trying to build frequency splitter (band splitter)

Discussion in 'Building With Reaktor' started by Tun, Jul 21, 2013.

  1. Tun

    Tun Forum Member

    Messages:
    199
    hi
    i am trying to build a frequency splitter to separate a signal into 4 bands to be processed individually, then mixed back together.
    i want to do this so that the bands can be split, mixed back together, then mixed with a phase inverted original of the signal and create silence (or as close to it as possible). what i really need is for the filters to add as little character to the sound as possible.
    i think i would need linear phase filters for this, but i cant find any and wouldnt know where to begin building them.

    i looked through the user library and saw bandsplitter, which uses a LP, which is then subtracted from the original signal to give two bands, and repeated to get as many bands as you need.
    this gives me the exact same sound as i started with, but i noticed that the bands arent split properly, and with steeper filters i actually get a more shallow slope on the "HP" section of the bands.

    i also tried some EZFFT stuff, but most of that goes over my head, and what i could take from it didnt seem to help me at all. for example, with the EZFFT brickwall filter (which i thought would be ideal to reverse engineer), it seems that with a sine wave attenuated by 100dB, it actually doesnt make much difference to the sine wave. not sure why, but thats for another thread.

    has anybody got any idea about how to create this band splitter, or a linear phase filter? or a link to a good band splitter or filter i can look at?


    thanks
    ---
    oh and i forgot to say. i would like to have an adjustable slope if i can too. this would just be a switch to select between different filter types (ideally 1-pole to 8-pole).
    ---
    sorry to keep adding to this already long thread, but i noticed in the band splitter included with reaktor that there is a unit delay on two of the bands.
    why is this? maybe that would solve my problem?
  2. Tun

    Tun Forum Member

    Messages:
    199
    edit:

    i managed to get a 1-pole and a 2-pole splitter working by experimenting with unit delays. i cant get the 4-pole version to work though :(
  3. Tr97

    Tr97 Forum Member

    Messages:
    50
    hi.

    maybe this thing is useful. it is not linear phase and the slope is fixed at 4 pole (24dB/oct). also the magnitude sum is not exactly one. so there is a bit of phasing if the split-frequencies are changed.

    it uses Linkwitz-Riley (double butterworth)-filters to do the split. this works perfect for a 2-band crossover but when extracting more bands phase problems occur if neighbouring splits get to near to each other.

    there is an unconnected 8-pole corecell inside but in this setup the phasing is too strong.

    Attached Files:

  4. Tun

    Tun Forum Member

    Messages:
    199
    thanks Tr97.
    i will have a look at this later
    ---
    just tried it out. its a nice filter, but it doesnt work for what im trying to do.
    i added all the bands and subtracted them from the original and there was still quite a lot of audio getting through still.

    i broke it down so it was just a single LP and HP and tried with that, but it still didnt work.

    thanks anyway
  5. Tr97

    Tr97 Forum Member

    Messages:
    50
    hi. next attempt.

    replaced the filters by linear phase filters (without fft) and added a "subtract original" button. seems to work now.

    Attached Files:

  6. Big Gnome

    Big Gnome Forum Member

    Messages:
    171
    The best implementation I've seen in Reaktor is the crossover in Flatblaster 2.
  7. Tun

    Tun Forum Member

    Messages:
    199
    thanks to both of you.
    i will look at those both now.

    @Tr97
    the problems with subtracting the original is that you end up with a curve that doesnt match, and is actually a very shallow curve, with a boost at the cutoff.
    if this has linear phase filters though it shouldnt be a problem, and in that case those things are gold dust! or so it seems. i couldnt find linear phase filters anywhere!
    anyway, i will go and check it out and report back.

    thank you both
    ---
    just checked those out. the flatblaster crossover is only 1-pole, which i have already built. thanks all the same though.

    as for the one Tr97 sent:
    do those filters subtract the low pass from the original sound? because the HP slope looks similar to the version i made with reaktors filters (using the same method). if so, have you got a HP version of the LP filter? because i think if the filters are linear phase i should be able to use them parallel and get a good result.

    thanks
  8. Tr97

    Tr97 Forum Member

    Messages:
    50
    hi.

    yes, the highpass filter is simply a delayed original minus the lowpass filtered signal. so the sum of the bands is a delayed copy of the input signal.

    the highpass is not as steep as the lowpass which is an approximation of a gaussian filter.
  9. Tun

    Tun Forum Member

    Messages:
    199
    damn. i was really hoping for the HP and LP to be 24dB/oct. this filter phase issue is killing me!
    i suppose i will have to stick with subtracting the filtered signal from the original.
    since its a widener im working on i will use a HP and subtract that to get a shallow LP. i think that makes more sense since the high end is less delicate when it comes to width.

    thanks everyone for your help

Share This Page