How to run CMSIS DSP on nRF52840?

[Kubek]

Hi there,
I'd like to make some FIR or IIR filter on my nRF52840 DK. I guess I'm supposed to:

1) Provide .a file into SES project's tree (or as Additional Input File):


2) Provide .h file(s). I used to provide just "arm_math.h" for STM32 projects, but saw in „fpu fft” example that these are two folders provided:

../../../../../../components/toolchain/cmsis/dsp/Include

../../../../../../components/toolchain/cmsis/include

3) Set "ARM_MATH_CM4" preprocessor definition.

4) In main() file include one .h file:

#include "arm_const_structs.h"
But I used to include "arm_math.h" for STM32 projects.

5) In main() file 
- Make two float32_t buffers - input and output
- Make instance of filter - arm_fir_instance_f32
- Init the filter instance - arm_fir_init_f32()
- FIlter the signal - arm_fir_f32()

I use the filter that I know works in Matlab, but here it seems to give me strange values on output - for example, from samples of value about 400, it makes an output about 1.66e-38. It's pretty small value. Maybe I'm just supposed to multiply that small float value, but it's strange - my filter shouldn't attenuate - it rather amplifies the signal in a few first samples.

Simultaneously, these are SoftDevice and ADC that work. Maybe it's a problem with multiplexing enough amount of CPU cycles for CMSIS DSP. I don't know, I had no problems with it on STM32 that only processed the signal.

[Einar Thorsrud]

Hi,

This should not be any problem, and you can use the ARM math library with or without the SoftDevice. You can see how it is done in the FPU FFT Example (which does essentially what you have described).

Perhaps it is a good idea to use this as reference, and copy the relevant parts to your project. Then you can verify that it still works and gives the expected result in your project before you modify it to your needs.