Knowledge-Distilled Neural Receiver for Jamming-Resistant OFDM-Based Low-Altitude Networks

Abstract

Reliable reception in orthogonal frequencydivision multiplexing (OFDM)-based low-altitude communication networks remains challenging, as unmanned aerial vehicles may be exposed to rapidly time-varying and severe jamming. Although deep learning-based receivers such as DeepRx demonstrate strong robustness, their high computational and memory requirements hinder deployment on resource-constrained airborne platforms. To address this issue, this paper proposes a lightweight jamming-resistant deep learning receiver for OFDM systems based on a teacher-student knowledge distillation (KD) framework, in which a compact student network learns to replicate the detection behavior of a high-capacity DeepRx teacher using bit-level distillation. Simulation results under multiple representative jamming scenarios show that the proposed receiver achieves bit-error-rate (BER) performance close to the teacher across a wide range of jamming-to-noise ratios (JNR), while consistently outperforming conventional linear minimum mean-square error (LMMSE) detection and a simple convolutional neural network (CNN) baseline. Meanwhile, more than $80 \%$ parameter reduction and over sixfold inference-speed improvement are achieved without noticeable performance degradation, enabling practical and reliable UAV communications in complex low-altitude electromagnetic environments.