[amrutmhatre90]

Has anyone compared HQplayer with streamers? Like Lumin, Aurender, Holo Red, EverSolo or Pi2es?

Stock HQplayer settings vs Upsampling.

 

[genefruit]

It's definitely a weird one! Just an hour of listening so far but to my ears is pleasing. Can't exactly articulate why though. As Jussi said don't change rate and modulator at the same time, i'm used to DSD512 so some of the changes I'm hearing might be simply from upping the rate.

I'm trying AHM7EC5L again but with Closed-form-16M with success. It could be simply the smoothness of Closed-form-16M taking over but the combo is pleasant to my ears. Addendum, the combination deadens some tracks so this might be a good combo to try if you have some that are simply too bright to enjoy.

 

[EMINENT]

Hallo.

Tonight, I played with overclock and was able to finally do PCM to DSD1024 x 48 with ASDM7ECv2 and 2 channel convolution.

 

[Triplefun]

@EMINENT I'm fascinated by your adventures and would like to know a bit more about your computer.i was looking at a 7950x3d but might change to the i9 14900k. The AMD is cooler but I believe I will get better results with the i9 and Holo may since I'm stuck with windows.What us your cooler and motherboard. Also what are your overclock results.

 

[GoldenSound]

So the new DSD modulators in HQP 5 are pretty incredible.
ASDM7EC-Super 512+fs achieves nearly 300dB SNR at DSD512, it's limited in accuracy by 24 bit input, not the modulator itself. Good lord

 

[taipan254]

So the new DSD modulators in HQP 5 are pretty incredible.
ASDM7EC-Super 512+fs achieves nearly 300dB SNR at DSD512, it's limited in accuracy by 24 bit input, not the modulator itself. Good lord

Thanks @GoldenOne - Would an objective comparison of the same upsampled input file via PGGB and HQP PCM upsampling (using some popular filters) make sense? If so, what value would such a comparison provide in your view?

 

[GoldenSound]

Thanks @GoldenOne - Would an objective comparison of the same upsampled input file via PGGB and HQP PCM upsampling (using some popular filters) make sense? If so, what value would such a comparison provide in your view?

So this post might go a little beyond the scope of this question but hopefully will be interesting for those curious about the differences.
For a PCM upsampling comparison there's a few things that we might want to look at.

Firstly we can look at the response of the filter itself by inputting 44.1khz white noise and observing the upsampled result. A 'perfect' filter would infinitely attenuate everything above 22.05khz and leave everything below 22.05khz untouched. Most real filters either attenuate/rolloff too slowly and so do not attenuate by 22.05khz, or they attenuate stuff below 22.05khz by a fair bit.

The two best filters in HQP are Sinc-L and Sinc-M/Mx.
Sinc-Mx attenuates at 21khz, and has very high attenuation.
Sinc-L attenuates at 22.05khz and has a bit less attenuation.

PGGB's approach is more similar to Sinc-L, being a 'halfband' filter (meaning it attenuates at a freq of half the sample rate). Though it uses the maximum number of taps/coefficients possible given the number of samples in a track. Often hundreds of millions. PGGB also has a more advanced noise shaper.

This is a comparison of the three options (all three set to -3dB to avoid intersample clipping):

As you can see, PGGB has the highest attenuation by far, thanks to a combination of the absurd number of taps, and also the noise shaper which is pretty impressive.

If we zoom in on the nyquist freq a fair bit we can see that Sinc-L and PGGB get pretty close to each-other in terms of cutoff steepness, as there are diminishing returns here the more taps you throw at it. Sinc-Mx cuts off earlier and is an 'apodizing' filter (we will come back to this) but with deeper attenuation and better attenuation between 22.05khz and about 23.5khz.

A 'perfect' filter would have infinite steepness, infinite attenuation, exactly at 22.05khz. Though this isn't possible in the real world as it'd require infinite computing power and an infinite number of samples.

So we can see that Sinc-Mx and Sinc-L are both excellent, though there is a tradeoff between bandwidth and time domain performance vs having better attenuation. PGGB however exceeds both.

We can test the theoretical limits of small signal accuracy by feeding something insanely low level. This is an example showing what each option provides when inputting -180dB 1khz tone and -301dB 6khz tone. (Outputting to a 24 bit file, input is 64 bit)

HQPlayer whilst the dynamic range is well....not exactly anything to complain about, nearly 300dB with a 24 bit file! PGGB does beat is here though by a significant margin. The -301dB tone can't quite be retrieved with HQP but PGGB can. In fact the jaggedy floor you can see isn't a limit of PGGB, but rather the fact that the input file is 'only' 64 bit.

If we input a 256 bit file (still output to 24 bit) to test the absolute limits then we get about 600dB dynamic range in a 24 bit file.....

Basically, HQP provides absolutely excellent upsampling and can be run in real time.
PGGB on the other hand is as close to a theoretically perfect reconstruction as you can get within the limits of available compute power, but the compute power required IS immense and therefore not only can you NOT run it realtime but it'll often take an hour or so to do an album.

The other thing to mention is that beyond the basic dynamic range, cutoff freq and steepness, there is a subjective aspect which is whether to use an apodizing or non-apodizing filter.
Apodization will attempt to remove/correct artifacts/problems/ringing that are present in the original file that may have been caused by filters in the ADC etc. Whether this is good/bad is debatable and arguably depends on the particular source material.
HQP has both apodizing and non-apodizing filters. (Sinc-M is apodizing, Sinc-L is not).
PGGB however uses a non-apodizing approach. This means that all the original samples are kept, and it does not attempt to correct any issues present in the recording.

