相关噪声下基于深度学习的LDPC码码率半盲识别算法

doi:10.12305/j.issn.1001-506X.2025.04.30

摘要/Abstract

摘要：

为了正确识别相关噪声下采用低密度奇偶校验码和高阶调制的无线通信系统的信道编码参数, 在已知候选码率集合和相应奇偶校验矩阵的假定下, 提出两种基于深度学习的码率半盲识别算法。所提神经网络由降噪子网络和码率识别子网络构成, 降噪子网络设计实数降噪子网络和复数降噪子网络。相比于实数降噪子网络, 复数降噪子网络以高复杂度为代价，获得更好的处理复信号的能力。进一步, 为了降低复数降噪子网络的复杂度, 提出一种基于网络剪枝技术的网络压缩算法。仿真实验结果表明, 通过使用联合优化降噪损失函数和码率识别损失函数的多任务学习策略: 一方面, 在相关噪声下提出的神经网络比传统算法具有更好的识别性能；另一方面, 当利用网络压缩算法将基于复数降噪子网络识别算法的复杂度降低到与基于实数降噪的子网络识别算法的复杂度相近时, 其性能仍优于基于实数降噪子网络的识别算法。

Abstract:

In order to correctly identify the channel coding parameters of low-density parity-check coded wireless communication systems with high-order modulation over correlated noise, two novel deep learning-based semi-blind identification algorithms for code rates are proposed under the assumption that the candidate set of code rates and the corresponding parity check matrices are known. The proposed neural networks consist of the denoising subnetwork and the code-rate identification subnetwork. Moreover, the denoising subnetwork designs the real-valued denoising subnetwork and the complex-valued denoising subnetwork. Compared to the real-valued denoising subnetwork, the complex-valued denoising subnetwork has a better ability to process complex-valued signals with the cost of high complexity. Furthermore, in order to reduce the complexity of the complex-valued denoising subnetwork, a novel network compression algorithm based on network pruning technology is proposed. Simulation results show that, by using the novel multi-task learning strategy which jointly optimizes the denoising loss function and the loss function corresponding to code-rate identification, on one hand, the proposed neural networks have better identification performance comparing with the traditional algorithms with correlated noise, on the other hand, the identification algorithm based on complex-valued denoising subnetwork still outperforms the identification algorithm based on real-valued denoising subnetwork when its complexity approaches to that of the identification algorithm based on real-valued denoising subnetwork by using the network compression algorithm.

Key words: correlated noise, blind identification of channel coding, low-density parity-check (LDPC) code, deep learning, complex-valued network

中图分类号:

TN92

袁磊, 杨艳娟, 郭毅, 戴鹏. 相关噪声下基于深度学习的LDPC码码率半盲识别算法[J]. 系统工程与电子技术, 2025, 47(4): 1335-1345.

Lei YUAN, Yanjuan YANG, Yi GUO, Peng DAI. Deep learning-based semi-blind identification algorithm for code rates of LDPC codes with correlated noise[J]. Systems Engineering and Electronics, 2025, 47(4): 1335-1345.

