任思睿, 黄铭. 基于改进的长短期记忆网络的调制识别算法[J]. 云南大学学报(自然科学版), 2021, 43(1): 39-45. doi: 10.7540/j.ynu.20200075
引用本文: 任思睿, 黄铭. 基于改进的长短期记忆网络的调制识别算法[J]. 云南大学学报(自然科学版), 2021, 43(1): 39-45. doi: 10.7540/j.ynu.20200075
REN Si-rui, HUANG Ming. A modulation classification algorithm based on modified LSTM network[J]. Journal of Yunnan University: Natural Sciences Edition, 2021, 43(1): 39-45. DOI: 10.7540/j.ynu.20200075
Citation: REN Si-rui, HUANG Ming. A modulation classification algorithm based on modified LSTM network[J]. Journal of Yunnan University: Natural Sciences Edition, 2021, 43(1): 39-45. DOI: 10.7540/j.ynu.20200075

基于改进的长短期记忆网络的调制识别算法

A modulation classification algorithm based on modified LSTM network

  • 摘要: 自动调制分类技术是无线通信技术中的一个重要研究领域,卷积神经网络以及长短期记忆网络(Long Short-Term Memory,LSTM)两种深度学习模型在基于特征的自动调制分类技术中得到了广泛的应用. 然而在实际应用中这两种模型都存在着一些问题,卷积神经网络模型在处理长时间依赖序列的分类任务时的准确率不佳,LSTM模型的时间性能会随着输入数据规模的增大显著下降. 针对以上问题,提出一种基于带有注意力机制的LSTM网络的调制识别算法. 首先,读取一定采样长度的信号原始数据,并通过长短期记忆网络提取信号特征;然后,利用注意力机制为学习到的特征分配权重以减少数据冗余;最后,分类器根据学习到的特征输出分类结果. 仿真实验结果表明,新算法能以较低的时间代价取得较高的准确率.

     

    Abstract: Automatic modulation classification technology is an important research field in wireless communication technology.Two deep learning models, convolutional neural network and long short-term memory network, have been widely used in feature-based automatic modulation classification technology. However, these two models have some problems in practical applications. The accuracy of the Convolution Neural Network model when it is classifying long-term dependence sequences is not good, and the time performance of the Long Short-Term Memory (LSTM) model will decrease significantly as the input data size increases. In response to these problems, this paper proposes a modulation recognition algorithm based on an LSTM model with attention mechanism. This algorithm first reads the original signal data sampled by a certain length, and extracts the signal features through the LSTM network.Then it uses the attention mechanism to assigns weights to the learned features to reduce data redundancy. Finally the classifier outputs the classification results according to the features learned by the LSTM network.The simulation experiment proves that the algorithm can achieve a high accuracy with a low time cost.

     

/

返回文章
返回