A very good tutorial about such an interesting topic....I am looking forward to see more videos from you Dr.Emil regarding communication systems hot topics and basic fundamentals as well. Overall, thank you so much Dr.Emil for your effort
Thank you Emil, I did not get how you dealt with the temporal nature of channel in the very last example application? How did you trained it so that it is general enough to get the right data over the fading channel, for instance?
This is precisely what the neural network learns how to deal with. We gather training data for different fading realizations and learn a mapping between received pilot signals and the channel states. The trained neural network outputs the channel estimates.
Thanks professor, Kindly, could you please explain how did you plot the detection regions of the NN results? How are the x and y coordinates of these regions are going from -8 to 8?
We made a fine grid of test points and then the NN classified each of them. We plotted the results using filled squares to get colored regions. We only show results for coordinate values between -8 and 8, it we could have easily increased the range.
Since the purpose of that example is to describe what deep learning should not be used for, using a purposely poorly trained network, we have decided to not share it. We don’t want to encourage anyone to improve on our poor solution since there is already a simple and optimal solution that is taught in the first course on communications.
![Ep 7. Machine Learning for Wireless [Wireless Future Podcast]](http://i.ytimg.com/vi/Mx-z6KQPJJc/mqdefault.jpg)
Thank you so much professor.
You are a gift to the researchers' community 🌹
Million thanks to the great professor. Greetings from Vietnam :)
A very good tutorial about such an interesting topic....I am looking forward to see more videos from you Dr.Emil regarding communication systems hot topics and basic fundamentals as well.
Overall, thank you so much Dr.Emil for your effort
In ahr mm wave algorithm, nn & dl are heavily used.
Could be useful for forward error correction and multipath compensation. Oh, you already kinda covered FEC for nonlineararity.
Yes, and some new aspect that is hard to model (such as nonlinearities) is needed if deep learning should be truly useful.
very good and useful. thanks Prof. Emil
Thank You Sir..
Amazing explanations
Thank you Emil,
I did not get how you dealt with the temporal nature of channel in the very last example application? How did you trained it so that it is general enough to get the right data over the fading channel, for instance?
This is precisely what the neural network learns how to deal with. We gather training data for different fading realizations and learn a mapping between received pilot signals and the channel states. The trained neural network outputs the channel estimates.
Thanks professor,
Kindly, could you please explain how did you plot the detection regions of the NN results? How are the x and y coordinates of these regions are going from -8 to 8?
We made a fine grid of test points and then the NN classified each of them. We plotted the results using filled squares to get colored regions. We only show results for coordinate values between -8 and 8, it we could have easily increased the range.
Can you share the dataset used in deep learning model
Since the purpose of that example is to describe what deep learning should not be used for, using a purposely poorly trained network, we have decided to not share it. We don’t want to encourage anyone to improve on our poor solution since there is already a simple and optimal solution that is taught in the first course on communications.
That is useful
All the math is giving way to DL :)