CN116669157A - A distributed high-efficiency power control method in a direct-to-device network - Google Patents

Abstract

The invention discloses a distributed high-energy-efficiency power control method in a D2D network, which belongs to the technical field of resource allocation in D2D communication. The control object of the present invention is a device direct network, including N D2D user pairs and K resource blocks, and each device has independent sending and receiving functions, and is connected to an adjacent single device through a D2D link for data communication. The present invention is based on a distributed high-energy-efficiency power control method in a direct-to-device network, establishes a direct-to-device network model and constructs an optimization problem aimed at maximizing network energy efficiency; adopts a continuous convex approximation SCA algorithm to obtain an optimal power control scheme, which is effective Improve the energy efficiency of the D2D network and reduce the complexity of calculation; combine the alternating direction multiplier method ADMM to distribute the calculation amount to each device, reduce the base station load and calculation overhead, and users can allocate resources more flexibly and improve communication performance.

Description

A distributed high-efficiency power control method in a device-to-device network

Technical Field

The present invention belongs to the technical field of resource allocation in device-to-device (D2D) communication, and relates to a distributed high-energy-efficiency power control method in a device-to-device network.

Background Art

In recent years, with the rapid development of science and technology, the widespread popularity of electronic devices such as smart phones has stimulated the explosive growth of high-speed wireless multimedia services. Users' demands for business volume, business types and service quality are constantly increasing, which poses greater challenges to the data transmission rate, business coverage and data transmission method selection of wireless communication systems. Device-to-device network D2D, as a communication technology that can effectively support the realization of the above requirements, is also considered to be one of the key technologies in 3GPP LTE-Advanced. In the D2D communication mode, since users can achieve short-distance direct communication, the channel quality is relatively higher and the loss during data transmission is smaller, D2D communication can achieve higher data transmission rate, lower power consumption and lower latency; and the base station can further improve coverage by controlling widely distributed user terminals, realize efficient use of spectrum resources, support more flexible network architecture and connection methods, and improve link flexibility and network reliability.

However, when using D2D communication technology, the same resources shared between two types of users will inevitably cause interference and performance loss, so it is necessary to perform reasonable power control on the D2D network to fully improve network energy efficiency. However, traditional centralized power control methods usually rely on central control nodes for calculations. The central base station collects and obtains channel information from each user to allocate resources between them. Its algorithm mechanism and computing performance have high requirements for the network infrastructure structure. Therefore, the overhead required for the operation and control of the base station will be very huge, and when the network scale is large, the computing cost will also increase significantly.

Summary of the invention

The main purpose of the present invention is to provide a distributed high-energy-efficiency power control method in a device-to-device network. Aiming at the problem of interference between user communications in a D2D network, based on the distributed high-energy-efficiency power control method in a device-to-device network, a device-to-device network model is established and an optimization problem with the goal of maximizing network energy efficiency is constructed; a continuous convex approximation SCA algorithm is used to obtain the optimal power control scheme, which effectively improves the energy efficiency of the D2D network and reduces the complexity of the calculation; the alternating direction multiplier method ADMM is combined to disperse the computational workload to each device, reducing the base station load and computational overhead, and users can allocate resources more flexibly to improve communication performance.

The objective of the present invention is achieved through the following technical solutions:

The present invention discloses a distributed high-energy-efficiency power control method in a device direct network, comprising the following steps:

Step 1: Construct a device-to-device network model. The distributed high-efficiency power control object is the device-to-device network. The device-to-device network includes N D2D user pairs, K resource blocks, and each device has independent sending and receiving functions. It is connected to an adjacent single device through a D2D link for data communication to reduce energy loss; construct a device-to-device network power loss model, considering the power loss of the RF power amplifier and the static circuit power loss caused by signal processing and active circuit blocks; construct a device-to-device network communication power control model, and constrain the transmission power and throughput to meet the channel bandwidth limitation and ensure user service quality requirements.

Step 1.1: Construct a device-to-device network model, in which the communication spectrum resources are provided in the form of resource blocks, and the D2D user pair set and the resource block set are respectively expressed as and Assume that _pn,k is the transmission power of the nth pair of D2D users when occupying the kth resource block, hn _,n,k and hn _,n′,k represent the channel coefficients from the receiving end of the nth link to the transmitting end of the nth link and the transmitting end of the n′th link respectively when using the kth resource block. Then, when the user equipment transmits data on the nth link using the kth resource block, the signal-to-interference-noise ratio γn _,k at the receiving end is expressed as:

Where σ ² is the variance of zero-mean additive white Gaussian noise. The maximum amount of data transmitted by the nth user is:

Where B is the bandwidth of a resource block. Then the total throughput R _tot of the device direct network is calculated according to formula (3):

Step 1.2: Construct a device-through network power loss model, taking into account the power loss of the RF power amplifier and the static circuit power loss caused by signal processing and active circuit blocks. The power loss of the RF power amplifier is closely related to the efficiency of the amplifier, given is the inverse of the power amplifier drain efficiency. The power loss P _m of the RF power amplifier is calculated according to formula (4):

Static circuit power loss is caused by signal processing and active circuit blocks, which require a certain amount of current to maintain their normal operation. The average value of the circuit power consumption is represented by the static variable _Pc , and the total circuit power consumption _Ps is expressed as:

_Ps ＝ _NPc (5)

Therefore, the device direct network power loss model is constructed as shown in formula (6).

