Decision-making method for multi-user power control in wireless networks

By modeling multi-user power control into n-person non-cooperative games in wireless networks, and introducing average field terms and variable learning rates in gradient rise algorithms, the WoLF-MFGA algorithm is designed, and the multi-user power control game is solved Nash balance problem achieves low-cost and efficient power control.

CN116822329BActive Publication Date: 2024-12-03GUIZHOU UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202310529999.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-05-11
Publication Date
2024-12-03
Estimated Expiration
2043-05-11

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively solve the problem of multi-user power control game in wireless networks, especially in the scenario of multi-person and multi-action, the gradient rise algorithm cannot be applied, and the existing algorithm is difficult to achieve low-cost power control and Nash balance.

Method used

Model the multi-user power control problem as a non-cooperative game of n-person, and introduce average field terms into the gradient rising algorithm of the WoLF criterion, design the WoLF-MFGA algorithm, and realize power control through variable learning rate and average field term update strategy Nash balance.

Benefits of technology

Through the WoLF-MFGA algorithm, the existence of Nash balance in multi-user power control game is proved, and an effective method is provided to achieve low-cost power control. The sufficient convergence of the algorithm is verified, and the simulation results show that it can be better The best response algorithm is to approximate the best response.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116822329B_ABST
    Figure CN116822329B_ABST
Patent Text Reader

Abstract

The present invention discloses a decision-making method for multi-user power control in a wireless network, comprising: modeling a multi-user power control problem as an n-person non-cooperative game, and proving the existence of the Nash equilibrium of the multi-user power control game; introducing a mean field term in a gradient ascent containing a WoLF criterion, and designing a mean field gradient ascent WoLF-MFGA algorithm containing a WoLF criterion; the WoLF-MFGA algorithm allows players to update strategies according to a current gradient and a variable learning rate; the present invention models the multi-user power control game problem as an n-person non-cooperative game.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of multi-user power control game technology, in particular to a decision method for multi-user power control in a wireless network. Background Art

[0002] The gradient descent algorithm is a method for solving optimization problems by gradually iterating and reducing the loss function value to minimize the loss function value. In the real world, players usually do not show "complete rationality". This may be due to the fact that the players' decisions are myopic, the decision update process is gradual, and the computing power is limited. This feature of decision-making is called "bounded rationality". Studies have shown that naive expectations and myopic adjustments can well represent the behavior of humans in choosing decisions without anticipating the future. In view of this, the gradient learning method is used as a general model to study the decision-making of players in bounded rationality. The gradient ascent algorithm is a common matrix game learning-related algorithm. In the general sum game of two people and two actions, Singh et al. proved that in the case of infinitesimal steps, the strategies of the players will converge to the Nash equilibrium or the average return in the whole process will converge to the limit value of the expected return of the Nash equilibrium. Subsequently, Bolwing et al. proposed an infinitesimal WoLF gradient ascent algorithm by introducing a variable learning rate (or WoLF criterion) into the infinitesimal gradient ascent algorithm, that is, when the player wins, the learning rate value is small, and when the player loses, the learning rate is large. They also proved that the algorithm converges to the Nash equilibrium in all dual-matrix games. Zinkevich et al. proposed a generalized infinitesimal gradient ascent algorithm, which extends the infinitesimal gradient ascent algorithm to games with two or more strategies and has universal consistency. For general and random games with two actions, Banerjee et al. proposed a WoLF learning algorithm based on strategy dynamics. However, since the gradient ascent algorithm requires the opponent's strategy to be completely known and is difficult to extend to general and multi-player and multi-action games, the algorithm cannot be applied to a large number of practical problems. To solve these problems, the mean field term is an effective method. When the number of players is high, the behavior of all other players is encapsulated in a mean field term. Each player only needs to know the value of the mean field term when making decisions. This is the overall impact of the player's strategy on his payoff function, rather than the individual strategies of other players. Gradient learning has been used to achieve Nash equilibrium in different types of games, but no one has yet used it to analyze n-person non-cooperative games and their applications.