My PERSONAL view is that I typically prefer non-apodizing. But there isn't a definitive 'correct' answer here.

Additionally there's the aspect of PCM vs DSD. This isn't really a question of which format is better, but rather which option will give you the best result on your DAC.
If you're using an R2R DAC, PCM may be better. If you're using a delta-sigma DAC (that also has genuine 1-bit DSD support) then using a great DSD modulator will be better than the modulator in your DAC and so DSD may be better.

HQP offers not just PCM upsampling but also high performance DSD modulators. PGGB is PCM only though. The PGGB dev actually mentioned he's not implemented DSD because Jussi's HQP modulators are so good that he currently can't improve on them.

Actually whilst writing this post I re-tested HQP's ASDM7EC-Super 512+fs modulator, but rather than upsampling AND modulating to DSD with HQP, I first upsampled in PGGB (outputting to 64 bit to test the theoretical limit) and then input the upsampled result into HQP to convert to DSD and this is what I got:

You can get better accuracy by upsampling with PGGB then doing the DSD modulation with HQP if you want to really push the limits of what can be done.

(I've left out checking phase accuracy & amplitude precision but might write that up later)

 

[EMINENT]

@EMINENT I'm fascinated by your adventures and would like to know a bit more about your computer.i was looking at a 7950x3d but might change to the i9 14900k. The AMD is cooler but I believe I will get better results with the i9 and Holo may since I'm stuck with windows.What us your cooler and motherboard. Also what are your overclock results.

Hailo.

I am happy to hear of your interest! Though, not much to it. Others have better systems too.

This is my basic OC with Asus AI suite. Mobo is Z790i with NZXT Kraken Z63 240mm radiator in Meshlicious case and Cooler Master V1100 psu . You can get better cooling and overclock with full size mobo and better cooling though.

Temps are real low when upsampling. With fans on blast, temps are between 40-45 both cpu and gpu.

Good luck with your quest.

 

[taipan254]

You can get better accuracy by upsampling with PGGB then doing the DSD modulation with HQP if you want to really push the limits of what can be done.

Thanks for the charts, @GoldenOne. It is not quite clear to me why it is preferential to upsample to PGGB first, then modulate to DSD. From the charts you show, the noise floor for PGGB 64 is below -450 db and can reproduce the small -301 db tone. If I look at the last image you pasted of the 64bit PGGB file upsampled to DSD, I see a noise floor of -350 to -400db. Why is the DSD modulation therefore more accurate? I'm clearly missing something!

 

[genefruit]

For those who are interested, the combination of using PGGB to upsample files offline then using HQPlayer to further convert to SDM and upsample is being discussed here. https://audiophilestyle.com/forums/...ew-of-math-and-magic/page/74/#comment-1262021

I gave it a try on a few files and one of the unexpected outcomes is I can now hit use Sinc-L, ASDM7ECv2 @ DSD1024X48. Still can't do Sinc-Mx. @EMINENT I expect this will be of interest to you. I'm probably calculating something wrong but this is 131070 taps X 1024 = 134M taps. I'm wrong, I'm sure.

 

[EMINENT]

For those who are interested, the combination of using PGGB to upsample files offline then using HQPlayer to further convert to SDM and upsample is being discussed here. https://audiophilestyle.com/forums/...ew-of-math-and-magic/page/74/#comment-1262021

I gave it a try on a few files and one of the unexpected outcomes is I can now hit use Sinc-L, ASDM7ECv2 @ DSD1024X48. Still can't do Sinc-Mx. @EMINENT I expect this will be of interest to you. I'm probably calculating something wrong but this is 131070 taps X 1024 = 134M taps. I'm wrong, I'm sure.

Haeloh.

Thank you. I shall revisit. As I tried before with usual settings but it sound the same as when 44.1 is upsampled to choice DSD. Will try this combo though as I do not remember if I ever did.

 

[1laraz]

So the new DSD modulators in HQP 5 are pretty incredible.
ASDM7EC-Super 512+fs achieves nearly 300dB SNR at DSD512, it's limited in accuracy by 24 bit input, not the modulator itself. Good lord

Could you please plot a comparison chart for ASDM7EC-Super and ASDM7EC-Super 512+fs both playing at DSD256?

 

[GoldenSound]

Could you please plot a comparison chart for ASDM7EC-Super and ASDM7EC-Super 512+fs both playing at DSD256?

Here you go!
I also included a 512 plot so it can be compared a bit more easily

 

[1laraz]

Here you go!
I also included a 512 plot so it can be compared a bit more easily

Thank you so much! So this is for the sweep tone at 1khz. I am wondering how the ASDM7EC-Super 512+fs modulator running at DSD256 behaves when the underlying content is hires in the range from 44khz to 88 khz.

 

[plumpudding2]

Here you go!
I also included a 512 plot so it can be compared a bit more easily

That's so odd, Jussi doesn't recommend running 512+fs modulators below their specified rate, but it seems the 512+fs modulator is already superior to "normal" super at 256x rate in the audible band. Not that it matters much as in either case you'll be far below the dac analog noise floor anyway.
Thanks so much for showing this by the way! I know everyone is asking you for more plots but I'm particularly interested in the s/n ratio of the new experimental AHM7EC5L at DSD1024, Jussi stated it's not suitable for digital volume control but hasn't mentioned how low it goes. Might be interesting if you find the time for more measurements