HQPlayer Convolution Question

[Miska]

5 hours ago, The Computer Audiophile said:

You've always said to use the highest sample rate filter possible in HQP. I usually upload a 352.8 filter. However, this means my filters have less resolution because they are created for 65,536 taps at 1x sample rates. The 352.8 filter still has 65,536 taps

 

Usually filter design tools adjust the filter length as necessary to represent the wanted response.

 

Important point to remember though is that 352.8k rate filter can cover 176.4 kHz wide frequency band. If your filter is created for 48 kHz, it can cover only 24 kHz worth of bandwidth. If not HF expanded, this would result in source content frequencies above 24 kHz getting lost. Filter with 176.4 kHz wide bandwidth is enough for all practical purposes without expansion (but can be still expanded if one wants).

 

5 hours ago, The Computer Audiophile said:

When I use Hang Loose Convolver, it upsamples the filter to 524,288 taps at 8x. 

 

HQPlayer scales the filters to needed rate no matter what the filter rate is and what the source rate is. And it is not same as simple upsampling of the filter. Instead a specific algorithm designed for this particular purpose is used.

 

5 hours ago, The Computer Audiophile said:

Question: What is the best way to use convolution filters in HQP, a 65,536 tap 352.8 kHz filter, a 524,288 tap 352.8 kHz filter, or are you doing some magic that makes this irrelevant?

 

If I understand filters correctly, a 65,536 filter at 48 kHz has a resolution of 0.732 Hz, whereas the same filter at 352.8 kHz has a resolution of 5.383 Hz. Still good, but not as good :~)

 

For many practical cases 64k taps is enough. That "resolution" means that you can have a brickwall filter that cuts from 0 dB to arbitrary attenuation within 0.732 Hz, or wihin 5.383 Hz.

 

How many taps your filter needs depends on how agressive (steep) frequency slopes the filter has. If it's rather gentle, not making drastic changes within very narrow few Hz bandwidth, it doesn't really need that many taps.

 

So far, for my convolution filters, I haven't really needed 64k taps. Usually it lingers around 16k taps or so.

 

Making drastic very narrow band changes usually sounds really bad. One reason is that such are extremely position sensitive. If your convolution filter has such, and you move your head by one inch, almost all such narrow steep corrections quickly become totally wrong. Higher the frequency the quicker.

 

[Miska]

29 minutes ago, Erik Haas said:

if I’m using Audiolense and pulling sweeps @ 96khz does it make sense to create filters at rates higher than this?

 

I have been generating filters @ 352.8khz but not sure if it’s worth doing so.

 

Yes, my recommendation is to generate filters for 352.8k. Then you don't need to use the "HF Expand" option and the filters sufficiently cover all PCM and DSD content.

 

[Miska]

14 hours ago, sledwards said:

@Miska I’m using 8 channels of convolution in the matrix pipeline and am observing a behavior that isn’t what I would expect. With a saved profile loaded and working, if I make a configuration change and hit ‘apply’, the values in the pipeline change to values of an old profile I used to use which was long ago deleted. Where is the program obtaining these values and why are they being loaded with a configuration change? I an running 5.5.0 Embedded.

 

It is same behaviour on both Embedded and Desktop. When you click Apply in Embedded, or OK in Desktop, the currently shown settings are saved as the unnamed default matrix setup that is loaded when you start HQPlayer, or select the "[Default]" profile in Client. When you click Load in either one, the profile is loaded as the currently active one, equivalent of selecting another profile in Client.