9. DSP

9.1. Overview

The dsp_demo sample project demonstrates the DSP CFFT (Complex Fast Fourier Transform) and CIFFT (Complex Inverse Fast Fourier Transform) calculations. It performs a 32-point FFT transform and validates the results through detailed error analysis.

The example includes: - Complex FFT (CFFT) calculation using radix-2 algorithm - Complex IFFT (CIFFT) calculation using radix-2 algorithm - Input signal generation and scaling - Detailed error analysis including MAE, RMSD, NRMSD, and SNR metrics - Result validation against golden reference data

9.2. Board Setting

No special settings are required.

9.3. Running the example

When the project runs correctly, the serial port terminal will output the following information:

--------------------
after CFFT_RD2, maxdiff= 0x0080 [0.00390816]
--------------------
MAE is 0.00079375, RMSD is 0.00176055, NRMSD is 0.00005502, MAXDIFF is 0.00390816, SNR is 67.12823486
CFFT_RD2 out scaleup by 64
after CIFFT_RD2, maxdiff= 0x0009 [0.00029564]
--------------------
MAE is 0.00012192, RMSD is 0.00014900, NRMSD is 0.00007450, MAXDIFF is 0.00029564, SNR is 73.52618408
CIFFT_RD2 out scaleup by 2
CIFFT_RD2 PASS

The output shows: - The maximum difference between calculated and golden results - Mean Absolute Error (MAE) - Root Mean Square Deviation (RMSD) - Normalized Root Mean Square Deviation (NRMSD) - Maximum Difference (MAXDIFF) - Signal-to-Noise Ratio (SNR) - Scaling factors used in the calculations

The test is considered successful if: - For CFFT: NRMSD is less than the threshold (0.00007 for normal compilation, 0.0002 for zcc) - For CIFFT: NRMSD is less than the threshold (0.0003 for normal compilation, 0.005 for zcc)