Intelligent reflecting surface assisted d2d communication optimization method

Abstract

The application relates to an intelligent reflecting surface assisted D2D communication optimization method and belongs to the technical field of wireless communication. First, an RIS assisted D2D communication system model is proposed, second, a deep neural network is trained to solve the optimization problem in the wireless communication system, and finally, a distributed DL based algorithm is used to find a feasible solution. The use of the intelligent reflecting surface in the method can reduce the interference caused by the LOS link and improve the channel quality.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology and relates to an intelligent reflective surface-assisted D2D communication system, in which an optimization scheme is designed using distributed deep learning. Background Technology

[0002] With the widespread commercialization of 5G, the next generation of cellular mobile communication systems is gradually attracting attention, with high bandwidth, low latency, high reliability, and low cost becoming key research focuses. However, the development of wireless communication has always been constrained by the scarcity and low utilization of spectrum resources. To address these prominent issues, numerous technologies have been introduced to improve and optimize network performance.

[0003] To meet the rapidly growing demand for data traffic and achieve seamless communication, Device-to-Device (D2D) communication technology is considered a promising approach. The application of D2D communication in cellular networks offers numerous benefits to end users and the network system. First, D2D communication can improve spectrum efficiency by reusing cellular user links, although some interference is introduced in the process, this can be eliminated through various means. Second, the short communication distance between two D2D users reduces communication latency and increases the overall network throughput; data transmission without base station intervention also significantly reduces network traffic load on the base station side. Third, D2D users can use less transmission power based on actual network conditions, improving the energy efficiency of D2D communication.

[0004] The capacity of future communication networks will further increase. From the current initial commercial operation of 5G, the energy consumption of base stations is a significant issue. Maintaining good operating costs while improving network performance is also a driving force for the sustainable development of operators. Reconfigurable Intelligent Surfaces (RIS), with their low energy consumption, low cost, programmability, and ease of deployment, are widely studied as a promising emerging technology. RIS is a planar array composed of a large number of reconfigurable passive reflective elements (such as phase shifters). It can independently introduce certain phase shifts into electromagnetic waves, intelligently manipulating them by appropriately adjusting their reflection coefficients to create a favorable propagation environment, especially when encountering congestion or severe fading. Research on metamaterials enables real-time configuration of RIS, which is essential for rapidly changing wireless communication environments. Moreover, RIS can be easily deployed on the surfaces of buildings or certain mobile devices, providing mobility and portability for practical architecture implementation. Compared to massively multi-input multiple-output (MIMO) technology, a significant advantage of using RIS to assist communication is the reduction of system energy consumption, enabling sustainable and green future communication.

[0005] Currently, there is considerable research on RIS (Reflector-Assisted Reflector System), but less research on RIS-assisted downlink vehicular communication systems. Furthermore, existing research mostly considers the continuous phase case of RIS, while in practical systems, discrete phase cases are more common for intelligent reflectors. In addition, the introduction of RIS phase optimization presents significant challenges to solving the system optimization problem. Current traditional algorithms employ semi-definite relaxation, iterative optimization, and other mathematical methods, resulting in high complexity and poor applicability.

[0006] The rapid rise of Machine Learning (ML) has gradually broken through the constraints of traditional algorithms. Deep Learning (DL), which can construct neural network structures with a large number of learnable parameters, has gradually become an emerging research hotspot in ML. Deep learning neural network models approximate complex original problems by constructing a nonlinear mapping between the original input data and the desired output, without requiring a strictly defined mathematical model. Furthermore, deep learning-based methods exhibit better adaptability and robustness compared to traditional algorithms in dealing with changes in the communication environment and channel data loss or corruption. This feature learning model can quickly adapt to environmental changes by updating parameters using new received data. In the long run, deep learning methods have lower computational complexity than model-based optimization methods. Moreover, neural networks can perform high-speed parallel computation, providing faster response times for wireless communication systems. Summary of the Invention

