high levels above 20kHz.

So I would say the Apodizing Counter is a great guide if you use for instance ps-gauss-short but the use of gauss-hb-s sure requires further attention. With an appropriate recording it‘s really great!

 

Yes, of course the regular hb reaches higher attenuation faster in the transition region since it also has higher attenuation. In that sense the hb-s is slower roll-off even though they reach the set attenuations roughly at same point.

 

[Kalpesh]

17 hours ago, Miska said:

 

It is something I actually made because someone asked what would be closest to Mola-Mola filter, and there weren't any. So this one fairly closely matches the one used by Mola-Mola. And it is very similar to most DAC chip filters (ESS etc) in terms of response as well. Relatively slow roll-off and fairly low attenuation.

 

Main difference to Mola-Mola and DAC chip filters though is that these are single stage to final output rate.

 

Since halfband filters are non-apodizing, those are only suitable for content where the Apod counter stays 0 (or close, like under 10).

 

it being hb xs

reading Bruno Putzeys quoted here https://www.stereophile.com/content/gramophone-dreams-55-mola-mola-tambaqui-da-processor-page-2 raised a few questions 

 

which, if any, HQP filters would you consider as euphonic, is BP suggesting anything longer than xtra short is euphonic ?

 

it turns out that a lot of filters out there sound really impressive, but only because they're heavily euphonic—not because they're sonically neutral....

 

does the following only relates to apodizing issues ? how it translates when picking HQP filters ?

 

To make matters worse, the optimum design differs for different sampling rates....

 

to the best of your knowledge are Holo DACS plagued by one of these issues (raises the interesting issue that, eventually, there is  more to the picture than noise level with some DACs when comparing SDM and PCM routes

 

one item was a completely signal-independent noise floor. If the noise floor modulates, that immediately gives the sound away as 'digital.' If you want to see a typical sigma-delta converter at its worst, feed it a constant code (ie DC). Some DC values produce clearly audible whistles. If you want to tease an R2R DAC in a similar manner, feed it a ramp signal. I wanted something where the noise floor is truly just a steady hiss, no matter what signal you put in. This is what drove the choice of single-edge PWM as the digital intermediate code.

 

last but not least, I don't know how to interpret the praise for the Tambaqui if its filter is as ordinary as I read here : closely matches the one used by Mola-Mola. And it is very similar to most DAC chip filters (ESS etc) in terms of response as well. the choice of single-edge PWM as the digital intermediate code?

[1laraz]

3 hours ago, Miska said:

 

 

 

 

I think it would be very helpful to have such charts for each filter.

Someone asked here not long ago for a comparison between filters based on absolute scales in time/frequency domain. 

[Zauurx]

Since version 5, with its lighter modulators, improvements, performance optimizations and acceptance of DSD256 as a "sweetspot" rather than 512 (Gustard A26), my server runs without a problem.
The i5-11500 is pushed to over 4ghz (maybe even too much) and I don't even use the "Multicore DSP" anymore.
The number of threads has dropped considerably!
The hardest part is tweaking the settings to keep ndis latency to a minimum.
I'll try DSD512 again, but with the A26, there's no gain (AKM chip limitation... as I understand it).

 

 

[bogi]

6 hours ago, Miska said:

It is also applicable to SDM, but in a different way. HQPlayer modulators are dithered, which also means that if you run same source data through the processing multiple times, you will never get same output bitstream twice.

If I well understand, 1bit SDM data cannot be dithered and you probably mean HQPlayer processing with PCM input and SDM conversion. You could mean that oversampling result (PCM signal at modulator rate) is dithered in HQPlayer before it is passed to delta sigma modulator. Am I guessing correctly? Does such a thing happen in DAC chips too?

 

Or do you use the dithering term in the context of noise shaping, something like implicit effect of noise shaping?

[bogi]

16 minutes ago, Zauurx said:

I'll try DSD512 again, but with the A26, there's no gain (AKM chip limitation... as I understand it).

 

With DSD512 input, if DSDD bit of AK4191EQ is set by firmware to 1, the chip has to downsample DSD512 (about 22.6 MHz) input to half rate (about 11.3), but it increases bith depth to some 5 to 7 bits ( if I well remember) binary weighted, so the conversion result may still contain more source audio related information than in the case of DSD256 input. Important is how it sounds for you (and measures).

