Electronics engineer. Signals at one end, machine learning at the other.
I work between hardware and software: a spectrum analyser on one side, a training loop on the other, and a fair amount of Python holding the middle together. The problems I like usually start in the physical layer and end as something someone can actually run.
I also have a habit of rebuilding the spreadsheet that everyone uses and nobody fully trusts. Whatever's pinned below is what that habit produced most recently.
Senior project. An SDR pipeline — USRP-2901, LabVIEW, PyTorch — that classifies 24 modulation types (ASK, PSK, FSK, QAM and OFDM) from real over-the-air I/Q, in real time.
The classifier is hierarchical rather than flat: single-carrier vs multi-carrier first, then the modulation within each branch. Both stages use a dual-stream 1D-ResNet reading time (I/Q) and frequency (FFT) together.
🏆 Third Place — Undergraduate Project Competition, SIU 2026 (34th IEEE Signal Processing and Communications Applications Conference)
📄 G. Deliktaş, E. Arslan, H. Polat, L. Özkan, S. Büyükçorak, "A Deep Learning Approach for SDR-Based Automatic Modulation Classification", in Proc. IEEE 34th SIU, 2026.
The paper covers where the project started — an earlier 12-class stage, before QAM, OFDM and the hierarchy. The final 24-class system that won the award isn't published as a repo: it's a team project and the dataset was collected on lab hardware, so it isn't mine alone to put online.
If you want the rest of it — the architecture, the results, why the OFDM branch was the hard part — email me.
Built during an RF test and field engineering internship, then cleaned up and published here.
| satcom-link-budget | LEO/GEO ground-station link budgets with an automated Word report generator. ITU-R propagation models, elevation sweeps, BER curves. Verified to ≤ 0.005 dB across 69 numerical checks. MATLAB Python |
| rf-level-diagram | Cascade power budget as a DAG, not a flat chain. Validated against a second solver written independently — 13/13 nodes identical to the bit — and against a manufacturer datasheet's G/T. 1058 tests. Python PySide6 |
| rf-block-diagram | Editor for RF block and cabling diagrams. Connectors drawn per IEEE Std 315, title-blocked vector PDF from A4 to A0. The layer that decides how things look imports no Qt, so 1011 tests run without a screen. Python PySide6 |
B.Sc. Electronics Engineering — Gebze Technical University, 2022–2026 Graduated with Honour Degree
Erasmus+ — Politechnika Poznańska, Poland — Computing and Telecommunications Machine learning, NLP and digital signal processing coursework.
Three internships along the way. One in RF test and field engineering at a SATCOM ground-station manufacturer — where the tools above came from. Two industrial ones before that: data transmission pipelines, communication-device integration, and PLC/automation basics. The unglamorous business of getting equipment to talk to each other, which turns out to be most of engineering.
- ML / Data — PyTorch · NumPy · Pandas · scikit-learn · CNN & ResNet architectures · dataset creation and labelling
- RF / Comms — SDR & USRP · digital modulation (ASK/PSK/FSK/QAM/OFDM) · I/Q acquisition and preprocessing · ITU-R P.618/P.676 · LEO & GEO link budgets
- Test & instrumentation — spectrum analyser · signal generator · LabVIEW · TX–RX test setup
- Languages & tooling — Python · MATLAB/Simulink · C · Git · LTspice
📧 guvendeliktas@gmail.com · 💼 linkedin.com/in/guven-deliktas · 📄 IEEE Xplore
Always happy to talk about RF, modulation, or why your link budget disagrees with your spectrum analyser. The modulation classification work isn't published as a repo, but the paper is linked above and my inbox is open for the rest of it.