[0007] The technical problem to be solved by this invention is:

[0008] To address the problem of poor mobility in traditional base stations, this invention provides a smart reflective surface-assisted D2D communication optimization method.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0010] A smart reflective surface-assisted D2D communication optimization method, characterized in that:

[0011] The processing steps at the drone base station transmitter are as follows:

[0012] Step 1: The transmitting end includes a drone base station, a RIS with M reflective elements, and K channels. Each channel contains one cellular user (CUE), meaning there are K cellular users and D D2D users, where multiple D2D users reuse the spectrum occupied by the cellular users; in addition, using... This indicates whether the d-th D2D user uses the k-th channel; that is, if the transmitter of the d-th D2D user transmits data through the k-th channel, then... ,otherwise Assume that each D2D user can only share spectrum with one cellular user at a time, i.e. Assuming all transceivers use a single antenna, the drone base station location is... The RIS position is D2D user pairs and cellular users are randomly distributed within a circle of radius R. l The location of each cellular user is , No. d Each DT position is denoted as The maximum distance between DT and the corresponding receiver DR is , No. d Each DR position is denoted as ;

[0013] Furthermore, assuming the RIS is equipped with One reflecting element, reflection coefficient matrix Represented as The first of this diagonal matrix Each element is represented as ,in Indicates the first The phase shift of each reflecting element, Indicates the first The amplitude of each reflecting element; in practice, the phase shift of each element can only take a finite number of discrete values, the first... Each discrete value is represented as ,in , Indicates the level used for quantizing phase shift. The number of bits; assuming RIS performs ideal reflection, so that the signal power of each reflecting element is lossless, i.e., the amplitude reflection coefficient. Meanwhile, assuming DT The transmit power is expressed as Its maximum value is and quantified as A walk of a distance, that is ;

[0014] Step 2: Generation of drone base station X bit information ;

[0015] Step 3: Modulate the information generated in Step 2. The modulated signal is... and satisfy ;

[0016] Step 4: Transfer the signal generated in step 3 s Send into the channel;

[0017] The steps for establishing a drone-to-user channel are as follows:

[0018] Step 5: When the signal transmitted by the drone base station propagates in free space without the obstruction of tall buildings, the drone can directly transmit the signal, thus forming a line-of-sight (LOS) link. As the signal continues to propagate, there is obstruction between the drone base station and the ground user, and it is also affected by electromagnetic wave reflection and scattering, thus forming a non-line-of-sight (NLOS) link.

[0019] Step 6: LOS and NLOS in the channel gain occur with a certain probability. Assume the ground node coordinates are... The coordinates of the UAV base station are The distance between the drone and the node is Then the LOS component probability of the node is:

[0020]

[0021] Where A and B are constants related to the environment. The NLOS component probabilities are: ;

[0022] Step 7: Drone to User d The channel gain is:

[0023]

[0024] in It is the path loss coefficient between the user and the drone link. It is the correlation coefficient with the NLOS link;

[0025] Step 8: Drones and the first l The distance between each CUE is ,Will Substitute into step 6 and In the process, the Rayleigh distribution coefficients are randomly generated. If the channel gain is: ;

[0026] Step 9: The distance between the drone and the RIS is Ignoring the differences in RIS position, Substitute into step 6 and In the process, the Rayleigh distribution coefficients are randomly generated. Then the channel gain is: ;

[0027] Step 10: Drones and the first d The distance between each DR is ,Will Substitute into step 6 and In the process, the Rayleigh distribution coefficient is randomly generated. Then the channel gain is: ;

[0028] Other link channel construction and processing steps are as follows:

[0029] Step 11: Considering the random Rayleigh distribution and path loss for other link channel gains, the path loss is:

[0030]

[0031] in It is the distance between nodes. These represent the path loss and path loss coefficient at 1m, respectively.

