Dynamic correlation method based on upper and lower bound network flows in cellular-free millimeter wave systems

Abstract

The present invention relates to a dynamic association method based on upper and lower bound network flows in a non-cellular millimeter wave system. The method aims at the dynamic association problem of a non-cellular millimeter wave large-scale MIMO system under the condition of dense users and dynamically changing communication environment, and proposes a dynamic association technology based on the idea of ​​phased decision-making and utilizing low-dimensional environmental perception information to solve the problem that the prior art fails to effectively handle user service needs, and the problem that the dynamic association algorithm has a large overhead, resulting in insufficient system dynamic association performance in reality. The method proposed in the present invention only utilizes low-dimensional environmental perception information, and the computational complexity is not high. Compared with traditional methods, the time complexity does not increase in an asymptotic sense, and the system dynamic association performance is effectively improved, and the dynamic communication environment can be adapted with low overhead. In addition, the system performance is evaluated using a statistical model, and the simulation results prove the superiority of the proposed method, which has certain value.

Description

Technical Field

[0001] The invention relates to a dynamic association method based on upper and lower bound network flow models in a non-cellular millimeter wave large-scale MIMO system, belonging to the technical field of mobile communications. Background Art

[0002] The millimeter wave band has abundant spectrum resources, which helps to meet the high throughput and low latency requirements of emerging applications. More innovative technologies are needed to pave the way for the development of post-5G and 6G technologies. Millimeter waves have the characteristics of high transmission loss. By deploying large-scale MIMO technology, beamforming gain can be provided for signal compensation, thereby achieving high spectrum efficiency. In the non-cellular millimeter wave large-scale MIMO system, a large number of distributed APs are deployed to use the same time-frequency resources to serve many users in the system. The hybrid beamforming architecture is an architecture with technical potential in the millimeter wave system, which allows the system to configure a large-scale antenna array while only requiring a limited number of RF chains. Analog beamforming provides directional gain, and digital beamforming provides diversity gain. An important advantage of hybrid beamforming technology is that the digital beamforming part only needs to consider equivalent channels. The equivalent channel consists of an air interface channel and an analog precoding matrix. Compared with the air interface channel, its dimension is significantly reduced, which can effectively reduce the complexity and system overhead of digital precoding. As for the construction of analog precoding, one method is to predefine the codebook and obtain the optimal beam through beam management.

[0003] An effective association strategy is an important component of the millimeter wave system. The naive dynamic association algorithm has an exponential time overhead. In the traditional millimeter wave system standard, the association scheduling is based on the idea of ​​minimizing the distance, which is converted into a linear programming problem with the goal of maximizing RSSI. Based on the idea of ​​staged decision-making, the dynamic association technology will affect the construction of subsequent equivalent channels, thus playing an important role in system performance. Therefore, low-overhead and effective dynamic association technology is becoming increasingly important. Summary of the invention

[0004] Technical problem: In view of the above problems, the present invention proposes a dynamic association method based on the upper and lower bound network flow model in a non-cellular millimeter wave large-scale MIMO system, which solves the problem that the existing technology does not effectively handle the quality of user communication services, and the problem that the dynamic association algorithm has a large overhead, resulting in insufficient dynamic association performance of the system in real situations. In addition, in view of the situation where the deployed AP is equipped with a small-scale antenna array, the proposed technology is improved to obtain another type of dynamic association technology based on environmental perception information, which effectively suppresses interference. In addition, the system performance is evaluated using a statistical model.

[0005] Technical solution: A dynamic association method based on the upper and lower bound network flow model of the present invention has the following specific steps:

[0006] The method is applied in a non-cellular millimeter wave massive MIMO system, and includes the following steps:

[0007] Step 1: Establish a system model. In a non-cellular millimeter wave system, consider the case of an access node AP first, and then generalize it; the number of antennas configured by the AP is N t , codebook That is, F represents N t A complex matrix with M rows and M columns, where M is the number of predefined beams, the number of user equipment UEs associated with the AP is K, and the corresponding UE set is It indicates that, assuming that the UE is configured with a single antenna;

[0008] Simulate precoding matrix N t A complex matrix with K rows and columns, each of which is selected from the codebook. If the beam index selected by the i-th user is b i , then construct a real number matrix with M rows and K columns in

[0009]

