Multiple-input-multiple-output non-orthogonal multiple access(MIMO-NOMA) technique involves the NOMA idea into MIMO systems, which can effectively improve the system energy efficiency and throughput. Nevertheless, complex spatial structures and rapidly varying spatial channels of the MIMO-NOMA system decrease the system's throughput and block its application. To address this issue, this paper proposes a deep-learning-based framework for MIMO-NOMA systems that employs an efficient Convolutional Deep Neural Network(CCDNN) tailored to MIMO-NOMA communications. The method comprises multiple convolutional and hidden layers and is able to solve the power-allocation problem through a dedicated algorithm, thereby improving both the transmission rate and energy efficiency of the MIMO-NOMA system. Numerical simulations are conducted for validation, and the results demonstrate that the proposed CCDNN framework outperforms conventional approaches, offering a viable solution for enhancing MIMO-NOMA power-allocation performance and laying a theoretical foundation for future research in this area.
目前,MIMO-NOMA系统的技术发展和应用为无线通信网络带来了新的机遇,但同时也面临诸多问题亟需解决,包括:①多用户分配机制采用的公平原则或基于效用最大化原则仍存在问题;②由于系统引入了更多的控制信道,导致系统复杂度增加,对系统性能的影响也更加明显;③在实际网络中,由于信道衰落、多径传播等原因,系统的性能指标与理论值有较大差异,这会进一步影响系统的性能;④由于MIMO-NOMA技术的标准化工作相对滞后、不够完善,会导致实际部署时,各设备之间的参数配置可能需要多次调整才能得到较好的性能。为了改进MIMO-NOMA系统的网络吞吐量及其保密性能,文献[8]针对3D-MIMO信道提出了一种基于机会型NOMA的分层安全模型,该模型可以根据不同的基站和用户提供若干种安全级别。此外,基于MIMO-NOMA系统,文献[9]将用户端天线动态地分为多个簇,并推导演化了功率分配方案以最大化小区容量,其中簇的数量不少于基站处发射天线的数量。需要注意的是,上述SIC模块需要获得每个用户端的信道状态信息(Channel state information,CSI),且MIMO系统的增益性能在很大程度上依赖于CSI的准确率。目前的功率分配方案几乎都是在假设系统处于完美CSI前提条件下设计的。然而,由于CSI具有超高的复杂度,MIMO-NOMA系统很难获得高精度CSI,因此会严重影响现有的功率分配方式。如前所述,相比于低复杂度等其他因素,虽然空间自由度是影响MIMO系统信道估计和预编码性能的关键因素之一 [10],但目前仅有少部分研究考虑了MIMO-NOMA系统的空间自由度因素。此外,在MIMO-NOMA系统中,功率分配机制的优化基本上都属于非确定性多项式复杂性问题(Non-deterministic polynomial complexity,NP),传统技术仅能求得次优解,导致现有的功率分配算法均有不同程度的局限性。
基于上述分析,开发对MIMO-NOMA系统的功率分配进行优化的新算法就显得尤为重要。近来,机器学习中的深度学习理论被广泛应用于处理大数据、求解非线性过程等难题。机器学习与处理大数据、求解非线性过程的关系在某种意义上是相辅相成的。一方面,大数据为机器学习算法的训练提供基础数据;另一方面,对于复杂的非线性问题,机器学习算法可以对大数据中的非线性特征进行建模,进而实现对复杂问题的求解。尽管基于深度学习的无线通信系统尚未成熟,但科研人员已经开展了一系列开创性工作,并已初步验证了该方法的良好性能[11]。文献[12]首次将深度学习的理念融入NOMA系统,并验证了深度学习在NOMA系统编码、解码、信道检测等方面的改进优势和有效性。此后,深度学习在无线通信领域引起了更为广泛的关注,部分学者将其应用于正交频分复用系统(Orthogonal frequency division multiplexing,OFDM)进行信道估计;部分学者将其引入大规模MIMO系统中进行信道估计及超分辨率到达方向(Super-resolution direction of arrivals,SDOA)估计,并得到了充分的证明。此外,无线通信中的深度学习方法在信道编码、毫米波(mm wave)、无人机(Unmanned aerial vehicles,UAVs)、OFDM接收机、认知无线电、移动用户卸载框架等多个领域中也进行了初步的相关研究[13]。
