HQ Player

[StreamFidelity]

43 minutes ago, Miska said:

But I still ask myself what purpose those graphs would actually serve? Which way it would be actually useful?

Sometimes pictures say more than words. It would help to understand what the descriptions of the filters in the manual mean. Something like slow- or sharp-roll-off.

 

I would not publish the diagrams in the manual. I would find a folder on the Signalyst server with filter diagrams useful. 

[Miska]

14 minutes ago, Zauurx said:

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 ??

 

It is length selection for the FFT filter. Now the filter is available also for SDM oversampling. The filter has been around for PCM for ages, and the configuration value has been in the settings file, but always been set to the default 512. I just thought I'd make it available in the settings dialog now that there is some extra space in the GUI. Default value of 512 is good. I personally wouldn't go below 256 or above 2048. It only affects steepness of the filter. Longer the filter, faster the roll-off. Smaller values gives gentler roll-off curve.

 

[Jud]

7 minutes ago, Miska said:

 

It is length selection for the FFT filter. Now the filter is available also for SDM oversampling. The filter has been around for PCM for ages, and the configuration value has been in the settings file, but always been set to the default 512. I just thought I'd make it available in the settings dialog now that there is some extra space in the GUI. Default value of 512 is good. I personally wouldn't go below 256 or above 2048. It only affects steepness of the filter. Longer the filter, faster the roll-off. Smaller values gives gentler roll-off curve.

 

 

So number of "taps."

[Miska]

7 minutes ago, Jud said:

So number of "taps."

 

Yes, exactly. Which is length of the filter in time. Thus "512" is 512 samples long in time. 512 / 44100 =  ~12 ms.

 

[Kalpesh]

3 hours ago, Miska said:

 

 

 

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.

 

Thank you very much for another very instructive post

on Wingless Angels (tested with Tambaqui by Stereophile) hb xs yields impressively fast yet deep percussions though you say it's ordinary !

on the Horowitz tested by Stereophile it lacks density imo

[dericchan1]

2 hours ago, Miska said:

 

It is length selection for the FFT filter. Now the filter is available also for SDM oversampling. The filter has been around for PCM for ages, and the configuration value has been in the settings file, but always been set to the default 512. I just thought I'd make it available in the settings dialog now that there is some extra space in the GUI. Default value of 512 is good. I personally wouldn't go below 256 or above 2048. It only affects steepness of the filter. Longer the filter, faster the roll-off. Smaller values gives gentler roll-off curve.

 

Hi Miska, this maybe a silly question. Let’s say I am already using a very long filter like Sinc Long or Sinc Mx, could increasing the fft value makes the filters even longer are they are different filters altogether?

[Miska]

8 minutes ago, dericchan1 said:

Hi Miska, this maybe a silly question. Let’s say I am already using a very long filter like Sinc Long or Sinc Mx, could increasing the fft value makes the filters even longer are they are different filters altogether?

 

It only affects filter called "FFT", nothing else.

 

[copy_of_a]

@Miska

Today I've played around with filters.

Transcoded some stuff (dirac impulses and music-cuts) through HQP Pro and found that poly-sinc-short (lp/mp) produces mirrored images.

Is ps-short not supposed to be apodizing?

44.1kHz/16 upsampled to 88.2kHz/24 (ps-short-lp):

[Miska]

21 minutes ago, copy_of_a said:

@Miska

Today I've played around with filters.

Transcoded some stuff (dirac impulses and music-cuts) through HQP Pro and found that poly-sinc-short (lp/mp) produces mirrored images.

Is ps-short not supposed to be apodizing?

 

It is apodizing, but as name says, it is short (very short) slow roll-off.

 

[copy_of_a]

7 minutes ago, Miska said:

It is apodizing, but as name says, it is short (very short) slow roll-off.

ok, thanks!

But itsn't the main purpose of apodizing filters to surpress images?

I've compared ps-short to ps-gauss short and the latter produces even a little less images (due to higher attenuation I assume) although it is non-apodizing.

I am confused 😐

[Miska]

36 minutes ago, copy_of_a said:

But itsn't the main purpose of apodizing filters to surpress images?

 

No, it is to correct errors in the source data and replace the ADC decimation filter's impulse response with alternative one.

 

All filters are supposed to suppress images.

 

For example gauss-xla and gauss-xl both suppress images. But only former one is apodizing.

 

36 minutes ago, copy_of_a said:

I've compared ps-short to ps-gauss short and the latter produces even a little less images (due to higher attenuation I assume) although it is non-apodizing.

