HQ Player

[Solstice380]

1 minute ago, The Computer Audiophile said:

Huh?

 

Will send PM.

[Miska]

1 minute ago, Solstice380 said:

 I don’t think so.  The clipping occurs in the software before the music is sent to the DAC.  You may lower the level of any clip artifacts, but they will still be there.  @Miska, correct?

 

Yes, value range from player to the DAC is limited. As an example 16-bit PCM has range from -32768 to +32767. There's no way to put bigger number in 16-bit integer. So the output needs to fit in the available range. The internal DSP pipeline in HQPlayer has practically unlimited range. So for output setting HQPlayer's volume to suitable value makes sure the values will fit in the output format. Otherwise it is enforced by the soft limiter.

 

[Miska]

9 minutes ago, Solstice380 said:

 My apologies this post.  It can be deleted by moderator or Chris. Thanks. 

 

I wrote a bit more lengthy response this time, no harm...

 

[Miska]

4 minutes ago, Solstice380 said:

@Miska I have the same behavior.  Is it possible that I noticed there is greater delay with conversions to higher sample rates?

 

It should be inverse, at lowest rates the delay is largest...

 

[Outlaw]

15 minutes ago, Miska said:

 

Does the same happen with local playback without NAA? What sampling rate are you using for output?

 

Yes same thing with local playback.44.1 dsd 256 output.

[Solstice380]

3 minutes ago, Miska said:

 

It should be inverse, at lowest rates the delay is largest...

 

 

I’ll watch more closely, it’s been a long time since I sent anything other than the max accepted rate to my DACs! I think since I started using HQP.  Is that simply because the higher rate has more samples to operate on?

[Outlaw]

19 minutes ago, Miska said:

 

Does the same happen with local playback without NAA? What sampling rate are you using for output?

 

Same local playback without NAA.44.1 dsd 256 poly sinc xtr mp

[Miska]

39 minutes ago, Outlaw said:

Same local playback without NAA.44.1 dsd 256 poly sinc xtr mp

 

How about poly-sinc-xtr-mp-2s?

[Miska]

40 minutes ago, Solstice380 said:

I’ll watch more closely, it’s been a long time since I sent anything other than the max accepted rate to my DACs! I think since I started using HQP.  Is that simply because the higher rate has more samples to operate on?

 

Some of the buffers are in bytes, so at higher rate they are relatively smaller - shorter amount of audio fits into same number of bytes.

 

[Yviena]

On 2/24/2019 at 8:13 PM, Miska said:

 

I'm curious if there's any difference in output rate between the one in DAC and what HQPlayer outputs? And what output rate from HQPlayer you used? At one point I was asking Soekris if they could provide one without digital filter to allow full speed external one but they said that they won't.

 

I used 352/384 khz output rate with closed form-M filter with HQplayer to bypass FIR1 in the DAC, the one i flashed to the DAC is a 44.1khz filter the dac internally oversamples it 8x before the FIR2 filters further upsample to  2.8/3.1 Mhz.

It would be really interesting  to see why this filter has a noticeable larger stage.

 

It is also possible to flash a NOS filter that will bypass FIR1/FIR2.

[Outlaw]

5 hours ago, Miska said:

 

How about poly-sinc-xtr-mp-2s?

Same issue

[Miska]

3 hours ago, Outlaw said:

Same issue

 

OK, really hard to say what it could be. You could try sending me HQPlayer log file over email and I can check if I can find something there...

 

[Miska]

4 hours ago, Yviena said:

I used 352/384 khz output rate with closed form-M filter with HQplayer to bypass FIR1 in the DAC, the one i flashed to the DAC is a 44.1khz filter the dac internally oversamples it 8x before the FIR2 filters further upsample to  2.8/3.1 Mhz.

It would be really interesting  to see why this filter has a noticeable larger stage.

 

It is also possible to flash a NOS filter that will bypass FIR1/FIR2.

 

I guess you are on a DIY version and not on the boxed regular DAC?

 

closed-form-M is quite different from those FIR filters. Since you cannot send PCM at 2.8/3.1 MHz rate over USB, can you flash it such that FIR1 is NOS and leave FIR2 intact? Then run something like poly-sinc-ext2 in HQPlayer to 352.8/384k which would be further upsampled to 2.8/3.1 by FIR2 in DAC?

 

Having the actual conversion section running at 352.8/384k vs 2.8/3.1M will have notable impact on amount of images left in the output spectrum. But from sound perspective the FIR1 is most critical. FIR2 matters only for moving the images further up.

