[sdiy] Looking for cheap lowpass filter algorithm

[Seb Francis]

Hi Scott,

Scott Gravenhorst wrote:
> So what are your suggestions regarding filter design, assuming you know you want certain
> rolloff characteristics, phase distortion, ripple, etc., do you start with a low number
> of taps (how many?) and keep physically testing/increasing the calculated coefficients
> until the ear says "OK"?  
>
> That seems rather hit or miss, or maybe I'm making too big a deal of this?  
>
> If you see a nice spectral response graph using many taps, what would make you want to
> try fewer taps when the graph looks horrible with them?
>
> I guess what I'm asking is given all this, what good is the graph then?  How truthful is
> it, i.e., on what graphical characteristics can I depend to give me reliable and useful
> information about my prospective design?
>
>
>
>   

Well I think if you require specific rolloff, phase, ripple 
characteristics then you can quite easily find the minimum number of 
taps necessary to achieve this - just use some software and keep 
increasing the number of taps until you get there.  Some software will 
provide you with an estimated number of taps required fulfil your 
requirements - this is a good starting point, but go lower as well as 
higher from this, and remember that even or odd number of taps can make 
a significant difference.  I would definitely recommend ScopeFIR 
software for this process, although it does have a couple of 
limitations: unless registered you can only export the coefficients for 
up to 32 taps, and you can't process audio with it - so you have to put 
the coefficients into something else to preview your filter with your ears.

I think with a bit of experience a graph will give you a good idea how 
it will sound.  In my case (since I had absolutely no experience with 
DSP filters) I was trying different things and listening to the results. 

What I found in this specific case was that less taps produced a 'nicer' 
sound.  I wrote in my previous post that I thought this had something to 
do with the fact that the phase didn't keep oscillating back and forth 
between -180 and 180 in the passband.  However, having thought about 
this some more I realise this is nonsense.  The graphs of FIR filters 
generally show a sawtooth pattern for the phase in the passband.  This 
is simply the time delay introduced by the filter taps - the phase shift 
that a particular delay introduces varies linearly with frequency - in 
other words there isn't any phase distortion.

So the only factor why it sounded nice must have been that it had a very 
gentle amplitude roll-off compared to higher order filters.  This fits 
with my experience with analog EQ: that I almost always prefer the sound 
of EQ with a very gentle, wide curve (although of course with analog 
EQ/filters there is phase distortion going on as well).

I didn't mind the fact that the filter makes sounds with high frequency 
content sound a bit 'muffled', since in this specific application people 
can just switch to 48kHz mode if they want good wideband frequency response.

I guess in most modern applications my 9-tap filter would not be 
considered adequate.  But then again the bandwidth of much old analog 
gear (that people still love and use to this day) would not be 
considered adequate by modern standards.  Poor frequency response and 
other forms of distortion and instability are not always a bad thing :)

Seb