Which HQP Filter are you using?

Hi Jussi,so did you suggest me to by an Audiophellio SE up to 384khz Dop up to 168 vs mine : 192khz Dop up to 64 for my Naim N dac ,doe’s it worth ?

Tks Giovanni.

I am enjoying this thread (and HQ Player) and I am always learning and I thought I would check my settings, to see if I should tweak.

I have HQ Player on my desktop (Roon on a NUC), upscaling to DSD 256 on my Topping D90 (fed via Allo USBridge Sig using Dietpi). I also use filters from Home Audio Fidelity making using of the Matrix pipeline in HQ Player, using 352kHz. Here are my exact settings below;

Does this all look correct? I have never changed DAC bits, should I leave at default? Is there anything else I should tweak?

If you use volume on Topping D90, its probably better to up-sample to PCM since DSD will be converted to PCM inside the DAC anyway to be able to perform digital volume.

Thanks. I use the D90 in DAC only mode (DSD) straight into my Musical Fidelity M6si via balanced XLR. I have tried the PCM, but generally prefer DSD, although I go for 256 (rather than 512) as there appears to be noise (according to the DAC chip documentation) above this level, I can then use the EC modulators so no real problem.

6 posts were split to a new topic: NAA on Pi; WiFi v Ethernet

2 posts were split to a new topic: Setting Convolution in HQP Embedded on Euphony

I’ve Moderated out the whole discussion about who should or shouldn’t be posting about what. It’s irrelevant and adds nothing to the thread.

Edit: Also shifted some discussions into their own threads.

For the Antipodes / HQPlayer users… update is available.


Hi Jussi,

Could you please link to a post where you describe the technical theory for the various filters? I am sure it is on here or CA somewhere but the threads are hundreds of pages long.

Even though WTA is a secret, do you have an educated guess of what it is doing? I have some knowledge of digital signal processing and my understanding is that the different algorithms are just different coefficient weights for each interpolated sample. If you had an infinite length filter the weights would be purely determined by the sinc function, but since there is a finite filter length you need to do some tapering towards the ends of the filter to avoid ringing artefacts.

There are different ways to taper the data and some methods cause smearing in the frequency domain, and therefore smearing of transients in the time domain. My guess is that WTA is some sort of multi-tapering, where the coefficients are calculated by a number of different sinusoidal tapers to preserve resolution in the frequency domain and temporal accuracy of transients in the time domain.

Is my understanding in the right ballpark?

I don’t really do that.

No, you are mixing things… :slight_smile:

Primary reason for smearing comes from length of the filter, transients affecting samples before and after the transient. Longer the filter, longer this time window is.

Since frequency and time and inversely proportional, challenge is to make the filter as short as possible in time domain while making it perform as well as possible in the frequency domain. At the same time. Better the design algorithm, closer to the impossible you can get.

MQA focuses entirely on time domain and thus it suffers a lot in frequency domain. Chord does the opposite. I think I explained this couple of messages earlier.

Just look at step response of the filter, and look at it in dB amplitude scale. That tells you about the transient response.

1 Like

@jussi_laako, with the EC (extra compensation) modulators, what does “compensation” refer to ?

7 posts were split to a new topic: Using a turntable input to HQP

I see that the latest Pro version includes a new filter – Sinc-L. Will it make its way into the Consumer version as well? If so, I wonder how does it compare to the Sinc-M filter.

Yes, next versions of Desktop and Embedded will include it.

It is quite a bit longer, but adaptive length just like the others. It is one million taps already at 8x ratio and with 256x ratio this 32 million taps.

But the filter design algorithm is totally different than sinc-S and sinc-M, this one has much lower attenuation, is non-apodizing, but steepness is extreme. So in other words, other extremely long filter, but different approach.

For PCM outputs it is not particularly heavy on CPU. For SDM outputs it gets heavy but GPU helps a lot. However, it uses a lot of GPU RAM, so in order to do DSD256 with it you need more than 8 GB of RAM on the graphics card. My RTX2080 (8 GB) runs out of memory with it for DSD256. But RTX2080Ti (11 GB) can do it with 8.4 GB memory usage and 13% GPU load.


Can’t wait to try it!

Interesting. I will probably just stick to PCM anyway but it does raise the question what spec my Antipodes CX is. All I know is that it’s a 6 core i7

Much simpler, what does it do to the sound and how does it change it in listening ?
May i use for files 16/44.1 ?
Tks Giovanni.

Thinking on buying a licence. Will this be on version 4.x or will be a 5.x release?

Current licenses are for 4.x. I have no information about 5.x.

@jussi_laako ; What is the filter on your RME ? :wink:
I think the filter inside has no impact over 192kb, but for my curiosity, what is the filter used on your RME (Sharp or NOS - PCM not direct DSD, of course)