P _tot = P _m + P _s (6)

Step 1.3: Construct a device-to-device network communication power control model. According to the device-to-device network total throughput R _tot and power loss P _tot obtained in steps 1.2 and 1.3, the device-to-device network energy efficiency η is calculated by formula (7):

Taking the device-to-device network communication power control model shown in formula (8) as the optimization objective and the channel bandwidth and throughput requirement lower limit as constraints, the distributed high-efficiency power control optimization problem of device-to-device network is constructed.

Wherein, the transmit power vector is represented by p = [p ₁ ,p ₂ , ..., p _N ] ^T , and p _n = [p _n,1 ,p _n,2 , ...,p _n,K ]; and They represent the maximum allowed total transmit power and minimum throughput of the nth user respectively, to ensure the communication service quality requirements of each user.

Step 2: Based on the device direct network communication power control model constructed in step 1, the properties of logarithmic functions are used to reconstruct the device direct network communication power control optimization problem into the difference form of two convex functions. The continuous convex approximation algorithm is used to approximate the reconstructed device direct network communication power control optimization problem into a continuous and differentiable convex function, thereby converting the optimization problem into a problem of solving a continuous convex function. Compared with directly solving the constructed non-convex optimization problem of device direct network communication power control, the computational difficulty is greatly reduced.

Step 2.1: To make the expression of the device-to-device network communication power control optimization problem more concise, let Using the properties of the logarithmic function in formula (2), the maximum transmission data volume _Rn of the nth user can be expressed as the following subtraction form:

R _n = _gn (p) _-hn (p) (9)

in,

The total throughput of the device direct network R _tot is reconstructed as:

Step 2.2: By introducing variables w = 1/P _tot and θ _n,k = wp _n,k , and θ = {θ _n,k }, the device direct network communication power control model constructed in step 1 is transformed into a device direct network communication power control model with a convex difference form as shown in formula (11):

Since g _n (p), h _n (p), g(p) and h(p) are all concave functions, the maximum transmission data volume R _n of the nth user and the total throughput R _tot of the device-to-device network are both differential forms of the two concave functions, thus proving that the transformed device-to-device network communication power control model The objective function of is a convex difference function, and its constraints all satisfy the convex difference constraints, so the converted device direct network communication power control model It is a convex difference programming problem and the power control model of direct network communication between devices The reconstructed device-to-device network communication power control optimization problem is approximated as a continuous and differentiable convex function using the continuous convex approximation SCA algorithm, thereby transforming the optimization problem into a problem of solving a continuous convex function to obtain a solution that satisfies the Karush-Kuhn-Tucker (KKT) condition of the device-to-device network communication power control model. In the tth iteration, the approximate convex function subproblem expression is:

Among them, θ ^(t) , and w ^(t) represent the optimal solutions θ, θn _,k and w of the t-th subproblem respectively.

Step 3: The objective function and constraints of the convex function subproblem approximated in the tth iteration are inseparable, so the alternating direction multiplier method ADMM cannot be directly applied to solve it. It is necessary to transform the approximate convex function subproblem and then construct the Lagrangian function of the approximate convex function subproblem to facilitate further solution.

Step 3.1: Introduce a new variable I _n,n′,k =G _n,n′,k θ _n′,k , Define auxiliary variables w _n , and where w _n , and are the local copies of w, In _,n′,k and In _′,n,k on device n, respectively, and specify The feasible set of constructing local variables in the tth round of iteration subproblems is:

in, and The optimal solution of the t-th round SCA subproblem and w ^(t) are calculated. Definition l＝{l _n }，ξ＝{ξ _n }， _Included The global variables corresponding to the elements in the approximation of the convex function subproblem Make the following equivalent transformations:

Step 3.2: Convex function subproblem obtained by equivalent transformation The augmented Lagrangian function of is shown in formula (15):

Where α＝{α _n }, β and λ＝{λ _n } are Lagrange multiplier vectors, and ρ>0 is called the penalty coefficient, which determines the size of the update step. and ξ _n properties, λ _n contains the Lagrange multiplier and Corresponding variables and w _n .

