Assume i have a discrete signal, which i want to up-sample by factor 100. Up-sampling using poly-phase algorithm sounds like a bad idea, in terms of cache low-efficiency due to low space-locality of the data.

Do you have any suggestions?



  • $\begingroup$ i have absolutely no idea what is meant by "low space-locality of the data". and having done bandlimited sample interpolation for a variety of different applications, i have little idea what's wrong with using the normal polyphase FIR interpolation. it's about the most efficient method if you want phase-linearity or "reversibility" (i.e. the interpolation result is the same whether you do it in the forward or backward direction). $\endgroup$ – robert bristow-johnson Aug 31 '14 at 17:57

Upsampling by 100 can be factored into upsampling by 2X, 2X, 5X and 5X for each output block (output blocks sized to be data cache resident), thus requiring a smaller phase table for each FIR interpolation kernel.

Some output points may either need to be overlapped across blocks or re-calculated, depending on computation speed vs cache load latency.


You can use a multirate structure where the stages have increasingly wider transition bands. This is very close to the situation answered in another question


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.