Posted on August 07, 2012 at 19:42Hello
I dont know why, but when I use FFT code, STM32F103RBT6 just stop running in totally different part of code - mostly in parts of STD library where asserts are defined. So I think it can corrupt memory or something in weird way. It can work for hours, but when I create some input to other parts of the software it stop working as I described.
For example - I reading data from sensor with I2C and running FFT every loop. In this loop I can do some additional inputs as PWM. When I dont change PWM signal it just work, but after few minutes of input it ends in Assert of I2C.
But when I dont use FFT, there is no problem with hangs.
Different implementation of the FFT works fine too, but that one which meets my requirements dont. This thing happened for me with very simple Kalman filter too (sw float).
Sources are in C. I use -0s, -thumb optimization, Summon GCC toolchain.
It seems like a compiler problem, I dont trust GCC very much since I found more bugs now.
I dont use any memory operations, all stuff is done by compiler.
Code of the FFT and Kalman is attached - both caused me a problems after 1 - 100 minutes. Do you know good implementation of a FFT with inverse FFT for 16bit/32bit integers? Radix 4 implementation by Ivan Mellen is not good, since it divide output by N and I need all 16bit range.
Important thing is, that after this hang main loop stop working, but interrupts from system timer, pwm, serial are working without any problem.
Thank you very much for any hints.
#stm32-fft-hang