[0032] Step 12: RIS and the first l The distance between each CUE is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ;

[0033] Step 13: RIS and the first d The distance between each DR is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ;No. d The distance between DT and RIS is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ;

[0034] Step 14: l The first CUE and the second d The distance between each DR is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ;

[0035] Step 15: d The first DT and the first The distance between DR users is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Let the loss matrix between DT and DR be denoted as ,in No. d Line number Column elements are denoted as Then the channel gain is: , of which d The channel gain between DT and DR is denoted as , No. d The first DT and the first The channel gain between the DRs is denoted as ;

[0036] The processing steps at the receiving end are as follows:

[0037] Step 16: k The signal-to-interference-plus-noise ratio (SINR) received by CUE in each channel is:

[0038]

[0039] in Indicates the UAV's transmission power. For the first d The transmit power of the DT, the first The additive white Gaussian noise at CUE in each channel follows the following rules: ;

[0040] Step 17: k The first channel dThe SINR received by each DR is:

[0041]

[0042] in and They represent from the first k Interference from other D2D users and CUE users on the first channel; d The additive white Gaussian noise at each DR follows: ;

[0043] Step 18: Therefore, the first k The first channel d The achievable rate of each DR is:

[0044]

[0045] Step 19: Finally, calculate the... k The sum rate of all DRs in each channel is:

[0046]

[0047] Step 20: Formulate the optimization problem as the following non-convex optimization problem:

[0048]

[0049] in , Indicates the first Use index vectors for D2D channels. This represents the transmit power of D DTs; This indicates the maximum interference limit caused to CUE. This represents the minimum SINR threshold of CUE. This represents the minimum achievable rate threshold for D2D users;

[0050] Step 21: Solve the above optimization problem using a distributed deep learning algorithm based on unsupervised learning; the specific implementation steps are as follows:

[0051] Step 22: To prevent the channel gain from being too large or too small and affecting the training results, a function should be used. The data is normalized, and the preprocessed channel gain is denoted as... ,Right now

[0052]

[0053] Step 23: Place the first All CSIs contained in a D2D are denoted as After data preprocessing, the preprocessed CSI is fed into the DNN network for CSI feedback, compressing it into... Bits, and together with the sigmoid layer, approximate a function that encodes the local CSI. Determine the CSI to be sent to the base station, denoted as ,Right now and ;

[0054] Step 24: BS will collect all local CSI data. and its own pre-processed CSI The data is fed into the DNN network for CSI feedback and compressed into... Bits, and together with the sigmoid layer, approximate a function that encodes the local CSI. ,Sure And broadcast it to all D2D pairs, i.e. ;

[0055] Step 25: By minimizing the loss function of the neural network designed in Steps 22 and 23, the maximum sum rate of D2D users is obtained; the custom neural network loss function is:

[0056]

[0057] in, The function to eliminate binarization error:

[0058]

[0059]

[0060] in, These are network hyperparameters.

[0061] A further technical solution of the present invention: The DNN network described in steps 23 and 24 has three layers, wherein the first layer contains 256 neurons, and the second and third layers each contain 128 neurons.

[0062] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0063] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0064] The beneficial effects of this invention are as follows:

[0065] This invention provides a smart reflector-assisted D2D communication optimization method that solves optimization problems in wireless communication systems by training a deep neural network (DNN). It finds feasible solutions using a distributed deep learning algorithm, without employing complex mathematical formulas and numerical optimization methods. This invention has the following three advantages:

[0066] First, a RIS-assisted D2D communication system model is proposed. Due to the flexibility and portability of UAVs, UAVs can improve the limitations of poor mobility of traditional base stations by serving as aerial base stations. The use of intelligent reflectors can reduce interference caused by LOS links and improve channel quality.

[0067] Secondly, the optimization scheme based on distributed deep learning (DL) can achieve near-optimal performance without solving complex optimization problems. The proposed distributed DL algorithm determines the optimal sharing strategy for local CSI, namely the optimal local coding strategy from a single user to the BS and the optimal coding strategy for the local CSI collected from the BS to the user. The proposed algorithm can handle the discrete situations commonly encountered in real-world systems, and for more complex multi-channel resource allocation problems, it employs more sophisticated CSI sharing strategies.

