Real-Time Classification of Low Probability of Intercept Signals Using an End-to-End Neural Network (EELPINet)

This paper presents an end-to-end neural network for real-time classification of low probability of intercept (LPI) radar signals using raw complex-valued inputs. Conventional LPI signal classification methods typically rely on time–frequency preprocessing, such as short-time Fourier transform or Wigner–Ville distribution, which introduces additional computational overhead and may distort signal characteristics under low signal-to-noise ratio (SNR) conditions. To address these limitations, the proposed method directly processes raw in-phase and quadrature signal sequences without explicit time–frequency analysis. The proposed network consists of a lightweight raw signal feature extractor and a sequence analysis module based on a state-space model, enabling efficient modeling of long-range temporal dependencies. In addition, a Gaussian-based padding strategy is employed to handle variable-length input signals while preserving statistical continuity. Experimental evaluations were conducted using a large-scale simulated LPI radar dataset covering multiple modulation types and SNR levels ranging from –16 dB to 10 dB. The proposed method achieved an average classification accuracy of 90.99% with low computational complexity, requiring only 1.62 GFLOPs and 3.73 ms inference time per sample. These results demonstrate that the proposed approach is effective and suitable for real-time LPI radar signal classification in electronic warfare environments.