基于模型剪枝动态调整压缩率的CSI反馈方法

doi:10.12305/j.issn.1001-506X.2023.08.37

摘要/Abstract

摘要：

针对现有基于深度学习(deep learning, DL)的信道状态信息(channel state information, CSI)反馈方案只能以固定压缩率(compression rate, CR)进行压缩反馈的问题，提出了一种基于模型结构化剪枝动态调整CR的CSI反馈方案。首先设计了残差信道特性注意(residual channel characteristic attention, RCCA)机制，并搭建了CSI反馈网络RCCA-Net，充分利用复信道矩阵实部虚部间的幅度相位依赖关系，进一步提高CSI反馈-重建质量。其次设计了RCCA-Prune方案，并以$\ell^2 $范数评估各神经元对最终结果的贡献度，通过模型结构化剪枝技术删除贡献度较小的神经元以实现CR的动态调整。仿真结果表明，所提的动态调整CR方案在不同CR下均有较佳的反馈性能，且在不同的测试环境中具有良好的泛化性。

关键词: 信道状态信息, 深度学习, 模型剪枝, 通道注意力

Abstract:

Aiming at the problem that the existing channel state information (CSI) feedback scheme based on deep learning (DL) can only perform compression feedback at a fixed compression rate (CR), a CSI feedback scheme for dynamically adjusting CR based on model pruning is proposed. Firstly, a residual channel characteristic attention (RCCA) mechanism is designed, and a CSI feedback network RCCA-Net is built, which makes full use of the amplitude-phase dependence relationship between the real parts and imaginary parts of the complex channel matrix to further improve CSI feedback-reconstruction quality. Secondly, the model RCCA-Prune is designed to dynamically adjust the CR, which uses the $\ell^2 $ norm to evaluate the contribution of each neuron to the final result, and deletes the neurons with less contribution through the model structured pruning technology to realize the dynamic adjustment of the compression rate. The simulation results show that the proposed dynamic adjustment CR scheme has better feedback performance under different CR, and has good generalization in different test environments.

Key words: channel state information, deep learning, model pruning, channel attention

中图分类号:

TN911.72

邵凯, 杜自群, 王光宇. 基于模型剪枝动态调整压缩率的CSI反馈方法[J]. 系统工程与电子技术, 2023, 45(8): 2615-2622.

Kai SHAO, Ziqun DU, Guangyu WANG. CSI feedback method for dynamically adjusting compression rate based on model pruning[J]. Systems Engineering and Electronics, 2023, 45(8): 2615-2622.