[0068] Finally, the computation time required for the trained DNN to obtain the optimal policy is much lower than that of traditional resource allocation schemes because the DNN only performs simple matrix operations. Moreover, the proposed algorithm determines the channel selection and resource allocation at the D2D user end and the optimal RIS phase at the UAV end, reducing the computational load on the UAV and thus reducing its energy consumption. Attached Figure Description

[0069] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0070] Figure 1 A coordinate map showing the locations of the drone, user, and RIS;

[0071] Figure 2 The diagram shows the principle of the proposed distributed deep learning algorithm. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0073] This invention provides a smart reflective surface-assisted D2D communication optimization method that uses an unmanned aerial vehicle (UAV) as an aerial base station (BS).

[0074] The processing steps at the drone base station transmitter are as follows:

[0075] Step 1: The transmitting end includes a drone base station, a RIS with M reflective elements, and K channels. Each channel contains one cellular user (CUE), meaning there are K cellular users and D D2D users, where multiple D2D users reuse the spectrum occupied by the cellular users. Furthermore, using... This indicates whether the d-th D2D user uses the k-th channel; that is, if the transmitter (DT) of the d-th D2D user transmits data through the k-th channel, then... ,otherwise Assume that each D2D user can only share spectrum with one cellular user at a time, i.e. Assuming all transceivers use a single antenna, the drone base station location is... The RIS position is D2D user pairs and cellular users are randomly distributed within a circle of radius R. l The location of each cellular user is , No. d Each DT position is denoted as The maximum distance between DT and the corresponding receiver DR is , No. d Each DR position is denoted as .

[0076] Furthermore, assuming the RIS is equipped with One reflecting element, reflection coefficient matrix Represented as The first of this diagonal matrix Each element is represented as ,in Indicates the first Phase shift of each reflecting element, Indicates the first The amplitude of each reflecting element. In practice, the phase shift of each element can only take a finite number of discrete values, the first... Each discrete value is represented as ,in , Indicates the level used for quantizing phase shift. The number of bits. In this invention, it is assumed that the RIS performs ideal reflection, thus the signal power of each reflecting element is lossless, i.e., the amplitude reflection coefficient. Meanwhile, assuming DT The transmit power is expressed as Its maximum value is And can be quantified as A walk of a distance, that is .

[0077] Step 2: Generation of drone base station X bit information ;

[0078] Step 3: Modulate the information generated in Step 2. The modulated signal is... and satisfy ;

[0079] Step 4: Transfer the signal generated in step 3 s Send into the channel;

[0080] The steps for establishing a drone-to-user channel are as follows:

[0081] Step 5: When the signal transmitted by the drone base station propagates in free space without the obstruction of tall buildings, the drone can directly transmit the signal, thus forming a line-of-sight (LOS) link. As the signal continues to propagate, there is obstruction between the drone base station and the ground user, and it is also affected by electromagnetic wave reflection and scattering, thus forming a non-line-of-sight (NLOS) link.

[0082] Step 6: LOS and NLOS in the channel gain occur with a certain probability. Assume the ground node coordinates are... The coordinates of the UAV base station are The distance between the drone and the node is Then the LOS component probability of the node is:

[0083]

[0084] Where A and B are constants related to the environment. The NLOS component probabilities are: .

[0085] Step 7: Drone to User d The channel gain is:

[0086]

[0087] in It is the path loss coefficient between the user and the drone link. It is the correlation coefficient with the NLOS link.

[0088] Step 8: Drones and the first l The distance between each CUE is ,Will Substitute into step 6 and In the process, the Rayleigh distribution coefficients are randomly generated. If the channel gain is: .

[0089] Step 9: The distance between the drone and the RIS is Ignoring the differences in RIS position, Substitute into step 6 and In the process, the Rayleigh distribution coefficients are randomly generated. Then the channel gain is: .

