An Improved CLDNN Framework with Spatial-Positional Attention for Robust Modulation Format Recognition

• Long Chen,
• Jiajun Liu,
• Xinyu Hu,

Vol. 19, No. 12, December 31, 2025
10.3837/tiis.2025.12.020, Download Paper (Free):

• cognitive radio
• automatic modulation format recognition
• convolutional long short-term memory neural network
• coordinate attention mechanism

Abstract

Modulation format recognition of radio signals is foundational to signal interception, monitoring, and detection. Traditional modulation format recognition algorithms mainly include computing likelihood functions and using feature extractors. However, computing likelihood functions has high complexity, and methods based on feature extraction require preprocessing of received signals. Owing to its strong ability to extract data features, deep learning has been applied to modulation format recognition. The convolutional long short-term memory neural network (CLDNN), by combining the spatial feature-extraction capability of convolutional neural networks (CNNs) and the temporal-sequence modeling capability of long short-term memory (LSTM), can efficiently process data with both spatial and temporal attributes and has been widely used in modulation-format recognition and related fields. However, for modulation format recognition, its key limitation is the insufficient extraction of spatial and positional features from modulated signals, which hinders recognition accuracy. Nevertheless, in modulation format recognition its extraction of the spatial and positional information of modulation signals is insufficient, which limits further improvement in recognition accuracy. To address the shortcomings of conventional modulation recognition methods and the limitations of CLDNN-based approaches, we propose an enhanced CLDNN architecture with Coordinated Attention Mechanism (CA-CLNN). The proposed method does not require preprocessing of the received signal, thereby reducing computational overhead and improving the speed of modulation format recognition. By improving the convolutional neural network structure, it enhances the capability for the received signal and enables efficient recognition in complex communication environments where multiple modulation formats coexist. Experimental results verify that under high Signal-to-Noise Ratio (SNR) conditions, the algorithm achieves a recognition rate exceeding 93% for 24 modulation formats, demonstrating its effectiveness. The source code is publicly available at https://github.com/liufanliu/CA-CLNN.

Statistics

Cite this article