Deep Reinforcement Learning Based Cognitive Equalization Algorithm Research in Underwater Communication

In the coming years, the Underwater Internet of Things is expected to bridge different technologies for sensing the ocean, allowing it to become an intelligent network of interconnected underwater objects with self-learning and intelligent computing capabilities. The key technology of the underwater network is underwater acoustic communication. In order to ensure the performance of the physical layer, channel equalization is usually adopted, the cognitive equalization algorithm is proposed based on the deep Q-network (DQN) to improve the selection of equalizer structure parameters and recursive algorithm parameters. First, the multi-scale time-varying underwater acoustic (UWA) channel model generates a certain number of UWA channels as the training set, and the cognitive equalizer can adaptively select the optimal number of taps and step length according to the channel impulse response (CIR) and signal-to-noise ratio (SNR) conditions of different UWA channels. Simulation results show that compared with the classical adaptive equalization algorithm, the trained cognitive equalizer not only has better generalization performance, but also can significantly reduce the bit error rate (BER) and shorten the channel equalization time, improving the equalization performance.