Designing time-invariant multi-path filter

[pfeatherstone]

I'm looking at different ways of designing multi-path filters for simulating an environment. Currently i'm using Jake's model to model a frequency selective fading channel. This works pretty well and is super configurable. (a base implementation is provided in GNURadio)

I'm also looking at more "dumb" methods for generating multi-path filters. I was thinking of generating an arbitrary frequency response, then using Parks-McClellan to give me the filter coefficients.
So algorithm would be:

• f = randint(low=0, high=0.5, num=21)
• a = randn(num=21)
• b = firpm(21, f, a)

Does this seem sensible?

[pfeatherstone]

Intuitively, I don't think this should work as a multi-path filter doesn't have "bands". Ideas?