Frequency band separation techniques

[Shroomz~>]

Hello,

I thought this subject might be an interesting one to discuss because there's more than one way to achieve band separation & it seems like the separation technique you choose would depend purely on the application (number of bands, device purpose etc). I've tried quite a few different methods now in the sdk which I'll detail if you like. I've had good results with all of them. I can see the benefits of certain methods for particular tasks, but thought it would be cool to discuss it & see what some of you guys think and have to offer on the subject. There's always the issue of cycle/memory cost as well, so that affects the method (module in the Scope paradigm) selection for particular tasks. What methods of band separation do you think are preferable & which do you prefer for given tasks? Ok, that's maybe a big question, but I think it's probably not & actually relatively simple for certain applications. Any thoughts? All input is welcome.

[MCCY]

You know, that I'm a big fan of broad seperation like I did in most of my devices, but I'm not that smart concerning alternative methods & their advantages. I remember reading something with "fir" on it recently, but I might remember wrongly. As I'm no expert in digital filters I have no idea how to implement/build fir-filters... Is it possible with sdk?

[tgstgs]

'... Is it possible with sdk?'

YES;

back to the phase!?
_________

good vibes from vienna

[hifiboom]

I think there is even a fir atom in the sdk package....

[MCCY]

Sounds cool. Could anybody explain in easy words, what fir is, b.t.w.?
Or was "back on phase" a hidden hint once again? It's nicer to ask here than reading in wiki.

[wolf]

fir: finite impulse response filter, basically done with the delayed signal added to the original signal.
iir: infinite impulse response filter, basically done with a delay feeded back & added to the original signal.

[hifiboom]

Sounds cool. Could anybody explain in easy words, what fir is, b.t.w.?
Or was "back on phase" a hidden hint once again? It's nicer to ask here than reading in wiki.

a lowpass is a mathematical technique to smoothen discrete values. quick discrete peaks are high freq. audio information.

fir uses a dedicated amount of delayed signals to calculate the new values.
iir uses the technique of recursive programming to calculate the new values.

mehr oder weniger sind alles mathematische Glättungsverfahren.
Glättung entspricht sozusagen einem Lowpass Filter.

[Shroomz~>]

Hey, thanks for all the comments so far!

You know, that I'm a big fan of broad seperation like I did in most of my devices

Yes, by broad separation I take it that you mean creating crossovers using configurations of HP & LP filters? Or maybe in a HP-BP-BP-LP setup?

You could also use a network of LP filters (6-24dB), sync inverters & 2>1 adders to create a nice separation circuit up to a certain No' of bands. It works on the theory that if you feed 2 LP filters with the same input signal, set one to say 300Hz & the other to say 3kHz then subtract the 300Hz filter output from the 3kHz filter ouput, you create a bandpass filter with a pass band of 300Hz - 3kHz.

I haven't tried the FIR approach though. Would that not be a bit much tax on the old dsps?

I guess there's alway's the parametric eq network as well, but I haven't investigated that as far up as 31 bands or anything like GRAPH EQ.

[Immanuel]

When you using lp/hp/bp filters, remember that the filter frequency is the -3dB point. Therefor put to identically dialed filters in series for a good and easily usable -6dB point.

[tgstgs]

@MCCY

its all about the phase you know . .

maybe now, with a 15 month delay,
you see my point in the funny phase discussion without emotion from a different phase position _;
---------
you can build your own filters or combine existing ones but dont invent the weel for a second time;

simple lowpass filter greetings

good vibes + (good vibes - 1sample) = cutofffrequency samplerate/2

[hifiboom]

good vibes + (good vibes - 1sample) = cutofffrequency samplerate/2

should be normalised to get the average value

goodvibes(t) / 2 + goodvibes(t-1) / 2 = cutofffrequency samplerate/2

so an add2N could be used in that case.

[MCCY]

its all about the phase you know . .

Yeah. Spend hours on that topic.

you can build your own filters or combine existing ones but dont invent the weel for a second time;

Now that was my point all the time, wasn't it? In the end all is about sound I'd say.

Fir seems to excede my knowledge for now (by far). From what I read here, in wiki & co. I'll have to learn quite a lot to understand those things on a more basic level... But I think best learning starts with interest

[tgstgs]

@hifiboom

always sending full good vibes no half from vienna

[hifiboom]

haha jeah,
my fault ! (nice one)

[sonolive]

hi,

'... Is it possible with sdk?'

YES;

DONE !!! a fully new concept/algo of IIR ... 11 Eq modes ... SDK maths and +++

Screen shots and analysis coming soon ... Stay Tuned ...

cheers
olive

[kylie]

Screen shots and analysis coming soon ... Stay Tuned ...

"soon" as in "scope5 coming soon" ?
nice to hear you have something new in preparation.

-greetings, markus-

[alexmcmahon]

The FIR atom is an enigma to me as is has no variables to configure - anyone know how to use it ? say to construct a FIR with 21 taps for example ?

Thanks

[alexmcmahon]

Another frequency band separation technique to consider is to use FFT, get rid of unwanted bands and do an IFFT

[Shroomz~>]

Thanks for getting us back on topic Alexmcmahon.

You could also connect the LP & HP outputs of a multimode filter to a x-fade module, getting a new filter type which has a controllable curve. In the 0 x-fade position the new filter will be a notch or bandreject filter.

[Neutron]

i bought an extra pulsar 1 with the explicit purpose of running crossover filters created on SDK to help design active crossovers and delays for amplifier-per-driver speakers.