Lightweight Automatic Modulation Classification Method Using Filter Pruning via Geometric Median

Automatic Modulation Classification (AMC) is a technique employed to determine the modulation type of signals in non-cooperative communication systems. It is crucial in military and civilian communications, offering significant potential for various applications. Recently, AMC methods based on deep learning (DL) are known for their outstanding performance. However, the high computational cost and large model size of DL-based AMC methods limit their application in edge Internet of Things (IoT) systems. Hence, this paper proposes a lightweight AMC method based on filter pruning via geometric median (FPGM). Specifically, we start by determining the geometric median of the filters in each layer, which effectively represents the filter distribution within that layer. Then, filters near the geometric median are identified and filtered out through the characteristics of the geometric median. Finally, the performance degradation of the model caused by the removal of filters can be restored through fine-tuning. Our experimental findings suggest that the proposed FPGM-based AMC method reduces the model size by 99%. At the same time, the classification accuracy is degraded by approximately 1.61%, which significantly outperforms other lightweight AMC methods.