Same here as @DSP_Geek ,.
I use only linear filters in a 6 way and convolve to 10hz
The only problem with doing convolution below 100hz is most these usb mics can’t measure that low….
Example the Dayton umm6 has a huge peak at 23hz
Also , you don’t want to linearize the system High-pass, so everything above that is fair game as long as your microphone is up to the task. A good half inch condensers should give you no problems! Or an earthworks (which I use)
The low frequency is where all the phase is, YES we want control over that especially!!! That where things make the most difference.
Some really tiny rooms (like cars) have so much room gain in the subsonics there is no system high pass. So I just leave the last little phase rise below 30hz alone ….or I’ll use a system wide 20hz minimum phase subsonic filter which cleans up the sub simultaneously and leave the added phase
It doesn’t cause pre ringing arbitrarily.
Now it comes with a caveat, sometimes a minimum phase crossover works better in LF . As long as all the phase is tracking the same direction on all drivers you can’t really sense it , but a phase rise below 100hz can sound more natural.
The biggest difference my FIRs make is getting relative polarity to work better between left and right and removing crossovers phase….. down in the sub bass, it isn’t very much noticed unless there multiple subs or multiple speakers trying to play and lots of different room issues, then an fir can be very powerful.
I use only linear filters in a 6 way and convolve to 10hz
The only problem with doing convolution below 100hz is most these usb mics can’t measure that low….
Example the Dayton umm6 has a huge peak at 23hz
Also , you don’t want to linearize the system High-pass, so everything above that is fair game as long as your microphone is up to the task. A good half inch condensers should give you no problems! Or an earthworks (which I use)
The low frequency is where all the phase is, YES we want control over that especially!!! That where things make the most difference.
Some really tiny rooms (like cars) have so much room gain in the subsonics there is no system high pass. So I just leave the last little phase rise below 30hz alone ….or I’ll use a system wide 20hz minimum phase subsonic filter which cleans up the sub simultaneously and leave the added phase
It doesn’t cause pre ringing arbitrarily.
Now it comes with a caveat, sometimes a minimum phase crossover works better in LF . As long as all the phase is tracking the same direction on all drivers you can’t really sense it , but a phase rise below 100hz can sound more natural.
The biggest difference my FIRs make is getting relative polarity to work better between left and right and removing crossovers phase….. down in the sub bass, it isn’t very much noticed unless there multiple subs or multiple speakers trying to play and lots of different room issues, then an fir can be very powerful.
Last edited:
@Viper_user yeah , with low taps I’ve only had good luck with rectangular window.
If your getting a better prediction with a window compared to rectangular then try it , I just don’t think it will work as good
If your getting a better prediction with a window compared to rectangular then try it , I just don’t think it will work as good
Thank you @wesayso that was a good read..
So it really depends where your centering the impulse and how long of an impulse your capture is if I understood that right.
So , I’ve obviously messed around with taps (samples) but I’ve never actually changed the FFT length….
Even with low taps, would that be something to try ?
So it really depends where your centering the impulse and how long of an impulse your capture is if I understood that right.
So , I’ve obviously messed around with taps (samples) but I’ve never actually changed the FFT length….
Even with low taps, would that be something to try ?
Pretty much, if you're not equalizing the drivers. I like making a highpass filter by subtracting a delayed impulse from an inversion of the corresponding low pass filter; you get both HPF and LPF for the price of one, and by inspection it becomes obvious the ringing of both filters cancels itself out on axis. Off axis is a different story, of course, so that's why I also like using filters as short as possible, such as this one:
A filter length of (SampleRate * 2 / CornerFrequency) + 1 has weights of zero at the beginning and end so ugly windowing artifacts are minimized.
Off-axis ringing of low frequency filters is no big deal so long as one pays attention to path lengths. Midrange and high filters can run into problems so some care is warranted with room treatment to avoid ringing bouncing off the wall into your ears with first reflections, but that's general good practice regardless of filter type and order.
A filter length of (SampleRate * 2 / CornerFrequency) + 1 has weights of zero at the beginning and end so ugly windowing artifacts are minimized.
Off-axis ringing of low frequency filters is no big deal so long as one pays attention to path lengths. Midrange and high filters can run into problems so some care is warranted with room treatment to avoid ringing bouncing off the wall into your ears with first reflections, but that's general good practice regardless of filter type and order.
@DSP_Geek (sorry can’t quote on next)
That’s super interesting…. So awhile back I thought for sure if someone was to use more taps then the entire length of the FFT, that would surly introduce some ringing, or temporal smearing or something….
Someone later explained that wasn’t the case, but what your saying maybe sorta says it could be true.
Yes , no ?
Thanks 🙏
That’s super interesting…. So awhile back I thought for sure if someone was to use more taps then the entire length of the FFT, that would surly introduce some ringing, or temporal smearing or something….
Someone later explained that wasn’t the case, but what your saying maybe sorta says it could be true.
Yes , no ?
Thanks 🙏
No FIR above 1 kHz? shouldn't a linear LR 12db at 4khz do better than a regular with phase rotation?
Cheers!
Because when I tune systems that are Pc based convolution, and can sample super high and can do ridiculously large tap count , i’m wondering if it messes things up to go higher than how big the impulse actually is.
Like what I’ve been doing is using mini DSP the 6144 taps at 48k as a base
So if I am sampling at 384k , I’ll multiply 6144 x8 and get 49,154 taps , that way I know I can get resolution to 10 Hz, and not exceed the length of the FFT…
I’ve been having good luck with that because when I add way more tabs sometimes it doesn’t work right, weird echoey reverb like sound comes out also
FIRs are perfectly fine above 1 kHz, of course, although I'm considering a passive 4th order between the low-mid and high-mid because I can use Stoopid Crossover Trix to smooth out the transition at 2 kHz, then another set of FIRs at 8 kHz because the crossover there wants to be sharp with delay matching otherwise phasing artifacts make themselves unwelcome.
This is entirely independent of whether FFTs or direct-form calculations are used to run the filter. I suspect if you were to use too short an FFT then the truncation would indeed mess up the filter response.
Hi,
i think too, ,
ripple are at "high" level. (-40dB)
is it a brickwall filter ? (Gibbs phenomenon ?)
not enough taps or too high Fs.
whatever the window choosen before calculation/convolution (time domaine or frequency domain).