[0003] One of the key issues in wireless network systems is multi-user power control. The main purpose of power control is to provide sufficient quality for each signal without causing unnecessary interference to other signals, that is, to maximize the utility of each user (player). In recent years, the problem of multi-user power control in wireless networks has been closely related to game theory. MacKenzie et al. showed that game theory is a suitable tool for solving various problems in wireless network communication systems. Yu et al. studied the problem of multi-user power control in frequency selective interference channels and proposed an iterative water-filling (IWF) algorithm to effectively achieve its Nash equilibrium. Furthermore, the convergence of the IWF algorithm shows that the Nash equilibrium of the two-behavior matrix game is unique. Yamashita et al. formulated the multi-user power control problem of digital subscriber lines as a nonlinear complementary problem, and used the Newton-type smoothing (NS) method to effectively calculate the Nash equilibrium solution, and concluded that the NS algorithm is more robust to strong interference than the existing synchronous IWF algorithm. In addition, Meshkati et al. used game theory models to study power control in multi-carrier code division multiple access systems and proposed an iterative distributed algorithm to achieve the existence and uniqueness of the Nash equilibrium of the power control game. Hao et al. designed a joint channel allocation and power control optimization algorithm based on non-cooperative games to reduce interference in wireless sensor networks and balance network energy consumption, and further proved that the algorithm converges to the Nash equilibrium. At the same time, many scholars have designed different learning algorithms to analyze multi-user power control games in wireless networks. Goodman and Mandayam referred to power control in wireless data transmission networks and proposed an algorithm in which the pricing function is proportional to the transmission power to maximize the utility of each user individually. Luo et al. proposed a convergence analysis of the IWF algorithm under more realistic channel settings and arbitrary number of users for multi-user power control in digital subscriber lines. He et al. proposed a projection neural network to solve the Nash equilibrium of multi-user power control optimization problems in modern digital subscriber lines. Tao et al. proposed a centralized power control algorithm based on projected gradient design to solve multi-user power control games in cognitive radio networks. In the literature, power control algorithms based on non-cooperative games are proposed. Gulzar et al. proposed an adaptive power control algorithm to deal with the power control problem of cognitive radio networks. Taghizadeh et al. proposed a mean field gradient ascent learning algorithm to solve the Nash equilibrium of n-person non-cooperative games in blockchain network mining, and obtained sufficient conditions for the convergence of the algorithm.

[0004] In the past two decades, wireless communications have become a broad application and research area, including cellular networks, satellite networks, wireless mesh networks, cognitive radio networks, wireless ad hoc and sensor networks, etc. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a decision-making method for multi-user power control in a wireless network. The present invention models the multi-user power control game problem as an n-person non-cooperative game.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a decision method for multi-user power control in a wireless network, comprising the following steps:

[0007] Step 1: Model the multi-user power control problem as an n-person non-cooperative game, and prove the existence of Nash equilibrium of the multi-user power control game;

[0008] Step 2: By introducing the mean field term into the gradient ascent with WoLF criterion, a mean field gradient ascent WoLF-MFGA algorithm with WoLF criterion is designed; the WoLF-MFGA algorithm allows players to update the strategy according to the current gradient and variable learning rate.

[0009] As a further improvement of the present invention, the step 1 is specifically as follows:

[0010] Assuming that all users are homogeneous and the unit price of wireless power usage is the same for each user, the main goal of each user is to maximize their own total utility; since the power transmitted by each user is random, the chance of a user winning in each round of power transmission is equal to the ratio of its transmission power to the total wireless power capacity:

[0011]

[0012] Where N = {1, ..., n} is the set of all users, n represents the number of all users, P i suc. is the probability distribution of the successful allocation of electricity to the i-th user and x i is the transmission power of user i on the carrier; let x i ∈X i ,in and

[0013] Define a utility function suitable for data applications; define the user's utility function as the ratio of its throughput and transmission power:

[0014]

[0015] where qi is the transmission power of the ith user, T i The throughput of the i-th user refers to the number of net information bits transmitted without error per unit time, that is, effective transmission; the throughput is expressed as:

[0016]

[0017] Where L is the number of signal bits, M is the total number of bits in the data packet; R i and l i are the transmission rate and signal-to-interference-noise ratio of the i-th user respectively; f(l i ) is the utility function of successfully transmitting power, i.e., the probability of receiving the transmitted power without error;

[0018] Assuming a transmission terminal has one or more bit errors, it will be retransmitted; f(l i ) is an increasing, continuous, S-shaped function with f(∞)=1, and requires f(0)=0 to ensure that p i =0 when A i =0;

[0019] Considering the cost of transmitting electricity, the payment function of user i can be expressed as:

[0020]

[0021] where p i is the unit price of electricity transmitted by the ith user, E i represents the expected revenue obtained by user i from successfully transmitting electricity;

[0022] The multi-user power control problem is modeled as an n-person non-cooperative game. By choosing how much power to transmit in the wireless network to maximize the utility of each user, the competitive relationship between multiple users is modeled as a multi-user power control game based on the user set, strategy set and payment function in the game model:

[0023] Γ= <N,X i ,J i > i∈N .

