CN-116782273-B - Intelligent reflecting surface-assisted D2D communication optimization method

Abstract

The invention relates to an intelligent reflection surface-assisted D2D communication optimization method, and belongs to the technical field of wireless communication. Firstly, a RIS-assisted D2D communication system model is provided, secondly, the optimization problem in a wireless communication system is solved by training a deep neural network, and finally, a feasible solution is found by adopting a distributed DL-based algorithm. The intelligent reflecting surface of the method can reduce interference caused by an LOS link and improve channel quality.

Inventors

• XU QIAN
• YOU QIAN
• WANG LING
• YANG XIN
• SUN WENBIN
• ZHANG ZHAOLIN
• SU JIA
• GONG YANYUN

Assignees

• 西北工业大学

Dates

Publication Date

20260508

Application Date

20230629

Claims (3)

1. 1. An intelligent reflection surface assisted D2D communication optimization method is characterized by comprising the following steps of: the processing steps at the transmitting end of the unmanned aerial vehicle base station are as follows: step 1, a transmitting end comprises an unmanned aerial vehicle base station, an RIS with M reflecting elements and K channels, wherein each channel comprises a cellular user CUE, namely K cellular users and D2D users, wherein the multiple D2D users reuse the frequency spectrum occupied by the cellular users, and in addition, the method comprises the following steps of Indicating whether the D-th D2D user uses the kth channel, i.e. if the D-th D2D user's transmitter transmits data over the kth channel Otherwise It is assumed that each D2D user can share spectrum with only one cellular user at the same time, i.e Assuming that all transceivers use a single antenna, the drone base station is located at RIS position is The D2D user pairs and the cellular users are randomly distributed in a circle with radius R, and the position of the first cellular user is The dT position is recorded as The maximum distance between DT and the corresponding receiver DR is The d-th DR position is noted as ; Further, assume that provision is made on RIS A reflection element, a reflection coefficient matrix Represented as The first diagonal matrix The individual elements are expressed as Wherein Represent the first The phase shift of the individual reflecting elements, Represent the first The amplitude of each reflecting element, in practice, the phase shift of each element takes only a limited number of discrete values, the first The discrete values are expressed as Wherein , Representing phase shift levels for quantization Assuming that RIS performs ideal reflection so that the signal power of each reflective element is lossless, i.e. amplitude reflection coefficient Meanwhile, assume DT The transmission power of (2) is expressed as Maximum value of it is And quantized into Discrete steps, i.e. ; Step2, the unmanned aerial vehicle base station generates X bit information ; Step 3, modulating the information generated in the step 2, wherein the modulated signal is And meet the following ; Step 4, sending the signal s generated in the step3 into a channel; the unmanned aerial vehicle to user channel construction processing steps are as follows: step 5, when the signals transmitted by the unmanned aerial vehicle base station are transmitted in free space, no tall building is shielded, the unmanned aerial vehicle can directly transmit the signals, a line-of-sight link LOS is formed at the moment, when the signals continue to be transmitted, shielding exists between the unmanned aerial vehicle base station and ground users, and the signals are influenced by electromagnetic wave reflection and scattering, so that a non-line-of-sight link NLOS is formed; step 6, LOS and NLOS in the channel gain appear according to probability, assuming the ground node coordinates are UAV base station coordinates are The distance between the unmanned plane and the node is The LOS component probability of the node is: wherein A, B is an environmental dependent constant, NLOS component probabilities are: ; and 7, the channel gain from the unmanned aerial vehicle to the user d is as follows: Wherein the method comprises the steps of Is the path loss coefficient between the user and the drone link, Is the correlation coefficient with the NLOS link; step 8, the distance between the unmanned plane and the first CUE is Will be Substituted into step 6 And In (1) randomly generating Rayleigh distribution coefficient If yes, the channel gain is: ; step 9, the distance between the unmanned plane and the RIS is Ignoring RIS position differences, one will Substituted into step 6 And In (1) randomly generating Rayleigh distribution coefficient The channel gain is: ; Step 10, the distance between the unmanned plane and the d-th DR is Will be Substituted into step 6 And In (1) randomly generating Rayleigh distribution coefficient The channel gain is: ; the other link channel construction processing steps are as follows: step 11, other link channel gains consider random Rayleigh distribution and path loss, and the path loss is as follows: Wherein the method comprises the steps of Is the distance between the nodes of the network, Respectively representing the path loss at 1m and the path loss coefficient; step 12 distance between RIS and the first CUE is Will be Substituting into the loss of the step 11 to randomly generate Rayleigh distribution coefficient The channel gain is: ; step 13 distance between RIS and dDR is Will be Substituting into the loss of the step 11 to randomly generate Rayleigh distribution coefficient The channel gain is: the distance between the dDT and the RIS is Will be Substituting into the loss of the step 11 to randomly generate Rayleigh distribution coefficient The channel gain is: ; step 14, the distance between the first CUE and the d DR is Will be Substituting into the loss of the step 11 to randomly generate Rayleigh distribution coefficient The channel gain is: ; step 15, dDT and dT The distance between DR users is Will be Substituting into the loss of the step 11 to randomly generate Rayleigh distribution coefficient The loss matrix between DT and DR is noted as Wherein Line d Column elements are noted as The channel gain is: Wherein the channel gain between the dT and DR is recorded as DDT and dT The channel gain between the DRs is noted as ; The processing steps at the receiving end are as follows: step 16, the signal-to-interference-plus-noise ratio SINR received by the CUE in the kth channel is as follows: Wherein the method comprises the steps of Representing the transmit power of the UAV, The transmission power of the dT is the dT Additive white gaussian noise at CUE in each channel obeys: ; Step 17, the SINR received by the d DR in the k channel is: Wherein the method comprises the steps of And Representing interference from other D2D users and CUE users on the kth channel, respectively, the additive white gaussian noise compliance at the D-th DR: ; step 18. Thus, the achievable rate of the d-th DR of the kth channel is: and 19, finally, calculating the sum rate of all DR in the kth channel as follows: step 20, expressing the optimization problem as the following non-convex optimization problem: Wherein the method comprises the steps of , Represent the first The index vector is used for the D2D channel, Representing the transmit power of D DTs; Representing the maximum interference limit imposed on the CUE, Representing the minimum SINR threshold of the CUE, Representing a minimum achievable rate threshold for the D2D user; And step 21, solving the optimization problem by adopting a distributed DL algorithm based on unsupervised learning, wherein the specific implementation steps are as follows: step 22, in order to prevent the channel gain from being too large or too small to influence the training result, the training result is passed through a function Normalize the data and record the pre-processed channel gain as I.e. Step 23, the first step All CSI contained by the D2D is noted as After data preprocessing is carried out on the system, the preprocessed CSI is fed to a DNN network for CSI feedback, and the system is compressed into the system Bits and together with sigmoid layers constitute a function of encoding local CSI Determining the CSI transmitted to the base station, denoted as I.e. And is also provided with ; Step 24 bs transmits all the collected local CSI And its own pre-processed CSI Feeding to DNN network for CSI feedback, compressing it into Bits and together with sigmoid layers constitute a function of encoding local CSI Determining And broadcast it to all D2D pairs, i.e ; Step 25, obtaining the maximum sum rate of the D2D user by minimizing the loss function of the neural network designed in step 22 and step 23, wherein the loss function of the self-defined neural network is as follows: Wherein, the To eliminate the function of the binarization error: Wherein, the Is a network super parameter.
2. 2. The intelligent reflector-assisted D2D communication optimization method according to claim 1, wherein the DNN network in step 23 and step 24 has three layers, wherein the first layer contains 256 neurons, and the second layer and the third layer each contain 128 neurons.
3. 3. A computer system comprising one or more processors, a computer-readable storage medium storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of claim 1.