[0090] Step 10: Drones and the first d The distance between each DR is ,Will Substitute into step 6 and In the process, the Rayleigh distribution coefficients are randomly generated. Then the channel gain is: .

[0091] Other link channel construction and processing steps are as follows:

[0092] Step 11: Considering the random Rayleigh distribution and path loss for other link channel gains, the path loss is:

[0093]

[0094] in It is the distance between nodes. These represent the path loss and path loss coefficient at 1m, respectively.

[0095] Step 12: RIS and the first l The distance between each CUE is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: .

[0096] Step 13: RIS and the first d The distance between each DR is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: . No. d The distance between DT and RIS is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: .

[0097] Step 14: l The first CUE and the second d The distance between each DR is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: .

[0098] Step 15: d The first DT and the first The distance between DR users is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Let the loss matrix between DT and DR be denoted as ,in No. d Line number Column elements are denoted as Then the channel gain is: , of which d The channel gain between DT and DR is denoted as , No. d The first DT and the first The channel gain between the DRs is denoted as .

[0099] The processing steps at the receiving end are as follows:

[0100] Step 16: k The signal-to-interference-plus-noise ratio (SINR) received by the CUE in each channel is:

[0101]

[0102] in This indicates the transmission power of the UAV. For the first d The transmit power of the DT, the first The additive white Gaussian noise at CUE in each channel follows the following rules: .

[0103] Step 17: k The first channel d The SINR received by each DR is:

[0104]

[0105] in and They represent from the first k Interference from other D2D users and CUE users on each channel. d The additive white Gaussian noise at each DR follows: .

[0106] Step 18: Therefore, the first k The first channel d The achievable rate of each DR is:

[0107]

[0108] Step 19: Finally, the number can be calculated. k The sum rate of all DRs in each channel is:

[0109]

[0110] Step 20: To improve the sum rate of D2D users while limiting interference to CUE, an optimization algorithm is needed to perform optimal phase design for RIS, power allocation at the D2D transmitter, and channel selection. The optimization problem can be formulated as the following non-convex optimization problem:

[0111]

[0112] in , Indicates the first Use index vectors for D2D channels. This represents the transmission power of D DTs. This indicates the maximum interference limit caused to CUE. This represents the minimum SINR threshold of CUE. This represents the minimum reachable rate threshold for D2D users.

[0113] Step 21: The optimization problem in the above equation is a non-convex mixed integer programming problem. The performance of the suboptimal solution obtained by mathematical iterative optimization methods cannot be guaranteed, and it usually has high complexity. Therefore, this invention proposes a distributed deep learning algorithm based on unsupervised learning.

[0114] The proposed distributed deep learning algorithm is as follows: Figure 2 As shown, the specific implementation steps are as follows:

[0115] Step 22: To prevent the channel gain from being too large or too small and affecting the training results, a function should be used. The data is normalized, and the preprocessed channel gain is denoted as... ,Right now

[0116]

[0117] Step 23: Place the first All CSIs contained in a D2D are denoted as After data preprocessing, the preprocessed CSI is fed into a DNN network (a three-layer network, with the first layer containing 256 neurons and the second and third layers each containing 128 neurons) for CSI feedback (CFM), compressing it into... Bits, and together with the sigmoid layer, approximate a function that encodes the local CSI. Determine the CSI to be sent to the base station, denoted as ,Right now and .

[0118] Step 24: BS will collect all local CSI data. and its own pre-processed CSI The data is fed into a DNN network (which has three layers, with the first layer containing 256 neurons and the second and third layers each containing 128 neurons) for CSI feedback, compressing it into... Bits, and together with the sigmoid layer, approximate a function that encodes the local CSI. ,Sure And broadcast it to all D2D pairs, i.e. .

