Re: [linux-audio-dev] Resampling

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: Re: [linux-audio-dev] Resampling
From: Juhana Sadeharju (kouhia_AT_nic.funet.fi)
Date: Tue Aug 14 2001 - 21:57:01 EEST


>From: Jussi Laako <jlaako_AT_pp.htv.fi>
>
>I've got lot of stability problems with high order IIR lowpass filters.
>Those tend accumulate significant DC offsets over time. OK for short samples
>but not for realtime streaming, IMHO.

I were reading about the stuff from Jon Dattorro's paper in
Journal of AES, 1988 Nov. (verify at www.aes.org). The paper
deals with implementation matters after the filter coefficients
are calculated. It seems to deal with second order IIR filters,
not higher order.

Paper suggest to use Direct Form I structure. Second order DFI structure
has four delay elements, and all forward and backward connections
are summed to one accumulator. Paper deals with 16/32 and 24/48 numbers
(<input, coefficients and output width>/<accumulator and multiplier
result width>). I simply use 32-bit float for the input, coefficients
and output, and 64-bit doubles for the accumulator and the multiplier
results. The accumulator result is then truncated directly to 32-bit
float output.

The tests would reveal for what filters the above system is an overkill
and for what filters it is needed. Filters working on low frequencies
usually needs more accuracy than filters working on higher frequencies.
I'm quite convinced 32-bit float is not enough if the implementation
structure is not planned in anyway, so, we have to make tests.

I remember reading somewhere that there is a structure where DC offset
gets removed off magically, but I don't remember details. That could
be tried with higher order IIRs. Olli of music-dsp mentions
a differential evolution person in his doc "http://www.student.oulu.fi/
 ~oniemita/DSP/deip.pdf"; the DE-guy sells a software for designing
higher order IIR filters. Check out if the manual or his papers tells
something about the implementation.

I have gone and still go through plenty of corner-stone papers in DSP.
I will keep the problem in mind and look for a solution.

Regards,

Juhana


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : Tue Aug 14 2001 - 21:58:56 EEST