Hi All, I'm doing some more research into digital filtering, and would like to present a poll to the DSP community. More specifically, I've read a lot about fast convolution/overlap add/save (OLA/OLS), and was wondering about real world applications of these techniques. From previous posts, it seems like the crossover point between FIRs and OLA/ OLS is anywhere from between 30 taps to 1k taps, depending on the hardware - meaning, OLA/OLS is more efficient than FIRs after 30 to 1k taps. So, I was wondering, if you have ever implemented OLA/OLS in hardware: 1. What was your particular application? 2. Did you implement OLA/OLS on a DSP or FPGA? Which one? A single DSP/ FPGA, or multiple? 3. What was the crossover point before you decided to use OLA/OLS? After how many FIR taps was it more efficient to use OLA/OLS? 4. How big was the FFT/IFFT that you used? 5. Did your application require real or complex data to be filtered? 6. Was your data Floating Point or Fixed Point? What was the precision? 7. Were your coefficients for OLA/OLS in the frequency domain Floating Point or Fixed Point? What was the precision? Thanks a lot for your response! cheers, Huang
Fast convolution/Overlap Add/Overlap Save poll
Started by ●September 11, 2007