Interests
I’m drawn to where software meets the machine. My work, so far, has touched this aspect from two different angles, automatic differentiation of Java bytecode with ADiJaC and true random numbers generated from radio noise.
Also, I tend to follow what’s going on in the open-source EDA world. And on top of that hardware description languages (HDLs), as well.
Publications
M. Barbu, D.-C. Trancă, and F.-A. Stancu, “True Random Number Generation from Very Low Frequency Radio Noise,” in 2025 24th RoEduNet Conference: Networking in Education and Research (RoEduNet), IEEE, 2025, pp. 1–6. doi:10.1109/RoEduNet68395.2025.11208358 · code
Abstract
Radio noise use in True Random Number Generation is a popular and well understood strategy, but relatively undermined in research. Given that Very Low Frequency radio streams record chaotic weather events from around the world (like thunderstorms), this novel entropy source sparked our interest. Our choice is supported by prior research in statistical modeling, along with an appropriate combination of both signal and statistical methods. Hence, we describe in detail our processing pipeline for engineering a TRNG from VLF radio noise. It yields high entropy, unbiased and identically distributed outputs that pass FIPS 140-2 tests, regardless of the chosen VLF stream source. To strengthen reproducibility, we have based all of our experiments on publicly available data. We also provide an open-source, modular command-line tool for building stream-based processing pipelines. Furthermore, we offer an example implementing a TRNG from FM radio noise found between standard transmission bands.
Education
- Master of Engineering, Advanced Computer Architectures, National University of Science and Technology POLITEHNICA Bucharest, 2023–2026. Thesis: Revising and Open-Sourcing ADiJaC for Parameter Estimation in Copolymerization Modeling, advised by Emil-Ioan Slușanschi.
- Copolymers can be modeled as walks through Markov chains, whose free parameters are found by fitting the model to a composition matrix derived from mass spectrometry. I used ADiJaC to compute the gradient of that fit’s cost function, so gradient descent can minimize the fit cost. A fit that took earlier solvers hours now takes under five minutes.
- To be published.
- Along the way I open-sourced ADiJaC, enhancing it with a build, a release pipeline, a Gradle plugin and documentation; see Projects.
- Copolymers can be modeled as walks through Markov chains, whose free parameters are found by fitting the model to a composition matrix derived from mass spectrometry. I used ADiJaC to compute the gradient of that fit’s cost function, so gradient descent can minimize the fit cost. A fit that took earlier solvers hours now takes under five minutes.
- Bachelor of Engineering, Computer Engineering, University POLITEHNICA of Bucharest, 2019–2023. Thesis: Radio Noise Processing for True Random Number Generation.
- High school, Tudor Vianu National College of Computer Science. Intensive Informatics class.
Teaching
- Teaching assistant, University POLITEHNICA of Bucharest, 2021–2024. Algorithm Analysis, Formal Languages and Automata, Analog Electronics, Digital Electronics, Programming Paradigms, and Computer Systems Architecture. I led labs and seminars, wrote course materials, shot video demos, and proposed and graded exams.
- Committee member, AcadNet Applied Informatics Olympiad, 2021–2024. I wrote contest problems for high school students, from local rounds to the national final.
- National Psycho-Pedagogical Training Program, the Romanian teacher training certification: level 1 in 2023, level 2 in 2026.
CV
You may find more information about my past work on LinkedIn.
You may reach out to me for a résumé.