Feature Bispectra Based Real Time FM Signal Recognition Using Adaptive Incremental Learning Feedforward Neural Network

On-line real time automatic communication signal¿ Frequency Modulated (FM) signal recognition has been a hot topic recently, which focuses on finding the characteristic feature in the noisy signal observations comprehensively identifying the the same or different version of transmitting devices with approximate performance parameters in modern electronic warfare. Direct use of Higher Order Statistics (HOS) becomes unavailable for this on-line application because of its huge computation time and memory space especially in the case of high frequency FM signal. This paper presents a novel view to improve the HOS analysis efficiency greatly by sub-sampling and its ability to detect the nonlinear parasitic amplitude modulation spontaneous frequency coupling while preserving the noise-contaminated feature and eliminating the random Gaussian noise. FM signal-related feature bispectra selection easily and significantly translates the 2-D feature matching pattern to a 1-D one applicable for an adaptive incremental learning Single-hidden Layer Feedforward Network (SLFN), which has been shown to be extremely fast with good generalization performance. Apart from the noise-to-signal level estimation as the expected learning accuracy, no other control parameters have to be manually chosen for this simple SLFN with adaptively incremental random hidden nodes one-by-one or chunk-by-chunk of fixed or varying chunk size. Simulation experiments show that this novel feature bispectra and adaptive incremental multi-class learning network outperform AIB, SB and BP, RBF in terms of computation time and recognition rate for on-line steady FM signal recognition.