Convolution filter , DSD and taps? Vs parametric EQ

[Snoopy]

I posted this in the roon forum but so far no response..


Im currently experimenting with Convolution filters for my Headphones. I Downloaded the 44.1khz and 48khz files from Github, added them to Roon and it works just fine. but I was wondering what the “taps” exactly mean, what are 20k taps exactly doing ? and how is a convolution filter different from a 10 band parametric EQ?

I noticed the Convolution Filters make everything more quite and I need a bit more volume. is it ok to disable volume levelling when using them instead of parametric EQ?

is it better to use the P-EQ for DSD output since I don’t have the matching 384khz files for the C-Filter?

I played around a bit and this is what my signal path looks like in my living room.
my other setup is able to do native DSD512.

Headphones are the meze liric and I downloaded the auto EQ wav files from oratory1990.

 

[voodooless]

I was wondering what the “taps” exactly mean, what are 20k taps exactly doing ?

The number of taps, coupled to the sample rate dictate the frequency resolution of the filter. The more taps, the higher the resolution. High-tap count is usually needed for low frequency correction.

and how is a convolution filter different from a 10 band parametric EQ?

Those are implemented as an IIR (infinite impulse response) and mimic more or less analoge EQ filters. Consequently they add a bit of phase shift, which FIR (convolution) filter don’t need to have. They can correct phase and response independently.

I noticed the Convolution Filters make everything more quite and I need a bit more volume. is it ok to disable volume levelling when using them instead of parametric EQ?

Generally no, neither is that a good idea for the parametric EQ. Usually the filter will have some boost area. To prevent clipping, the overall volume is lowered. Better keep it that way and just turn up the volume a bit more.

is it better to use the P-EQ for DSD output since I don’t have the matching 384khz files for the C-Filter?

No problem: either Roon will resample the filter to the correct rate, or you can provide your own up samples version.

 

[Snoopy]

So using the convolution filter is more accurate than a parametric EQ? The thing with the "taps" is that the same as chord is doing ?

 

[voodooless]

So using the convolution filter is more accurate than a parametric EQ?

Not always. If you don’t have enough taps an IIR filter may be more effective. Also note that a FIR filter introduces a delay, so for video, high tap coins filters can be a problem. That’s why for instance Dirac uses a hybrid form.

The thing with the "taps" is that the same as chord is doing ?

Yes it is.

 

[Snoopy]

Not always. If you don’t have enough taps an IIR filter may be more effective. Also note that a FIR filter introduces a delay, so for video, high tap coins filters can be a problem. That’s why for instance Dirac uses a hybrid form.

Yes it is.

How much are enough? Redbook seems to be around 20k taps. DSD 299k.
Don't think there is a way to increase this ?

 

[voodooless]

How much are enough?

That depends on the requirements of your filter. But for a higher sample rate, you’ll need more taps to accomplish the same thing.

20k taps is plenty. I would not worry about it.

 

[Jimbob54]

I may not get the subtleties of this but there is an obvious huge advantage to using a configurable PEQ to a convolved filter . You can tweak the PEQ to taste.

Im sure you can create a convolved filter from a tweaked EQ profile, but not the other way round.

 

[Snoopy]

That depends on the requirements of your filter. But for a higher sample rate, you’ll need more taps to accomplish the same thing.

20k taps is plenty. I would not worry about it.

I'm just trying to understand how this works and what it does really. Since you mentioned Dirac I assume if I would use this with loudspeakers the filter would be more complex (bigger room, more reflections).

Wouldnt this increase the required amount of taps? But with headphones this would be tiny in comparison?

Because I tried 44.1khz, 96khz, 192khz files... and all I get are 20k - 37k taps but with DSD I get 299k taps.

But how would the chord Dave archive anything useful if it doesn't know with what headphones or loudspeakers it's actually working?

 

[voodooless]

But how would the chord Dave archive anything useful if it doesn't know with what headphones or loudspeakers it's actually working?

It doesn’t care. The only function of that filter is to remove as much of the out of band content after oversampling. So at 44.1 kHz, it will remove basically everything above 22.05 kHz.

Reconstruction filter - Wikipedia

en.m.wikipedia.org

 

[voodooless]

I may not get the subtleties of this but there is an obvious huge advantage to using a configurable PEQ to a convolved filter . You can tweak the PEQ to taste.

That is just a user interface problem. One could have a parametric EQ generate a FIR filter. It’s just vastly more complex. RePhase will do it for instance.

Im sure you can create a convolved filter from a tweaked EQ profile, but not the other way round.

Not without major caveats. FIR filters can do things IIR filters cannot do.

 

[voodooless]

I'm just trying to understand how this works and what it does really. Since you mentioned Dirac I assume if I would use this with loudspeakers the filter would be more complex (bigger room, more reflections).

Wouldnt this increase the required amount of taps? But with headphones this would be tiny in comparison?

No, it would not. With an IIR filter, every change you make to the response, adds at least one more IIR filter block. FIR filters are more holistic. They cover the whole frequency and phase band at the same time, and the numbers of taps determines the how accurately you can make corrections. The number of corrections doesn’t really matter. For instance, 1024 taps @48 kHz gives you 47 hz of resolution. If you want the same resolution at 96 kHz, you’ll need double the amount of taps. That’s what Roon is showing you.

 

[Snoopy]

No, it would not. With an IIR filter, every change you make to the response, adds at least one more IIR filter block. FIR filters are more holistic. They cover the whole frequency and phase band at the same time, and the numbers of taps determines the how accurately you can make corrections. The number of corrections doesn’t really matter. For instance, 1024 taps @48 kHz gives you 47 hz of resolution. If you want the same resolution at 96 kHz, you’ll need double the amount of taps. That’s what Roon is showing you.

I have to admit this is going a little bit over my head. Maybe because English is not my first language, and the more I read about this stuff the more complex it gets. Machine learning, photo editing etc. That stuff pops up when I Google convolution filters.

I'm switching between between parametric EQ and convolution with my headphones. Both those settings are from oratory1990 and target the same frequency response.

But the PEQ sounds wider but I get more of the headphone feeling without crossfeed on.

But that's only my first impressions. After a few albums.

The convolution filter sounds alot more like loudspeakers or PEQ with crossfeed on.

 

[Jimbob54]

That is just a user interface problem. One could have a parametric EQ generate a FIR filter. It’s just vastly more complex. RePhase will do it for instance.

Not without major caveats. FIR filters can do things IIR filters cannot do.

I stand thoroughly corrected

 

[Snoopy]

No, it would not. With an IIR filter, every change you make to the response, adds at least one more IIR filter block. FIR filters are more holistic. They cover the whole frequency and phase band at the same time, and the numbers of taps determines the how accurately you can make corrections. The number of corrections doesn’t really matter. For instance, 1024 taps @48 kHz gives you 47 hz of resolution. If you want the same resolution at 96 kHz, you’ll need double the amount of taps. That’s what Roon is showing you.

So I just did measurements with HouseCurve for iOS yesterday for my loudspeakers. And I get 2 paths, 131.000 taps with a 44.1khz file.

If I Play the same file through my headphone setup with convolution filters for the headphones I get 5000 taps.

Obviously this increases with higher sample rates for DSD.

But it seems after all that the Taps increase with a bigger space (living room) Vs headphones?

Kinda want to get DIRAC ART when it becomes available after the Storm Audio exclusive deal ends

 

[voodooless]

But it seems after all that the Taps increase with a bigger space (living room) Vs headphones?

No such relation exists.