search

helloMRDJ / autoencoder-for-the-Physical-Layer

Using Keras to validate the simulation results according to Paper : "An Introduction to Deep Learning for the Physical Layer"
29 stars 18 forks source link

amplitude constraint #1

Open doloresgarcia opened 4 years ago

doloresgarcia commented 4 years ago

Hi! Thank you for uploading your code! It has been very useful. Have you thought about how to implement the amplitude constraint that they mention in the paper ( |x_i|<1) for all i? I would like to obtain the 16 QAM constellation as a result of the training but that is not possible with the energy constraint or the average power constraint. I thought maybe an option would be to change the activation function of the last layer of the encoder to a tanh to limit the outputs to [-1,1]. But it does not seem to converge to 16 QAM then. any ideas?

:)

helloMRDJ commented 4 years ago

@doloresgarcia Hello, I think maybe you could try peak power constraint by using clip(x,-1,1) in the last layer of the encoder.