2015-11-19 06:36 AM
I have installed all DSP source files on the project and f32 fft works perfectly, but when I switch to q31 or q15, while the input buffer shows got data, the output buffers all values (the frequency spectrum) are 0. All the files were replaced to the same files for q15 or q31. Has anyone encountered this?
#fft-stm32f42015-11-19 10:48 PM
A foolish mistake, since the amplitude is now in q15, the sensitivity is a little smaller (less bits in the amplitude). Since the sensitivity was smaller the fft did not detect so much noise, thus outputing 0's. Needed to apply bigger amplitude movement on the sensor.
2015-11-19 11:33 PM
For a M4 with FPU, using q15 or q31 makes limited sense.
Unlike (some) DSPs, these formats are not natively supported on the Cortex M core and need to be emulated. On M0/M3 cores, this is still faster than a single-precision floating point emulation.2015-11-20 08:33 AM
I tried to achieve minimal time for FFT, q15 and q31 took less time than f32, for the same sample quantity.
2015-11-20 09:38 AM
>I tried to achieve minimal time for FFT, q15 and q31 took less time than f32, for the same sample quantity.
Built the f32 code to actually use the FPU ?