Automated Satellite Communication Testbed for Complex Interference Signal Separation Using Real-Time AI-Driven Data: Satellite communication interference mitigation testbed using AI driven data

Automated Satellite Communication Testbed for Complex Interference Signal Separation Using Real-Time AI-Driven Data

Petersen, Sigurd Sándor

4. semester

Electronic Systems, Master

2025

2025-06-04

83

This thesis builds a flexible, reprogrammable testbed that emulates a satellite communication link, including a ground station, a satellite transponder, and a controllable interference source. The platform automates measurement campaigns and collects high‑quality data to train AI‑based interference mitigation. The system is validated with common modulation schemes (BPSK, QPSK, and 8PSK—ways of encoding data in a radio signal), showing signal quality with an error vector magnitude (EVM) of 0.22% ± 0.005% and a modulation error rate up to 51.87%. These measurements indicate excellent signal integrity under both low and highly increased interference. The work also implements a U‑Net neural network to separate interference from raw, mixed I/Q signals in real time without preprocessing (I/Q are the in‑phase and quadrature components used to represent radio signals). The U‑Net is trained on extensive real‑world datasets with over 200,000 samples for each combination of gain and frequency settings. It produces denoised signals close to the reference, achieving correlation over 98% for QPSK, 96% for BPSK, and up to 98.45% for 8PSK—even at the highest interference levels. The results demonstrate robust interference suppression, high reconstruction fidelity, and strong generalization. Together, the validated automated testbed and the U‑Net architecture meet the goal of enabling real‑time, AI‑driven separation of complex interference in satellite communication links.

Denne kandidatafhandling udvikler en fleksibel, omprogrammerbar testplatform, der kan efterligne en satellitkommunikationsforbindelse med jordstation, satellittransponder og en kontrollerbar interferenskilde. Platformen muliggør automatiske målekampagner, der indsamler data af høj kvalitet til AI-baseret reduktion af interferens. Systemet er valideret med almindelige modulationsformer (BPSK, QPSK og 8PSK – måder at indkode data i et radiosignal), hvor signalets kvalitet er målt med en Error Vector Magnitude (EVM) på 0,22% ± 0,005% og en modulationsfejlraten op til 51,87%. Disse målinger bekræfter en fremragende signalintegritet under både lav og kraftigt øget interferens. Derudover implementeres en U‑Net-neuralt netværk til at adskille interferens fra rå, blandede I/Q‑signaler i realtid uden forbehandling (I/Q er de in‑fase og kvadraturkomponenter, der repræsenterer radiosignaler digitalt). U‑Net‑modellen er trænet på omfattende, virkelige datasæt med over 200.000 prøver for hver kombination af forstærknings- og frekvensindstillinger. Modellen producerer et støjreduceret signal tæt på reference, med korrelation over 98% for QPSK, 96% for BPSK og op til 98,45% for 8PSK – selv ved de højeste interferensniveauer. Resultaterne viser robust undertrykkelse af interferens, høj rekonstruktionsnøjagtighed og stærk generalisering. Den validerede, automatiserede testplatform og U‑Net-arkitekturen opfylder dermed målet om at muliggøre AI‑drevet, realtids separation af kompleks interferens i satellitkommunikationsforbindelser.

[This apstract has been rewritten with the help of AI based on the project's original abstract]

Satellite Communication ; Complex Interference ; Signal Separation ; Real-Time AI ; Real-Time AI-Driven Data ; Interference mitigation ; Testbed ; U-Net architecture ; Complex interference ; Satellite communication links ; Emulating satellite communication system ; Reprogrammable testbed ; Interference source ; AI-based interference mitigation

Download PDF
View record in AAU Student Projects

Petersen, Sigurd Sándor:

• Dielectric properties of building materials for propagation studies at mmWave frequency (2023)