[0119] Step 25: By minimizing the loss function of the neural network designed in Steps 22 and 23, the maximum sum rate of D2D users is obtained. The custom neural network loss function is:

[0120] in, The function to eliminate binarization error:

[0121]

[0122]

[0123] in, These are network hyperparameters.

[0124] Example 1:

[0125] The processing steps at the drone base station transmitter are as follows:

[0126] Step 1: The transmitting end includes a drone base station, a RIS with 4 reflectors, 2 channels, and 3 D2D users. All equipment is equipped with a single antenna. The drone base station is located at... m, RIS position is m, cellular users (CUE) and D2D users are randomly distributed within a circle with a radius of 10m, and the position of the first D2D transmitter (DT) is denoted as m. m, the distance from the corresponding receiver DR is 5m, and the position of the kth CUE is The base station's maximum transmit power is 30W, and the DT's maximum transmit power is... noise variance In addition, power quantization level Number of phase quantization bits ,but Path loss index between user and drone link Correlation coefficient with NLOS link Path loss coefficient at 1m in other links The path loss index under this link .

[0127] Step 2: The drone base station generates 128 bits of information;

[0128] Step 3: Modulate the information generated in Step 2. The modulated signal is... and satisfy ;

[0129] Step 4: Transfer the signal generated in step 3 s Send into the channel;

[0130] The signals transmitted by the drone base station will form LOS and NLOS links. The drone-to-user channel construction process in the k-th channel is as follows:

[0131] Step 5: LOS and NLOS in the channel gain occur with a certain probability. Assume the ground node coordinates are... The coordinates of the UAV base station are The distance between the drone and the node is Then the LOS component probability of the node is:

[0132]

[0133] in The NLOS component probabilities of a node are: .

[0134] Step 6: Path loss coefficient between user and drone link Correlation coefficient with NLOS link Then the drone will reach the user. i The channel gain is:

[0135]

[0136] Step 7: The distance between the drone and the CUE in the k-th channel is ,but , , Randomly generate Rayleigh distribution coefficients Then the channel gain is: .

[0137] Step 8: The distance between the drone and the RIS is If we ignore the differences in RIS position, then , , Randomly generate Rayleigh distribution coefficients Then the channel gain is: .

[0138] Step 9: Drones and the first d The distance between D2D users is ,but , , Randomly generate Rayleigh distribution coefficients Then the channel gain is: .

[0139] The other link channel construction and processing steps are as follows:

[0140] Step 10: Path loss coefficient at 1m The path loss index under this link For other link channel gains, considering a random Rayleigh distribution and path loss, the path loss is:

[0141]

[0142] Step 11: The distance between RIS and the CUE of the k-th channel is ,but Randomly generate Rayleigh distribution coefficients Then the channel gain is: .

[0143] Step 12: RIS and the first k The distance between the d-th DR in each channel is ,but Randomly generate Rayleigh distribution coefficients Then the channel gain is: . No. k The distance between the d-th DT and RIS in each channel is ,but Randomly generate Rayleigh distribution coefficients Then the channel gain is: .

[0144] Step 13: The CUE in the k-th channel and the... d The distance between DTs is ,but Randomly generate Rayleigh distribution coefficients Then the channel gain is: .

[0145] Step 14: d The first DT and the first The distance between DR users is ,use Let the loss matrix be denoted as , then the ... d OK Column elements Randomly generate Rayleigh distribution coefficients Then the channel gain is: , of which d The channel gain between DT and DR is denoted as , No. d The first DT and the first The channel gain between the DRs is denoted as .

