Deep belief networks based radar signal classification system

A threat library is used in most of the existing electronic warfare systems to identify or execute jamming against various radar signals. The conventional method uses frequency, pulse repetition interval, and pulse width sampled from the pulse description word column as characteristics of a signal. Such sampling technique cannot effectively model each radar signal when dealing with a complex signal array. In this paper, a new deep belief network model is proposed to generate a more efficient threat library for radar signal classification. The proposed model consists of independent restricted Boltzman machines (RBMs) of frequency, pulse repetition interval, pulse width respectively, and a RBM which fuses the result again. The performance of the existing system and the proposed system is evaluated by testing the signals with measurement errors and insufficient information. As a result, the proposed system shows more than 6% performance improvement over the existing system. © Springer-Verlag GmbH Germany, part of Springer Nature 2018.