void FIfConvSI (const float x[], float y[], int Nout, const float h[], int Ncof, int mr, int Nsub, int Ir)


Filter a signal with an FIR filter (sample rate change)


This procedure convolves a set of filter coefficients with an array of data. Optionally, the input signal can be interpolated before filtering, and the output signal can be subsampled.

The process of interpolation is equivalent to the conceptual notion of inserting Ir-1 zeros between each element of the input array. This increased rate signal array is convolved with the filter coefficients. Since the increased rate signal array has imbedded zeros, only every Ir'th filter coefficient is involved in producing a given output point. These coefficients can be considered to constitute a sub-filter. Sub-filters are used in a round-robin fashion to produce successive output points.

The process of subsampling is equivalent to keeping only every Nsub'th output point. The filter slides along Nsub elements at a time in the increased rate (interpolated) signal array for every point that is stored in the output array.

The input array is x[.]. The first lmem = (Ncof-1)/Ir samples of x[.] are past inputs. Let the increased rate array be xi[.]. Then, xi[l*Ir]=x[l], with other elements in xi[.] being zero. The relationship between indices of x[l] and xi[m] is m=l*Ir+mr, or l=floor(m/Ir), mr=m-l*Ir. The first output point is calculated with filter coefficient h[0] aligned with sample xi[Ir*lmem+mr], or equivalently, h[mr] aligned with sample x[lmem]. The last output point is calculated with h[0] aligned with the sample xi[Ir*lmem+mr+(Nout-1)*Nsub].

If Nout output values are to be calculated, the input array x[.] must have Nx elements, where Nx is determined from

  lmem*Ir+mr+(Nout-1)*Nsub <= Ir*Nx-1
From this relationship,
  Nx = lmem + ceil((mr+1+(Nout-1)*Nsub)/Ir)
     = lmem + floor((mr+(Nout-1)*Nsub)/Ir) + 1
Conversely if the input array has Nx elements, the number of output samples that can be calculated as
  Nout = floor((Ir*(Nx-lmem)-1-mr)/Nsub) + 1
       = ceil((Ir*(Nx-lmem)-mr)/Nsub).


-> const float x[]
Input array of data. Let lmem=(Ncof-1)/Ir. The first output point is calculated as follows
  y[0] = h[mr]*x[lmem] + h[mr+Ir]*x[lmem-1]
                       + h[mr+2*Ir]*x[lmem-2] + ...
The array x must have at least lmem+((Nout-1)*Nsub+mr)/Ir+1 elements.
<- const float y[]
Array of output samples
-> int Nout
Number of output samples to be calculated
-> const float h[]
Array of Ncof filter coefficients
-> int Ncof
Number of filter coefficients
-> int mr
Filter coefficient offset. The first output point has filter coefficient h[mr] aligned with data element x[lmem]. Normally mr is in the range 0 to Ir-1.
-> int Nsub
Subsampling ratio. Only every Nsub'th filtered output is calculated for the interpolated sequence and stored in the output array.
-> int Ir
Interpolating ratio. Conceptually, Ir-1 zeros are inserted between each element of the input array to create an interpolated sequence before convolving with the filter coefficients.

Author / revision

P. Kabal / Revision 1.17 2005/02/01

See Also


Main Index libtsp