I am confused 😐

 

Yeah, it is not exactly black and white on borderline cases like these. I should actually mark gauss-short as apodizing as well. For some reason I have not... The two are quite similar - but different.

 

[copy_of_a]

38 minutes ago, Miska said:

 

No, it is to correct errors in the source data and replace the ADC decimation filter's impulse response with alternative one.

 

All filters are supposed to suppress images.

 

For example gauss-xla and gauss-xl both suppress images. But only former one is apodizing.

 

 

Yeah, it is not exactly black and white on borderline cases like these. I should actually mark gauss-short as apodizing as well. For some reason I have not... The two are quite similar - but different.

 

Makes snese - many thanks 👍

[sumitc]

@Miska I have multi-channel output from HQPlayer to my surround sound processor which is limited to 192kHz processing.  The SSP digital outputs are connected to a dac8Pro.  Most of the content that I am playing back is 48kHz or 44.1kHz.  I have been using Gauss1, but from the documentation several NS options, 1/4/9, could also get used.  From a theoretical standpoint, what would the best option be here?  I am also wondering how NS in HQPlayer would interact from presumably some additional NS that would be done in the ESS DAC.  Thx!

[Miska]

2 minutes ago, sumitc said:

@Miska I have multi-channel output from HQPlayer to my surround sound processor which is limited to 192kHz processing.  The SSP digital outputs are connected to a dac8Pro.  Most of the content that I am playing back is 48kHz or 44.1kHz.  I have been using Gauss1, but from the documentation several NS options, 1/4/9, could also get used.

 

If you are just doing conversion to 192k output, without additional processing such as volume control (more than the small about -3 dB headroom gain) or room correction or something else, using TPDF or Gauss1 dither is good option. You can also gain some increase in audio band SNR with noise-shapers, but for such case the additional headroom is not so critical since the SNR is anyway limited by the source content.

 

2 minutes ago, sumitc said:

 From a theoretical standpoint, what would the best option be here?  I am also wondering how NS in HQPlayer would interact from presumably some additional NS that would be done in the ESS DAC.  Thx!

 

Since the HQPlayer PCM shapers are relatively gentle, and in this case additionally bandwidth limited, there's no interaction with the ESS modulator's noise shaping. They are on purpose designed such way that there are no negative interactions.

 

[Miska]

Here are plots for poly-sinc-short:

 

And poly-sinc-gauss-short:

 

Both are specified for -192 dB stop-band attenuation matching 32-bit PCM accuracy. This way they are kept as short as possible while reaching very good reconstruction accuracy.

 

For standard 19.1 kHz image attenuation test tone used for example by Stereophile (see below) where image lands at 25 kHz, pretty much full attenuation is reached in both cases. And response is down by -26 dB at 22.05 kHz.

 

If we compare for example to Mola-Mola Tambaqui, which reaches about -36 dB attenuation by 25 kHz using such test tone (blue plot):

So Mola-Mola falls in category of slow roll-off. Although still not nearly as slow as MQA filter which is down only by -18 dB at same point. Or ESS slow roll-off which is down by -12 dB. These are however only down by -3 dB at 22.05 kHz (green line), so these are halfband filters and thus will have notable leak around Nyquist.

 

[paopawdecarabao]

Hi,

 

I am new here and I am interested on HQplayer. I am more having fun playing music from digital atm. I am only streaming on Qobuz through my Lumin u1 mini to a tube DAC. But I want to change my lumin to the Denafrips Arce Streamer which is a Hqplayer NAA and get a Denafrips Dac as well.

 

I have a Macbook Air M1 8gb to plan to try with HQ player. How do I connect my MBA to the Arve streamer? MBA usb hardwired directly to the arce streamer or is there an option I can wireless stream from HQplayer using Qobuz to the Arce streamer?

 

sorry if I am kinda new to this because my Lumin has an app that I could just stream through my network wireless using Qobuz.

 

i am just confused if I just hardwire my laptop to the arce streamer then to the dac. Please advise. 

[Kalpesh]

33 minutes ago, Miska said:

Here are plots for poly-sinc-short:

 

And poly-sinc-gauss-short:

 

Both are specified for -192 dB stop-band attenuation matching 32-bit PCM accuracy. This way they are kept as short as possible while reaching very good reconstruction accuracy.

 

For standard 19.1 kHz image attenuation test tone used for example by Stereophile (see below) where image lands at 25 kHz, pretty much full attenuation is reached in both cases. And response is down by -26 dB at 22.05 kHz.