针对NOMA系统由时空信道快速变化导致系统吞吐量下降,以及功率分配机制优化等问题,本文首先创新性地将NOMA理念融入MIMO系统;其次将经典的最大比值合并(Maximum ratio combining,MRC),即依据阵列增益,通过合并多个信号路径改善接收端的信号质量,检测向量作为MIMO系统的自由度;最后将深度学习引入MIMO-NOMA系统,提出了适用于MIMO-NOMA的高效深度学习通信卷积方法CCDNN。
如图1所示,本文主要讨论经典MIMO-NOMA系统的下行链路。其中基站采用具有M根天线和D个多天线用户的均匀线性阵列(Uniform linear array,ULA),假设该下行链路的衰落环境为瑞利衰落信道(Rayleigh fading channel)。本文中每个移动用户配备有Nr根天线,并且基站没有用户的上、下行链路信息。
算法1中,输入由仿真环境参数、信道矢量 hm 、预编码矩阵 P 等3部分组成,而训练集由信道矢量 hm 和预编码矩阵 P 共同组成,并假定涉及的无线信道会受到无线环境各类衰落因子、高斯白噪声AWGN等因素共同影响作用。同时,设定网络的误码率阈值,最终CCDNN网络训练结束,对网络参数和每层的权重进行更新。
为了评估本文CCDNN框架的计算复杂度,采用乘法累加计数(Multiplication and addition counts,MAC)作为度量,比较网络参数[23,24]。MAC可以在网络训练过程中,以乘法和加法的形式度量参数的数量。为了直观地评估本文算法的计算复杂度,本节将本文所提CCDNN框架与基于LSTM NOMA的方案[25]、基于深度学习辅助的MIMO功率分配方案[26]中使用的功率分配数据集进行比较(见表6)。
SaadW, BennisM, ChenM Z. A vision of 6G wireless systems: applications, trends, technologies, and open research problems[J]. IEEE Netw, 2020, 34(3): 134-142.
[2]
ZhangH Z, ZhouT, XuT H, et al. Remote interference discrimination testbed employing AI ensemble algorithms for 6G TDD networks[J]. Sensors, 2023, 23(4): No.2264.
[3]
AkyildizI F, KakA, NieS T. 6G and beyond: the future of wireless communications systems[J]. IEEE Access, 2020, 8: 133995-134030.
[4]
ZhuL P, XiaoZ Y, XiaX G, et al. Millimeter-wave communications with non-orthogonal multiple access for B5G/6G[J]. IEEE Access, 2019, 7: 116123-116132.
[5]
JiaoR C, DaiL L.On the max-min fairness of beam space MIMO-NOMA[J]. IEEE Transactions on Signal Processing, 2020, 68: 4919-4932.
[6]
HuangH J, GuoS, GuiG, et al. Deep learning for physical-layer 5G wireless techniques: opportunities, challenges and solutions[J]. IEEE Wireless Commun-ications, 2020, 27(1): 214-222.
[7]
KaurJ, SinghM L.User assisted cooperative relaying in beam space massive MIMO NOMA based systems for millimeter wave communications[J]. China Communications, 2019, 16(6): 103-113.
[8]
HeH T, WenC K, JinS, et al. Deep learning-based channel estimation for beamspace mmwave massive MIMO systems[J]. IEEE Wireless Communication letters, 2018, 7(5): 852-855.