[Zauurx]

51 minutes ago, bogi said:

 

With DSD512 input, if DSDD bit of AK4191EQ is set by firmware to 1, the chip has to downsample DSD512 (about 22.6 MHz) input to half rate (about 11.3), but it increases bith depth to some 5 to 7 bits ( if I well remember) binary weighted, so the conversion result may still contain more source audio related information than in the case of DSD256 input. Important is how it sounds for you (and measures).

 

This completes Jussy's comment on the operation of the chip  AK4499EX works (?) like here :

https://community.roonlabs.com/t/best-native-dsd-dacs-for-use-with-hqplayer/132298/231?

 

Quote

DSD 512 is not native rate for the AK4499EX chip which always operates at 128fs or 256fs speed. It cannot operate at 512fs or higher, or 64fs or lower. So any other rate requires rate down-conversion for the converter chip. This results in worse audio band SNR performance at DSD512 because some of the ultrasonic noise is aliased back to audio band. Distortion remains unchanged though.

 

In the end, it's up to me to test and see or rather listen if I like it better ?

[Miska]

4 hours ago, Kalpesh said:

which, if any, HQP filters would you consider as euphonic, is BP suggesting anything longer than xtra short is euphonic ?

 

it turns out that a lot of filters out there sound really impressive, but only because they're heavily euphonic—not because they're sonically neutral....

 

No, length of the filters cannot make sound more or less "euphonic". Leaky filters could sound "euphonic", like NOS.

 

4 hours ago, Kalpesh said:

does the following only relates to apodizing issues ? how it translates when picking HQP filters ?

 

To make matters worse, the optimum design differs for different sampling rates....

 

It has mostly to do with 1x rate filters vs Nx rate filters. For that reason there's for example the poly-sinc-gauss-hires filter.

 

1x rates are particularly difficult to deal with, hires sources are quite a bit easier in many ways.

 

4 hours ago, Kalpesh said:

to the best of your knowledge are Holo DACS plagued by one of these issues (raises the interesting issue that, eventually, there is  more to the picture than noise level with some DACs when comparing SDM and PCM routes

 

one item was a completely signal-independent noise floor. If the noise floor modulates, that immediately gives the sound away as 'digital.' If you want to see a typical sigma-delta converter at its worst, feed it a constant code (ie DC). Some DC values produce clearly audible whistles. If you want to tease an R2R DAC in a similar manner, feed it a ramp signal. I wanted something where the noise floor is truly just a steady hiss, no matter what signal you put in. This is what drove the choice of single-edge PWM as the digital intermediate code.

 

When you run Holo DACs in NOS mode from external upsampling, those things don't depend on the DAC itself. Since it is not doing any DSP, just converting digital to analog. It all depends on quality of the algorithms.

 

Funny thing that both Bruno and Rob talk about same things, but absolutely disagree on the correct approach... 😀

 

4 hours ago, Kalpesh said:

last but not least, I don't know how to interpret the praise for the Tambaqui if its filter is as ordinary as I read here : closely matches the one used by Mola-Mola. And it is very similar to most DAC chip filters (ESS etc) in terms of response as well.

 

It's filter is pretty ordinary.

 

For some reason filters get a lot of attention, but modulators rarely get any mention. Although that is other 50% of the DAC digital side performance.

 

[Miska]

4 hours ago, 1laraz said:

I think it would be very helpful to have such charts for each filter.

Someone asked here not long ago for a comparison between filters based on absolute scales in time/frequency domain. 

 

But is it really helpful? Or just creates some preconception? And how many readers would get useful information out of those, without misinterpretations etc?

 

[Miska]

1 hour ago, Zauurx said:

Since version 5, with its lighter modulators, improvements, performance optimizations and acceptance of DSD256 as a "sweetspot" rather than 512 (Gustard A26), my server runs without a problem.
The i5-11500 is pushed to over 4ghz (maybe even too much) and I don't even use the "Multicore DSP" anymore.

 

Clocks get pushed higher up, because more work is on fewer cores.

 

1 hour ago, Zauurx said:

The number of threads has dropped considerably!

 

As result, there's less workload distribution to different cores... Threads are the mean to place work on different cores.

 

Having some threads existing doesn't yet mean anything itself. Because for example in above case, most of those threads are sleeping and thus not being executed. They are woken up only when there is work for them.

 

1 hour ago, Zauurx said:

I'll try DSD512 again, but with the A26, there's no gain (AKM chip limitation... as I understand it).

