A jamming cognition approach based on multi-task learning

Jamming signal cognition plays a crucial role in communication, control, and early warning within complex electromagnetic environments, providing key decision-making support for subsequent jamming suppression. Therefore, efficient and reliable jamming cognition is particularly critical. However, existing jamming cognition methods primarily follow a serial cognitive architecture where signal type is identified first, followed by parameter estimation, leading to suboptimal overall timeliness. To address this issue, a parallel jamming cognition method based on multi-task learning is proposed, which can simultaneously identify the jamming signal types and estimate the interference parameters. This algorithm is developed within a multitask framework characterized by hard parameter sharing. It incorporates a shared-layer network to extract the correlation information between jamming signals and their corresponding parameters. Furthermore, distinct independent task-layer networks are employed to capture the distinguishing features among various jamming signals. This approach facilitates the simultaneous identification of signal types and the estimation of their parameters. Moreover, to prevent the network from being dominated by a single task, which could hinder the effective optimization of difficult tasks, an improved multi-gradient descent algorithm is used for joint optimization of jamming recognition and parameter estimation tasks. Simulation results show that the proposed method significantly outperforms the LSTM and SKNet baseline algorithms in jamming recognition accuracy at low jamming-to-noise ratio. For the parameter estimation task, the algorithm is able to achieve a normalized root-mean-square error of 10^ - 2 for centre-frequency parameter estimation when the jamming-to-noise ratio is greater than 10 dB, which is better than that of traditional algorithms and single-task algorithms. Lastly, compared to the jamming cognition process that follows a serial architecture, the proposed method reduces the jamming cognition time by 40%, which effectively improves the timeliness of the jamming cognition process.