[SciPy-dev] Apparently bug in filter_design.py

[Joachim Saul]

Hi there,

while using a IIR filter designed using scipy.signal.iirfilter, an
error occurs within lp2bp() and potentially in the other similar
routines as well, where the line

    aprime[Dp-j] = val

occurs. Since val can be complex, and aprime be e.g. of double type, an
error may be (and im my situation is) issued:

TypeError: can't convert complex to float; use abs(z)

Which, in turn, doesn't quite make sense, at least *if* a real
part of the values in aprime may be negative.

A quick fix would be to modify lp2bp():

replace
    bprime = Num.zeros(Np+1,artype)
    aprime = Num.zeros(Dp+1,artype)
by
    bprime = Num.zeros(Np+1, 'D')
    aprime = Num.zeros(Dp+1, 'D')

and just before "return" add
    bprime = bprime.astype(artype)
    aprime = aprime.astype(artype)

This would force aprime and bprime to be always double precision
complex (which should never harm) and at the end cast to the
originally wanted type. Since this is used in filter design only,
the overhead though the conversion would be negligible.

Any comments?

Cheers,
Joachim