 

So the idea would be to run FIR1 in HQPlayer and FIR2 in the DAC.

 

Holo Spring 2 allows sending 1.4/1.5 MHz PCM over USB and this removes all measurable images from the output...

[Yviena]

8 hours ago, Miska said:

 

I guess you are on a DIY version and not on the boxed regular DAC?

 

closed-form-M is quite different from those FIR filters. Since you cannot send PCM at 2.8/3.1 MHz rate over USB, can you flash it such that FIR1 is NOS and leave FIR2 intact? Then run something like poly-sinc-ext2 in HQPlayer to 352.8/384k which would be further upsampled to 2.8/3.1 by FIR2 in DAC?

 

Having the actual conversion section running at 352.8/384k vs 2.8/3.1M will have notable impact on amount of images left in the output spectrum. But from sound perspective the FIR1 is most critical. FIR2 matters only for moving the images further up.

 

So the idea would be to run FIR1 in HQPlayer and FIR2 in the DAC.

 

Holo Spring 2 allows sending 1.4/1.5 MHz PCM over USB and this removes all measurable images from the output...

No it's their commercial dac1541 i have, there are serials pins for flashing firmware/filters on the pcb.

 

Looking at the dac filter i have loaded any sample rate that is 352/384 disables the FIR1 so it's just FIR2 working so I don't think it would matter if i flash NOS to other sample rates.

[Miska]

3 minutes ago, Yviena said:

No it's their commercial dac1541 i have, there are serials pins for flashing firmware/filters on the pcb.

 

OK, good to know that it is possible to modify it! They didn't admit such possibility something like two years ago at Munich High-End when I talked to them...

 

4 minutes ago, Yviena said:

Looking at the dac filter i have loaded any sample rate that is 352/384 disables the FIR1 so it's just FIR2 working so I don't think it would matter if i flash NOS to other sample rates.

 

OK, then it sounds like it is good to go with the current setup. Easier to compare at least. Looking for similar one, based on your description, you could try bunch of the poly-sinc linear phase filters except poly-sinc-hb. Plus sinc-M. To check which one you find to have most similar properties.

 

[Yviena]

26 minutes ago, Miska said:

 

OK, good to know that it is possible to modify it! They didn't admit such possibility something like two years ago at Munich High-End when I talked to them...

 

 

OK, then it sounds like it is good to go with the current setup. Easier to compare at least. Looking for similar one, based on your description, you could try bunch of the poly-sinc linear phase filters except poly-sinc-hb. Plus sinc-M. To check which one you find to have most similar properties.

 

 

Hmm i did try poly-sinc-ext2,poly-sinc,poly-sinc-shrt,xtr and sinc-M, but they still sound kind of closed in compared to the one flashed, the soekris line of dac is very mirror/aliasing picky though so maybe that's one of the reasons.

[Miska]

38 minutes ago, Yviena said:

Hmm i did try poly-sinc-ext2,poly-sinc,poly-sinc-shrt,xtr and sinc-M, but they still sound kind of closed in compared to the one flashed, the soekris line of dac is very mirror/aliasing picky though so maybe that's one of the reasons.

 

ext2 attenuates >240 dB by Nyquist:

 

And xtr2 is designed to reach about the same by little bit higher to avoid images in DAC output:

 

Do you know how much does the FIR1 attenuate?

 

So ext2 is your best bet, keeps 32-bit bit data clean of any images by ~8-bit margin. I can push things over 240 dB if needed...

 

[Yviena]

No idea but the guy who made it will publish txt source file for the filter in some days, it's probably made it matlab so it'll be relatively easy to check what the filter does, but he did say he is sacrificing B/W for supression on the lower side of FS/2 not over as many other filters usually do, and that he's fully committing to the sampling theorem.

But i see no reason why software upsampling can't be made to sound the same, or better with it's much higher processing power.

[Miska]

11 minutes ago, Yviena said:

No idea but the guy who made it will publish txt source file for the filter in some days, it's probably made it matlab so it'll be relatively easy to check what the filter does, but he did say he is sacrificing B/W for supression on the lower side of FS/2 not over as many other filters usually do, and that he's fully committing to the sampling theorem.

But i see no reason why software upsampling can't be made to sound the same, or better with it's much higher processing power.

 

The question is, how much... So ext2 is flat to 20 kHz and reaches -240 dB by fs/2.

 

You could also compare against poly-sinc-mqa which begins to roll off earlier, but not as steep and doesn't have as much attenuation. But gives relatively very short impulse response in time domain.

