[embedded] cortex m4f DSP Library

Use CM4 Built in kernel FPU You need to add compilation options when compiling ：--cpu Cortex-M4.fp, To enable FPU Instruction set , Otherwise, ordinary Thumb2 Instruction implementation DSP function . At the same time, you need to add macro definitions ：ARM_MATH_CM4, This is in CMSIS DSP Library Of arm_math.h It uses .CMSIS DSP Library Can be in ARM The official website of ：https://developer.arm.com/tools-and-software/embedded/cmsishttps://developer.arm.com/tools-and-software/embedded/cmsis

If you use MDK Words , Directly in Runtime Environment Add DSP library ,Floating Point Hardware choice Single Precision, Predefined macros plus ARM_MATH_CM4, In this way, the project is automatically added DSP library ：

 

The Library above uses FPU Instruction set generated DSP library , The following is not used FPU Instruction set ......

DSP Library tests ：

static float sin_out[256];
static float fft_mag[256];
static float fft_out[256];

static void DSP_Test()
{
    #define FFT_LENGTH  256
    
    arm_rfft_fast_instance_f32 S;

    //  Create a t=2S Sampling data within the clock , Number of sampling points FFT_LENGTH, Sampling period T=t/FFT_LENGTH, sampling frequency F=1/T
    for(uint16_t i = 0; i < FFT_LENGTH; i++)
    {
        sin_out[i] = arm_sin_f32(i*2*3.1416f/128) + 0.3f*arm_sin_f32(i*2*3.1416f/16); //
    }
    
    for(uint16_t i = 0; i < FFT_LENGTH; i++)
    {
        LOGI("%.4f\r\n",sin_out[i]);
        DDL_DelayUS(1000);
    }
    
    arm_rfft_fast_init_f32(&S, FFT_LENGTH);         // initialization 1024 spot FFT Parameters 
    arm_rfft_fast_f32(&S,sin_out,fft_out,0);        // Calculation FFT
    arm_cmplx_mag_f32(fft_out,fft_mag, FFT_LENGTH); // Calculate the modulus 

    for(uint16_t i = 0; i < FFT_LENGTH / 2; i++)
    {
        LOGI("%4d %4.4f %4.4f\r\n", i, (float)i * 128 / FFT_LENGTH, fft_mag[i]);//frequency = N * sample frequency / FFT length
        DDL_DelayUS(1000);
    }
}

It turns out that Excel Drawing on top , Sample point image （1Hz Sine wave and 8Hz Sine wave superposition ）：

 FFT The result of the calculation is ：

  The two peaks correspond to 1Hz and 8Hz, Consistent with the frequency of the sampling point .