Step 4: Using the alternating direction multiplier method ADMM, local variables, global variables and Lagrange multipliers are updated in sequence to achieve the ability to solve optimization problems in a distributed environment. The penalty coefficient is dynamically updated, and the value of the penalty coefficient is adaptively adjusted according to the results of each round of iteration to further improve the convergence performance of the ADMM algorithm. The update of local variables is independent of other nodes and is calculated in parallel locally; global variables are used for information exchange and coordination between nodes; Lagrange multipliers are used to process constraints. Through alternating updates, the distributed high-efficiency power control algorithm can gradually approach the optimal solution and achieve efficient optimization solutions in a distributed environment.

Step 4.1: Initialize variables Initialize the SCA subproblem iteration number t in step 2;

Step 4.2: Initialize ADMM iteration round j;

Step 4.3: Update local variables Local variables are local variables on each distributed node. Each node is responsible for solving the sub-problem related to itself and performing local optimization according to the characteristics of the problem. The update of local variables is performed without considering the information of other nodes. Each node independently solves its own optimal solution and performs calculations locally in parallel. Update local variables according to formula (16):

Step 4.4: Update the global variables {ξ ^(j+1) , l ^(j+1) }. Global variables are variables shared among all nodes and are used to store information exchange and coordination between nodes. In the ADMM algorithm, global variables are usually used to pass shared information about constraints and problems between nodes. The update of global variables is achieved through communication and information exchange between nodes. Using the local variables obtained in step 4.3, update the global variables in the following manner:

Since formula (17) is an unconstrained quadratic optimization problem, the closed-form solution of the global variable ξ ^(j+1) is calculated according to formula (19) and formula (20):

In order to derive the closed-form solution of l ^(j+1) as shown in formula (18), the global variable l ^(j+1) is transformed into the form of a square term as shown in formula (21):

in,

By setting the first-order derivative of formula (21) to zero, the closed-form solution of the global variable l ^(j+1) is written as:

Step 4.5: Update the Lagrange multiplier Lagrange multipliers are used to introduce constraints into the optimization problem and gradually adjust their own values through alternating updates. Each node has a corresponding Lagrange multiplier, which is used to process the constraints related to the node. Through the update of Lagrange multipliers, the ADMM algorithm exchanges and coordinates information between the global and local to obtain the optimal result. The update rules of Lagrange multipliers are shown in formulas (23)-(25):

Step 4.6: Update the penalty coefficient according to formula (26):

Among them, μ>0, τ ^incr >1 and τ ^decr >1 can be selected in combination with different structures and complexities of the problem by adaptively adjusting the size of the penalty coefficient according to the result of the current iteration.

Step 4.7: Convergence condition judgment, if the ADMM convergence condition is met (∈ is a very small positive number): Then continue to step 4.8; otherwise, let j = j + 1 and jump to step 4.3 to continue updating and iterating;

Step 4.8: If |θ ^(t+1) -θ ^(t) |≤ε or t≥T is satisfied, where ε is the convergence error and T is the maximum number of iterations, the algorithm terminates; otherwise, continue with step 4.2.

Through the above distributed high-efficiency power control algorithm steps, the converted device direct network communication power control model is finally obtained. The local KKT point, that is, {θ ^* ,w ^* }

Step 5: Based on the optimal solution {θ ^* , w ^* } obtained in step 4, p ^* = θ ^* / w ^* is calculated to finally obtain a power control result that optimizes the network energy efficiency EE.

It also includes step six: according to the distributed high-efficiency power control optimization result of the device direct network obtained in step five, the optimal communication power allocation in the D2D network is realized, which can reduce the energy consumption of the device during the communication process, improve the communication quality between devices, and enhance the signal coverage, thereby reducing resource loss, improving network capacity and performance, and providing users with a more reliable communication experience.

Beneficial effects:

1. The present invention discloses a distributed high-energy-efficiency power control method in a device-to-device network, constructs a device-to-device network communication power control model based on a D2D network, considers the maximum allowed total transmission power and minimum throughput constraints, and selects the optimal power control scheme to improve network energy efficiency while meeting user service quality requirements.

2. The present invention discloses a distributed high-energy-efficiency power control method in a device direct network. The continuous convex approximation SCA algorithm is used to iteratively approximate the constructed device direct network communication power control problem by utilizing sub-problems in a standard convex optimization form, thereby reducing the difficulty and complexity of the calculation, making the calculation of the high-energy-efficiency power control scheme in the device direct network more efficient.

3. The present invention discloses a distributed high-energy-efficiency power control method in a device direct network. Through the ADMM algorithm, tasks or data are distributed to multiple nodes for execution. Each node only needs to deal with local sub-problems, and calculations can be performed in parallel, which can effectively improve computing efficiency and reduce the computing burden of the central control unit. Even if a node fails or fails, the entire system can continue to operate. In contrast, centralized algorithms impose too much computing burden on central nodes, and have poor processing capabilities for sudden errors. A single node failure may cause the entire system to crash. The present invention discloses a distributed high-energy-efficiency power control method in a device direct network with higher reliability, making it more robust in the face of hardware failures, network interruptions or other abnormal situations.

