深度学习驱动的非正交多址系统联合检测与解码
100 浏览量
更新于2024-08-27
收藏 338KB PDF 举报
本文主要探讨了"深度学习在非正交多址（Non-Orthogonal Multiple Access, NOMA）系统中的联合检测与解码"这一前沿课题。随着近年来深度学习技术在通信领域广泛应用，研究人员致力于利用这些技术提升系统的效率和性能。作者 Fuqiang Sun、Kai Niu 和 Chao Dong 来自中国北京邮电大学的通用无线通信国家重点实验室，他们针对NOMA系统提出了新颖的方法。
NOMA是一种允许多个用户在同一资源块上进行同时传输的技术，通过非正交编码使得接收端能够区分信号并实现多用户的高效共享。传统的多用户检测方法如Message Passing Algorithm (MPA) 和 Belief Propagation (BP) 算法可能在复杂度和性能上存在局限性。
在本文中，作者构建了一个基于深度学习的联合检测器和解码器。首先，他们将神经网络应用于MPA，通过模仿该算法的工作原理，设计神经网络结构来处理多用户检测任务。同时，他们也构建了基于BP算法的解码器，以优化信息的准确传递。这两个组件被巧妙地连接在一起，形成了一个更大的神经网络模型。
关键在于，作者并未简单地将BP和MPA的结构机械拼接，而是对各自的因子图的边赋予权重，并运用深度学习技术进行训练。这样做的目的是让网络能够自动学习和优化这些权重，从而在保持或超越传统方法的基础上，进一步提高系统在检测和解码过程中的性能。这种方法的优势在于能够适应复杂的信道条件，减少手动调优的需要，并有可能发掘出传统方法未发现的性能优化空间。
相比于传统的MPC多用户检测和BP解码策略，深度学习联合检测与解码方法展示了显著的优势，包括降低计算复杂性、增强抗干扰能力以及可能的性能提升。这项研究对于推动NOMA系统在5G和未来通信网络中的实际应用具有重要意义，预示着深度学习在无线通信领域的深度集成和创新可能成为一种趋势。
Deep Learning Based Joint Detection and Decoding
of Non-Orthogonal Multiple Access Systems
Fuqiang Sun, Kai Niu, and Chao Dong
Key Laboratory of Universal Wireless Communications, Ministry of Education
Beijing University of Posts and Telecommunications, Beijing, China 100876
Email: {sunfuqiang, niukai, dongchao}@bupt.edu.cn
Abstract—In recent years, many studies have applied deep
learning techniques to the field of communication. Many of them
have achieved reduced complexity or improved performance. In
this paper, we apply deep learning techniques to joint multiuser
detection and decoding of non-orthogonal multiple access system.
We construct the neural network detector and decoder based on
message passing algorithm (MPA) and belief propagation (BP)
algorithm, respectively. And the decoder is cascaded with the
detector to form a larger neural network. To achieve this, we
assign weights to the edges of the respective factor graphs and
then use deep learning approaches to train these weights to get
better performance. Compared with the conventional method (i.e.
MPA for multiuser detection and BP for decoding), the proposed
method improves the joint detection and decoding performance,
while the computational complexity is not greatly increased.
I. INTRODUCTION
After decades of development, many excellent training
methods and neural network structures have been invented in
the field of deep learning to accomplish various tasks [1],
[2], [3]. In some areas, deep learning methods outperform
traditional methods, especially in computer vision and speech
processing [4]. In the last decade, there have been continuous
attempts to apply deep learning methods to the field of com-
munications. And many satisfactory results [5], [6], [7]have
demonstrated that this method is feasible.
The fifth-generation wireless communication requires high-
er spectral efficiency, lower latency, and massive connec-
tions. Current orthogonal multiple access technologies hardly
meet these requirements [8]. Non-orthogonal multiple ac-
cess(NOMA) is a feasible solution to meet these requirements
[8], [9]. Unlike orthogonal multiple access, an orthogonal
resource can be allocated to multiple users in NOMA system.
