Digital EQ

Hi,

I'm working on a digital EQ to be simulated on a TMS320C6713 dsk board.
Initially I am developing a three band EQ to be implemented via
convolution. y(n) = k(n) * h(n)

The way my group at university have got an ouput is by averaging the
output of the three bands. This obviously does not give an accurate
output and is quite disappointing in audio fidelity.

One specific thing I wanted to ask about was the different FIR filter
types available in Matlabs FDATool. Initially I have used the
Equiripple filter which gave reasonable cut-off for the low pass and
high pass and little ripple but the filter order could not go higher
than about 120.

We have been informed by our lecturers to try and get a 200 order
filter for each band...does this sound correct in audio terms? Our
faith in our lecturers has diminished somewhat as they have given us
little guidance on our DSP course and I am now seeking advice/support
from elsewhere.

One other filter type I have tried is the Windowing FIR - specifially
the hamming window which give great performance (little to no ripple in
the pass band and good cut off slope) for low pass, high pass & band
pass at 200 order.

Are these hamming filters good for audio use? Any help/advice
appreciated! 
Thanks, Alex

"mr_alex" <dancedynamix@hotmail.com> wrote in message
news:1141047886.882888.217780@i40g2000cwc.googlegroups.com...
> Hi,
>
> I'm working on a digital EQ to be simulated on a TMS320C6713 dsk board.
> Initially I am developing a three band EQ to be implemented via
> convolution. y(n) = k(n) * h(n)
>
> The way my group at university have got an ouput is by averaging the
> output of the three bands. This obviously does not give an accurate
> output and is quite disappointing in audio fidelity.
>
> One specific thing I wanted to ask about was the different FIR filter
> types available in Matlabs FDATool. Initially I have used the
> Equiripple filter which gave reasonable cut-off for the low pass and
> high pass and little ripple but the filter order could not go higher
> than about 120.
>
> We have been informed by our lecturers to try and get a 200 order
> filter for each band...does this sound correct in audio terms? Our
> faith in our lecturers has diminished somewhat as they have given us
> little guidance on our DSP course and I am now seeking advice/support
> from elsewhere.
>
> One other filter type I have tried is the Windowing FIR - specifially
> the hamming window which give great performance (little to no ripple in
> the pass band and good cut off slope) for low pass, high pass & band
> pass at 200 order.
>
> Are these hamming filters good for audio use? Any help/advice
> appreciated!
> Thanks, Alex

Sound like they are after a fast convolution to me. The Chassaing book has
an example of this.

Regards, Howard

mr_alex wrote:

> Hi,
>
> I'm working on a digital EQ to be simulated on a TMS320C6713 dsk board.
> Initially I am developing a three band EQ to be implemented via
> convolution. y(n) = k(n) * h(n)
>
> The way my group at university have got an ouput is by averaging the
> output of the three bands. This obviously does not give an accurate
> output and is quite disappointing in audio fidelity.
>
> One specific thing I wanted to ask about was the different FIR filter
> types available in Matlabs FDATool. Initially I have used the
> Equiripple filter which gave reasonable cut-off for the low pass and
> high pass and little ripple but the filter order could not go higher
> than about 120.
>
> We have been informed by our lecturers to try and get a 200 order
> filter for each band...does this sound correct in audio terms? Our
> faith in our lecturers has diminished somewhat as they have given us
> little guidance on our DSP course and I am now seeking advice/support
> from elsewhere.
>
> One other filter type I have tried is the Windowing FIR - specifially
> the hamming window which give great performance (little to no ripple in
> the pass band and good cut off slope) for low pass, high pass & band
> pass at 200 order.
>
> Are these hamming filters good for audio use? Any help/advice
> appreciated!

Hello Alex.

You were here with the "Confused" thread a little while ago, no? I want
to ask a question: want kind of filter do you want to implement?
Usually, in audio eq one uses so called "parametric" (the other eq type
would be "graphic", but then you need a lot more than three bands).
Lowpass / highpass etc. are only one part of the filter types. In
addition to these types, one often uses shelving or peaking filters
(this is for example discussed here:
http://www.wpi.edu/Academics/Depts/HUA/TT/TTHandbook/sound/eq.html).

Audio engineers are accustomed to specifying filter type (lowpass,
highpass, shelving, peaking), Q, boost and frequency as filter
parameters. If you really want to programm an audio eq with any
practical relevance, then you need to implement these parameters to
specify the filter. In the link I gave you in the other thread (r b j's
cookbook) you see how to calculate the coefficients of a biquad
section, given the filter parameters. Such biquads are usually
implemented with recursive (IIR) filters, and connnected in series (and
not parallel) - ie each biquad's output goes into the next biquad's
input. Each band needs four (or five, depending) coefficients, and they
are computationally extremely efficient.

In contrast, one typically uses FIR filters in audio work only when you
are interested in either very steep filter slopes or the phase response
of the filter. I don't think either is really what you are after (there
is a short discussion on this here:
http://www.weiss.ch/eq1/images/brochureEQ1-LP.PDF).

Of course, maybe the course just requires you to implement an FIR
filter, and your teacher is trying to package the FIR filter as audio
equalizer to make it an interesting project. Ask your assistant: is FIR
a requirement, or can the equalizer also be implemented with IIR
filters?

BTW: Depending on the sample rate and the frequency of interest for the
filter, even 200 FIR coefficients don't get you the kind of performance
you expect from an audio equalizer. The last time I did a quick
back-of-the-envelope calculation, I think I arrived at 8192 as a more
or less sufficent number of FIR taps to emulate a given parametric IIR
equalizer setting (at 96kHz sampling rate). However, the advantage of
the FIR approach is that you can choose the number of filter taps
independently of the number of filter bands. This number of taps is
only realizable using frequency domain filtering (unless you have a
Cell Processor at your disposal :-).

Regards,
Andor