4. In traditional ADMM algorithms, the penalty coefficient is usually fixed and does not change with the iteration process. This may cause the algorithm to converge slowly or fail to converge in some cases. This is because when the constraints of the original problem are strong, the fixed penalty coefficient may cause the solution to the dual problem to be inaccurate, thereby affecting the convergence of the entire algorithm. In contrast, the distributed high-efficiency power control method in the device direct network disclosed in the present invention adopts a dynamically updated penalty coefficient and achieves convergence with fewer iterations. Dynamic value better balances the relationship between the original problem and the dual problem by adaptively adjusting the size of the penalty coefficient according to the result of the current iteration, and can adapt to different structures and complexities of the problem, so that the distributed algorithm can converge faster, and can improve the convergence performance of the model compared to fixed value.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solutions in the specific implementation manner of the present invention, the drawings required for describing the embodiments are briefly introduced below.

FIG1 is a flow chart of a distributed high energy efficiency power control method in a device direct access network according to the present invention.

FIG2 is a schematic diagram of a device-to-device network communication system model;

3 is a comparison diagram of the simulation convergence curves of the ADMM algorithm using the dynamic update penalty coefficient and the fixed value penalty coefficient in combination with Example 1 of the present invention "A distributed high-efficiency power control method in a device direct access network";

FIG. 4 is a schematic diagram showing a comparison of the results of using the distributed method and the centralized power control method of the present invention in Example 1.

DETAILED DESCRIPTION

In order to make the purpose and technical solution of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

Embodiment 1:

The present invention provides a distributed high-energy-efficiency power control method in a device-to-device network. Consider the D2D communication scenario shown in Figure 1, where each device is connected to a single adjacent device via a D2D link for data communication. Assume that the D2D pairs are evenly distributed in the network, and the distance between D2D users is 10m, the base station coverage radius is 25m, the path loss index is 3.8, the carrier frequency is 2GHz, the reference distance is 100m, the subcarrier bandwidth is 180kHz, the noise power spectrum density is -174dBm/Hz, the shadow variance is 8dB, and the shadow fading obeys a log-normal distribution. The fixed value ρ and the dynamic initial value ρ ⁽⁰⁾ are both set to 1.1. The drain efficiency and average circuit power consumption P _c of the power amplifier are 0.38 and 10W respectively. Aiming at the problem of excessive load and high cost of the central unit caused by the centralized algorithm, the present invention proposes a distributed algorithm based on the combination of ADMM and SCA in the D2D communication background with the goal of maximizing network EE. The specific method includes the following steps:

Step 1: Construct a device-to-device network model. The distributed high-efficiency power control object is the device-to-device network. The device-to-device network includes N D2D user pairs, K resource blocks, and each device has independent sending and receiving functions. It is connected to an adjacent single device through a D2D link for data communication to reduce energy loss; construct a device-to-device network power loss model, considering the power loss of the RF power amplifier and the static circuit power loss caused by signal processing and active circuit blocks; construct a device-to-device network communication power control model, and constrain the transmission power and throughput to meet the channel bandwidth limitation and ensure user service quality requirements.

Step 1.1: Construct a device-to-device network model, in which the communication spectrum resources are provided in the form of resource blocks, and the D2D user pair set and the resource block set are respectively expressed as and Assume that _pn,k is the transmission power of the nth pair of D2D users when occupying the kth resource block, hn _,n,k and hn _,n′,k represent the channel coefficients from the receiving end of the nth link to the transmitting end of the nth link and the transmitting end of the n′th link respectively when using the kth resource block. Then, when the user equipment transmits data on the nth link using the kth resource block, the signal-to-interference-noise ratio γn _,k at the receiving end is expressed as:

Where σ ² is the variance of zero-mean additive white Gaussian noise. The maximum amount of data transmitted by the nth user is:

Where B is the bandwidth of a resource block. Then the total throughput R _tot of the device direct network is calculated according to formula (3):

Step 1.2: Construct a device-through network power loss model, taking into account the power loss of the RF power amplifier and the static circuit power loss caused by signal processing and active circuit blocks. The power loss of the RF power amplifier is closely related to the efficiency of the amplifier, given is the inverse of the power amplifier drain efficiency. The power loss P _m of the RF power amplifier is calculated according to formula (4):

Static circuit power loss is caused by signal processing and active circuit blocks, which require a certain amount of current to maintain their normal operation. The average value of the circuit power consumption is represented by the static variable _Pc , and the total circuit power consumption _Ps is expressed as:

_Ps ＝ _NPc (5)

Therefore, the device direct network power loss model is constructed as shown in formula (6).

