Effect of free-space attenuation on detection performance

Determine and demonstrate the effect of free-space propagation attenuation on the detection performance of the presented low-cost UAV radio-frequency detection and classification system, which uses VGG11_BN–VGG19_BN convolutional neural networks trained on 2D spectrograms derived from 74.9 ms IQ segments at 14 Msps in the 2.4 GHz ISM band, by systematically varying transmitter–receiver distance under controlled line-of-sight field conditions.

Background

The field test at Zurich Lake evaluated the mobile detection system using a directional circularly polarised antenna and a USRP B210 at distances of approximately 110 m, 340 m, 560 m, and 670 m, with antenna angles of 0°, 90°, and 180°. The models (VGG11_BN–VGG19_BN) achieved mean balanced accuracy around 0.80, with most misclassifications occurring between noise and drone classes.

Although Table \ref{tab:field_test_acc_distance} reports a slight accuracy decrease with distance, the authors note that the longest available line-of-sight distance (670 m) was insufficient to clearly reveal the impact of free-space path loss on system performance. A controlled paper across larger distances and conditions is needed to establish the relationship between path loss and detection accuracy.