[0024] As a further improvement of the present invention, the step 2 is specifically as follows:

[0025] The gradient strategy update formula is as follows:

[0026]

[0027] in is a variable learning rate parameter, is a constant; at the kth iteration, the step size of the algorithm in the gradient direction is in The update rules are as follows:

[0028]

[0029] Mean field game theory is used to help analyze the behavior of a large number of users by encapsulating their behavior into mean field terms. The mean field terms are used for estimation, and the update formula is as follows:

[0030]

[0031] where Z(k) is the mean field term of the kth iteration;

[0032] Substituting the mean field term into the gradient strategy update formula, the probability distribution of the i-th user successfully transmitting power is obtained as follows:

[0033]

[0034] At this time, the payment function of user i is rewritten as:

[0035]

[0036] Substitute the rewritten payment function into the gradient policy update formula and re-obtain the gradient learning policy update formula as follows:

[0037]

[0038] The strategy update of each user needs to be obtained through the function of the mean field term. The update equation of WoLF mean field term gradient ascent learning is rewritten as the fixed point iteration formula as follows:

[0039]

[0040] In the multi-user power control game Γ, each user encapsulates the influence of other users with the mean field term and ignores his own influence on the mean field term.

[0041] Since the strategies of all players are to influence the expected utility of other players through the probability of success, the present invention models the power control problem between users in a wireless network as an n-person non-cooperative game to analyze the strategic behavior of the players. The Nash equilibrium point of the game exists and is unique. The present invention considers the gradient learning strategy of the players while retaining the privacy information of the players as a bounded rational learning model. Therefore, in order to achieve the Nash equilibrium of the game, a new mean field gradient ascent (WoLF-MFGA) algorithm with variable learning rate is proposed by introducing variable learning rate and mean field terms into the gradient ascent algorithm. In addition, the variable gradient based on multi-user decision-making behavior ensures that low-cost power control can be achieved between users. This is a new method that provides an effective method for equilibrium analysis of power control games in wireless networks.

[0042] The beneficial effects of the present invention are:

[0043] 1. The present invention models the multi-user power control problem as an n-person non-cooperative game and uses some existing analysis tools to prove the existence of Nash equilibrium of the multi-user power control game, that is, there is at least one mixed strategy Nash equilibrium in the strategy space.

[0044] 2. The WoLF-MFGA gradient ascent algorithm is designed by combining the mean field term with the gradient ascent algorithm with a variable learning rate. In addition, the Banach compression fixed point theorem is used to give sufficient conditions for the convergence of the WoLF-MFGA algorithm.

[0045] 3. In order to achieve the Nash equilibrium of the multi-user power control game, some simulation results show that the WoLF-MFGA algorithm can better approximate the best response algorithm. In order to analyze the convergence of the proposed algorithm, the sensitivity results of the WoLF-MFGA algorithm are studied. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a display diagram of the multi-user power control game process in an embodiment of the present invention;

[0047] Figure 2 It is a schematic diagram of the total utility of the multi-user power control game under different learning parameters of the WoLF-MFGA algorithm and the BR algorithm in an embodiment of the present invention;

[0048] Figure 3 Schematic diagram of the total utility of the multi-user power control game under different Ai distributions of the WoLF-MFGA algorithm and the BR algorithm in an embodiment of the present invention;

[0049] Figure 4 The WoLF-MFGA algorithm and the BR algorithm in the embodiment of the present invention are shown in FIG. iSchematic diagram of the total utility of the multi-user power control game under distribution;

[0050] Figure 5 Schematic diagram of the evolution of total utility under different learning rate parameters in an embodiment of the present invention;

[0051] Figure 6 is a schematic diagram of the evolution of total utility under different wireless network scale parameters in an embodiment of the present invention;

[0052] Figure 7 A of the user in the embodiment of the present invention i Schematic diagram of the increase ratio at 400 generations. DETAILED DESCRIPTION

[0053] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0054] Example

[0055] The main purpose of this embodiment is to explore the strategic decision-making of multi-user power control in wireless networks from the perspective of mean field game theory.

