Speaker
Description
Wideband spectrum sensing and cost-effective RF measurement are increasingly important as wireless systems require larger volumes of clean, reliable RF data. Machine-learning methods for signal detection and classification are especially dependent on training datasets collected from realistic measurements. Commercial software-defined radios (SDRs) offer a low-cost platform for scalable data collection and controlled dataset generation, but their measurements are affected by hardware-dependent impairments such as internal leakage, IQ imbalance, frequency offsets, tuning-dependent gain variation, and band-edge attenuation. If left uncharacterized, these effects can be mistaken for real spectral activity or can distort features used by downstream sensing algorithms.
This paper presents a calibration-aware SDR sensing pipeline for accurate RF measurement and wideband spectrum sensing that optimizes operating conditions to reduce measurement distortion and normalizes receiver-dependent artifacts by applying hardware corrections before signal detection and comparison. The pipeline is evaluated using controlled over-the-air transmissions, AERPAW measurements, and laboratory spectrum analyzer references. Commercial SDRs across multiple cost ranges are compared to assess the relationship between affordability, measurement fidelity, and sensing performance.
Analysis shows that the sensing pipeline detects over-the-air signals while suppressing hardware-induced artifacts and improving agreement with reference measurements in observed frequency location and received power trends. The final evaluation quantifies frequency error, power agreement, artifact suppression, and detection and classification reliability across the tested devices, supporting the use of calibrated low-cost SDRs for trustworthy spectrum sensing experiments and RF machine-learning dataset generation.
| Talk Length | 15 Minutes |
|---|---|
| Link to Open Source Code | github.com/NCStateIMPRESSLab/sdr-ss |
| Acknowledge | Acknowledge In-Person |