Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[whoareyou]

I just started using this feature and all seems to be working ok, but I have question about performance.
When I run convolution UI shows I have 2.0x real time performance (using SSE).

I've had no issues, but after reading some forum comments, it seems like this should not be working properly?

This is with 6 channel / 11 paths and at 352.8khz (SACD).  Filter length of 65536.

No stuttering. Nothing seems to be wrong / sound bad (actually sounds great).  I've generated new filters and done all kinds of stuff on the computer (for hours) while music is playing, and filters are active.
So far, not even a hiccup.

But reading through forum, I notice people saying performance should be much higher when using convolution?

Just wondering if I am missing something obvious, and was hoping someone could provide their thoughts?

This is with I5 6 core, 8th gen processor.  JRiver uses about 11% processor when convolution is running.

Thanks

[mattkhan]

Is your question about the fact it reports 2x realtime? I would think this is due to the high sample rate.

If everything runs smoothly then I wouldn't worry.

Having said that you realise your filter length is relatively short for such a sample rate? About 10Hz if it is linear phase

[whoareyou]

Is your question about the fact it reports 2x realtime? I would think this is due to the high sample rate.

If everything runs smoothly then I wouldn't worry.

Having said that you realise your filter length is relatively short for such a sample rate? About 10Hz if it is linear phase

Yes question is about the 2x realtime.  I'm not sure what it is really reporting, or why it may be indication of playback issues.

Having said that you realise your filter length is relatively short for such a sample rate? About 10Hz if it is linear phase

No I did not. 
Filter is minimum phase.
This is a bit of a learning curve for me and filter length is a bit of mystery.  I've seen some calculations that I need to wrap my head around but ....
Any suggestions for sample rates vs filter length?

[mattkhan]

Yes question is about the 2x realtime.  I'm not sure what it is really reporting, or why it may be indication of playback issues.

basically if playback is stable then no problem, you don't have that much headroom though so if you intended to add more paths or raise sample rate further then you might get a problem (unless you get a faster CPU)

Filter is minimum phase.
This is a bit of a learning curve for me and filter length is a bit of mystery.  I've seen some calculations that I need to wrap my head around but ....
Any suggestions for sample rates vs filter length?

it just depends what resolution you need, i.e. what you're attempting to do with the filter

frequency resolution for a minimum phase fir filter is sample rate / taps so 352800 / 65536 = ~ 5.4Hz

this is not great for use at the low end (under 100Hz or so) but is fine higher up

[whoareyou]

frequency resolution for a minimum phase fir filter is sample rate / taps so 352800 / 65536 = ~ 5.4Hz

this is not great for use at the low end (under 100Hz or so) but is fine higher up

How is the 100Hz low end boundary derived from the 5.4Hz frequency resolution?
I've seen similar calculation, but I don't understand this.

Thanks

[mattkhan]

How is the 100Hz low end boundary derived from the 5.4Hz frequency resolution?
I've seen similar calculation, but I don't understand this.

I said "say 100Hz or so" not boundary, there isn't a fixed frequency at which a frequency resolution becomes a problem, it just depends what you're trying to do. For example, correcting a modal peak requires the filter to be precisely defined at that frequency (which can mean to 0.1Hz if you want to be really exact) so clearly a 5Hz resolution will be pretty useless in this case. Alternatively say you wanted a 100Hz crossover, will 5Hz resolution be sufficient for this? perhaps it will be workable but I would guess it will be an approximation rather than exact.

[whoareyou]

I said "say 100Hz or so" not boundary, there isn't a fixed frequency at which a frequency resolution becomes a problem, it just depends what you're trying to do. For example, correcting a modal peak requires the filter to be precisely defined at that frequency (which can mean to 0.1Hz if you want to be really exact) so clearly a 5Hz resolution will be pretty useless in this case. Alternatively say you wanted a 100Hz crossover, will 5Hz resolution be sufficient for this? perhaps it will be workable but I would guess it will be an approximation rather than exact.

Ok.  That helps very much.

I really appreciate the explanations.

[mattkhan]

if you do want to combine IIR (i.e. peq in dsp studio) with convolution then you may find https://yabb.jriver.com/interact/index.php/topic,129609.0.html useful as an easier way to manage the filters and to avoid having to think about the vagaries of the jriver filter definitions

[whoareyou]

This looks very interesting

Thanks again.