At the transmitting end, messages from different users are sent
non-orthogonally and interference information is actively in-
troduced. Compared to OMA system, the receiver complexity
is increased in NOMA system, but higher spectral efficiency
can be achieved. Like linear codes, such as BCH codes and
low-density parity-check (LDPC) codes, a NOMA system can
be represented by a factor graph in which there are two types
of nodes, variable nodes (VNs) corresponding to the users, and
function nodes (FNs) corresponding to the orthogonal resource
elements (REs). Message passing algorithm can be performed
in this factor graph to do multiuser detection.
This work is supported by the National Natural Science Foundation of
China (No. 61671080, No.61601047).
The message passing algorithm is essentially the same
as the belief propagation (BP) algorithm commonly used in
LDPC decoding. In the two algorithms, messages are both
transmitted over edges of the factor graph. In [5], the author
assign weights to the edges of the Tanner graph that represents
corresponding linear code. And then these weights are trained
by deep learning method. Inspired by this, We suspect that
the weights on the edges of the NOMA factor graph may
affect the performance of multiuser detection. So we design
a neural network multiuser detector which is an alternative
representation of the factor graph. The nodes in the neural
network multiuser detector is corresponding to messages trans-
mitted over edges. Due to the multi-ary modulation, multiple
messages are delivered over one edges of the factor graph
when the node is updated.
In this paper, we proposed a scheme of joint detection
and decoding of NOMA systems based on deep learning.
Unlike the fully connected neural network, the proposed neural
network is constructed based on the corresponding factor
graph and is a sparse neural network. And if only multi-user
detection is performed, the gain from channel coding cannot be
utilized. So we design the multi-user detector and decoder as
neural networks to form a large neural network. And then use
deep learning technologies to train the whole neural network
to get better performance.
The remainder of the paper is organized as follows. Section
II briefly describes the system model of NOMA. We present
the proposed deep neural network in Section III. The numerical
results and conclusion are presented in Section IV and Section
V.
II. SYSTEM MODEL
Consider a simplified uplink non-orthogonal multiple access
system where V users multiplex onto F orthogonal resource
elements. For vth user, each K-bit input binary data u
v
is
channel-coded into an N-bit coded block c
v
.
c
v
= u
v
G (1)
Where G is the generator matrix. Given the modulation order
J, each user’s N-bit codeword block will be transmitted in
N
J
time slots. For the tth time slot, t = 1, 2, · · ·
N
J
, every J
bits in c
v
of the vth user constitute a J-bits vector b
t
v
and are
978-1-5386-4920-6/18/$31.00 ©2018 IEEE
下载后可阅读完整内容，剩余4页未读，立即下载
2018-11-23 上传
2021-05-22 上传
2021-02-08 上传
2021-05-27 上传
2021-05-26 上传
2011-12-01 上传
2021-05-30 上传
2021-08-11 上传
weixin_38686557
- 粉丝: 4
- 资源: 930
我的内容管理
展开
最新资源
- JavaScript实现的高效pomodoro时钟教程
- CMake 3.25.3版本发布：程序员必备构建工具
- 直流无刷电机控制技术项目源码集合
- Ak Kamal电子安全客户端加载器-CRX插件介绍
- 揭露流氓软件：月息背后的秘密
- 京东自动抢购茅台脚本指南：如何设置eid与fp参数
- 动态格式化Matlab轴刻度标签 - ticklabelformat实用教程
- DSTUHack2021后端接口与Go语言实现解析
- CMake 3.25.2版本Linux软件包发布
- Node.js网络数据抓取技术深入解析
- QRSorteios-crx扩展：优化税务文件扫描流程
- 掌握JavaScript中的算法技巧
- Rails+React打造MF员工租房解决方案
- Utsanjan：自学成才的UI/UX设计师与技术博客作者
- CMake 3.25.2版本发布，支持Windows x86_64架构
- AR_RENTAL平台：HTML技术在增强现实领域的应用