[0146] The processing steps at the receiving end are as follows:

[0147] Step 15: k The SINR received by CUE in each channel is:

[0148]

[0149] Step 16: k The first channel d The SINR received by each DR is:

[0150]

[0151] Step 17: d The achievable rate of each DR is:

[0152]

[0153] Step 18: k The sum rate of all DRs for each channel is:

[0154]

[0155] Step 19: Minimum SINR threshold for CUE Cellular users can reach the minimum data rate threshold Therefore, resource allocation is formulated as a nonconvex optimization problem as follows:

[0156]

[0157] Step 20: This invention proposes a deep learning algorithm based on unsupervised learning to solve this optimization problem. The specific optimization process is as follows: Figure 2 As shown. The entire training process consists of 1000 episodes. The algorithm inputs the current channel information. To prevent the channel gain from being too large or too small and affecting the training results, it should be controlled through a function. The data is normalized, and the preprocessed channel gain is denoted as... ,Right now

[0158]

[0159] The specific implementation steps of the proposed distributed deep learning algorithm are as follows:

[0160] Step 21: D2D All included CSIs are denoted as After data preprocessing, it is fed into a DNN network (which has three layers, with the first layer containing 256 neurons and the second and third layers each containing 128 neurons) for CSI feedback, compressing it into... Bits, and together with the sigmoid layer, approximate a function that encodes the local CSI. The CSI to be sent to the BS is specified as follows: ,Right now and .

[0161] Step 22: BS will collect all local CSI data. and its own pre-processed CSI The data is fed into the basic DNN module for CSI feedback and compressed into... Bits, and together with the sigmoid layer, approximate a function that encodes the local CSI. ,Sure And broadcast it to all D2D pairs, i.e. .

[0162] Step 23: Obtain the maximum sum rate of D2D users by minimizing the loss function of the entire neural network. (Settings) Then the custom neural network loss function is:

[0163]

[0164] in, The function to eliminate binarization error:

[0165]

[0166]

