基于迁移元学习的调制识别算法

doi:10.12382/bgxb.2022.0583

摘要/Abstract

摘要：

针对基于深度学习的调制识别算法在仅有几个带标签信号样本时无法训练的问题,通过模型无关元学习算法提高网络的泛化性能,以使网络对仅有几个训练样本的待测信号实现准确识别。同时对深度神经网络进行预训练以降低元学习阶段网络的训练难度,并根据迁移学习思想,通过引入可学习的缩放偏移参数来迁移预训练所得网络参数,减少学习新类信号所需训练的网络参数量,当面对新类信号的识别任务时通过少量信号样本微调网络就能实现准确识别。实验结果表明,算法在新类信号训练样本仅有5个时最高可实现93.5%的识别准确率。

关键词: 调制识别, 模型无关元学习, 迁移学习, 深度神经网络

Abstract:

For the problem that the modulation recognition algorithm based on deep learning can not be trained when there are only a few labeled signal samples, the model-agnostic meta-learning algorithm is used to improve the generalization performance of the network so that the network can accurately recognize the signals to be recognized with only a few training samples. At the same time, the deep neural network is pre-trained to reduce the training difficulty of the network at the meta learning stage. According to the idea of transfer learning, the amount of network parameters required for learning new class signals is reduced by introducing the learnable scaling offset parameters to migrate the network parameters obtained from the pre training. When facing the recognition task of new class signals, the accurate recognition can be achieved by finely tuning the network through a small number of signal samples. The experimental results show that the algorithm can achieve a recognition accuracy of 93.5% when there are only 5 training samples.

Key words: modulation recognition, model-agnostic meta-learning, transfer learning, deep neural network

中图分类号:

TN975

庞伊琼, 许华, 张悦, 朱华丽, 彭翔. 基于迁移元学习的调制识别算法[J]. 兵工学报, 2023, 44(10): 2954-2963.

PANG Yiqiong, XU Hua, ZHANG Yue, ZHU Huali, PENG Xiang. Modulation Recognition Algorithm Based on Transfer Meta-Learning[J]. Acta Armamentarii, 2023, 44(10): 2954-2963.

图/表 12

参考文献 21

[1]	赵雄文, 郭春霞, 李景春. 基于高阶累积量和循环谱的信号调制方式混合识别算法[J]. 电子与信息学报, 2016, 38(3): 674-680.
	ZHAO X W, GUO C X, LI J C. Mixed recognition algorithm for signal modulation schemes by high-order cumulants and cyclic spectrum[J]. Journal of Electronics & Information Technology, 2016, 38(3): 674-680. (in Chinese)
[2]	李东瑾, 杨瑞娟, 董睿杰. 基于联合投影字典学习的辐射源调制识别[J]. 兵工学报, 2020, 41(7): 1464-1472. doi: 10.3969/j.issn.1000-1093.2020.07.026
	LI D J, YANG R J, DONG R J. Emitter signal modulation recognition based on joint projection dictionary learning[J]. Acta Armamentarii, 2020, 41(7): 1464-1472. (in Chinese) doi: 10.3969/j.issn.1000-1093.2020.07.026
[3]	江伟华, 童峰, 王彬, 等. 采用主分量分析的非合作水声通信信号调制识别[J]. 兵工学报, 2016, 37(9): 1670-1676. doi: 10.3969/j.issn.1000-1093.2016.09.017
	JIANG W H, TONG F, WANG B, et al. Modulation recognition method of non-cooperation underwater acoustic communication signals using principal component analysis[J]. Acta Armamentarii, 2016, 37(9): 1670-1676. (in Chinese)
[4]	O’SHEA T J, CORGAN J, CLANCY T C. Convolutional radio modulation recognition networks[C]// Proceedings of International Conference on Engineering Applications of Neural Networks. Cham, Germany:Springer, 2016.
[5]	O’SHEA T J, ROY T, CLANCY T C. Over-the-air deep learning based radio signal classification[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12(1): 168-179. doi: 10.1109/JSTSP.2018.2797022 URL
[6]	KUMAR Y, SHEORAN M, JAJOO G, et al. Automatic modulation classification based on constellation density using deep learning[J]. IEEE Communications Letters, 2020, 24(6): 1275-1278. doi: 10.1109/COML.4234 URL
[7]	LIU K, GAO W J, HUANG Q H. Automatic modulation recognition based on a DCN-BiLSTM network[J]. Sensors, 2021, 21(5): 1577. doi: 10.3390/s21051577 URL
[8]	王洋, 冯永新, 宋碧雪, 等. DP-DRCnet卷积神经网络信号调制识别算法[J]. 兵工学报, 2023, 44(2): 545-555. doi: 10.12382/bgxb.2021.0620
	WANG Y, FENG Y X, SONG B X, et al. A modulation recognition algorithm of DP-DRCnet convolutional neural network[J]. Acta Armamentarii, 2023, 44(2): 545-555. (in Chinese) doi: 10.12382/bgxb.2021.0620
[9]	ZHUANG F Z, QI Z Y, DUAN K Y, et al. A comprehensive survey on transfer learning[J]. Proceedings of the IEEE, 2020, 109(1): 43-76. doi: 10.1109/PROC.5 URL
[10]	YU X, LI D Z, WANG Z Y, et al. A deep learning method based on convolutional neural network for automatic modulation classification of wireless signals[J]. Wireless Networks, 2019, 25(7): 3735-3746. doi: 10.1007/s11276-018-1667-6
[11]	BU K, HE Y, JING X J, et al. Adversarial transfer learning for deep learning based automatic modulation classification[J]. IEEE Signal Processing Letters, 2020, 27: 880-884. doi: 10.1109/LSP.97 URL
[12]	LIANG Z, TAO M L, WANG L, et al. Automatic modulation recognition based on adaptive attention mechanism and ResNeXt WSL model[J]. IEEE Communications Letters, 2021, 25(9): 2953-2957. doi: 10.1109/LCOMM.2021.3093485 URL
[13]	HUISMAN M, VAN RIJN J N, PLAAT A. A survey of deep meta-learning[J]. Artificial Intelligence Review, 2021, 54(6): 1-59. doi: 10.1007/s10462-020-09852-3
[14]	李凡长, 刘洋, 吴鹏翔, 等. 元学习研究综述[J]. 计算机学报, 2021, 44(2): 422-446.
	LI F C, LIU Y, WU P X, et al. A survey on recent advances in meta-learning[J]. Chinese Journal of Computers, 2021, 44(2): 422-446. (in Chinese)
[15]	FINN C, ABBEEL P, LEVINE S. Model-agnostic meta-learning for fast adaptation of deep networks[C]// Proceedings of International Conference on Machine Learning. Ithaca, NY, US: Cornell University, 2017: 1126-1135.
[16]	SNELL J, SWERSKY K, ZEMEL R. Prototypical networks for few-shot learning[C]// Proceedings of the 31st Annual Conference on Neural Information Processing Systems. Long Beach, CA, US: Cornell University, 2017, 30: 4077-4087.
[17]	SUNG F, YANG Y X, ZHANG L, et al. Learning to compare:relation network for few-shot learning[C]// Proceedings of the 31st IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, US:IEEE, 2018: 1199-1208.
[18]	VINYALS O, BLUNDELL C, LILLIARAP T, et al. Matching networks for one shot learning[C]// Proceedings of Advances in Neural Information Processing Systems 29. Barcelona, Spain:Neural Information Processing Systems, 2016, 29: 3630-3638.
[19]	KOCH G R, ZEMEL R, SALAKHUTDINOV R. Siamese neural networks for one-shot image recognition[C]// Proceedings of the ICML Deep Learning Workshop. Lille, France, Computer Science, 2015.
[20]	CHEN Y T, HOFFMAN M W, COLMENAREJO S G, et al. Learning to learn by gradient descent by gradient descent[C]// Proceedings of Advances in Neural Information Processing Systems 29.Sydney, Australia: Neural Information Processing Systems, 2016.
[21]	朱克凡, 王杰贵, 刘有军. 小样本条件下基于数据增强和WACGAN的雷达目标识别算法[J]. 电子学报, 2020, 48(6): 1124-1131. doi: 10.3969/j.issn.0372-2112.2020.06.012
	ZHU K F, WANG J G, LIU Y J. Radar target recognition algorithm based on data augmentation and WACGAN with a limited training Data[J]. Acta Electronica Sinica, 2020, 48(6): 1124-1131. (in Chinese)