P _tot = P _m + P _s (6)

Step 1.3: Construct a device-to-device network communication power control model. According to the total throughput R _tot and power loss P _tot of the device-to-device network obtained in steps 1.2 and 1.3, the energy efficiency η of the device-to-device network is calculated by formula (7):

Taking the device-to-device network communication power control model shown in formula (8) as the optimization objective and the channel bandwidth and throughput requirement lower limit as constraints, the distributed high-efficiency power control optimization problem of device-to-device network is constructed.

Wherein, the transmit power vector is represented by p = [p ₁ ,p ₂ , ..., p _N ] ^T , and p _n = [p _n,1 ,p _n,2 , ...,p _n,K ]; and They represent the maximum allowed total transmit power and minimum throughput of the nth user respectively, to ensure the communication service quality requirements of each user.

Step 2: Based on the device direct network communication power control model constructed in step 1, the properties of logarithmic functions are used to reconstruct the device direct network communication power control optimization problem into the difference form of two convex functions. The continuous convex approximation algorithm is used to approximate the reconstructed device direct network communication power control optimization problem into a continuous and differentiable convex function, thereby converting the optimization problem into a problem of solving a continuous convex function. Compared with directly solving the constructed non-convex optimization problem of device direct network communication power control, the difficulty of calculation is greatly reduced.

Step 2.1: To make the expression of the device-to-device network communication power control optimization problem more concise, let Using the properties of the logarithmic function in formula (2), the maximum transmission data volume _Rn of the nth user can be expressed as the following subtraction form:

R _n = _gn (p) _-hn (p) (9)

in,

The total throughput of the device direct network R _tot is reconstructed as:

Step 2.2: By introducing variables w = 1/P _tot and θ _n,k = wp _n,k , and θ = {θ _n,k }, the device direct network communication power control model constructed in step 1 is transformed into a device direct network communication power control model with a convex difference form as shown in formula (11):

Since g _n (p), h _n (p), g(p) and h(p) are all concave functions, the maximum transmission data volume R _n of the nth user and the total throughput R _tot of the device-to-device network are both differential forms of the two concave functions, thus proving that the transformed device-to-device network communication power control model The objective function of is a convex difference function, and its constraints all satisfy the convex difference constraints, so the converted device direct network communication power control model It is a convex difference programming problem and the power control model of direct network communication between devices The reconstructed device-to-device network communication power control optimization problem is approximated as a continuous and differentiable convex function using the continuous convex approximation SCA algorithm, thereby transforming the optimization problem into a problem of solving a continuous convex function to obtain a solution that satisfies the Karush-Kuhn-Tucker (KKT) condition of the device-to-device network communication power control model. In the tth iteration, the approximate convex function subproblem expression is:

Among them, θ ^(t) , and w ^(t) represent the optimal solutions θ, θn _,k and w of the t-th subproblem respectively.

Step 3: Convex function subproblem approximated in the tth iteration The objective function and constraints of are inseparable, so the alternating direction multiplier method ADMM cannot be directly applied to solve it. It is necessary to perform appropriate mathematical transformation on the approximate convex function subproblem and then reconstruct the approximate convex function subproblem. The Lagrangian function is convenient for further solution.

Step 3.1: Introduce a new variable I _n,n′,k =G _n,n′,k θ _n′,k , Define auxiliary variables w _n , and where w _n , and are the local copies of w, In _,n′,k and In _′,n,k on device n, respectively, and specify The feasible set of constructing the local variable in t rounds of iterative subproblems is:

in, and The optimal solution of the SCA subproblem in round t can be obtained by and w ^(t) are calculated. Definition l＝{l _n }，ξ＝{ξ _n }， _Included The global variables corresponding to the elements in the approximation of the convex function subproblem Make the following equivalent transformations:

Step 3.2: Convex function subproblem obtained by equivalent transformation The augmented Lagrangian function of is shown in formula (15):

Where α＝{α _n }, β and λ＝{λ _n } are Lagrange multiplier vectors, and ρ>0 is called the penalty coefficient, which determines the size of the update step. and ξ _n properties, λ _n contains the Lagrange multiplier and Corresponding variables and w _n .

Step 4: Using the alternating direction multiplier method ADMM, local variables, global variables, and Lagrange multipliers are updated in sequence to achieve the ability to solve optimization problems in a distributed environment. The dynamically updated penalty coefficient is used to adaptively adjust the value of the penalty coefficient according to the results of each round of iterations to further improve the convergence performance of the ADMM algorithm. The update of local variables is independent of other nodes and is calculated in parallel locally; global variables are used for information exchange and coordination between nodes; Lagrange multipliers are used to process constraints. Through this alternating update method, the distributed high-efficiency power control algorithm can gradually approach the optimal solution and achieve efficient optimization solutions in a distributed environment.