[9]
WuJ Y, XuT H, ZhouT, et al. Feature-based spectrum sensing of NOMA system for cognitive IoT networks[J]. IEEE Internet of Things Journal, 2023, 10(1): 801-814.
[10]
WangJ P, CaoF, HeX Y, et al. Multi carrier system joint receiving method based on MAI and ICI[J]. Journal of Jilin University(Engineering and Technology Edition), 2018, 41(6):1793-1797.
LiaoYong, YangZhi-jing, LiXue. Research progress of potential applications of AI in 6G air interface physical layer[J]. Journal of Beijing University of Posts and Telecommunications, 2022, 45(6): 21-30.
[13]
Tweneboah-KoduahS, AffumE A, AgyekumK A P, et al. Performance of cooperative relay NOMA with large antenna transmitters[J]. Electronics, 2022, 11(21): No.3482.
[14]
AlraddadyF, AhmedI, HabtemicailF. Robust hybrid beam-forming for non-orthogonal multiple access in massive MIMO downlink[J]. Electronics, 2022, 11(1): No.75.
[15]
TangS Y, MaZ, XiaoM, et al. Hybrid transceiver design for beamspace MIMO-NOMA in code-domain for mmWave communication using lens antenna array[J]. IEEE Journal on Selected Areas in Communications, 2020, 38(9): 2118-2127.
HuXiang-ge, WuGuang-fu, LiTao, et al. Power allocation scheme for MIMO-NOMA system based on QoE[J]. Journal of Beijing Unirersity of Posts and Telecommunications, 2021, 44(4): 62-67.
[18]
WangJ P, ZouN Y, ZhangY C, et al. Study on downlink performance of multiple access algorithm based on antenna diversity[J]. ICIC Express Letters, 2015, 9(4): 1221-1225.
[19]
AlimoD, SaitoM. Beam selection for mm-wave massive MIMO systems using ACO & combined digital precoding under hybrid transceiver architecture[J]. IEICE Communications Express, 2020, 9(6): 170-175.
[20]
QianL P, WuY, YuN N, et al. Learning driven NOMA assisted vehicular edge computing via underlay spectrum sharing[J]. IEEE Transactions on Vehicular Techndogy, 2021, 70(1): 977-992.
[21]
LuY X, ChengP, ChenZ, et al. Deep autoencoder learning for relay-assisted cooperative communication systems[J]. IEEE Transactions on Communications, 2020, 68(9): 5471-5488.
[22]
YeN, LiX M, YuH X, et al. Deep NOMA: A unified framework for NOMA using deep multi-task learning[J]. IEEE Transactions on Wireless Communications, 2020, 19(4): 2208-2225.
[23]
SaadM, BennisM, ChenM. A vision of 6G wireless systems: applications, trends, technologies, and open research problems[J]. IEEE Network, 2020, 34(3): 134-142.
[24]
LuY X, ChengP, ChenZ, et al. Deep multi-task learning for cooperative NOMA: system design and principles[J]. IEEE Journal on Selected Areas in Communications, 2021, 39(1): 61-78.
[25]
SabujS R, AsieduD K P, LeeK J, et al. Delay optimization in mobile edge computing: Cognitive UAV-assisted eMBB and mMTC services[J]. IEEE Transactions on Cognitive Communications and Networking, 2022, 8(2): 1019-1033.
[26]
CorgnatiL, BertaM, KokkiniZ. Assessment of OMA gap-filling performances for multiple and single coastal HF radar systems: validation with drifter data in the ligurian sea[J]. Remote Sensing, 2024, 16(13): 2458-2471.
[27]
WangX, WangH T, DongZ L. A NOMA-VLC power allocation scheme for multi-user based on sparrow search algorithm[J]. Optoelectronics Letters, 2023, 7: 1327-1336.
[28]
TanY Z, ChenB. Joint channel estimation and power allocation for the CRS-NOMA[J]. Chinese Journal of Electronics, 2020, 29(1): 177-182.