Two features make the use of interpolation filter difficult to implement in parallel. First, the filters are different for each frequency to interpolate and this frequency varies with the output frequency and wave-number, according to the mapping function. Secondly, although the filter length is fixed, the filter is applied at different locations in the input space. Thus, the interpolation is not a convolution of the input space with a fixed filter, but the application of various weighted masks at irregular locations of this space. Therefore, neither the frequency axis nor the wave-number axis can be used as a parallel dimension.

11/16/1997