图/表 12

图1

图2

图3

图4

表1

图5

图6

表2

图7

表3

图8

图9

参考文献 27

1	LIU Z Y , ZHANG L , DING Z . Exploiting bi-directional channel reciprocity in deep learning for low rate massive MIMO CSI feedback[J]. IEEE Wireless Communications Letters, 2019, 8 (3): 889- 892. doi: 10.1109/LWC.2019.2898662
2	SHEN W Q , DAI L L , LI Y , et al. Channel feedback codebook design for millimeter-wave massive MIMO systems relying on lens antenna array[J]. IEEE Wireless Communications Letters, 2018, 7 (5): 736- 739. doi: 10.1109/LWC.2018.2818130
3	LIN Y P, WU P H, PHOONG S M. Statistics-aided codebook designs for limited feedback beamforming systems over millimeterwave channels[C]//Proc. of the IEEE 23rd International Conference on Digital Signal Processing, 2018.
4	SHEN W Q , DAI L L , ZHANG Y , et al. On the performance of channel-statistics-based codebook for massive MIMO channel feedback[J]. IEEE Trans.on Vehicular Technology, 2017, 66 (8): 7553- 7557. doi: 10.1109/TVT.2017.2656908
5	ADHIKARY A , NAM J , AHN J Y , et al. Joint spatial division and multiplexing—the large-scale array regime[J]. IEEE Trans.on Information Theory, 2013, 59 (10): 6441- 6463. doi: 10.1109/TIT.2013.2269476
6	JIANG Z , MOLISCH A F , CAIRE G , et al. Achievable rates of FDD massive MIMO systems with spatial channel correlation[J]. IEEE Trans.on Wireless Communications, 2015, 14 (5): 2868- 2882. doi: 10.1109/TWC.2015.2396058
7	ADHIKARY A , Al SAFADI E , SAMIMI M K , et al. Joint spatial division and multiplexing for mm-wave channels[J]. IEEE Journal on Selected Areas in Communications, 2014, 32 (6): 1239- 1255. doi: 10.1109/JSAC.2014.2328173
8	JOUNG J , KURNIAWAN E , SUN S . Channel correlation mo-deling and its application to massive MIMO channel feedback reduction[J]. IEEE Trans.on Vehicular Technology, 2016, 66 (5): 3787- 3797.
9	CUI Y M, FANG Y L. Research on PCA data dimension reduction algorithm based on entropy weight method[C]//Proc. of the 2nd International Conference on Machine Learning, Big Data and Business Intelligence, 2020: 392-396.
10	LIU J E, WANG K, LIU H B. Distributed compressed sensing of doubly selective channel in massive MIMO systems[C]//Proc. of the World Conference on Computing and Communication Technologies, 2021: 21-25.
11	GE Y, ZENG Z, ZHANG T, et al. Spatio-temporalcorrelated channel feedback for massive MIMO systems[C]//Proc. of the IEEE/CIC International Conference on Communications in China, 2018.
12	WEN C K , SHIH W T , JIN S . Deep learning for massive MIMO CSI feedback[J]. IEEE Wireless Communications Letters, 2018, 7 (5): 748- 751. doi: 10.1109/LWC.2018.2818160
13	LI X Y , WU H M . Spatio-temporal representation with deep neural recurrent network in MIMO CSI feedback[J]. IEEE Wireless Communications Letters, 2020, 9 (5): 653- 657. doi: 10.1109/LWC.2020.2964550
14	GUO J , WEN C K , JIN S , et al. Convolutional neural network-based multiple-rate compressive sensing for massive MIMO CSI feedback: design, simulation, and analysis[J]. IEEE Trans.on Wireless Communications, 2020, 19 (4): 2827- 2840. doi: 10.1109/TWC.2020.2968430
15	WANG Z , LI Y , WANG C , et al. A-OMP: an adaptive OMP algorithm for underwater acoustic OFDM channel estimation[J]. IEEE Wireless Communications Letters, 2021, 10 (8): 1761- 1765. doi: 10.1109/LWC.2021.3079225
16	ZHANG Z Y , CAI X , LI C G , et al. One-bit quantized massive MIMO detection based on variational approximate message passing[J]. IEEE Trans.on Signal Processing, 2017, 66 (9): 2358- 2373.
17	LIU T T , XING L , SUN Z G . Study on convergence of plug-and-play ISTA with adaptive-kernel denoisers[J]. IEEE Signal Processing Letters, 2021, 28, 1918- 1922. doi: 10.1109/LSP.2021.3111594
18	HU J, SHEN L, SUN G. Squeeze-and-Excitation networks[C]//Proc. of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 7132-7141.
19	LOHIT S , KULKARNI K , KERVICHE R , et al. Convolutional neural networks for noniterative reconstruction of compressively sensed images[J]. IEEE Trans.on Computational Imaging, 2018, 4 (3): 326- 340. doi: 10.1109/TCI.2018.2846413
20	MOUSAVI A, DASARATHY G, BARANIUK R G. Deepcodec: adaptive sensing and recovery via deep convolutional neural networks[C]//Proc. of the 55th Annual Allerton Conference on Communication, Control, and Computing, 2017: 744-744.
21	YAO H T , DAI F , ZHANG S L , et al. DR2-net: deep residual reconstruction network for image compressive sensing[J]. Neurocomputing, 2019, 359, 483- 493. doi: 10.1016/j.neucom.2019.05.006
22	赵小强, 宋昭漾. 多级跳线连接的深度残差网络超分辨率重建[J]. 电子与信息学报, 2019, 41 (10): 2501- 2508. doi: 10.11999/JEIT190036
	ZHAO X Q , SONG Z Y . Super-resolution reconstruction of deep residual network with multi-level skip connections[J]. Journal of Electronics & Information Technology, 2019, 41 (10): 2501- 2508. doi: 10.11999/JEIT190036
23	TANG Y H , WANG Y H , XU Y X , et al. Scop: scientific control for reliable neural network pruning[J]. Advances in Neural Information Processing Systems, 2020, 33, 10936- 10947.
24	HE Y, LIU P, WANG Z W, et al. Filter pruning via geometric median for deep convolutional neural networks acceleration[C]//Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 4340-4349.
25	VADERA S , AMEEN S . Methods for pruning deep neural networks[J]. IEEE Access, 2022, 10, 63280- 63300. doi: 10.1109/ACCESS.2022.3182659
26	MEIJERINK A , MOLISCH A F . On the physical interpretation of the Saleh-Valenzuela model and the definition of its power delay profiles[J]. IEEE Trans.on Antennas and Propagation, 2014, 62 (9): 4780- 4793. doi: 10.1109/TAP.2014.2335812
27	ALKHATEEB A. DeepMIMO: a generic deep learning dataset for millimeter wave and massive MIMO applications[EB/OL]. [2022-05-05]. https://deepmimo.net/.

