[Paper Review] A Radio Signal Modulation Recognition Algorithm Based on Residual Networks and Attention Mechanisms

Ruisen Luo, Tao Hu+4 more|arXiv (Cornell University)|Sep 27, 2019

Wireless Signal Modulation Classification10 references4 citations

TL;DR

This paper proposes a deep learning-based radio signal modulation recognition algorithm that integrates residual networks (ResNet) and self-attention mechanisms to improve accuracy and training stability. By leveraging IQ data frames of 16QAM, 2FSK, and 4FSK signals under Gaussian white noise, the model achieves over 93% average recognition accuracy, demonstrating high robustness and practical value in real-time applications.

ABSTRACT

To solve the problem of inaccurate recognition of types of communication signal modulation, a RNN neural network recognition algorithm combining residual block network with attention mechanism is proposed. In this method, 10 kinds of communication signals with Gaussian white noise are generated from standard data sets, such as MASK, MPSK, MFSK, OFDM, 16QAM, AM and FM. Based on the original RNN neural network, residual block network is added to solve the problem of gradient disappearance caused by deep network layers. Attention mechanism is added to the network to accelerate the gradient descent. In the experiment, 16QAM, 2FSK and 4FSK are used as actual samples, IQ data frames of signals are used as input, and the RNN neural network combined with residual block network and attention mechanism is trained. The final recognition results show that the average recognition rate of real-time signals is over 93%. The network has high robustness and good use value.

Motivation & Objective

• To address the challenge of inaccurate modulation type recognition in complex radio signal environments.
• To improve deep neural network performance by mitigating gradient vanishing in very deep architectures.
• To enhance feature learning and training convergence through the integration of attention mechanisms.
• To develop a robust, real-time modulation recognition system suitable for practical wireless communication applications.

Proposed method

• The proposed method employs a residual network (ResNet) architecture to stabilize training in deep networks and alleviate gradient disappearance.
• An attention mechanism is incorporated to dynamically emphasize important signal features in the IQ data, accelerating gradient descent and improving feature representation.
• The model is trained on 10 standard modulation types—MASK, MPSK, MFSK, OFDM, 16QAM, AM, FM—under additive white Gaussian noise (AWGN).
• IQ data frames of actual signals (16QAM, 2FSK, 4FSK) are used as input for end-to-end training.
• The network combines recurrent neural network (RNN) components with residual blocks and attention modules to model temporal and spectral signal characteristics.
• The architecture is optimized through backpropagation with adaptive learning, enabling effective feature learning in noisy conditions.

Experimental results

Research questions

• RQ1 Can a residual network with attention mechanisms improve modulation recognition accuracy in noisy radio signal environments?
• RQ2 How does the integration of attention mechanisms affect the convergence speed and stability of deep neural networks in modulation recognition?
• RQ3 To what extent does the proposed model generalize across different modulation types under Gaussian white noise?
• RQ4 Does the combination of residual blocks and attention mechanisms outperform standard RNN or CNN-based approaches in real-time signal recognition?

Key findings

• The proposed model achieves an average recognition accuracy of over 93% on real-time signals, demonstrating high performance under noisy conditions.
• The integration of residual blocks effectively mitigates gradient vanishing in deep network architectures, enabling stable training.
• The attention mechanism significantly accelerates gradient descent and enhances feature learning by focusing on discriminative signal components.
• The model exhibits strong robustness across multiple modulation types, including 16QAM, 2FSK, and 4FSK, under additive white Gaussian noise.
• The method shows practical utility due to its high accuracy and real-time classification capability, suitable for deployment in dynamic radio environments.