[0010] At this time, the simulated precoding matrix A=FB is obtained; for the i-th user, the transmitted data stream is That is i Following a complex Gaussian distribution with a mean of 0 and a variance of 1, the digital precoding is represents a complex column vector with dimension K; the air interface channel between the i-th user and the AP is It contains the line-of-sight path, and the received noise random variable is Define y i is the received signal of the i-th user

[0011]

[0012] Obtain the equivalent channel by beam scanning to avoid direct air interface channel estimation and define h i is the equivalent channel between the ith user and the AP, that is,

[0013]

[0014] Expanded to the scenario where the number of APs is N, the AP set is AP k represents the kth AP, and its associated user set is The size of the user set is The simulated precoding matrix is Represents an N t Row K kThe complex matrix of columns, and the channel between the i-th user is If the kth AP is associated with the ith user, the transmitted data flow is represents digital precoding, then the received signal of the i-th user is expressed as

[0015]

[0016] The equivalent channel between the i-th user and the k-th AP is expressed as

[0017]

[0018] Then the signal received by the i-th user can be expressed by the equivalent channel as

[0019]

[0020] Let R i represents the achievable rate of the i-th user. The general form of the dynamic association problem is

[0021]

[0022] Step 2: According to the traditional scheme, the dynamic association problem is modeled as a cost flow:

[0023] First, the concepts of network and network flow are explained: a network is a directed graph in is a collection of nodes, is a set of edges. For an edge e = (u, v), u and v are called the endpoints of e. Each edge in the graph They are combined with a non-negative integer weight c(u,v), which represents the capacity of the edge (u,v); by additionally defining two unequal nodes in the graph as source nodes With the meeting point Such a network is called a flow network;

[0024] The flow in the flow network G is defined as a function f, also called network flow, where f(u,v) is defined in a binary pair A real function that satisfies the following three properties: capacity limitation, skew symmetry, and flow conservation;

[0025] for f(u,v) represents the flow of the edge, c(u,v)-f(u,v) represents the remaining capacity of the edge, and the flow of the entire network is That is, the sum of the outflow from the source point;

[0026] The traditional millimeter wave network is based on the basic idea of ​​maximizing the received signal strength indication RSSI to solve the dynamic association strategy. The problem is modeled as the following minimum cost flow form

[0027]

[0028] Among them, N RF Represents the number of configured RF chains, R i,j represents the RSSI measured by the i-th AP for the j-th user, X i,j If 1 is set, it means that the i-th AP is associated with the j-th user; otherwise, it means that the i-th AP is not associated with the j-th user.

[0029] Step 3: Considering the user service priority, the traditional model is expanded into a minimum cost flow model with upper and lower bounds:

[0030] When the environmental interaction information feedback shows that a high-priority user is still not satisfied, the number of links serving this user can be increased according to the service strategy. Assuming that according to the service strategy, the at least number of links required to be provided for the jth user is d(j), the problem has the following form:

[0031]

[0032] For deployed APs with small-scale antenna arrays, large beam widths, and high probability of LOS line-of-sight, the RSSI index in the objective function is converted into relative physical distance.

[0033]

[0034] Among them, D i,j represents the relative physical distance between the i-th AP and the j-th user;

[0035] Step 4: Build a capacity-cost network based on the problem:

[0036] According to the problem modeling, a capacity-cost network with upper and lower capacity constraints is established, and according to the flow conservation property, it is transformed into a general form of flow network. After checking the satisfying rows of the lower bound conditions, the minimum cost maximum flow is solved in the residual network. Finally, a dynamic association strategy is constructed based on the flow function;

[0037] Step 5: Solve the dynamic association solution based on the continuous shortest path algorithm:

[0038] Apply the minimum cost flow algorithm based on the continuous shortest path to solve the minimum cost maximum flow of s′-t′. If s′-t′ is not full flow, that is, it does not satisfy At this time, there is no feasible solution that meets the capacity lower bound condition, and the network needs to be adjusted according to the user priority strategy;

[0039] In the residual network, the minimum cost flow algorithm based on the continuous shortest path is applied to solve the st minimum cost maximum flow, and a dynamic association strategy is constructed according to the network flow, that is, if f(a i,u j ) is full flow, user j is associated with the i-th AP.