图/表 13

图1

图2

表1

图3

表2

图4

图5

表3

图6

图7

表4

表5

图8

参考文献 30

1	梁应敞, 谭俊杰. 智能无线通信技术研究概况[J]. 通信学报, 2020, 41 (7): 1- 17.
	LIANG Y C , TAN J J . Research overview of smart wireless communication technology[J]. Communications Letters, 2020, 41 (7): 1- 17.
2	YANG W T , DU H Y . Semantic communications for future internet: fundamentals, applications, and challenges[J]. IEEE Communications Surveys & Tutorials, 2023, 25 (1): 213- 250.
3	KIM K , TUN Y K , MUNIR M S , et al. Deep reinforcement learning for channel estimation in RIS-aided wireless networks[J]. IEEE Communications Letters, 2023, 27 (8): 2053- 2057. doi: 10.1109/LCOMM.2023.3280821
4	ROSSEEL J , MANNONI V . Decoding short LDPC codes via BP-RNN diversity and reliability-based post-processing[J]. IEEE Trans.on Communications, 2022, 70 (12): 7830- 7842. doi: 10.1109/TCOMM.2022.3218821
5	TIAN J J , PEI Y Y , HUANG Y D , et al. Modulation-constrained clustering approach to blind modulation classification for MIMO systems[J]. IEEE Trans.on Cognitive Communications and Networking, 2018, 4 (4): 894- 907. doi: 10.1109/TCCN.2018.2879370
6	司海飞, 胡兴柳, 史震. 基于联合特征参数提取的非合作信号调制识别算法[J]. 通信学报, 2020, 41 (7): 172- 185.
	SI H F , HU X L , SHI Z . Non-cooperative signal modulation identification algorithm based on joint feature parameter extraction[J]. Communications Letters, 2020, 41 (7): 172- 185.
7	PENG S L , SUN S J , YAO Y D . A survey of modulation classification using deep learning: signal representation and data preprocessing[J]. IEEE Trans.on Neural Networks and Learning Systems, 2022, 33 (12): 7020- 7038. doi: 10.1109/TNNLS.2021.3085433
8	SHEA T J O , 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
9	ZHANG Z F , LUO H , WANG C , et al. Automatic modulation classification using CNN-LSTM based dual-stream structure[J]. IEEE Trans.on Vehicular Technology, 2020, 69 (11): 13521- 13531. doi: 10.1109/TVT.2020.3030018
10	CHE S L , ZHANG M Q , LI Y . A rank-iteration based blind identification method for convolutional codes[J]. IEEE Communications Letters, 2023, 27 (5): 1252- 1256. doi: 10.1109/LCOMM.2023.3257870
11	CHANG Y Y , ZHANG W , WANG H , et al. Fast blind recognition of BCH code based on spectral analysis and probability statistics[J]. IEEE Communications Letters, 2021, 25 (10): 3170- 3174. doi: 10.1109/LCOMM.2021.3100348
12	王垚, 王翔, 杨国栋. 基于编码矩阵结构特征的非删余极化码参数盲识别算法[J]. 通信学报, 2022, 43 (2): 22- 33.
	WANG Y , WANG X , YANG G D . A blind identification algorithm for non-censored polarisation code parameters based on structural features of the coding matrix[J]. Communications Letters, 2022, 43 (2): 22- 33.
13	TIAN X , WU H C . Novel blind identification of LDPC codes using average LLR of syndrome a posteriori probability[J]. IEEE Trans.on Signal Processing, 2014, 62 (3): 632- 640. doi: 10.1109/TSP.2013.2293975
14	WU Z J , ZHANG L M , ZHONG Z G . Blind recognition of LDPC codes over candidate set[J]. IEEE Communications Letters, 2019, 24 (1): 11- 14.
15	LI L Q , HUANG Z P , ZHOU J . A two-stage maximum a posterior probability method for blind identification of LDPC codes[J]. IEEE Signal Processing Letters, 2021, 28, 111- 115. doi: 10.1109/LSP.2020.3047334
16	LI L Q , HUANG Z P , ZHOU J . Blind identification of LDPC codes based on a candidate set[J]. IEEE Communications Letters, 2021, 25 (9): 2786- 2789. doi: 10.1109/LCOMM.2021.3083921
17	DEHDASHTIAN S , HASHEMI M , SALEHKALEYBAR S . Deep-learning-based blind recognition of channel code parameters over candidate sets under AWGN and multi-path fading conditions[J]. IEEE Wireless Communication Letters, 2021, 10 (5): 1041- 1045. doi: 10.1109/LWC.2021.3056631
18	SHARMA S K , CHATZINOTAS S , OTTERSTEN B . SNR estimation for multi-dimensional cognitive receiver under correlated channel/noise[J]. IEEE Trans.on Wireless Communications, 2013, 12 (12): 6392- 6405. doi: 10.1109/TWC.2013.103113.130523
19	LIANG F , SHEN C , WU F . An iterative BP-CNN architecture for channel decoding[J]. IEEE Journal of Selected Topics in Signal Processing, 2018, 12 (1): 144- 159. doi: 10.1109/JSTSP.2018.2794062
20	ZHANG L , YANG X L , LIU H , et al. Efficient residual shrinkage CNN denoiser design for intelligent signal processing: modulation recognition, detection, and decoding[J]. IEEE Journal on Selected Areas in Communications, 2022, 40 (1): 97- 111. doi: 10.1109/JSAC.2021.3126074
21	曹廷晖. 复数神经网络研究[D]. 西安: 西安电子科技大学, 2019.
	CAO T H. Research on complex neural network[D]. Xi'an: Xidian University, 2019.
22	YA T , LIN Y , HOU C B , et al. Complex-valued networks for automatic modulation classification[J]. IEEE Trans.on Vehicular Technology, 2020, 69 (9): 10085- 10089. doi: 10.1109/TVT.2020.3005707
23	LUO Q P , ZHAO M M , CHEN Z J , et al. Complex-valued convolution and frequency global filter for automatic modulation recognition[J]. IEEE Communications Letters, 2023, 27 (7): 1779- 1783. doi: 10.1109/LCOMM.2023.3271633
24	XIAO C H , YANG S Y , FENG Z X . Complex-valued depthwise separable convolutional neural network for automatic modulation classification[J]. IEEE Trans.on Instrumentation and Measurement, 2023, 72, 2522310.
25	LIU Z, LI J, SHEN Z Q, et al. Learning efficient convolutional networks through network slimming[C]//Proc. of the IEEE International Conference on Computer Vision, 2017: 2755-2763.
26	PARK S J . Bitwise log-likelihood ratios for quadrature amplitude modulations[J]. IEEE Communications Letters, 2015, 9 (6): 921- 924.
27	COCHACHIN F , BOUTILLON E , DECLERCQ D . Sign-preserving min-sum decoders[J]. IEEE Trans.on Communications, 2021, 69 (10): 6439- 6454. doi: 10.1109/TCOMM.2021.3099173
28	WANG Y , GUI G , GACANIN H , et al. An efficient specific emitter identification method based on complex-valued neural networks and network compression[J]. IEEE Journal on Selected Areas in Communications, 2021, 39 (8): 2308- 2317.
29	LIN M, CHEN Q, YAN S C. Network in network[C]//Proc. of the IEEE International Conference on Learning Representations, 2014.
30	杨洁, 董标, 付雪. 基于轻量化分布式学习的自动调制分类方法[J]. 通信学报, 2022, 43 (7): 135- 143.
	YANG J , DONG B , FU X . Lightweight decentralized learning-based automatic modulation classification method[J]. Journal on Communications, 2022, 43 (7): 135- 143.