 

And by the way, when comparing, are you playing at -3 dBFS volume setting, through HQPlayer, with/without HQPlayer's filter? So that you get matched levels...

 

[Yviena]

26 minutes ago, Miska said:

 

The question is, how much... So ext2 is flat to 20 kHz and reaches -240 dB by fs/2.

 

You could also compare against poly-sinc-mqa which begins to roll off earlier, but not as steep and doesn't have as much attenuation. But gives relatively very short impulse response in time domain.

 

And by the way, when comparing, are you playing at -3 dBFS volume setting, through HQPlayer, with/without HQPlayer's filter? So that you get matched levels...

 I volume matched it with a sound meter i have, the hqplayer filters are  2-3dBFS louder compared to the dac filter so when comparing i usually set the volume at -3 for dac filter, and -6 for hqplayer filters.

[Miska]

3 hours ago, Yviena said:

 I volume matched it with a sound meter i have, the hqplayer filters are  2-3dBFS louder compared to the dac filter so when comparing i usually set the volume at -3 for dac filter, and -6 for hqplayer filters.

 

OK, I see, so the DAC is applying some headroom probably.

[asdf1000]

4 hours ago, Miska said:

And xtr2 is designed to reach about the same by little bit higher to avoid images in DAC output:

 

From the manual, for ext2 (PCM) I see "completely cutting off by Nyquist frequency (non-halfband)."

 

Does that also achieve avoiding images in DAC output?

[Miska]

7 minutes ago, Em2016 said:

From the manual, for ext2 (PCM) I see "completely cutting off by Nyquist frequency (non-halfband)."

 

Does that also achieve avoiding images in DAC output?

 

It achieves avoiding images within fs/2 (Nyquist) band of the filter output rate.

 

So regarding DAC output, it depends on the output rate. Since upsampling doesn't happen to infinite sampling rate, there is always some output rate. Images appear around multiples of digital filter output sampling rate. These images are then supposed to be filtered out by analog reconstruction filter at DAC output. Since such filter is usually low order, the idea behind digital filters is to move those images to start at high frequency to make the analog filter reach enough attenuation by the frequency where images start to appear in order to remove them.

 

Typical DAC chip filter outputs at 352.8 or 384 kHz rate, so there are usually some left-over images at multiples of this rate. If upsampling filter works instead at 705.6/768k rate, the images are now twice higher and the analog filter as reached higher attenuation by then, and so on. (you can see example plots of this in the ADI-2 thread, 352.8k output from the on-chip filter vs 705.6k output from HQPlayer)

 

For example Holo Spring 2 R2R section doesn't have images I could detect when input rate is 1.4/1.5 MHz.

 

In above discussion with Yviena, there's a second stage filter in the DAC that further upsamples after 352.8/384k input. This is like HQPlayer's -2s filters (with some differences on how higher than 44.1/48k rates are handled). Soekris DAC in question is R2R type and with those final output rates similar to Holo Spring 2 and MSB DACs.

 

However, when output is let's say DSD512, HQPlayer runs upsampling digital filters to 22.5792 MHz rate. Quite a bit higher than in above cases. So images would be at multiples of that rate. But the analog filter has cut out everything long before such frequency is reached. And any images would be anyway buried in the modulator's noise shaping noise way before. So with SDM DAC's, if the upsampling ratio is high enough, only job for the analog filter is to remove the noise shaping noise, instead of images. If the ratio is limited to let's say typical on-chip filter at 352.8k, then the analog filter needs to deal with both, images and the noise shaping noise.

 

[Miska]

4 hours ago, Yviena said:

 I volume matched it with a sound meter i have, the hqplayer filters are  2-3dBFS louder compared to the dac filter so when comparing i usually set the volume at -3 for dac filter, and -6 for hqplayer filters.

 

By the way, one thing worth checking is "DAC Bits" setting in HQPlayer. I'm assuming you are using TPDF dither? To set exactly correct value, one would need linearity sweep measurement of the DAC. But 20 bits is good starting point, that is for example what Holo Spring manages before starting to deviate. So you get best accuracy by setting DAC Bits to correct value and using TPDF/Gauss1 dither, or a noise shaper. In case of 352.8/384k output rate NS9 and you can also try NS5.

 

[tboooe]

5 minutes ago, Miska said:

But 20 bits is good starting point, that is for example what Holo Spring manages before starting to deviate.

Am I reading this right..for Holo Spring set DAC Bits in HQP to 20?