[0040] The step 4 specifically includes:

[0041] Step 4.1: Order is the weight matrix of the bipartite graph. In the RMCFD algorithm with the minimum cost flow based on RSSI, W = R. In the DMCFD algorithm with the minimum cost flow based on relative physical distance, W = D. Add source and sink points s and t, construct node sets for AP and UE, and design the function add-edge(u,v,d(u,v),c(u,v),w(u,v)) to update the network structure, which represents an edge from node u to node v. The lower bound of the edge capacity is d(u,v), the upper bound of the edge capacity is c(u,v), and the cost of the edge is w(u,v). Considering that RSSI has negative values, and in user-intensive scenarios, the RF chain of the AP is generally fully utilized, based on the equivalence of the problem, a fixed offset is added to each element of the weight matrix to ensure that any element is non-negative. At this time, for the initial flow network, edges with non-zero capacity and negative weights are not processed.

[0042] Step 4.2: Indicates AP i At most one data stream is provided to user j, and its contribution to the objective function is f(a i ,u j )×W i,j , if the environmental perception information fed back by the previous dynamic association shows that the channel quality is poor (a i ,u j ) is correct, then set c(a i ,u j )=0; where a i Representing AP i ,u j represents the jth user;

[0043] Step 4.3: Indicates AP i The upper bound of the data flow is N RF ; Indicates that the user is served by any number of APs;

[0044] Step 4.4: add-edge(t,s,0,∞,0) to construct a network of passive points and sinks. At this time, we can solve the problem according to the solution of the minimum cost flow of the passive sink and the lower bound.

[0045] Step 4.5: Add auxiliary source points and sink points s′, t′, consider the capacity lower bound, and add auxiliary edges according to the classification discussion;

[0046] Assuming that any edge e(u,v) already has the flow of d(u,v), for any node in the flow network The initial inflow minus the initial outflow is x, and it can be processed in three cases if necessary: ​​if x=0, the flow conservation property is satisfied and no additional edge is needed; if x>0, it means that the node inflow is too large, and an additional edge with a capacity of x is required from s′ to v; if x<0, it means that the node outflow is too large, and an additional edge with a capacity of -x is required from v to t′.

[0047] In the non-cellular millimeter wave system, a user communication service quality QoS model is established to calculate the achievable rate and of the system under different user densities, and to judge the effectiveness of the dynamic association scheme. The Monte Carlo method is used, and the main steps are as follows:

[0048] Step 1: A simple simulation is performed on the user QoS situation. Specifically, users in the scenario will have different QoS, and there will also be sudden message demands or long data waiting queues. The system has a set of user service priority decision strategies, which dynamically adjusts the priority of the served users through the continuous changes in the communication environment. Users with higher priorities are modeled as demand users in the above algorithm. Users are randomly distributed, and a certain density of high-priority users is set;

[0049] Step 2: For the scenario where the number of APs is N, the AP set is AP k The associated user set is The associated user set size is The simulated precoding matrix is The channel between user i is If AP k Associated with user i, the transmitted data flow is represents digital precoding, B represents channel bandwidth, then for the i-th user, the achievable communication rate is R i

[0050]

[0051] The system and rate are

[0052] Beneficial effects: The method proposed in the present invention only utilizes low-dimensional environmental perception information, and the computational complexity is not high. Compared with traditional methods, the time complexity does not increase in an asymptotic sense. It effectively handles user communication service quality constraints, suppresses signal interference between users, improves system performance, and can adapt to dynamic communication environments with low overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1It is a dynamic association technology framework diagram.

[0054] Figure 2 This is a simulation diagram comparing the performance of several types of dynamic association algorithms under different user density scenarios.

[0055] Figure 3 This is a simulation diagram comparing the performance of several types of dynamic correlation algorithms when the system is configured with antenna arrays of different sizes.

[0056] Figure 4 This is a simulation diagram of the effect of different positioning accuracies on the performance of the dynamic association algorithm based on relative physical distance when there is a positioning error. DETAILED DESCRIPTION

[0057] The present invention is described in detail below in conjunction with examples:

[0058] Consider the downlink of a non-cellular millimeter-wave massive MIMO system, assuming that the system is deployed in a 20m×20m square area, and the number of APs is configured to be 4. Each AP is configured with a small-scale antenna array. The present invention is set as a uniform linear array of 16 antennas. At the same time, each AP is configured with 8 RF chains, and there is a LOS between the user and the AP. The number of users in the area is set to [20,400], and these users are randomly and evenly distributed in the square area. A simple simulation was performed for user service priority. Specifically, users in the scenario will have different QoS, and there will also be sudden message demands or long data waiting queues. Assume that the system has a set of user service priority determination strategies, which dynamically adjusts the priority of the served users through the continuous changes in the communication environment. The set of users with higher priority is The system needs to serve it as much as possible, that is, adjust

