FIR filter with complex data and real filter coefficients

Kandle__Douglas — Mon, 10 Dec 2018 20:33:41 GMT

The most common case I have using the FIR filtering functions is with complex time series data and real coefficients. However all of the functions I've found to use on complex data have complex filter coefficients. I've tried splitting the real and imaginary parts into two arrays, filtering each individually, then putting them back together into a complex array -- and I've converted the real coefficients to complex and performed a complex-complex FIR filter operation. Both yield the same answers. However the complex FIR is about 3 times the execution speed (which is a problem for me). It seems that having to split the real and imaginary parts and then putting them back together again should be totally unnecessary. Is there anyway around these extra steps while still getting the speed of doing scalar multiplies during the filter execution?

Hello,

Tatyana_B_Intel — Tue, 11 Dec 2018 07:20:17 GMT

Hello,

A couple releases ago it was introduced special FIR API for complex data and real coefficients - ippsFIRSR32f_32fc, ippsFIRMR32f_32fc.

Please try to use it. Here are the links to reference manual pages for single-rate and multi-rate FIRs.

https://software.intel.com/en-us/ipp-dev-reference-firsr

https://software.intel.com/en-us/ipp-dev-reference-firmr

If you have any questions please let me know.

Best regards,

Tatyana

topic Hello, in Intel® Integrated Performance Primitives

FIR filter with complex data and real filter coefficients

Hello,