 

Yes, because the AK4499EX can only operate at 5.6 or 11.2 MHz. So DSD512 needs to be converted by the AK4191 to half the sampling rate before sending it out to AK4499EX. It means that it is not DSD Direct anymore. Direct only works at 128x and 256x speeds.

 

[bogi]

14 minutes ago, Zauurx said:

This completes Jussy's comment on the operation of the chip  AK4499EX works (?) like here

 

Yes. He discussed 4499EX in that quote and I discussed 4191EQ. Jussi measured A26 it so he knows what's the sweet spot from technical point of view.

 

  

14 minutes ago, Zauurx said:

In the end, it's up to me to test and see or rather listen if I like it better ?

 

Of course, listening is base. It's about what you like to hear on your chain. It is very useful to know also technical details and to search for correlation between them and the sound you hear. Improvements attempts may be done based on technical information, but on the end it is you who decides if you like it or not, if it is worth the money or not.

[Miska]

1 hour ago, bogi said:

If I well understand, 1bit SDM data cannot be dithered

 

Yes it can. And HQPlayer modulators are dithered. Number of bits doesn't matter. We are not talking about PCM bits here.

 

1 hour ago, bogi said:

You could mean that oversampling result (PCM signal at modulator rate) is dithered in HQPlayer before it is passed to delta sigma modulator.

 

No, because there's no word length reduction involved.

 

1 hour ago, bogi said:

Am I guessing correctly?

 

No...

 

1 hour ago, bogi said:

Does such a thing happen in DAC chips too?

 

I don't think so. They already cut so many corners in oversampling area that it would be pretty pointless.

 

1 hour ago, bogi said:

Or do you use the dithering term in the context of noise shaping, something like implicit effect of noise shaping?

 

Also HQPlayer's PCM noise-shapers are dithered. Plain noise-shaper would always produce same data for same input. That would be bad.

 

[bogi]

2 minutes ago, Miska said:

HQPlayer modulators are dithered. Number of bits doesn't matter. We are not talking about PCM bits here.

 

Thanks! I hear about dithering of SDM data for the first time. Do you know about any other products or solutions, which use SDM dithering?

[bogi]

8 minutes ago, Miska said:

Plain noise-shaper would always produce same data for same input. That would be bad.

 

I'm curious, why is that bad?

[Miska]

3 minutes ago, bogi said:

I'm curious, why is that bad?

 

Because on repetitive data it would produce repetitive patterns that create spurious tones. In a slightly similar way as you get from truncation/rounding instead of proper dithering.

 

I have also seen such in my measurements from various devices...

 

[1laraz]

28 minutes ago, Miska said:

 

But is it really helpful? Or just creates some preconception? And how many readers would get useful information out of those, without misinterpretations etc?

 

Well, users periodically ask for these charts and you provide them. All is given now is the summary table with subjective description of filters ("space" etc). 

[semente]

15 hours ago, audiofool said:

A Dither Question:

Would there be any advantage or disadvantage to re-dither 44/16 cd source before processing with hqplayer to sdm output?

My thinking is similar to using apodizing filters to fix issues, maybe adding dither is good in some cases? 

 

 

 

I think that HQPlayer does it for you when you convert sample rates, as it is required for any up- or down-conversion.

Most HQPlayer filters are apodizing.

[juniorbudel]

Here, version 5.2 made the CPU load a little lighter, but it was still not possible to run a 48khz rate with the EC7 V3 super modulator with upsampler for DSD256 in MM M2 Pro, now a crash occurs every 30 seconds, which already It's better than what I had before, I'll get it in the next updates!

[Miska]

1 hour ago, bogi said:

Thanks! I hear about dithering of SDM data for the first time. Do you know about any other products or solutions, which use SDM dithering?

 

Yes, some do. Some others don't. So it varies.

[Miska]

1 hour ago, 1laraz said:

Well, users periodically ask for these charts and you provide them. All is given now is the summary table with subjective description of filters ("space" etc). 

 

Yes, originally it didn't have subjective description either. But those subjective descriptions are based on theory. And one cannot really deduct those from the graphs even if I would add the graphs. I have considered adding the graphs. But I still ask myself what purpose those graphs would actually serve? Which way it would be actually useful?

[Zauurx]

Please, can you explain the impact of the FFT setting? and its impact depending on the choice of modulators/filters/convolution/SDM/music ?
512 / 1024 or much more ??

 

https://en.wikipedia.org/wiki/Fast_Fourier_transform