FFT and floating point data

Status
Not open for further replies.
A

adamski

Active member
I would like to port some code to the Teensy 3.5/6 from a mobile/desktop project, that is all floating point arithmetic, and does some FFT based processing.

Are there any examples of doing floating point FFT on the 3.6? I realise the Audio Library is all fixed point.

Or would I need to convert from ints to floats before doing my FP processing?
 
You will most probably have to rewrite the wrappers for the corresponding DSP instructions (dspinst.h) to use their float equivalents.
 
Would it be a bad idea to use a general FFT implementation such as OouraFFT rather than the one in the Audio Library? I will be pulling in data from the ADC, and converting each value to a float as it comes in.
 
adamski said:
Would it be a bad idea to use a general FFT implementation such as OouraFFT rather than the one in the Audio Library? I will be pulling in data from the ADC, and converting each value to a float as it comes in.
Click to expand...

most likely bad idea,
FFT are very CPU intensive activities, and Teensy uses
a) code optimized for build-in DSP (all 3.x Teensies and integer format)
b) optimized code for FPU (3.5/6 Teensies)

Code is anyhow compiled and optimized by CMSIS.

Where possible you should use 16bit FFT as it uses special dual-word operations.
 
Ok, thanks. I'll try the Audio Library then, and convert the 16bit integer results to floats which should not be too CPU intensive with the FPU.
 
Status
Not open for further replies.
Back
Top