Step 4.1: Initialize variables Initialize the SCA subproblem iteration number t in step 2;

Step 4.2: Initialize ADMM iteration round j;

Step 4.3: Update local variables Local variables are local variables on each distributed node. Each node is responsible for solving the sub-problem related to itself and performing local optimization according to the characteristics of the problem. The update of local variables is performed without considering the information of other nodes. Each node independently solves its own optimal solution and performs calculations locally in parallel. Update local variables according to formula (16):

Step 4.4: Update the global variables {ξ ^(j+1) , l ^(j+1) }. Global variables are variables shared among all nodes and are used to store information exchange and coordination between nodes. In the ADMM algorithm, global variables are usually used to pass shared information about constraints and problems between nodes. The update of global variables is achieved through communication and information exchange between nodes. Using the local variables obtained in step 4.3, update the global variables in the following manner:

Since formula (17) is an unconstrained quadratic optimization problem, the closed-form solution of the global variable ξ ^(j+1) is calculated according to formula (19) and formula (20):

In order to derive the closed-form solution of l ^(j+1) as shown in formula (18), the global variable l ^(j+1) is transformed into the form of a square term as shown in formula (21):

in,

By setting the first-order derivative of formula (21) to zero, the closed-form solution of the global variable l ^(j+1) is written as:

Step 4.5: Update the Lagrange multiplier Lagrange multipliers are used to introduce constraints into the optimization problem and gradually adjust their own values through alternating updates. Each node has a corresponding Lagrange multiplier, which is used to process the constraints related to the node. Through the update of Lagrange multipliers, the ADMM algorithm exchanges and coordinates information between the global and local to obtain the optimal result. The update rules of Lagrange multipliers are shown in formulas (23)-(25):

Step 4.6: Update the penalty coefficient according to formula (26):

Among them, μ>0, τ ^incr >1 and τ ^decr >1 can be selected in combination with different structures and complexities of the problem by adaptively adjusting the size of the penalty coefficient according to the result of the current iteration.

Step 4.7: Convergence condition judgment, if the ADMM convergence condition is met (∈ is a very small positive number): Then continue to step 4.8; otherwise, let j = j + 1 and jump to step 4.3 to continue updating and iterating;

Step 4.8: If |θ ^(t+1) -θ ^(t) |≤ε or t≥T is satisfied, where ε is the convergence error and T is the maximum number of iterations, the algorithm terminates; otherwise, continue with step 4.2.

Through the above distributed high-efficiency power control algorithm steps, the converted device direct network communication power control model is finally obtained. The local KKT point, that is, {θ ^* ,w ^* }

Step 5: Based on the optimal solution {θ ^* , w ^* } obtained in step 4, p ^* = θ ^* / w ^* is calculated to finally obtain a power control result that optimizes the network energy efficiency EE.

In this example, Figure 3 shows the convergence curve of using the ADMM algorithm to solve the SCA convex optimization subproblem. Compared with the results obtained by the traditional centralized convex optimization algorithm, the ADMM algorithm can obtain the same optimal solution to the SCA convex optimization subproblem in a distributed manner. In addition, in the traditional ADMM algorithm, the fixed penalty coefficient may lead to inaccurate solutions to the dual problem, thereby affecting the convergence of the entire algorithm. In contrast, the method of dynamically updating the penalty coefficient uses fewer iterations to achieve convergence, which improves the convergence performance of the algorithm.

The results of the distributed method of the present invention and the centralized power control method are compared as shown in Figure 4. After the proposed distributed algorithm finally converges, an optimal solution close to that of the centralized algorithm is obtained. In practical applications, the distributed algorithm disperses the computational workload to each device, each node only needs to process local sub-problems, and the computation can be performed in parallel, which effectively improves the computational efficiency, reduces the computational load of the central node, and achieves network energy efficiency close to that of the centralized method.

The specific description above further illustrates the purpose, technical solutions and beneficial effects of the invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims (6)

