South-Central Minzu University,a. College of Electronic and Information Engineering; b. Hubei Key Laboratory of Intelligent Wireless Communications; c. Hubei Engineering Research Center of Intelligent Internet of Things Technology,Wuhan 430074,China
Show less
文章历史+
Received
Accepted
Published
2024-04-18
Issue Date
2025-10-27
PDF (1675K)
摘要
针对蜂窝车联网(Cellular-Vehicle to Everything,C-V2X)通信场景下无线信道受到干扰导致通信过程中可能存在信息丢失的情况,通过对联邦分布式随机梯度下降(Federated Learning-Distributed Stochastic Gradient Descent,FL-DSGD)进行抗干扰模型更新机制的优化以减少上述通信链路不可靠情况的影响.该方案首先建立车辆与基站的通信链路及传输模型参数;然后在通信链路不可靠导致模型参数在传输过程中部分缺失的情况下,根据链路可靠性混合权重矩阵,利用车辆上存储的本地模型以及基站存储的全局模型参与当前轮次联邦学习的模型更新,以填充丢失的模型参数.仿真结果表明:在通信链路不可靠的情况下,FL-DSGD方案达到90%的训练准确率以及85%的测试准确率所需的通信轮次约为分布式基线方案所需通信轮次的50%.
Abstract
Aiming at the possible information loss in the communication process caused by interference to the wireless channel in the Cellular Vehicle to Everything (C-V2X) communication scenario, the impact of the above unreliable communication link is reduced by optimizing the anti-interference model update mechanism of the Federated Learning Distributed Stochastic Gradient Descent (FL-DSGD). Firstly, the communication link between the vehicle and the base station and the transmission model parameters are established; Then, when the communication link is unreliable, leading to partial loss of model parameters in the transmission process, according to the link reliability mixed weight matrix, the local model stored on the vehicle and the global model stored in the base station are used to participate in the model update of the current round of federated learning to fill in the missing model parameters. Simulation results show that when the communication link is unreliable, the communication rounds required for FL-DSGD scheme to achieve 90% training accuracy and 85% test accuracy are about 50% of the communication rounds required for the distributed baseline scheme.
随着道路上联网车辆的不断增多,车载通信已成为一个重要的研究领域[1].为了满足车联网(Vehicle to Everything,V2X)通信的严格要求,C-V2X需要共享频谱资源,从而实现车对基础设施(Vehicle to Infrastructure,V2I)和车对车(Vehicle to Vehicle,V2V)的同步通信.因此,如何在有限的频谱资源内减少干扰并同时保证V2I链路的信道容量和V2V通信的可靠性,成为V2X通信亟待解决的一个重要问题.
传统的优化方法已经被用来应对V2X通信中的干扰问题[2-5].在文献[2]中,提出了一种无线电资源管理(Radio Resource Management,RRM)算法,保证了基于端到端(Device to Device,D2D)的V2X系统的延迟和可靠性要求.基于类似的V2X框架,文献[3]提出了一个优化问题,在只考虑缓慢变化的大规模衰减信道的情况下为D2D车辆系统设计一个高效的频谱和功率分配算.此外,在文献[4-5]中研究了排队延迟对吞吐量和可靠性的影响.
WANGJ, LIUJ, KATON. Networking and communications in autonomous driving: A survey[J]. IEEE Communications Surveys & Tutorials, 2019, 21(2): 1243-1274.
[2]
SUNW, STRÖME G, BRÄNNSTRÖMF, et al. Radio resource management for D2D-based V2V communication[J]. IEEE Transactions on Vehicular Technology, 2016, 65(8): 6636-6650.
[3]
LIANGL, LIG Y, XUW. Resource allocation for D2D-enabled vehicular communications[J]. IEEE Transactions on Communications, 2017, 65(7): 3186-3197.
[4]
MEIJ, ZHENGK, ZHAOL, et al. A latency and reliability guaranteed resource allocation scheme for LTE V2V communication systems[J]. IEEE Transactions on Wireless Communications, 2018, 17(6): 3850-3860.
[5]
LIUC F, BENNISM. Ultra-reliable and low-latency vehicular transmission: An extreme value theory approach[J]. IEEE Communications Letters, 2018, 22(6): 1292-1295.
[6]
MCMAHANH B, MOOREE, RAMAGED, et al. Communication-efficient learning of deep networks from decentralized data[C]//Artificial Intelligence and Statistics, Fort Lauderdale:PMLR,2017: 1273-1282.
[7]
LIT, SAHUA K, ZAHEERM, et al. Federated optimization in heterogeneous networks[J]. Proceedings of Machine learning and systems, 2020, 2: 429-450.
[8]
FALLAHA, MOKHTARIA, OZDAGLARA. Personalized federated learning with theoretical guarantees: A model-agnostic meta-learning approach[J]. Advances in Neural Information Processing Systems, 2020, 33: 3557-3568.
[9]
DINHC T, TRANN H, NGUYENT D. Personalized federated learning with Moreau envelopes[J]. Advances in Neural Information Processing Systems, 2020, 33:21394-21405.
[10]
LIT, HUS, BEIRAMIA, et al. Ditto: Fair and robust federated learning through personalization[C]//International Conference on Machine Learning. Virtual:PMLR, 2021: 6357-6368.
[11]
YUANK, LINGQ, YINW. On the convergence of decentralized gradient descent[J]. SIAM Journal on Optimization, 2016, 26(3): 1835-1854.
[12]
ASSRANM, LOIZOUN, BALLASN, et al. Stochastic gradient push for distributed deep learning[C]//International Conference on Machine Learning. Long Beach:PMLR, 2019: 344-353.
[13]
LIANX, ZHANGW, ZHANGC, et al. Asynchronous decentralized parallel stochastic gradient descent[C]//International Conference on Machine Learning. Stockholm:PMLR, 2018: 3043-3052.
[14]
TANGH, GANS, ZHANGC, et al. Communication compression for decentralized training[J]. Advances in Neural Information Processing Systems, 2018, 31: 7663-7673.
[15]
KOLOSKOVAA, STICHS U, JAGGIM. Decentralized stochastic optimization and gossip algorithms with compressed communication[C]//International Conference on Machine Learning. Long Beach:PMLR, 2019: 3478-3487.