Preventing Unauthorised Drone Communications Through Real-Time Spectrum Analysis
Аннотация
On the one hand, the impact of unregulated drone activities on restricted and/or sensitive zones has clearly become one of the significant security issues, especially in cities, airports, government buildings and facilities, as well as at sites hosting socially substantial events. Conventional detection procedures, such as radar-based procedures and optical tracking, tend to be limited by factors including line-of-sight constraints, high internal configuration costs, and vulnerability to environmental pollutants and threats, including environmental disturbances, which are often overlooked in the pursuit of real-time threat minimisation. To overcome this, we propose a real-time drone detection and mitigation algorithm that utilises RF sensors and machine learning based on software-defined radios (SDRs). The system constantly monitors the 2.4 GHz and 5.8 GHz ISM bands, receiving live data and displaying signals from drones as well as other wireless-transmitting devices. With the help of sophisticated signal preprocessing, time-domain and frequency-domain characteristics, such as spectral power, modulation patterns, and signal bursts, are obtained and normalised. This is fed into a convolutional neural network classifier to distinguish between illegal drone communications and legal RF activity. The control module, upon detecting illicit drone signals, initiates mitigation measures that may include alarming or RF band jamming, with minimal interference to the operations of nearby wireless systems. As demonstrated by numerous commercial drone prototype operations conducted under simulated city conditions, the detection rate is 96.4%, with an average latency of 180 ms and an accuracy exceeding 92% in environments with heavy RF congestion near Wi-Fi, Bluetooth, BLE, and Zigbee transmissions. The reliability of the system has been statistically validated through over 30 repeat trials, with a confidence range of -2.1 per cent at a detection rate. It has been found, based on its features in radius-frequency fingerprinting, that it outperforms radar and RF fingerprinting in terms of real-time responsiveness, strength, and economic conditions. The obtained findings verify the suggestion of SDR and a machine-learning-based framework as a viable, scalable, extensive, and feasible solution for real-time drone tracking and elimination, enhancing security in remote and heavily populated zones, particularly in terms of operational dependability and minimal external interference.