[0059] The implementation method of the present invention in the system is specifically as follows:

[0060] (1) Establishing a system model. This study considers a non-cellular millimeter wave system:

[0061] First consider the case of one AP, and then generalize it. The number of antennas configured on the AP is N t , codebook The number of predefined beams is M, the number of UEs associated with the AP is K, and the corresponding UE set is To simplify the problem description, we assume that the UE is equipped with a single antenna. Simulated precoding matrix Each column of is selected from the codebook. If the beam index selected by user i is b i , then construct the matrix

[0062]

[0063] At this time, A=FB. For user i, the transmitted data flow is Digital precoding is The air interface channel between user i and AP is It contains the line-of-sight path. The received noise random variable is Define y i is the received signal of user i.

[0064]

[0065] In millimeter wave systems, as the size of antenna arrays continues to increase, the size of the air interface channel matrix increases, which increases the channel estimation overhead. The analog-digital hybrid precoding architecture effectively reduces the complexity of system deployment, but it also brings certain difficulties to directly estimate the air interface channel. In order to overcome this problem, the equivalent channel is obtained by beam scanning, avoiding direct air interface channel estimation. Define h i is the equivalent channel between user i and AP, that is

[0066] Expanded to the scenario where the number of APs is N, the AP set is AP k The associated user set is in The simulated precoding matrix is The channel between user i is If AP k Associated with user i, the transmitted data flow is represents digital precoding. Then the received signal of user i is expressed as

[0067]

[0068] User i and AP k The equivalent channel between

[0069]

[0070] Then the user received signal can be expressed by the equivalent channel as

[0071]

[0072] (2) According to the traditional solution, the dynamic association problem is modeled as a cost flow:

[0073] Traditional millimeter wave networks solve dynamic association strategies based on maximizing RSSI. At the same time, strategies are developed based on the idea of ​​minimizing distance, which is transformed into maximizing RSSI. The problem modeling is as follows:

[0074]

[0075] Among them, N RF Represents the number of configured RF chains, R i,j Indicates AP i RSSI measured for user j, X i,j A value of 1 indicates that the i-th AP is associated with the j-th user, whereas a value of 1 indicates that the i-th AP is not associated with the j-th user.

[0076] (3) Considering the user service priority, the traditional model is expanded into a minimum cost flow model with upper and lower bounds:

[0077] Assuming that according to the service policy, the minimum number of links that need to be provided for user j is d(j), then the problem has the following form:

[0078] For deployed APs with small antenna arrays, large beam widths, and high probability of LOS, another type of dynamic association technology based on environmental perception information is deployed to effectively suppress interference. Specifically, the RSSI indicator in the objective function is transformed into a relative physical distance

[0079]

[0080] Among them, D i,j Indicates AP i The relative physical distance from user j.

[0081] (4) Establish a capacity-cost network based on the problem:

[0082] 4.1: Order is the weight matrix of the bipartite graph. In RMCFD, W = R, and in DMCFD, W = D. Add source and sink points s, t, and construct node sets for AP and UE, where a i Representing AP i ,u j represents user j. Design a function add-edge(u,v,d(u,v),c(u,v),w(u,v)) to update the network structure, which represents an edge from node u to node v. The lower bound of the edge capacity is d(u,v), the upper bound of the edge capacity is c(u,v), and the cost of the edge is w(u,v). Considering that RSSI has negative values, and in user-intensive scenarios, the AP's RF chain is generally fully utilized, based on the equivalence of the problem, a fixed offset can be added to each element of the weight matrix to ensure that any element is non-negative; at this time, for the initial flow network, there is no need to specially handle edges with non-zero capacity and negative weights.

[0083] 4.2: Indicates AP i At most one data stream is provided to user j, and its contribution to the objective function is f(a i ,u j )×W i,j , if the environmental perception information fed back by the previous dynamic association shows that the channel quality is poor (a i ,u j ) is correct, then set c(a i ,u j )=0.

[0084] 4.3: Indicates AP i The upper bound of the data flow is N RF ;

[0085] Indicates that the user can be served by any AP.

[0086] 4.4: add-edge(t,s,0,∞,0) constructs a network of passive points and sinks. At this time, we can solve the problem according to the solution of the minimum cost flow of the upper and lower bounds of passive sinks.