[0056] First, the multi-user power control problem is modeled as an n-player non-cooperative game, and the existence of Nash equilibrium of the multi-user power control game is proved. Then, by introducing the mean field term in the gradient ascent with WoLF criterion, a new mean field gradient ascent with WoLF criterion (WoLF-MFGA) algorithm is designed. The WoLF-MFGA algorithm allows players to update strategies based on the current gradient and a variable learning rate. Subsequently, this embodiment studies the sufficient convergence conditions of the proposed WoLF-MFGA algorithm. In addition, the WoLF-MFGA algorithm is used to achieve the Nash equilibrium of the multi-user power control game, and some simulation results are given. Finally, in order to analyze the convergence of the proposed algorithm, this embodiment studies the sensitivity results of the WoLF-MFGA algorithm.

[0057] A decision-making method for multi-user power control in wireless networks considers the decision-making problem of multi-user power control game in wireless networks, and the multi-user power control game problem can be modeled as an n-person non-cooperative game. Since gradient ascent ensures that low-cost and high-benefit power control can be achieved between multi-user decision-making behaviors, designing a variable learning rate mean field gradient ascent (WoLF-MFGA) algorithm to achieve the Nash equilibrium of the n-person non-cooperative game is one of the research hotspots worthy of attention.

[0058] 1. N-person non-cooperative game in multi-user power control problem:

[0059] Assuming that all users are homogeneous and the unit price of wireless power usage is the same for each user, the main goal of each user is to maximize their own total utility. Since the power transmitted by each user is random, the chance of a user winning in each round of power transmission is equal to the ratio of its transmission power to the total wireless power capacity.

[0060]

[0061] Where N = {1, ..., n} is the set of all users, n represents the number of all users, is the probability distribution of the successful allocation of electricity to the i-th user and x i is the transmission power of user i on the carrier. Let x i ∈X i ,in and

[0062] The above equation represents the probability of winning for a single user, and the expected portion of what a user in a radio network receives from each carrier can be expressed in the same way. This is because the proportion of the reward that a user who is part of the radio network receives from the carrier's transmitted power is proportional to the proportion of their transmitted power to the wireless transmission power.

[0063] In order to model the power control problem as an n-person non-cooperative game, we first need to define a utility function suitable for data applications. Obviously, a higher signal-to-interference-and-noise ratio level at the receiver output usually leads to a lower bit error rate, thereby achieving a higher throughput. However, to achieve a higher signal-to-interference-and-noise ratio level, the user terminal needs to transmit at a higher power, which in turn leads to a lower battery life. This trade-off can be quantified by defining the user\utility function as the ratio of its throughput to the transmission power, that is:

[0064]

[0065] where q i is the transmission power of the ith user. i The throughput of the i-th user refers to the number of net information bits transmitted without error per unit time (sometimes referred to as effective transmission). The throughput can be expressed as:

[0066]

[0067] Where L is the number of signal bits and M is the total number of bits in the packet. i and l i are the transmission rate and signal-to-interference-noise ratio of the i-th user, respectively. i ) is the utility function of successfully transmitting power, that is, the probability of receiving the transmitted power without error.

[0068] Assuming a transmission has one or more bit errors, it will be retransmitted. i ) is an increasing, continuous, S-shaped function with f(∞)=1, and requires f(0)=0 to ensure that p i =0 when A i =0.

[0069] By considering the cost of transmitting electricity, the payment function of user i can be expressed as:

[0070]

[0071] where p i is the unit price of electricity transmitted to the i-th user (electricity charges and other maintenance costs), E i Represents the expected benefit obtained by user i from successfully transmitting electricity. In a multi-user power system, each user may be affected by the strategies of other users, and with limited perception capabilities, it is difficult for each user to collect status information of all other users. This leads to the consideration of establishing a unified price calculated by the terminal to determine how much power to transmit, where the price is the same as the unit of measurement used by the utility function. The multi-user power control problem is modeled as an n-person non-cooperative game, by choosing how much power to transmit in the wireless network to maximize the utility of each user. Therefore, the competitive relationship between multiple users can be modeled as a "multi-user power control game" based on the user set, strategy set, and payment function in the game model:

[0072] Γ= <N,X i ,J i > i∈N ,

[0073] The multi-user power control problem is modeled as an n-person non-cooperative game. The multi-user power control game process is shown as follows: Figure 1 shown.