Description

Intelligent reflecting surface-assisted D2D communication optimization method Technical Field The invention belongs to the technical field of wireless communication, and relates to an optimization scheme design through distributed deep learning in an intelligent reflection surface-assisted D2D communication system. Background With The widespread commercialization of The fifth generation cellular mobile communication system (The 5th Generation Telecommunication, 5G), the next generation cellular mobile communication system is also attracting attention, and high bandwidth, low latency, high reliability, low cost, etc. are further The focus of research. However, the development of wireless communication is always limited by the shortage of spectrum resources and low utilization, and in order to solve these outstanding problems, numerous technical means are introduced to promote and optimize the network performance. In order to meet the rapidly growing demand for data traffic and to achieve seamless communication, device-to-Device (D2D) communication technology is considered as a promising technology. The use of D2D communication capabilities in cellular networks provides a number of benefits to end users and network systems. Firstly, the D2D communication multiplexing cellular user link can improve the frequency spectrum efficiency, although a certain interference is introduced in the process, the interference can be eliminated by a plurality of means, secondly, the communication time delay between the users can be reduced and the throughput of the whole network can be improved because the communication distance between the two D2D users is short, the network traffic load at the base station side is greatly avoided because the data is not forwarded through the base station, and thirdly, the D2D users can use less transmission power according to the actual network state when transmitting, and the energy efficiency of the D2D communication is improved. The capacity of the future communication network will further increase, and from the operation of the current 5G primary business, the energy consumption of the base station is a non-negligible problem, and maintaining good operation cost while improving network performance is also a sustainable development motive force of operators. The intelligent reflection surface (Reconfigurable Intelligent Surface, RIS) is widely researched as an emerging technology with an extremely application prospect by the characteristics of low energy consumption, low cost, programmability, easy deployment and the like. RIS is a planar array of a large number of reconfigurable passive reflecting elements (e.g., phase shifters) that can be independently introduced into electromagnetic waves with certain phase shifts that are intelligently manipulated by appropriately adjusting their reflection coefficients to create a good propagation environment, especially when blocking or severe fading is encountered. Research into metamaterials enables real-time configuration of RIS, which is essential for a rapidly changing wireless communication environment. Moreover, RIS can be easily deployed to the surface of a building or some mobile device, providing mobility and portability for architectural implementations in practical scenarios. One significant advantage of using RIS-assisted communication over large-scale multiple-input multiple-output (Multiple Input Multiple Output, MIMO) techniques is that system energy consumption can be reduced, enabling sustainable green future communications. There are currently more research on RIS, and less research on RIS-assisted download communication systems. Moreover, in the existing research, the RIS continuous phase condition is mostly considered, but in a practical system, the discrete phase condition is more common for the intelligent reflecting surface. In addition, because RIS phase optimization is introduced, the method has great challenges for solving the system optimization problem, and the currently used traditional algorithm solves the problem through semi-positive relaxation, iterative optimization and other mathematical methods, so that the method has high solving complexity and poor applicability. The rapid rise of machine learning (MACHINE LEARNING, ML) gradually breaks through the constraint of the traditional algorithm, and deep learning (DEEP LEARNING, DL) can construct a neural network structure with a large number of learnable parameters, so that the neural network structure becomes an emerging research hotspot of ML. The deep-learned neural network model approximates complex primitive problems without requiring a strictly defined mathematical model by constructing a nonlinear mapping between the raw input data and the desired output. In addition, the deep learning-based method is more adaptive and robust than the conventional algorithm in terms of communication environment change and channel data loss damage, and the characteristic learn