[0087] 4.5: Add auxiliary source and sink points s′, t′, consider the lower bound of capacity, and add auxiliary edges based on the classification discussion.

[0088] Assuming that any edge e(u,v) already has the flow of d(u,v), for any node in the flow network The initial inflow minus the initial outflow is x. If necessary, it can be processed in three cases: if x = 0, the flow conservation is satisfied and no additional edge is needed; if x > 0, it means that the node inflow is too large, and an additional edge with a capacity of x is required from s' to v; if x < 0, it means that the node outflow is too large, and an additional edge with a capacity of -x is required from v to t'

[0089] (5) Solve the dynamic association scheme based on the continuous shortest path algorithm:

[0090] Apply the minimum cost flow algorithm based on the continuous shortest path to solve the minimum cost maximum flow of s′-t′. If s′-t′ is not full flow, that is, it does not satisfy At this time, there is no feasible solution that meets the capacity lower bound condition, and the network needs to be adjusted according to the user priority strategy.

[0091] In the residual network, the minimum cost flow algorithm based on the continuous shortest path is applied to solve the st minimum cost maximum flow. The continuous shortest path algorithm is a type of implementation of the minimum cost flow algorithm, that is, the shortest augmenting path is continuously solved in the residual network of the flow network. A potential energy function is set for the nodes in the flow network, and the unit cost of all edges in the flow network is converted to a non-negative value, so that the Dijkstra algorithm can be applied to find the augmenting path with the minimum cost in the flow network. According to the network flow, a dynamic association strategy is constructed, that is, if f(a i ,u j ) is full, then user j is associated with AP i .

[0092] (6) Apply Monte Carlo method to analyze system performance

[0093] A simple simulation was conducted for the user QoS situation. Specifically, users in the scenario will have different QoS, and there will also be sudden message demands or long data waiting queues. This study does not consider and simulate various indicator constraints in detail. It can be assumed that the system has a set of user service priority determination strategies, and dynamically adjusts the priority of the served users through the continuous changes in the communication environment. Users with higher priorities are modeled as demand users in the above algorithm, that is, the system needs to serve them as much as possible. Users are randomly distributed, and a certain density of high-priority users is set.

[0094] For the scenario where the number of APs is N, the AP set is AP k The associated user set is in The simulated precoding matrix is The channel between user i is If AP k Associated with user i, the transmitted data flow is represents digital precoding. B represents channel bandwidth. For user i, the achievable communication rate is R i

[0095]

[0096] The above shows the whole process of performing non-cellular massive MIMO dynamic association and performance analysis using the method of the present invention.

[0097] Figure 1The following is a framework diagram of the dynamic association technology. It can be seen that the system first models the dynamic association problem as a minimum cost flow problem with upper and lower bounds through environmental perception information, such as RSSI or location information, and establishes a flow network; then the continuous shortest path algorithm is applied to solve it. Specifically, auxiliary points and edges are first constructed to check the existence of the solution, that is, whether there is a flow that satisfies the capacity lower bound constraint. If so, the minimum cost maximum flow is found in the residual network; finally, a dynamic association strategy is constructed based on the obtained network flow.

[0098] Figure 2 The figure shows the performance comparison of several types of dynamic association algorithms under different user density scenarios. It can be seen that as the user density in the scenario increases, the probability of increasing the number of users with better channel quality or higher service priority increases, and the system and rate increase accordingly. Two types of greedy algorithms for comparison are designed in the simulation. The simulation results show that as the number of users with higher service priority and random distribution increases, since the AP needs to provide services to these users as much as possible, the proposed algorithm based on environment perception can effectively cooperate with APs to handle signal interference between users, thereby achieving a significant performance improvement.

[0099] Figure 3 The figure shows the performance comparison of several types of dynamic association algorithms when the system is configured with antenna arrays of different sizes. It can be seen that APs deploy antenna arrays of different sizes, resulting in different beam widths and different interference effects between users. The dynamic association method based on relative physical distance, which aims to suppress signal interference between users as much as possible, performs best in scenarios with small-scale antenna arrays, but performs worse than the dynamic association technology based on RSSI in scenarios with larger-scale antenna arrays.