训练集、验证集与测试剂	信号调制样式
训练集	OOK、8ASK、QPSK、16PSK、16APSK、64APSK、16QAM、64QAM、256QAM、AM-SSB-SC
验证集	128QAM、AM-SSB-WC、FM、OQPSK、GMSK、128APSK、BPSK
测试集	4ASK、8PSK、32PSK、32APSK、32QAM、AM-DSB-SC、AM-DSB-WC

训练集、验证集与测试剂	信号调制样式
训练集	OOK、8ASK、QPSK、16PSK、16APSK、64APSK、16QAM、64QAM、256QAM、AM-SSB-SC
验证集	128QAM、AM-SSB-WC、FM、OQPSK、GMSK、128APSK、BPSK
测试集	4ASK、8PSK、32PSK、32APSK、32QAM、AM-DSB-SC、AM-DSB-WC

方法	特征提取网络	5-way
方法	特征提取网络	1-shot	5-shot	20-shot
MAML	ConvNet	50.44±0.84	69.62±0.71	79.05±0.86
	ResNet	27.69±0.45	40.45±0.78	41.53±0.57
MAML+迁移预训练	ConvNet	47.32±0.42	68.21±0.65	77.92±0.51
	ResNet	82.44±0.74	91.26±0.5	92.74±0.47
MAML+迁移预训练+[ψ₁,ψ₂](本文算法)	ConvNet	48.56±0.34	70.38±0.24	79.82±0.28
	ResNet	85.1±0.72	93.5±0.49	95.88±0.36

方法	特征提取网络	5-way
方法	特征提取网络	1-shot	5-shot	20-shot
MAML	ConvNet	50.44±0.84	69.62±0.71	79.05±0.86
	ResNet	27.69±0.45	40.45±0.78	41.53±0.57
MAML+迁移预训练	ConvNet	47.32±0.42	68.21±0.65	77.92±0.51
	ResNet	82.44±0.74	91.26±0.5	92.74±0.47
MAML+迁移预训练+[ψ₁,ψ₂](本文算法)	ConvNet	48.56±0.34	70.38±0.24	79.82±0.28
	ResNet	85.1±0.72	93.5±0.49	95.88±0.36

元学习算法	特征提取网络	5-way 5-shot	5-way 1-shot
PN算法	ConvNet	69.16±0.34	62.93±0.56
	ResNet	84.89±0.28	74.34±0.38
RN算法	ConvNet	70.73±0.42	64.03±0.57
	ResNet	85.52±0.75	75.53±0.68
MN算法	ConvNet	63.34±0.23	58.34±0.26
	ResNet	75.02±0.42	69.93±0.45
SN算法	ConvNet	60.34±0.52	43.52±0.74
	ResNet	67.32±0.56	55.43±0.46
Meta-learner LSTM算法	ConvNet	62.34±0.62	54.37±0.53
	ResNet	72.36±0.43	63.33±0.52
本文算法	ConvNet	68.23±0.54	55.32±0.39
	ResNet	93.5±0.49	85.1±0.72