Projects

Here is a compilation of my personal projects, created for various reasons—whether for enjoyment, academic pursuits, or a blend of both.

Convolutional and Turbo Codes for deep-space communications

A C implementation of the CCSDS (Consultative Committee for Space Data Systems) standard for Turbo Codes, a class of high-performance error-correction codes used in deep-space communications. The implementation includes a library that handles the main building blocks of a turbo code: convolutional codes, interleavers, encoders, and their decoding algorithms. Viterbi and BCJR algorithms are used for the decoding of convolutional codes, but are also used in the decoding algorithm for turbo codes. A few programs and scripts to benchmark them are also included.

The source code is available on GitHub along with some benchmark results.

Wireless MAC protocols simulator

The scope of this project is to compare different channel access policies in wireless networks. It consists of an discrete-event network simulator written in Java that simulates a wireless channel with a given number of transmitters/receivers. The IEEE 802.11 MAC protocol controls the access to the channel with a Distributed coordination function (DCF) [3], but some alternatives have been proposed in the literature. Here the DCF is compared to two new approaches with constant-time medium access resolution.

A fun project that covers a lot of different topics: software engineering, numerical optimization and wireless communications.

Source code and report available on GitHub.

YouTube video summarizer

A classic example of engineers turning a 10-minute task into a 3-hour job. An interesting video popped up in my YouTube feed, but I quickly realized that the simple topic it was supposed to cover was drawn out for an eternity: all I wanted was a quick summary. Inspired by the army of e/acc accounts on my Twitter X feed, I wanted to get my feet wet with LLMs and some AI tools, so I took the opportunity to load some credits (read: $) in my OpenAI account, download a local version of Whisper (whisper.cpp developed by the great Georgi Gerganov) and start building a system that summarizes YouTube videos. The approach is fairly simple: using yt-dlp, check if there are english subtitles available for the video, if there are, download them and send them to ChatGPT to get a summary, otherwise download the audio track, get a transcription from the Whisper model, and then send it to ChatGPT. I am experimenting with local models to replace ChatGPT (mainly Llama-2), but running them on an old CPU results in very low tokens/sec.

Source code on GitHub.

Numerical solver for space-division-multiplexed optical fiber transmission

Project done for my PhD program involving the simulation of light transmission in few-mode optical fibers under the effect of optical nonlinearities, namely Kerr and Raman amplification. The purpose was to study the effect of random birefringence on the properties of Raman amplification for this new type of optical fibers.

The results have been published in this paper:

The project contains Python and MATLAB scripts for:

  1. Running the physics simulation using COMSOL to extract the fiber modes given the fiber index profile
  2. Processing the mode profiles and computing the overlap integrals to determine linear and nonlinear coupling coefficients
  3. Running the experiments and collecting results for many random birefringence processes

The heavy lifting, i.e. the code for solving the nonlinear coupled differential equations using a split-step algorithm, is implemented in C++ with the Intel oneAPI Math Kernel Library (MKL) and compiled in a python module using the pybind11 library.

Finally, the code is packaged and run on a HPC cluster using Singularity containers.

Source code available on GitHub.

PyNLIN

Also related to the topic of optical fibers. A pure Python implementation of some mathematical models for the computation of the nonlinear interference noise accumulating during the transmission of light in optical fibers. Code is automatically dockerized and Singularity-ized (?) in the CI/CD pipeline, and pushed to GitHub's container registry. It is then a lot easier in the HPC environment to simply pull the image and run it, instead of building it locally without root access, or, even worse, building it on my personal machine and uploading gigabytes of data with my sub-par internet connection.

Source code available on GitHub.

Publications derived from this project:

  1. F. Lorenzi, G. Marcon et al., "Model for Nonlinear Interference Noise in Raman-amplified WDM Systems," 2022 European Conference on Optical Communication (ECOC), Basel, Switzerland, 2022, pp. 1-4
  2. F. Lorenzi, G. Marcon et al., "Nonlinear Interference Noise in Raman-Amplified WDM Systems," in Journal of Lightwave Technology, vol. 41, no. 20, pp. 6465-6473, 15 Oct.15, 2023

LastTip

This is a fun little project that was born with me realizing that the music obsessed person that I once was, the one that would spend nights searching for obscure bands and new genres, has been replaced by the Spotify algorithm. With this, a lot of the music that I once listened to has been left aside collecting digital dust. Fortunately, I have been tracking all the music I listen to on last.fm since 2009 (actually a few years earlier, but stupid me decided he didn't identify with a metalhead teenager anymore, and made a new account).

The solution I came up with was a Flask application that would fetch albums from my last.fm profile and gave me back a random suggestion, along with a Spotify link to the suggested album. I then built a Telegram bot that talks to the API I built to get recommendations directly from Telegram, for convenience. Later, I added the option to also get recommendations from the list of albums saved in my Spotify account, as I realized that most of the interesting stuff I discover gets added to my library and promptly forgotten.

Get some recommendations at this link. Or go directly to my last.fm profile.

Source code as always on GitHub.