1. A distributed high-efficiency power control method in a device-to-device network, characterized in that it comprises the following steps: Step 1: Construct a device-to-device network model. The distributed high-efficiency power control object is the device-to-device network. The device-to-device network includes N D2D user pairs, K resource blocks, and each device has independent sending and receiving functions. It is connected to an adjacent single device through a D2D link for data communication to reduce energy loss; construct a device-to-device network power loss model, considering the power loss of the RF power amplifier and the static circuit power loss caused by the signal processing and active circuit blocks; construct a device-to-device network communication power control model, and constrain the transmission power and throughput to meet the channel bandwidth limitation and ensure the user service quality requirements; Step 2: Based on the device direct network communication power control model constructed in step 1, the properties of the logarithmic function are used to reconstruct the device direct network communication power control optimization problem into the difference form of two convex functions, and the continuous convex approximation algorithm is used to approximate the reconstructed device direct network communication power control optimization problem into a continuous and differentiable convex function, thereby converting the optimization problem into a problem of solving a continuous convex function; Step 3: The objective function and constraint of the convex function subproblem approximated in the tth iteration are inseparable, so the alternating direction multiplier method ADMM cannot be directly applied to solve it. It is necessary to transform the approximate convex function subproblem and then construct the Lagrangian function of the approximate convex function subproblem to facilitate further solution; Step 4: Using the alternating direction multiplier method ADMM, local variables, global variables and Lagrange multipliers are updated in sequence to achieve the ability to solve optimization problems in a distributed environment; the penalty coefficient is dynamically updated to adaptively adjust the value of the penalty coefficient according to the results of each round of iterations to further improve the convergence performance of the ADMM algorithm; the update of local variables is independent of other nodes and is calculated in parallel locally; global variables are used for information exchange and coordination between nodes; Lagrange multipliers are used to process constraints; through alternating updates, the distributed high-efficiency power control algorithm can gradually approach the optimal solution and achieve efficient optimization solutions in a distributed environment; Step 5: Based on the optimal solution {θ ^* , w ^* } obtained in step 4, p ^* = θ ^* / w ^* is calculated to finally obtain a power control result that optimizes the network energy efficiency EE. 2. A distributed high-efficiency power control method in a device direct network as described in claim 1, characterized in that: it also includes step six, according to the distributed high-efficiency power control optimization result of the device direct network obtained in step five, to achieve optimal communication power allocation in the D2D network, which can reduce the energy consumption of devices during communication, improve the communication quality between devices, enhance signal coverage, thereby reducing resource loss and improving network capacity and performance. 3. A distributed high-efficiency power control method in a device-to-device network as claimed in claim 1 or 2, characterized in that: the implementation method of step 1 is: Step 1.1: Construct a device-to-device network model, in which the communication spectrum resources are provided in the form of resource blocks, and the D2D user pair set and the resource block set are respectively expressed as and Assume that _pn,k is the transmission power of the nth pair of D2D users when occupying the kth resource block, hn _,n,k and hn _,n′,k represent the channel coefficients from the receiving end of the nth link to the transmitting end of the nth link and the transmitting end of the n′th link respectively when using the kth resource block. Then, when the user equipment transmits data on the nth link using the kth resource block, the signal-to-interference-noise ratio γn _,k at the receiving end is expressed as: Where σ ² is the variance of zero-mean additive white Gaussian noise; the maximum amount of data transmitted by the nth user is: Wherein, B is the bandwidth of a resource block; then the total throughput R _tot of the device direct network is calculated according to formula (3): Step 1.2: Construct a device-through network power loss model, taking into account the power loss of the RF power amplifier and the static circuit power loss caused by signal processing and active circuit blocks; the power loss of the RF power amplifier is closely related to the efficiency of the amplifier, given is the inverse of the power amplifier drain efficiency. The power loss P _m of the RF power amplifier is calculated according to formula (4): Static circuit power loss is caused by signal processing and active circuit blocks, which require a certain amount of current to maintain their normal operation; the average value of the circuit power consumption is represented by the static variable _Pc , and the total circuit power consumption _Ps is expressed as: _Ps ＝ _NPc (5) Therefore, the device direct network power loss model is constructed as shown in formula (6); P _tot = P _m + P _s (6) Step 1.3: Construct a device-to-device network communication power control model. According to the device-to-device network total throughput R _tot and power loss P _tot obtained in steps 1.2 and 1.3, the device-to-device network energy efficiency η is calculated by formula (7): Taking the device-to-device network communication power control model shown in formula (8) as the optimization target and the channel bandwidth and throughput requirement lower limit as the constraints, the distributed high-efficiency power control optimization problem of the device-to-device network is constructed. Wherein, the transmit power vector is represented by p = [p ₁ ,p ₂ , ..., p _N ] ^T , and p _n = [p _n,1 ,p _n,2 , ...,p _n,K ]; and They represent the maximum allowed total transmit power and minimum throughput of the nth user respectively, to ensure the communication service quality requirements of each user. 4. A distributed high-efficiency