[0167] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A smart reflective surface-assisted D2D communication optimization method, characterized in that: The processing steps at the drone base station transmitter are as follows: Step 1: The transmitting end includes a drone base station, a RIS with M reflective elements, and K channels. Each channel contains one cellular user (CUE), meaning there are K cellular users and D D2D users, where multiple D2D users reuse the spectrum occupied by the cellular users; in addition, using... This indicates whether the d-th D2D user uses the k-th channel; that is, if the transmitter of the d-th D2D user transmits data through the k-th channel, then... ,otherwise Assume that each D2D user can only share spectrum with one cellular user at a time, i.e. Assuming all transceivers use a single antenna, the drone base station location is... The RIS position is D2D user pairs and cellular users are randomly distributed within a circle of radius R. l The location of each cellular user is , No. d Each DT position is denoted as The maximum distance between DT and the corresponding receiver DR is , No. d Each DR position is denoted as ; Furthermore, assuming the RIS is equipped with One reflecting element, reflection coefficient matrix Represented as The first of this diagonal matrix Each element is represented as ,in Indicates the first The phase shift of each reflecting element, Indicates the first The amplitude of each reflecting element; in practice, the phase shift of each element can only take a finite number of discrete values, the first... Each discrete value is represented as ,in , Indicates the level used for quantizing phase shift. The number of bits; assuming RIS performs ideal reflection, so that the signal power of each reflecting element is lossless, i.e., the amplitude reflection coefficient. Meanwhile, assuming DT The transmit power is expressed as Its maximum value is and quantified as A walk of a distance, that is ; Step 2: Generation of drone base station X bit information ; Step 3: Modulate the information generated in Step 2. The modulated signal is... and satisfy ; Step 4: Transfer the signal generated in step 3 s Send into the channel; The steps for establishing a drone-to-user channel are as follows: Step 5: When the signal transmitted by the drone base station propagates in free space without the obstruction of tall buildings, the drone can directly transmit the signal, thus forming a line-of-sight (LOS) link. As the signal continues to propagate, there is an obstruction between the drone base station and the ground user, and it is also affected by electromagnetic wave reflection and scattering, thus forming a non-line-of-sight (NLOS) link. Step 6: LOS and NLOS in the channel gain occur according to probability. Assume the ground node coordinates are... The coordinates of the UAV base station are The distance between the drone and the node is Then the LOS component probability of the node is: Where A and B are constants related to the environment. The NLOS component probabilities are: ; Step 7: Drone to User d The channel gain is: in It is the path loss coefficient between the user and the drone link. It is the correlation coefficient with the NLOS link; Step 8: Drones and the first l The distance between each CUE is ,Will Substitute into step 6 and In the process, the Rayleigh distribution coefficient is randomly generated. If the channel gain is: ; Step 9: The distance between the drone and the RIS is Ignoring the differences in RIS position, Substitute into step 6 and In the process, the Rayleigh distribution coefficient is randomly generated. Then the channel gain is: ; Step 10: Drones and the first d The distance between each DR is ,Will Substitute into step 6 and In the process, the Rayleigh distribution coefficient is randomly generated. Then the channel gain is: ; The other link channel construction and processing steps are as follows: Step 11: Considering the random Rayleigh distribution and path loss for other link channel gains, the path loss is: in It is the distance between nodes. These represent the path loss and path loss coefficient at 1m, respectively. Step 12: RIS and the first l The distance between each CUE is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ; Step 13: RIS and the first d The distance between each DR is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ;No. d The distance between DT and RIS is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ; Step 14: l The first CUE and the second d The distance between each DR is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Then the channel gain is: ; Step 15: d The first DT and the first The distance between DR users is ,Will Substitute the values ​​into the loss function from step 11 and randomly generate the Rayleigh distribution coefficients. Let the loss matrix between DT and DR be denoted as ,in No. d Line number Column elements are denoted as Then the channel gain is: , of which d The channel gain between DT and DR is denoted as , No. d The first DT and the first The channel gain between the DRs is denoted as ; The processing steps at the receiving end are as follows: Step 16: k The signal-to-interference-plus-noise ratio (SINR) received by CUE in each channel is: in Indicates the UAV's transmission power. For the first d The transmit power of the DT, the first The additive white Gaussian noise at CUE in each channel follows the following rules: ; Step 17: k The first channel d The SINR received by each DR is: in and They represent from the first k Interference from other D2D users and CUE users on the first channel; d The additive white Gaussian noise at each DR follows: ; Step 18: Therefore, the first k The first channel d The achievable rate of each DR is: Step 19: Finally, calculate the... k The sum rate of all DRs in each channel is: Step 20: Formulate the optimization problem as the following non-convex optimization problem: in , Indicates the first Use index vectors for D2D channels. This represents the transmit power of D DTs; This indicates the maximum interference limit caused to CUE. This represents the minimum SINR threshold of CUE. This represents the minimum achievable rate threshold for D2D users; Step 21: Solve the above optimization problem using a distributed deep learning algorithm based on unsupervised learning; the specific implementation steps are as follows: Step 22: To prevent the channel gain from being too large or too small and affecting the training results, a function should be used. The data is normalized, and the preprocessed channel gain is denoted as... ,Right now Step 23: Place the first All CSIs contained in a D2D are denoted as After data preprocessing, the preprocessed CSI is fed into the DNN network for CSI feedback, compressing it into... Bits, together with the sigmoid layer, constitute the function that encodes the local CSI. Determine the CSI to be sent to the base station, denoted as ,Right now and ; Step 24: BS will collect all local CSI data. and its own pre-processed CSI The data is fed into the DNN network for CSI feedback and compressed into... Bits, together with the sigmoid layer, constitute the function that encodes the local CSI. ,Sure And broadcast it to all D2D pairs, i.e. ; Step 25: By minimizing the loss function of the neural network designed in Steps 22 and 23, the maximum sum rate of D2D users is obtained; the custom neural network loss function is: in, The function to eliminate binarization error: in, These are network hyperparameters.

2. The intelligent reflective surface-assisted D2D communication optimization method according to claim 1, characterized in that... The DNN network described in steps 23 and 24 has three layers, with the first layer containing 256 neurons and the second and third layers each containing 128 neurons.

3. A computer system, characterized in that... include: One or more processors, a computer-readable storage medium for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of claim 1.

Images