[0100] Figure 4 The figure shows the comparison of the performance of the dynamic association algorithm based on relative physical distance with different positioning accuracies when there is a positioning error. It can be seen that with the continuous improvement of positioning accuracy, the system performance is also continuously improving. When the positioning error is within 1m, the performance close to the perfect location information can be obtained, and there is only a relatively obvious performance loss in the ultra-dense network. Therefore, the proposed algorithm has a certain degree of robustness and can effectively combat the situation where there is a certain error in positioning accuracy.

[0101] The present invention proposes a dynamic association method based on upper and lower bound network flow models in a non-cellular millimeter-wave large-scale MIMO system. The proposed method only utilizes low-dimensional environmental perception information and has low computational complexity. Compared with traditional methods, the time complexity does not increase in an asymptotic sense, which effectively improves system performance and can adapt to dynamic communication environments with low overhead. In addition, a statistical model is used to evaluate system performance, and simulation results prove the superiority of the proposed method, which has certain value.

Claims

1. A dynamic association method based on upper and lower bound network flows in a non-cellular millimeter wave system, characterized in that: The method is applied in a non-cellular millimeter wave massive MIMO system, and includes the following steps: Step 1: Establish a system model. In a non-cellular millimeter wave system, consider the case of an access node AP first, and then generalize it; the number of antennas configured by the AP is N t , codebook That is, F represents N t A complex matrix with M rows and M columns, where M is the number of predefined beams, the number of user equipment UEs associated with the AP is K, and the corresponding UE set is It indicates that, assuming that the UE is configured with a single antenna; Simulate precoding matrix N t A complex matrix with K rows and columns, each of which is selected from the codebook. If the beam index selected by the i-th user is b i , then construct a real number matrix with M rows and K columns in At this time, the simulated precoding matrix A=FB is obtained; for the i-th user, the transmitted data stream is That is i Following a complex Gaussian distribution with a mean of 0 and a variance of 1, the digital precoding is represents a complex column vector with dimension K; the air interface channel between the i-th user and the AP is It contains the line-of-sight path, and the received noise random variable is Define y i is the received signal of the i-th user Obtain the equivalent channel by beam scanning to avoid direct air interface channel estimation and define h i is the equivalent channel between the ith user and the AP, that is, Expanded to the scenario where the number of APs is N, the AP set is AP k represents the kth AP, and its associated user set is The size of the user set is The simulated precoding matrix is Represents an N t Row K k The complex matrix of columns, and the channel between the i-th user is If the kth AP is associated with the ith user, the transmitted data flow is represents digital precoding, then the received signal of the i-th user is expressed as The equivalent channel between the i-th user and the k-th AP is expressed as Then the signal received by the i-th user can be expressed by the equivalent channel as Let R i represents the achievable rate of the i-th user. The general form of the dynamic association problem is Step 2: According to the traditional scheme, the dynamic association problem is modeled as a cost flow: First, the concepts of network and network flow are explained: a network is a directed graph in is a set of nodes, ε is a set of edges, for edge e = (u, v), u and v are called the endpoints of e, and each edge (u, v) ∈ ε in the graph is associated with a non-negative integer weight c(u, v), representing the capacity of the edge (u, v); by additionally defining two unequal nodes in the graph as source nodes With the meeting point Such a network is called a flow network; The flow in the flow network G is defined as a function f, also called network flow, where f(u,v) is defined on the tuple (u,v), A real function that satisfies the following three properties: capacity limitation, skew symmetry, and flow conservation; For (u,v)∈ε, f(u,v) represents the flow of the edge, c(u,v)-f(u,v) represents the remaining capacity of the edge, and the flow of the entire network is That is, the sum of the outflow from the source point; The traditional millimeter wave network is based on the basic idea of ​​maximizing the received signal strength indication RSSI to solve the dynamic association strategy. The problem is modeled as the following minimum cost flow form Among them, N RF Represents the number of configured RF chains, R i,j represents the RSSI measured by the i-th AP for the j-th user, X i,j If 1 is set, it means that the i-th AP is associated with the j-th user; otherwise, it means that the i-th AP is not associated with the j-th user. Step 3: Considering the user service priority, the traditional model is expanded into a minimum cost flow model with upper and lower bounds: When the environmental interaction information feedback shows that a high-priority user is still not satisfied, the number of links serving this user can be increased according to the service strategy. Assuming that according to the service strategy, the at least number of links required to be provided for the jth user is d(j), the problem has the following form: For deployed APs with small-scale antenna arrays, large beam widths, and high probability of LOS line-of-sight, the RSSI index in the objective function is converted into relative physical distance. Among them, D i,j represents the relative physical distance between the i-th AP and the j-th user; Step 4: Build a capacity-cost network based on the problem: According to the problem modeling, a capacity-cost network with upper and lower capacity constraints is established, and according to the flow conservation property, it is transformed into a general form of flow network. After checking the satisfiability of the lower bound condition, the minimum cost maximum flow is solved in the residual network, and finally a dynamic association strategy is constructed according to the flow function. Step 5: Solve the dynamic association solution based on the continuous shortest path algorithm: Apply the minimum cost flow algorithm based on the continuous shortest path to solve the minimum cost maximum flow of s′-t′. If s′-t′ is not full flow, that is, it does not satisfy At this time, there is no feasible solution that meets the capacity lower bound condition, and the network needs to be adjusted according to the user priority strategy; In the residual network, the minimum cost flow algorithm based on the continuous shortest path is applied to solve the st minimum cost maximum flow, and a dynamic association strategy is constructed according to the network flow, that is, if f(a i ,u j ) is full flow, user j is associated with the i-th AP.

