I read on Cool Edit help that FIR filter is better compared to IIR which could produce ringing due to phase shifting. But it's not mentioned whether FFT is good or not.

Check this thread:

http://www.hydrogenaudio.org/forums/index....c=2418&hl=phase

Especially note the comments by Frank Klemm

I would say that Cool Edit's FFT Filter screen is an UI to generate FIR filters with specific parameters quickly.

FIR filtering is basically convolution of a signal with a generated impulse response, aka the filter kernel. In this case FFT is a method that speeds up the convolution process considerably.

And probably Cool Edit's manual oversimplified matters a bit when it stated that FIR is better than IIR. Not quite always.

as far as i understand every IIR filter can be approximated "close enough" with a long enough FIR filter.

so FIR can do whatever IIR can do.
but FIR can do many nasty things too.

so it strongly depends on the so-called filter kernel (the impulse response) of the FIR filter if sound better or worse or just the same as a IIR with the same frequenzy/amplitude function.

bye,
--hustbaer

I think IIR filter is bad!! (for image processing too..) because it is non-linear phase.. Which means, the delay of the filter is not a constant for the entire spectrum.. Some frequencies could be delayed longer than others resulting in
terrible output quality..

For Best Quality, you have to ensure that the filters are linear-phased..

nooooooooooooo!

for image processing, yes, but never for audio.
with audio-data, it's important that you have very little pre-ringing, if possible zero!

with linear-phase filters you have plenty pre-ringing - and that's the bad thing with audio.
mathematically the result is "cleaner" - but it just sounds crappy!

no natural phenomenon causes significant pre-ringing - they all behave like IIR filters with little or no pre-ringing, and much post ringing.

if you read the article linked above (EDIT: i mean "thread" not "article" of course) carefully you will find the same info there - only a little bit more detailed.

please never ever use linear-phase filters when you can use minimum-phase filters.

one exception maybe - very very small oversampling filters.
lets say no more than 8 or 9 taps.

bye,
--hustbaer

Well, with 44.1 KHz sampling, a FIR filter with 80 taps would produce less than 1 ms pre-ringing, which I doubt would be audible, thanks to temporal backwards masking. Going up to 128 taps would be around 1.5 ms, which would be around the limits of threshold of audibility.

Also, you can design FIR filters with non-symmetrical ringing that have little pre-ringing.

IIR filters are normally used, maybe for speech type of audio..
I have not heard of IIR filters in high end audio systems??
They are mostly FIRs..

Having the output delays of the IIR filter as a function of frequency is bad! I am
sure it is noticeable.. (wouldn't it??)

In fields of image processing, it is bad!!

you can not compare image processing with audio processing .

in a typical "sound reproduction system" there are so many many components that alter phase and introduce frequency-dependant delays, that it really does not matter if some IIR filter adds some more.

of course it always depends on where you want to use that filter, how much attenuation per octave you want (or how small a "Q" when it comes to EQs) and some other parameters.

e.g. an IIR brickwall-filter for CD-audio should be absolutely no problem.

it can sometimes make sense to use a linear-phase (zero-phase) FIR filter, but most of the times the pre-ringing produced by such a filter is more disturbing that the frequency-dependant delya of a linear-phase filter.

most DSP studio-monitors that feature FIR filters to compensate for the IR of the drivers and use them as crossovers have a recommended minimum-phase setting, because the spot where the different pre-ringing artifacts played be the different drivers cancel each other out would be very small in linear-phase mode.

and many digital equalizers use IIR filters also because they need far less computing power.

imho the downside with IIR filters is that they need high floating-point precision to calculate precisely - but then again, you can always use a minimum-phase FIR filter to get (nearly) the same result.

so - i know the problems with pre-ringing - it's simply audible when it's too long, but i do not really know any problems with (moderate) phase-shifts or phase-dependant delays...
so maybe if you have some information i have not, you let me know

bye,
--hustbaer

FIR means Finite Impulse Response.
IIR means Infinite Impulse Response.

In other words, in a quantized digital domain, a long enough FIR should be approximately equivalent to an IIR filter, like hustbaer said.

In other words, again, there is next to no difference between the two from an implementation standpoint. There are some notation differences (IIRs tend to be written as recursive FIRs), but the process is fundamentally identical. You will not use an IIR filter as an EQ in anything digital/discrete (which is the only place where either has meaning), as it would take an infinite amount of processor time to calculate it.

To answer the topic question: neither; but you'll never use an IIR filter. You'll only ever approxmate using FIRs.

Sorry to resurrect an oldish thread, but that comment is nonsense!

What gives a digital IIR filter its "infinite" response is the feedback within the structure. With the FIR structure, once your impulse has walked its way down the length of the filter the output is back to zero, for the IIR it will keep going until the response decays below the numeric resolution (well, ideally it would, but it might hit a low level limit cycle depending on the implementation architecture - then the response really would be infinite, although not in the manner intended ).

You would be hard pressed to find a digital mastering desk anywhere that does not use IIR EQ filters.

The main advantage of IIR is that it just gets better attenuation for a given number of filter taps, ie. they are cheaper to implement. Cycles these days are plentifull though, and the 0 latency FFT based methods of implementing FIR filtering have evened the odds quite a bit.

I dont really understand why people still use mixed output from band pass filters for parametric EQ ... IIR or FIR, the stuff hopening in cross over regions just is not pretty. Why not "just" take the amplitude spectrum desired by the user and calculate a minimum phase filter from it for FFT based filtering?

So, the conclusion is that IIR or minimum-phase FIR filter is better than linear-phase FIR filter for EQ.

If that so, Adobe Audition/Cool Edit help should be revised

edit: I experiment a bit with Cool Edit FFT filter, it uses linear-phase FIR. Wouldn't it audibly worse than IIR? considering the pre-echo of FFT size 16385 is already more than 185 ms.