[0074] 2. WoLF-MFGA algorithm:

[0075] In the real world, players are usually of limited rationality, that is, they lack foresight. Therefore, the gradient learning method is used as a universal model of limited rationality to analyze the decision-making behavior of players. The gradient strategy update formula is as follows:

[0076]

[0077] in is a variable learning rate parameter (WoLF criterion), is a constant.

[0078] The step size of the algorithm in the gradient direction is The main idea of ​​the WoLF criterion is to adjust the learning rate slowly and carefully when the player wins the game, and to learn quickly when the player loses or performs poorly. The update rules are as follows:

[0079]

[0080] Since the payment of user i is affected by the strategies of other users, it is neither realistic nor calculable to solve the Nash equilibrium of the multi-user power control game when facing a large number of users. At this time, the traditional analysis method may face the problem of high computational complexity or even difficulty in calculation, but the mean field game theory helps analyze the behavior of a large number of users by encapsulating the behavior of a large number of users into mean field terms, which can be estimated using mean field terms. The update formula is as follows:

[0081]

[0082] where Z(k) is the mean field term for the kth iteration. A basic assumption behind the mean field game is that the number of players is large enough so that changes in the strategies of individual players will not significantly change the payoff value of the mean field game. This assumption is valid in the multi-user power control game in wireless networks, and the number of users is large enough. Therefore, each user only needs to estimate the total transmission power of the network, i.e., the mean field term Z(k), at each iteration.

[0083] Substituting formula (2) into formula (1), we get the probability distribution of the i-th user successfully transmitting power as follows:

[0084]

[0085] At this time, the payment function of user i is rewritten as:

[0086]

[0087] Substituting formula (3) into (1), we can get the gradient learning strategy update formula as follows:

[0088]

[0089] Obviously, the strategy update of each user needs to be obtained through the function of the mean field term. Therefore, the update equation (3) of WoLF mean field term gradient ascent (WoLF-MFGA) learning can be rewritten as the fixed point iteration formula as follows:

[0090]

[0091] In the multi-user power control game Γ, each user encapsulates the influence of other users with the mean field term and ignores his own influence on the mean field term, so the mean field equilibrium is different from the best response equilibrium.

[0092] In summary, the WoLF-MFGA algorithm is designed by introducing the WoLF criterion and the mean field term into the gradient ascent algorithm. The implementation process of the WoLF-MFGA algorithm is shown in Algorithm 1, which shows the alternating updates between the mean field term and the user strategy.

[0093]

[0094] 3. Numerical simulation:

[0095] The multi-user power control game is modeled as an n-player non-cooperative game, and the proposed variable learning rate mean field gradient ascent (WoLF-MFGA) learning algorithm is evaluated through numerical simulation. First, the WoLF-MFGA algorithm is compared with the best response (BR) algorithm considering the complete information of the players' payoff functions, which is a good evaluation scheme. Secondly, the sensitivity analysis of the WoLF-MFGA algorithm to the parameters is considered, and the convergence of the WoLF-MFGA algorithm is studied in the dynamic situation of actual wireless networks.

[0096] The players update their strategies iteratively according to their best response functions and Algorithm 1. For simplicity, all simulation parameters are set as shown in Table 1 unless otherwise specified.

[0097]

[0098] Table 1 Parameter settings in the algorithm

[0099] The parameter selection is based on the multi-user power control problem in wireless networks. i In order to illustrate the accessibility, ease of consumption and energy value of the power control system, it is assumed that the expected utility A i The average distribution is in the range of 70,000 to 90,000. i It represents the unit price of electricity transmission, assuming that the unit price of electricity transmission is uniformly distributed in the range of 1 to 9. All simulations are run 6 times, and the average results are described.

[0100] Comparison of the WoLF-MFGA algorithm and the Best Response (BR) algorithm:

[0101] In the following figures, the total utility obtained by the Variable Learning Rate Mean Field Gradient Ascent (WoLF-MFGA) algorithm and the Best Response (BR) algorithm in wireless network power control is compared. The total utility is defined as the sum of all user benefits. In all figures, The choices of are small enough to ensure the convergence of the proposed WoLF-MFGA algorithm. The total utility is calculated in the wireless network and converges to the Nash equilibrium of the corresponding n-person non-cooperative game.

