Optical fiber communications, machine learning, and robotics.
My doctoral research focused on applying model-based neural networks to equalization in coherent optical fiber communication systems. I developed learned digital back-propagation (LDBP) algorithms that combine classical DSP structure with deep learning optimization, reducing receiver complexity while maintaining performance in dispersion-managed long-haul and metro systems. I also designed low-complexity convolutional neural network equalizers targeting practical transceiver constraints. This work was conducted within the EU H2020 MSCA REAL-NET network.
During my master’s studies, I investigated low-complexity automatic modulation classification (AMC) algorithms using statistical moments and likelihood maximization. I implemented and compared machine-learning-based and parametric classifiers under realistic fading and noise conditions, culminating in an IEEE Communications Letters publication.
I am currently developing AI-assisted segment detection and PID gain scheduling methods for competitive line-following robots using IR sensor arrays and physics-based simulation. This work explores where machine learning should be inserted within the sensing and control pipeline—sensor processing, segment recognition, or action generation—without replacing classical PID control.