| 摘要: |
| 在遥测信道环境中，经过信道估计与信道均衡处理后，理想的高斯信道下检测性能最优的最大似然序列检测算法面临误码率性能下降严重的问题。结合遥测调制信号符号间的记忆性特点，提出了基于数据预处理的双向长短期记忆（Bidirectional Long Short-term Memory，BiLSTM）网络的信号检测方法。依据符号间记忆长度，对均衡后的序列进行符号级重编码，以及特征级窗口化预处理，将一维序列的局部记忆特征转化为结构化输入。消融实验结果表明，该预处理可降低各信噪比下的误码率，从而有效提升后续网络的检测性能。在误码率为1×10-4时，相较于传统的最大似然序列检测方法，该方法能够提供约4 dB的误码率增益。 |
| 关键词: 〖JP2〗遥测信道 连续相位调制信号 最大似然序列检测 深度学习 双向长短期记忆（BiLSTM）网络〖JP〗 |
| DOI：10.20079/j.issn.1001-893x.250209001 |
|
| 基金项目: |
|
| Telemetry Channel CPM Signal Detection Based on BiLSTM with Data Preprocessing |
| WANG Le,HUANG Qiuxia,WU Kaifeng |
| (School of Artificial Intelligence and Computer Science，North China University of Technology，Beijing 100144，China) |
| Abstract: |
| In the telemetry channel，after channel estimation and equalization，the maximum likelihood sequence detection algorithm，which has the optimal detection performance under Gaussian channel，faces a serious challenge of significant decline in bit error rate(BER).Considering the memory characteristics between symbols of the telemetry modulation signal，the authors propose a detection method based on the bidirectional long short-term memory(BiLSTM) with data preprocessing.According to the inter-symbol memory length，the equalized sequence is transformed to structured inputs by symbol recoding and feature windowing preprocessing.The results of the ablation experiment demonstrate that the method reduces the BER under various signal-to-noise ratios(SNRs)，and effectively enhances the detection performance of subsequent networks.At a BER of 1×10-4，the proposed method achieves about 4 dB gain compared with the traditional maximum likelihood sequence detection method. |
| Key words: telemetry channel CPM signal maximum likelihood sequence detection deep learning bidirectional long short-term memory(BiLSTM) network |
《电讯技术》编辑部 