|
Excerpt Description
Threading filtering operations above Intel IPP is very straightforward. In most cases, threading at the signal level is either elementary or virtually impossible. The former is generally easy because there are no data dependencies between result y[n] and result y[n+i]; examples include FIR and convolution. The latter is generally impractical because there are un-predictable dependencies between individual samples in the result; examples of this include frequency-domain filtering using the FFT, recursive filtering using an IIR, and learning filters such as the LMS-FIR.
| 
