How is deep learning applied in telecommunications?

Hello everyone,

I work in the field of telecommunications. My goal is to plot BER values against certain SNR values. I send a certain number of bits to MATLAB and determine how many of them are true and how many are false. For example, 0 dB -> 0.1668; 5 dB -> 0.0518; 0.0101.

The code I wrote in MATLAB works for SNR in decibels, I am trying to find the BER value for each SNR value. In addition to classical communication, I want to do these operations with Deep Learning in MATLAB.

For example, it will be very good for my work to do the following:

1) For example, I want to have the following Maximum-Likelihood process done by Deep learning, but I don't know how to do it.

        for loop_h=1:nT
            for loop=1:M
            
                Metrics(loop,loop_h) = norm(y-H(:,loop_h)*ss(1,loop))^2; %Here, "ss" stands for M-QAM modulation symbols and H stands for Rayleigh fading channel
            
            end  
        end

2) I would like to have channel estimation if possible. For example, I express the Rayleigh channel as follows. How can I predict this with Deep learning in MATLAB?

        H=sqrt(0.5).*(randn(nR,nT) + 1i*randn(nR,nT));

Any suggestions on how I can do these operations, a book, a piece of code, etc.? I would appreciate it if you could share it with me. Thanks in advance for your suggestions and help.

Answer 1

[Santosh Fatale]

Hi Fatih,

I believe the following example could be useful for you to understand the usage of Deep Learning for performing channel estimation.

https://in.mathworks.com/help/5g/ug/deep-learning-data-synthesis-for-5g-channel-estimation.html

Answer 2

[hejeh]

Hey — newbie here too, but this sounds doable!

Quick noob-friendly tips:

• Make a labeled dataset in MATLAB of received vectors y (with known ss and H) and train a small neural net to classify which QAM symbol was sent — that replaces your ML metric loop.
• For channel estimation, send pilot symbols and train a regression network to predict H from the noisy pilot y_pilot.
• Use MATLAB’s Deep Learning Toolbox (trainNetwork, layer definitions) and start with a tiny FC or CNN model so it trains fast.
• Keep it simple: synthetic data → train → test BER vs SNR the same way you do now.

Good luck — post your toy code/results and people here will help you tune it!