2. The dynamic association method based on upper and lower bound network flows in a non-cellular millimeter wave system according to claim 1, characterized in that: The step 4 specifically includes: Step 4.1: Order is the weight matrix of the bipartite graph. In the RMCFD algorithm with the minimum cost flow based on RSSI, W = R. In the DMCFD algorithm with the minimum cost flow based on relative physical distance, W = D. Add source and sink points s and t, construct node sets for AP and UE, and design the function add-edge(u,v,d(u,v),c(u,v),w(u,v)) to update the network structure, which represents an edge from node u to node v. The lower bound of the edge capacity is d(u,v), the upper bound of the edge capacity is c(u,v), and the cost of the edge is w(u,v). Considering that RSSI has negative values, and in user-intensive scenarios, the RF chain of the AP is generally fully utilized, based on the equivalence of the problem, a fixed offset is added to each element of the weight matrix to ensure that any element is non-negative. At this time, for the initial flow network, edges with non-zero capacity and negative weights are not processed. Step 4.2: Indicates AP i At most one data stream is provided to user j, and its contribution to the objective function is f(a i ,u j )×W i,j , if the environmental perception information fed back by the previous dynamic association shows that the channel quality is poor (a i ,u j ) is correct, then set c(a i ,u j )=0; where a i Representing AP i ,u j represents the jth user; Step 4.3: Indicates AP i The upper bound of the data flow is N RF ; Indicates that the user is served by any number of APs; Step 4.4: add-edge(t,s,0,∞,0) to construct a network of passive points and sinks. At this time, we can solve the problem according to the solution of the minimum cost flow of the passive sink and the lower bound. Step 4.5: Add auxiliary source points and sink points s′, t′, consider the capacity lower bound, and add auxiliary edges according to the classification discussion; Assuming that any edge e(u,v) already has the flow of d(u,v), for any node in the flow network The initial inflow minus the initial outflow is x, and it can be processed in three cases if necessary: ​​if x=0, the flow conservation property is satisfied and no additional edge is needed; if x>0, it means that the node inflow is too large, and an additional edge with a capacity of x is required from s′ to v; if x<0, it means that the node outflow is too large, and an additional edge with a capacity of -x is required from v to t′.

3. The dynamic association method based on upper and lower bound network flows in a non-cellular millimeter wave system according to claim 1, characterized in that: In the non-cellular millimeter wave system, a user communication service quality QoS model is established to calculate the achievable rate and of the system under different user densities, and to judge the effectiveness of the dynamic association scheme. The Monte Carlo method is used, and the main steps are as follows: Step 1: A simple simulation is performed on the user QoS situation. Specifically, users in the scenario will have different QoS, and there will also be sudden message demands or long data waiting queues. The system has a set of user service priority decision strategies, which dynamically adjusts the priority of the served users through the continuous changes in the communication environment. Users with higher priorities are modeled as demand users in the above algorithm. Users are randomly distributed, and a certain density of high-priority users is set; Step 2: For the scenario where the number of APs is N, the AP set is AP k The associated user set is The associated user set size is The simulated precoding matrix is The channel between user i is If AP k Associated with user i, the transmitted data flow is represents digital precoding, B represents channel bandwidth, then for the i-th user, the achievable communication rate is R i The system and rate are

Images