[0102] Figure 2 is the total utility of the multi-user power control game of the WoLF-MFGA algorithm and the BR algorithm under different learning parameters, where Depend on Figure 2 It can be seen that the total utility of the WoLF-MFGA algorithm and the best response algorithm is determined by all users. are set to 0.3, 0.5, and 0.8, respectively. Since best response learning is the optimal solution in theory, it is likely to achieve higher utility than the proposed WoLF-MFGA algorithm. Smaller, WoLF-MFGA learning can better approximate the best response learning when is larger, which has little effect on the total utility of the WoLF-MFGA algorithm, but the best response fluctuates greatly, which means that the best response learning method is not stable. However, considering the incomplete information of users in the competitor model, the equilibrium solution of the power control game in the wireless network is very close to the solution of the best response. The main reason for the fluctuation is the randomness of the selected parameters. It is not difficult to see that no matter how the learning parameters change, the total utility value of the WoLF-MFGA algorithm is very stable, that is, the WoLF-MFGA algorithm is insensitive to the parameters and has good stability. In summary, the equilibrium solution of the WoLF-MFGA algorithm is as close as possible to the optimal solution of the multi-user power control game. Then, as the number of users increases, the competitive behavior of users in the network becomes more intense, resulting in a decrease in total utility.

[0103] Figure 3 is the total utility of the multi-user power control game of the WoLF-MFGA algorithm and the BR algorithm under different Ai distributions; Figure 3 A comparison of different ranges of A is shown i The total utility of A. i The value ranges of are [20000,40000], [40000,60000], [60000,80000], [80000,100000], [100000,120000] and [120000,140000]. Assume that in each case, the user's reward A i uniformly distributed within a given range. Then, as A i The i-th user successfully transmits power and gets rewarded, which leads to a higher total utility of the entire wireless network. i The relationship between A and B is linear, so total utility increases almost as A iLinear growth.

[0104] Figure 4 The WoLF-MFGA algorithm and the BR algorithm have different q i Total utility of multi-user power control game under distribution; Figure 4 It shows that for different q i The total utility of the range wireless network for the WoLF-MFGA algorithm and the BR algorithm. In each case, q i are uniformly distributed within a given range. i Corresponding to the unit price of wireless power control, it is easy to see that the total utility increases with q i At the same time, the total utility is expressed relative to q i Scope of reciprocal behavior.

[0105] Sensitivity of WoLF-MFGA algorithm convergence:

[0106] In order to analyze the convergence of the proposed algorithm, some parameters were studied in this paper. N and A i In all simulations, the range of iterations is chosen large enough to show the convergence of the proposed WoLF-MFGA algorithm.

[0107] Figure 5 is the evolution of total utility under different learning rate parameters, Equal to 0.1, 0.3, 0.5, 0.7, 0.8 and 1; Figure 5 It can be seen that when the variable learning rate parameter When a small change occurs, it can be seen that the smaller the value of the learning rate change, the slower the convergence speed of the proposed WoLF-MFGA algorithm. However, when the variable learning rate changes more, it means that the fluctuation of the user's total utility value is smaller, and the WoLF-MFGA algorithm converges faster.

[0108] Figure 6 The evolution of total utility under different wireless network scale parameters is set to 100, 300, 600, 900, 1200 and 1400; Figure 6 It shows that with the increase of iterations, users of different sizes in the WoLF-MFGA algorithm tend to a stable point in the wireless network. The more users there are in the multi-user power control game in the wireless network, the higher the total utility. Conversely, the fewer users there are in the wireless network, the lower the total utility.

[0109] Depend on Figure 7 It can be seen that, assuming A i =7×10 4 , p i~uniform(1,9). After about 100 iterations, the total utility can converge. In order to analyze the proposed WoLF-MFGA algorithm with the parameter A i behavior when it changes, so A i It increases by 60% at iteration 400. This sudden change in parameter value causes the wireless network system to move to a new equilibrium.