 

If we compare for example to Mola-Mola Tambaqui, which reaches about -36 dB attenuation by 25 kHz using such test tone (blue plot):

So Mola-Mola falls in category of slow roll-off. Although still not nearly as slow as MQA filter which is down only by -18 dB at same point. Or ESS slow roll-off which is down by -12 dB. These are however only down by -3 dB at 22.05 kHz (green line), so these are halfband filters and thus will have notable leak around Nyquist.

 

The real challenge is trying to figure out how a bunch of parameters affect reconstruction and conversion of digital so that it sounds drastically different and that truth seems out of reach. I doubt I could peak a filter and guess how it will fulfil my pleasure with given track by looking at graphs... But I avidly keep taking your insights

[Miska]

43 minutes ago, Kalpesh said:

The real challenge is trying to figure out how a bunch of parameters affect reconstruction and conversion of digital so that it sounds drastically different and that truth seems out of reach. I doubt I could peak a filter and guess how it will fulfil my pleasure with given track by looking at graphs... But I avidly keep taking your insights

 

You see that's why I have not been putting numbers and graphs to the manual. But instead overall textual descriptions and suggestions. Some number, or a graph don't really give a good answer. Answer is more complex, it would be long series of mathematical formulas.

 

[1laraz]

15 minutes ago, Miska said:

 

You see that's why I have not been putting numbers and graphs to the manual. But instead overall textual descriptions and suggestions. Some number, or a graph don't really give a good answer. Answer is more complex, it would be long series of mathematical formulas.

 

For general educational purposes it is still better to see once these "slow roll-offs" once rather than to read about it many times and guessing what that should mean. I agree that the charts and numbers may be not suitable for the manual, but for some support page on the Signalyst website for those who are interested in it.

[Miska]

11 minutes ago, 1laraz said:

For general educational purposes it is still better to see once these "slow roll-offs" once rather than to read about it many times and guessing what that should mean. I agree that the charts and numbers may be not suitable for the manual, but for some support page on the Signalyst website for those who are interested in it.

 

It would be useful to combine those with a suitable time domain graphs. Otherwise it becomes way too much focused on just frequency domain. But I prefer different kind of graphs for those rather than the basic impulse response plot...

 

[StreamFidelity]

34 minutes ago, Miska said:

But I prefer different kind of graphs for those rather than the basic impulse response plot...

Based on your publications in this forum, I have made a drawing. Corrections are welcome.

 

 

It shows symbolically, which frequencies are let through (passband), from when the blocking effect (cutoff) starts and how long it takes (transition range), until the lowpass filter develops its full effect (stopband). If there is a better representation for it just give it to us.

 

There are really bad filters, which for example start too early or too late with the cutoff. Or do not attenuate enough. Mostly implemented in DACs. 

 

Another interesting question is the effectiveness of the modulators, which are supposed to be 50%. Can this be shown in a diagram?

[Miska]

7 minutes ago, StreamFidelity said:

Based on your publications in this forum, I have made a drawing. Corrections are welcome.

 

That explains the basic parameters (at least if you add passband ripple) in a traditional way. But it doesn't go into other differences between different filters that affect how it sounds.

 

7 minutes ago, StreamFidelity said:

Another interesting question is the effectiveness of the modulators, which are supposed to be 50%. Can this be shown in a diagram?

 

One could show some basic figures in a similar way. It becomes sort of inverse of the graph you did. But it would tell even less about differences between modulators because they are much more complex animals than filters.

 

[Kalpesh]

4 hours ago, Miska said:

 

You see that's why I have not been putting numbers and graphs to the manual. But instead overall textual descriptions and suggestions. Some number, or a graph don't really give a good answer. Answer is more complex, it would be long series of mathematical formulas.

 

do those formulas include (or even are made of) integral calculus ? I don't believe in (reasonable) rolloff accounting much for audible differences etc. I'm strictly profane there and am probably mathematically wrong (does not matter if in the end I learn, understand better)but the way I explain to myself the differences in sound comes from balance made (by you as of HQP) in the brackets/range of integrals, that in given cpu time, as well as window frame (in reference of what I understood of long filters caveats)not everything can be computed and that is what differs, shifting the consistency of the reproduced sound from an area to another, thus explaining personal and genre preferences, more than attenuation or rolloff. In instance do you play with the definition of "band limited" ? 

 

bonus question : what should we notice as bad wrong etc when apodizing counter goes berserk (542 counts in just over 4 minutes) ? I just enjoyed a track (sung pop with rich layers, minimum machine drumming ) with Sinc Medium and I preferred it over any apodizing I tried (psg long, xla, ext2...) I tried for its deeper yet more precise bass line ; is apodizing a factor that does not matter, passed a certain age whenever discussing whatever happens above 1X K Hz is moot ?

