Interference-mitigated multitask learning CNN based on statistical prior weights and feature-wise linear modulation

Abstract

Global Navigation Satellite System (GNSS) signals are susceptible to intentional interference. Existing interference cognition methods typically address only a single task—either classification or parameter estimation—and are prone to negative transfer in multitask learning. To tackle these issues, this paper proposes a dual-branch multitask learning convolutional neural network (CNN) based on statistical prior weights and feature-wise linear modulation. Built upon a multitask learning backbone, the proposed method initializes the weights of the two tasks using statistical prior weights, while simultaneously introducing a Feature-Wise Linear Modulation (FiLM) module at the bifurcation of shared features to tailor feature representations for distinct tasks. Simulation results demonstrate that, under low INR conditions, the proposed method still maintains high interference classification accuracy while keeping the parameter estimation error at a low level, exhibiting favorable stability and practicality.