Conv2D	卷积核数目	(卷积核大小, 步长)
1	128	(1×1, 1)
2	128	(2×2, 1)
3	128	(2×2, 1)
4	128	(2×2, 1)

Conv2D	卷积核数目	(卷积核大小, 步长)
1	64	(1×9, 1)
2	32	(1×3, 1)
3	16	(1×3, 1)
4	1	(1×15, 1)

参数名	参数值
信道编码方式	LDPC码
调制方式	8PSK, 16QAM
码长	648
码率集合	1/2, 2/3, 3/4, 5/6
相关系数	0, 0.5, 0.9
8PSK训练SNR/dB	2, 3, 4, 5, 6
16QAM训练SNR/dB	3, 4, 5, 6, 7
每SNR每码率码字数	10⁵
批处理大小	256
训练轮数	100

相关系数	调制方式	网络	r	χ
相关系数	调制方式	网络		0.2	0.5	0.8
0.5	8PSK	RJDI		1/2	0.90	0.40	0.45
			2/3	2.78	2.55	2.60
			3/4	4.40	4.15	4.20
			5/6	5.35	5.32	5.34
		CJDI	1/2	0.36	0.35	0.25
			2/3	2.50	2.10	1.90
			3/4	3.40	3.25	3.00
			5/6	3.88	3.81	3.79
	16QAM	RJDI	1/2	1.56	1.32	1.28
			2/3	3.60	3.56	3.54
			3/4	5.18	5.17	5.15
			5/6	5.98	5.96	5.94
		CJDI	1/2	1.18	1.06	1.03
			2/3	3.50	3.40	3.20
			3/4	4.75	4.60	4.48
			5/6	5.15	5.00	4.98
0.9	8PSK	RJDI	1/2	-3.15	-4.06	-3.86
			2/3	-0.40	-1.30	-1.20
			3/4	2.25	0.75	0.90
			5/6	4.60	2.20	2.40
		CJDI	1/2	-4.85	-4.95	-5.00
			2/3	-3.60	-3.75	-3.80
			3/4	-2.65	-2.80	-2.90
			5/6	-2.15	-2.20	-2.30
0.9	16QAM	RJDI	1/2	-0.90	-1.70	-1.85
			2/3	3.80	0.87	0.84
			3/4	5.50	3.10	2.80
			5/6	6.38	4.70	4.60
		CJDI	1/2	-3.20	-3.46	-3.48
			2/3	-1.45	-1.55	-1.60
			3/4	-0.30	-0.38	-0.40
			5/6	0.31	0.22	0.18

网络	调制	相关系数	P_r	参数量	浮点运算次数
RJDI	8PSK	-	-	8 835	4.009M
RJDI	16QAM	-	-	8 835	3.007M
CJDI	8PSK	-	-	17 783	7.612M
CJDI	16QAM	-	-	17 783	5.709M
PJDI	8PSK	0.5	0.35	8 719	3.677M
		0.5	0.8	1 037	392.448K
		0.9	0.35	8 839	3.731M
		0.9	0.8	1 133	435.456K
	16QAM	0.5	0.35	8 311	3.494M
		0.5	0.8	1 169	338.688K
		0.9	0.35	8 455	2.669M
		0.9	0.8	1 157	334.656K