sydney222 / python_csinet Goto Github PK

What is the matrix DATA_HtestFin_all.mat? How was this obtained and why is it used specially for cosine similarity?
Cant we use the data DATA_Htestin.mat for cosine similarity calculation?

Hello!

Could you specify a license for this project?

I do not know where is a part of the code to normalize the data input into the range [0,1] as mentioned in the paper. And then, why are cosine similarity and NMSE calculated from the reconstructed complex data which you subtracted the predicted real and imaginary data by 0.5? How do you reconvert the normalized data to original data to calculate NMSE and rho?

could you tell me how to create the data?

This part of your code seems to be wrong. x_test_C and x_hat_C are the angular domain channels, it should be the frequency domain channels X_test and X_hat instead.

That could explain your unused dimension change code as well.

As I can see, the fequency domain channel is the proper one to choose for final NMSE calculation. Besides, you don't have to re-calculate angular domain MSE, it is how your define your loss function.

Of course it could be my misunderstanding, it would be very kind of you to correct me if so.