系统参数	取值
基站天线数	32×32
用户天线数	1
系统带宽/GHz	0.5
OFDM子载波数	1 024
OFDM采样因子	1
多径数目	3

参数	CR
参数	1/4	1/8	1/16	1/32
模型参数量/个	2 103 086	1 054 254	529 838	267 630
模型大小/M	8.96	4.95	2.95	1.95
Saleh-Valenzuela/dB	-17.21	-16.46	-16.13	-15.97
DeepMIMO/dB	-12.92	-10.87	-9.96	-9.55

模型	参数	CR
模型	参数	1/4	1/8	1/16	1/32
RCCA-Prune	模型参数量/个	2 103 086	1 054 254	529 838	267 630
RCCA-Prune	模型大小/M	8.96	4.95	2.95	1.95
SM-CsiNet+	模型参数量/个	2 103 072	2 365 984	2 431 904	2 448 480
SM-CsiNet+	模型大小/M	8.97	9.96	10.22	10.28

[1]	汪萌, 诸兵. 不确定性建模在2D和3D目标检测中的应用[J]. 系统工程与电子技术, 2023, 45(8): 2370-2376.
[2]	崔天舒, 王栋, 黄振. 面向空间认知通信的轻量化网络自动调制分类方法[J]. 系统工程与电子技术, 2023, 45(7): 2220-2226.
[3]	姜雨, 袁琪, 胡志韬, 吴薇薇, 顾欣. 基于气象因素的机场进离港延误预测[J]. 系统工程与电子技术, 2023, 45(6): 1722-1731.
[4]	陈洋, 廖灿辉, 张锟, 刘建, 王鹏举. 基于自监督对比学习的信号调制识别算法[J]. 系统工程与电子技术, 2023, 45(4): 1200-1206.
[5]	张晔, 侯毅, 欧阳克威, 周石琳. 单变量序列数据分类方法综述[J]. 系统工程与电子技术, 2023, 45(2): 313-335.
[6]	邵正途, 许登荣, 徐文利, 王晗中. 基于LSTM和残差网络的雷达有源干扰识别[J]. 系统工程与电子技术, 2023, 45(2): 416-423.
[7]	李瑞泽, 张双辉, 刘永祥. 基于卷积ADMM网络的高效结构化稀疏ISAR成像方法[J]. 系统工程与电子技术, 2023, 45(1): 56-70.
[8]	韩啸, 陈世文, 陈蒙, 杨锦程. 基于互易点学习的LPI信号开集识别[J]. 系统工程与电子技术, 2022, 44(9): 2752-2759.
[9]	张立民, 谭凯文, 闫文君, 张聿远. 基于多级跳线残差网络的雷达辐射源识别[J]. 系统工程与电子技术, 2022, 44(7): 2148-2156.
[10]	金国栋, 薛远亮, 谭力宁, 许剑锟. 基于孪生神经网络的目标跟踪算法进展研究[J]. 系统工程与电子技术, 2022, 44(6): 1805-1822.
[11]	赵晓枫, 徐叶斌, 吴飞, 牛家辉, 蔡伟, 张志利. 基于全局感知机制的地面红外目标检测方法[J]. 系统工程与电子技术, 2022, 44(5): 1461-1467.
[12]	禹永植, 张春红, 郝海. 非完美CSI情况下大规模MIMO系统的下行链路能效优化[J]. 系统工程与电子技术, 2022, 44(5): 1694-1700.
[13]	邹虹, 白陈阳, 何鹏, 崔亚平, 王汝言, 吴大鹏. 基于分布式深度学习的边缘服务放置策略[J]. 系统工程与电子技术, 2022, 44(5): 1728-1737.
[14]	陈冬, 句彦伟. 基于语义分割实现的SAR图像舰船目标检测[J]. 系统工程与电子技术, 2022, 44(4): 1195-1201.
[15]	孙晶明, 虞盛康, 孙俊. 基于深度学习的HRRP识别姿态敏感性分析[J]. 系统工程与电子技术, 2022, 44(3): 802-807.