Timeline for Identifying 40Hz frequency shift
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 9, 2015 at 12:53 | comment | added | Carl Witthoft | To be more accurate: you can do a DFT (discrete Fourier Transform) for any length input. The time-saving "collapses" that Cooley-Tukey and many enhancements thereof require powers of 2 length, or products of powers of small primes. | |
| Jun 9, 2015 at 8:48 | comment | added | JRE | OK. Cool. Learn something new every day. | |
| Jun 9, 2015 at 8:47 | comment | added | Nils Pipenbrinck | There are fast algorithms for all powers of two and for all prime numbers. Every block-size N can be handled by doing a FFT-chain based on N's prime-factors. Most FFT-libs out there only deal with pow2 as you said, and that makes people think that pow2 is required. If you want a block-size of 1280 for example you can start with FFTs of size 5 followed by FFTs with block-size 256. That'll be faster than rounding up to the next power of two. | |
| Jun 9, 2015 at 8:37 | comment | added | JRE | Most FFT implementations do only use powers of 2. Some of the FFT libraries I've seen will accept any blocksize, but may switch to a DFT internally or use some other method of achieving the given block size. I'm not enough of a mathematician to argue whether or not the FFT can actually use other block sizes, I just know that it isn't generally done. | |
| Jun 9, 2015 at 8:31 | comment | added | Nils Pipenbrinck | A common misunderstanding: You can do a FFT at any blocksize. Powers of two just happen to be the fastest and leanest to implement. I agree with you other conclusions though :-) | |
| Jun 9, 2015 at 7:59 | history | answered | JRE | CC BY-SA 3.0 |