>I have a couple of other reasons why it would be good to keep the 'old'
>filters:
>
>--  they are of historical significance. 'tone' and 'reson' are
>described as such in all the literature from at least Dodge and Jerse
>onwards.
>
>-- they are relatively simple - which mean that dsp innocents like me
>have a batting chance of understanding them! The code is not only
>functional, it is very educational.


I don't think anyone is suggesting getting rid of tone and reson.  They were
discussing instability problems with filter2.  When the poles get close to
the unit circle the finite precision representation of the numbers causes
them to be greater than one which causes the resonse to "blow-up".  Higher
order filters use coefficients which are very small in magnitude.  My
question is are the problems due to rounding errors or due to some other
problem in the code?  There was some dicusion on comp.dsp recently on how to
compensate for the rounding errors but I didn't look at it in detail.

Regards,
Hans Mikelson