Hi Jayanth,
I assume the part you'd like to play with is the PFB_FIR since the next
step is to just follow with a FFT. Here's my Python code for simulating
the pfb_fir. It's based on this note by Parsons and Werthheimer
http://setiathome.ssl.berkeley.edu/~aparsons/papers/2003-06_pfb_32.html.
def pfb_fir(x):
N = len(x) # x is the incoming data time stream.
taps = 4
L = 512 # Points in subsequent FFT.
bin_width_scale = 1.0
dx = math.pi/L
X = numpy.array([n*dx-taps*math.pi/2 for n in range(taps*L)])
coeff = numpy.sinc(bin_width_scale*X/math.pi)*numpy.hanning(taps*L)
y = np.array([0+0j]*(N-taps*L))
for n in range((taps-1)*L, N):
m = n%L
coeff_sub = coeff[L*taps-m::-L]
y[n-taps*L] = (x[n-(taps-1)*L:n+L:L]*coeff_sub).sum()
return y
Sean McHugh
> Hi,
>
> I'm getting started on implementing a polyphase filter bank in CUDA. But
> before getting into it, it's probably a good idea if I play around with
> some
> simple implementation in a relatively more user-friendly environment such
> as
> MATLAB. Does anybody have a bare-bones MATLAB (or even C) implementation
> of
> a polyphase filter bank that you wouldn't mind sharing?
>
> Appreciate any help! Cheers,
>
> Jayanth
>
