Compressed wide spectrum sensing scheme based on BP network

This paper proposes a compressed sensing (CS) scheme to reconstruct and estimate the signals. In this scheme, the framework of CS is used to break the Nyquist sampling limit, making it possible to reconstruct and estimate signals via fewer measurements than that is required traditionally. However, the reconstruction algorithms based on CS are normally non-deterministic polynomial hard (NP-hard) in mathematics, which makes difficulties in obtaining real-time analysis-results. Therefore, a new compressed sensing scheme based on back propagation (BP) neural network is proposed under an assumption that every sub-band is the same. In this new scheme, BP neural network is added into detection process, replacing for signal reconstruction and decision-making. By doing this, heavy calculation cost in reconstruction is moved into pre-training period, which can be done before the real-time analysis, bringing about a sharp reduction in time consuming. For simplify, 1-bit quantification is taken on compressed signals. Simulations demonstrate the performance enhancement in the proposed scheme: compared with normal CS-based scheme, the proposed one presents a much shorter response time as well as a better robustness performance to noise via fewer measurements.