power control method in a device-to-device network as claimed in claim 3, characterized in that: the implementation method of step 2 is: Step 2.1: To make the expression of the device-to-device network communication power control optimization problem more concise, let Using the properties of the logarithmic function in formula (2), the maximum transmission data volume _Rn of the nth user can be expressed as the following subtraction form: R _n = _gn (p) _-hn (p) (9) in, The total throughput of the device direct network R _tot is reconstructed as: Step 2.2: By introducing variables w = 1/P _tot and θ _n,k = wp _n,k , and θ = {θ _n,k }, the device direct network communication power control model constructed in step 1 is transformed into a device direct network communication power control model with a convex difference form as shown in formula (11): Since g _n (p), h _n (p), g(p) and h(p) are all concave functions, the maximum transmission data volume R _n of the nth user and the total throughput R _tot of the device-to-device network are both differential forms of the two concave functions, thus proving that the transformed device-to-device network communication power control model The objective function of is a convex difference function, and its constraints all satisfy the convex difference constraints, so the converted device direct network communication power control model It is a convex difference programming problem and the power control model of direct network communication between devices have the same optimal solution; the reconstructed device-to-device network communication power control optimization problem is approximated as a continuous and differentiable convex function using the continuous convex approximation SCA algorithm, thereby transforming the optimization problem into a problem of solving a continuous convex function to obtain a solution that satisfies the Karush-Kuhn-Tucker (KKT) condition of the device-to-device network communication power control model; in the tth round of iteration, the approximate convex function subproblem expression is: Among them, θ ^(t) , and w ^(t) represent the optimal solutions θ, θn _,k and w of the t-th subproblem respectively. 5. A distributed high-efficiency power control method in a device-to-device network as claimed in claim 4, characterized in that: the implementation method of step 3 is: Step 3.1: Introduce a new variable I _n,n′,k =G _n,n′,k θ _n′,k , Define auxiliary variables w _n , and where w _n , and are the local copies of w, In _,n′,k and In _′,n,k on device n, respectively, and specify The feasible set of constructing local variables in the tth round of iteration subproblems is: in, and The optimal solution of the t-th round SCA subproblem and w( ^t ) are calculated; definition l＝{l _n }，ξ＝{ξ _n }， _Included The global variables corresponding to the elements in the approximation of the convex function subproblem Make the following equivalent transformations: Step 3.2: Convex function subproblem obtained by equivalent transformation The augmented Lagrangian function of is shown in formula (15): Where α = {α _n }, β and λ = {λ _n } are Lagrange multiplier vectors, ρ>0 is called the penalty coefficient, which determines the size of the update step; and ξ _n properties, λ _n contains the Lagrange multiplier and Corresponding variables and w _n . 6. A distributed high-efficiency power control method in a device-to-device network as claimed in claim 5, characterized in that: the implementation method of step 4 is: Step 4.1: Initialize variables Initialize the SCA subproblem iteration number t in step 2; Step 4.2: Initialize ADMM iteration round j; Step 4.3: Update local variables Local variables are local variables on each distributed node. Each node is responsible for solving the sub-problem related to itself and performing local optimization according to the characteristics of the problem. The update of local variables is performed without considering the information of other nodes. Each node independently solves its own optimal solution and performs calculations in parallel locally. The local variables are updated according to formula (16): Step 4.4: Update the global variables {ξ ^(j+1) , l ^(j+1) }; global variables are variables shared among all nodes and are used to store information exchange and coordination between nodes; in the ADMM algorithm, global variables are usually used to transmit shared information of constraints and problems between nodes, and the update of global variables is achieved through communication and information exchange between nodes; using the local variables obtained in step 4.3, update the global variables in the following manner: Since formula (17) is an unconstrained quadratic optimization problem, the closed-form solution of the global variable ξ ^(j+1) is calculated according to formula (19) and formula (20): In order to derive the closed-form solution of l ^(j+1) as shown in formula (18), the global variable l ^(j+1) is transformed into the form of a square term as shown in formula (21): in, By setting the first-order derivative of formula (21) to zero, the closed-form solution of the global variable l ^(j+1) is written as: Step 4.5: Update the Lagrange multiplier Lagrange multipliers are used to introduce constraints into the optimization problem and gradually adjust their own values through alternating updates. Each node has a corresponding Lagrange multiplier, which is used to process the constraints related to the node. Through the update of Lagrange multipliers, the ADMM algorithm exchanges and coordinates information between the global and local to obtain the optimal result. The update rules of Lagrange multipliers are shown in formulas (23)-(25): Step 4.6: Update the penalty coefficient according to formula (26): Among them, μ>0, τ ^incr >1 and τ ^decr >1 can be selected based on the different structures and complexities of the problem by adaptively adjusting the size of the penalty coefficient according to the result of the current iteration; Step 4.7: Convergence condition judgment, if the ADMM convergence condition is met (∈ is a very small positive number): Then continue to step 4.8; otherwise, let j = j + 1 and jump to step 4.3 to continue updating and iterating; Step 4.8: If |θ ^(t+1) -θ ^(t) |≤ε or t≥T is satisfied, where ε is the convergence error and T is the maximum number of iterations, the algorithm terminates; otherwise, continue to step 4.2; Through the above distributed high-efficiency power control algorithm steps, the converted device direct network communication power control model is finally obtained. The local KKT point of , i.e., {θ ^* ,w ^* }.