[Miska]

1 hour ago, Kalpesh said:

shifting the consistency of the reproduced sound from an area to another, thus explaining personal and genre preferences, more than attenuation or rolloff. In instance do you play with the definition of "band limited" ? 

 

Band limiting always creates a compromise, thus for any 1x rate (44.1 / 48k) you have a challenge to balance between time and frequency domains (ok, except maybe solo piano or similar band limited solo instrument). Since these are 1/x related. It is impossible to reach perfect result in both since they are inversely related to each other. My ultimate goal has been to come up with something in between that get as close as possible to the impossible. So you have range of different balances from time domain focus to frequency domain focus.

 

If your recording is non-band limited, like a true DSD recording, and in many cases true DXD or 705.6/768k recording, you don't have this issue. And many times also 192k is enough to avoid any band limiting. While 96k is not yet enough.

 

Overdoing filter in any aspect is not optimal. Like MQA focused so much on transients that they made filter so short that it ended up cutting tens of kHz off resulting in severe slew rate limiting. While overdoing frequency domain with too long filter means that everything around the transient affects the transient by spreading the energy.

 

Genre relation comes from the signal composition differences of different genres and different recording techniques. If you have a dry multi-track studio recording of rock - it doesn't really have much space information. But it is very demanding in terms of transients and dense frequency content. Absolute opposite are classical music recordings like the ones made by 2L in churches, where you have a lot of space information, practically no transients and relatively sparse frequency composition. This also shifts focus requirements of the filter.

 

One thing that matters a lot is shape of the filter's roll-off curve.

 

1 hour ago, Kalpesh said:

bonus question : what should we notice as bad wrong etc when apodizing counter goes berserk (542 counts in just over 4 minutes) ? I just enjoyed a track (sung pop with rich layers, minimum machine drumming ) with Sinc Medium and I preferred it over any apodizing I tried (psg long, xla, ext2...) I tried for its deeper yet more precise bass line ; is apodizing a factor that does not matter, passed a certain age whenever discussing whatever happens above 1X K Hz is moot ?

 

This is what I mean by personal preferences. Different people tend to focus on different aspects of the music, and different people are sensitive to different signal properties. So what the manual says is rough guidance what could be suitable for what purpose. It is not any universal absolute truth that would apply to everyone. There is no such thing as one size fits all filter. People are different, and people listen for different kinds of music. With different kind of equipment. Also rest of the system plays role here.

 

I'm personally allergic to any high frequency errors. When listening casually I also tend to focus on clarity, definition and correctness of highs and high frequency attacks. Listening with some stock DACs makes me feel bad, and can even cause my ears start ringing because the top end is so bad. This is worst when combined with some amplifiers and metal dome tweeters that tend to have +20 dB resonance spike just above 20 kHz.

 

When I listen something for development, I have trained myself to listen critically for errors in different areas. I know what things to look for and what kind of sonic errors lossy compressions and similar cause.

 

Apodizing counter is awareness thing. It tells you that the content contains errors that could be fixed with an apodizing filter. Similar way as the metering feature is nice for detecting fake hires and similar, and help you decide if you want to engage the "20 kHz cut-the-crap filter".

 

[copy_of_a]

9 hours ago, Miska said:

 

You see that's why I have not been putting numbers and graphs to the manual. But instead overall textual descriptions and suggestions. Some number, or a graph don't really give a good answer. Answer is more complex, it would be long series of mathematical formulas.

While I also find graphs enlightening quite often I am finding your way to describe the filters extremely helpful!

For example: if you look at the graphs of ps-gauss (white) and ps-xtr-short (blue) one could think ps-xtr-short reproduces more "space".

But due to the "design" of the filters infact ps-xtr-short clearly sounds more intimate - you really focus on the tones of each instrument. It sounds incredibly detailed. So "Timbre" is an accurate way to describe the characteristic of this filter (IMO). I love this filter when listening to headphones (with Rock/Pop/Electronic/Jazz/Accoustic)

 

Then there's of course also personal perception. Me for instance I am having a hard time to agree on the different descriptions for ps-gauss and ps-gauss-long. Looking at the frequency response plots I wouldn't expect that I could hear a difference between the two at all. The description in the manual states, "timbre" and "transients" for both and for the long version also "space".

Strangley I perceive it the other way around: I am finding the ps-gauss filter more "open" and "airy" than ps-gauss-long. Only plausible explanation I can think of is I am sensitive to the longer ringing in ps-gauss-long (which in itself has no "long" ringing of course, but somewhat longer than ps-gauss). Maybe that doesn't make sense at all, but that's what I am hearing.

 

*edit: added gauss vs gauss-long below for reference