Multitask Learning-Based Modulation and Signal Type Recognition for Space-Ground Integrated Networks

Abstract

A space-ground integrated network (SGIN) will be a future network for the heterogeneous convergence of space- and ground-based networks. The SGIN envisions satellites with the capability to adapt to various communication protocols, enabling the convergence of diverse network systems. A crucial aspect of satellite payload in the SGIN is the recognition of satellite signal types and their modulation modes, which substantially enhances the processing of heterogeneous wireless signals at the baseband processing. A multitask learning (MTL) model-based convolutional neural network (CNN) architecture is proposed, which addresses the recognition problem. An MTL model-based CNN architecture is proposed, which addresses the recognition problem. This model is composed of 3 key components: A multi-input part that processes in-phase/quadrature (IQ) complex signals and power spectral density data, a set of shared part that facilitates the model’s efficiency and mitigate overfitting, and a multitask output part capable of concurrently recognizing signal types and modulation modes. Furthermore, we have developed a dataset that encompasses 5 satellite signal protocols, i.e., signal type, and 6 modulation modes, derived from the digital video broadcasting (DVB) protocols and the tracking, telemetry, and command (TT&C) standards of the consultative committee for space data systems (CCSDS). This dataset also takes into account the impact of the additive white Gaussian noise (AWGN) channel and land mobile satellite (LMS) channel models. Simulation experiments validate the effectiveness of the proposed MTL model in accurately identifying various satellite signal protocols and modulation techniques.