Papers submitted to IEEE journals

Since last my ECOC paper on end-to-end learning unfortunately got rejected. But fear not, we have not stopped working since we got those news, quite the contrary. Big thanks to co-authors Francesco Da Ros, Darko Zibar and Mikkel N. Schmidt.

End-to-end learning in bandwidth limited optical channels

The first paper is a direct extension of our ECOC submission, where we try to learn FIR filters for pulse-shaping and receiver-side filtering in optical communication links. Crucially, we learn the filters jointly with the use of stochastic gradient descent, which shows a significant improvement over the single-sided optimization schemes we compare to. Since ECOC, we have added,

  • A non-linear modulator, namely the electro absorption modulator (EAM)
  • Wavelength division multiplexing (WDM)
  • More details and more simulations

The paper is currently under review, so fingers crossed!

Links:

  • Link to arXiv
  • Link to code. This repository evolves over time. Next step is do some gradient-free stuff to make it more experimentally feasible.

Blind non-linear equalization

The second paper is a side-project that I have been working on for a while. We take leap in the work by Caciularu and Burshtein on variational autoencoders for receiver side blind equalization. We extended their channel model from a finite impulse response filter to a second order Volterra series, and still maintaining a analytical evidence lower bound (ELBO). The model performs on par with the supervised equalizers in most scenarios, and highlights the importance of appropriate channel assumptions in unsupervised equalization frameworks.

Links:

  • Link to arXiv
  • Link to repository
  • A. Caciularu and D. Burshtein, “Blind Channel Equalization Using Variational Autoencoders,” in 2018 IEEE International Conference on Communications Workshops (ICC Workshops), May 2018, pp. 1–6. doi: 10.1109/ICCW.2018.8403666.

Other resources