TMS320F28P550SJ: About arc deteciton signal

Hi Alfred,

May I know in more detail what type of feature vector you want to design for arc fault detection?

Current arc fault models which are provided by TI have fixed size of 256 input feature vector, if you are interested to expand / reduce the size of input feature vector then you need to design your own neural network architecture. 
You can refer to 
1.1. What is Tiny ML Tensorlab? — TI Tiny ML Tensorlab 1.3.0 documentation

To understand how to design your own neural network in TI's ecosystem.

In default configuration our reference design already uses Log Spectral Power (LSP) as its feature not raw FFT bins.
After the 1024-point FFT computes the magnitude spectrum, run_feature_extraction() groups every 4 consecutive FFT bins, sums their magnitudes,
and applies 10*logf(sum) - fft_offset to produce 32 LSP features per frame.
Eight frames are concatenated into a 256-element int8 vector fed to the Neural Network.