Implementation separated autoencoder in an OFDM system by Deep Network Designer

I want to implement an autoencoder that separated to an encoder in tramsmitter and a decoder in receiver in an OFDM system . The joint loss function of system is sum of two separated loss functions(L1: reduction of BER, L2: reduction of PAPR). The L2 loss function is only related to encoder and L1 is related to both encoder and decoder. I don't know how design a DNN with a black box between two separated part of that by Deep Network designer? Also I don't know how can I reduce the PAPR that is only related to encoder(i.e. the method of backpropagation in this structure)?

Answer 1

[Aditya Patil]

You should be able to write custom layers, and import them into Deep Network Designer. However, currently it is not possible to implement the architecture you mentioned with only Deep Network Designer.

Answer 2

[Ethem]

You can implement this using custom layers and custom training loops. Check out these links:

https://www.mathworks.com/help/deeplearning/ug/train-network-using-custom-training-loop.html

https://www.mathworks.com/help/comm/ug/autoencoders-for-wireless-communications.html

[CHANNARAM]

I aim to implement a Peak-to-Average Power Ratio (PAPR) reduction in MIMO-OFDM systems using a Deep Learning technique, specifically a Convolutional Autoencoder (CAE). The objective is to reduce PAPR, enhance Bit Error Rate (BER) performance, and improve power spectral density. This project will leverage deep learning-based encoding and decoding techniques to optimize signal transmission efficiency in MIMO-OFDM systems. can i know how to implement this type of project.

[Ethem]

There are many examples using AI in communication systems here:

https://www.mathworks.com/help/comm/ai-for-wireless.html

There are several application examples here:

https://www.mathworks.com/help/comm/applications.html

There are also OFDM and autoencoder examples here:

https://www.mathworks.com/help/comm/autoencoding.html