[0110] In this embodiment, the behavioral decision of the multi-user power control problem in a wireless network is studied from the perspective of mean field game theory. For the n-person non-cooperative game in multi-user power control, a variable learning rate mean field gradient ascent (WoLF-MFGA) algorithm is designed by introducing a variable learning rate (WoLF criterion) and a mean field term method into the gradient ascent algorithm, and is applied to realize the Nash equilibrium of the n-person non-cooperative game. The convergence of WoLF-MFGA is proved with the help of nonlinear analysis tools, and sufficient conditions for the convergence of the WoLF-MFGA algorithm are given. Simulation experiments show that by adjusting some appropriate parameters, the WoLF-MFGA algorithm can better approximate the exact solution of the game, that is, the exact solution corresponding to the best response algorithm. In order to verify the convergence of the proposed algorithm, the WoLF-MFGA algorithm is applied to realize the Nash equilibrium of the multi-user power control game in the wireless network, and the sensitivity of the algorithm to the parameters is analyzed. Furthermore, the analysis method provided in this embodiment is extended to other scenarios, such as shared resources, multi-agent cooperation, blockchain and network economy, which is considered to be future work.

[0111] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A decision method for multi-user power control in a wireless network, characterized in that: The following steps are involved: Step 1: Model the multi-user power control problem as an n-person non-cooperative game, and prove the existence of Nash equilibrium of the multi-user power control game; Step 2: By introducing the mean field term into the gradient ascent with WoLF criterion, the mean field gradient ascent WoLF-MFGA algorithm with WoLF criterion is designed; the WoLF-MFGA algorithm allows the players to update the strategy according to the current gradient and the variable learning rate; The step 1 is specifically as follows: Assuming that all users are homogeneous and the unit price of wireless power usage is the same for each user, the main goal of each user is to maximize their own total utility; since the power transmitted by each user is random, the chance of a user winning in each round of power transmission is equal to the ratio of its transmission power to the total wireless power capacity: Where N = {1, ..., n} is the set of all users, n represents the number of all users, P i suc. is the probability distribution of successful power allocation to the ith user, x i is the transmission power of user i on the carrier; let x i ∈X i ,in and Define a utility function suitable for data application; The utility function is the ratio of its throughput to the transmission power: where q i is the transmission power of the ith user, T i The throughput of the i-th user refers to the number of net information bits transmitted without error per unit time, that is, effective transmission; the throughput is expressed as: Where L is the number of signal bits, M is the total number of bits in the data packet; R i and l i are the transmission rate and signal-to-interference-noise ratio of the i-th user respectively; f(l i ) is the utility function of successfully transmitting power, i.e., the probability of receiving the transmitted power without error; Assuming a transmission terminal has one or more bit errors, it will be retransmitted; f(l i ) is an increasing, continuous, S-shaped function with f(∞)=1, and requires f(0)=0 to ensure that p i =0 o'clock Considering the cost of transmitting electricity, the payment function of user i is expressed as: where p i is the unit price of electricity transmitted by the ith user, E i represents the expected revenue obtained by user i from successfully transmitting electricity; The multi-user power control problem is modeled as an n-person non-cooperative game. By choosing how much power to transmit in the wireless network to maximize the utility of each user, the competitive relationship between multiple users is modeled as a multi-user power control game based on the user set, strategy set and payment function in the game model: The step 2 is specifically as follows: The gradient strategy update formula is as follows: in is a variable learning rate parameter, is a constant; at the kth iteration, the step size of the algorithm in the gradient direction is in The update rules are as follows: Mean field game theory is used to help analyze the behavior of a large number of users by encapsulating their behavior into mean field terms. The mean field terms are used for estimation, and the update formula is as follows: in is the mean field term of the kth iteration; Substituting the mean field term into the gradient strategy update formula, the probability distribution of the i-th user successfully transmitting power is obtained as follows: At this time, the payment function of user i is rewritten as: Substitute the rewritten payment function into the gradient policy update formula and re-obtain the gradient learning policy update formula as follows: The strategy update of each user needs to be obtained through the function of the mean field term. The update equation of WoLF mean field term gradient ascent learning is rewritten as the fixed point iteration formula as follows: In the multi-user power control game Γ, each user encapsulates the influence of other users with the mean field term and ignores his own influence on the mean field term.

Citation Information

Patent Citations

  • Dynamic wireless network selection method based on evolutionary game theory

    CN105120468A

  • Network selection method based on non-cooperative game

    CN107949025A