Network slice resource allocation method with deterministic service quality guarantee

Abstract

The invention provides a network slice resource allocation method with a deterministic service quality guarantee, which comprises the following steps of: (1) constructing a cellular-free large-scale multiple input multiple output (CF-mMIMO) network architecture which takes a user as a center and is provided with multiple slices; (2) constructing a resource allocation optimization problem by taking maximization of the sum rate of all users as a target based on the service quality requirement of each user equipment; (3) eliminating non-convexity of a short packet rate formula through a successive rate lower bound maximization algorithm; (4) eliminating the non-convexity of the signal to interference plus noise ratio formula through a continuous convex approximation (SCA) algorithm; (5) optimizing the bandwidth resource and the power resource through an alternate optimization method; (6) optimizing the user slice correlation variable by adopting a simulated annealing algorithm; and (7) calculating the maximum sum rate according to the optimal allocation result. According to the method, the sum rate is remarkably improved, the transmission delay is reduced, the resource utilization efficiency is improved, and the method has good expansibility and practical value.

Description

Technical Field

[0001] The present invention relates to the field of 6G wireless communication technologies, and particularly to a network slice resource allocation method with guaranteed deterministic quality of service, which is applicable to scenarios such as smart grids and autonomous driving that require ultra-low latency and high reliability. Background Art

[0002] With the rapid development of wireless communication technologies, the sixth-generation mobile communication technology (6G) has become a hot topic in research and development. 6G networks need to support performance metrics such as ultra-low latency, ultra-high reliability, and massive access to meet the complex requirements of vertical industries such as intelligent manufacturing, autonomous driving, and remote healthcare. Network slicing technology is a virtualization-based resource management technology that enables customized support for different service requirements by creating multiple logically isolated virtual networks (slices) on the same physical network. The need for network slicing in 6G is particularly urgent because traditional "one-size-fits-all" network designs are difficult to meet the service differentiation requirements in highly dynamic and complex scenarios. Through network slicing, 6G can effectively overcome problems such as low utilization of physical network resources and insufficient QoS guarantee, and provide isolation and deterministic quality of service guarantee for different service scenarios.

[0003] As a deep integration of power and information and communication technologies, the smart grid is an important support for the transformation of the power industry towards high efficiency, safety, stability, and intelligence. With the continuous increase in the communication network requirements of the smart grid, the limitations of the fifth-generation public network technology (5G) have gradually emerged. Although 5G provides ultra-high bandwidth and low-latency connection capabilities, it is still difficult to fully meet the extremely high reliability and strict latency requirements of the smart grid in control-type services. For example, critical tasks such as grid protection and load control require "fiber-level" performance, including millisecond-level latency and nearly 100% data transmission reliability. Moreover, the network slicing technology of 5G still has deficiencies in resource isolation and hard quality guarantee, especially in complex scenarios with multi-slice sharing, where resource competition and uncertainty problems are difficult to avoid. This gap makes it difficult for 5G to meet the hard expectations of the smart grid for the communication quality of control-type services.

[0004] The emergence of 6G provides an important opportunity to solve the above problems. Through more sophisticated network slicing management and optimized architecture design, 6G can flexibly allocate network resources according to the diverse needs of smart grids. Especially in control business scenarios, 6G can provide stronger resource exclusivity and ensure ultra-reliable and low-latency communication services through customized high-priority slices. In addition, 6G introduces sub-6GHz, millimeter wave and terahertz band technologies in frequency band selection. By using its ultra-large spectrum bandwidth, it significantly improves the data transmission rate and network capacity, and provides hardware foundation support for low-latency and high-reliability control services. However, in smart grid communication networks, resource allocation based on network slicing still faces many challenges. Due to the diversity of smart grid services, the QoS requirements of different services vary significantly, and it is difficult for existing resource allocation methods to formulate efficient resource allocation strategies based on these differences. In addition, in the unique network architecture of 6G, how to reasonably allocate frequency and power resources based on network slicing technology through access selection and resource optimization mechanisms to meet the stringent requirements of smart grid control services for reliability and low latency is still a hot issue that needs to be studied urgently.

[0005] In summary, the existing 5G network slicing technology has deficiencies in resource isolation and hard QoS guarantee, and it is difficult to meet the millisecond latency and ultra-high reliability requirements of smart grid control services. Traditional resource allocation methods do not take into account the short packet rate characteristics of URLLC services, and resource competition is prone to occur in multi-slice sharing scenarios. The present invention solves the above problems by combining the CF-mMIMO architecture with a new optimization algorithm. Summary of the invention

[0006] Technical problem: The present invention provides a network slice resource allocation method with deterministic service quality guarantee to solve the problems in the prior art, such as unreasonable resource allocation leading to resource waste, failure to apply the actual 6G network scenario architecture, and failure to consider the QoS constraints of URLLC services. This method, on the premise of meeting the deterministic QoS requirements of each user, aims to maximize the sum rate of all users, realizes the reasonable allocation of system resources, improves the sum rate, reduces the transmission delay and avoids the waste of resources.

[0007] Technical solution: The technical problem to be solved by the present invention is to provide a network slice resource allocation method with deterministic service quality guarantee. On the basis of ensuring the user's QoS requirements, the optimization goal is to maximize the sum rate of all users. The optimization problem is solved by the successive rate lower bound maximization algorithm, SCA algorithm and simulated annealing algorithm to achieve reasonable resource allocation.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] A network slice resource allocation method with deterministic quality of service guarantee, comprising the following steps: constructing a multi-slice user-centric cell-free massive multiple input multiple output (CF-mMIMO) network architecture and defining the connection relationship between access points (APs) and user equipment (UEs); constructing a resource allocation optimization problem with the goal of maximizing the sum rate of all users based on the QoS requirements of each UE;

[0010] decomposing the complex non-convex problem; eliminating the non-convexity of the short packet rate formula through the successive rate lower bound maximization algorithm; eliminating the non-convexity of the signal-to-interference-plus-noise ratio formula using the SCA algorithm; alternately optimizing to solve the optimal bandwidth resource allocation and power resource allocation; solving the optimal user slice association variable through the simulated annealing algorithm; calculating the maximum sum rate according to the optimal resource allocation result.

[0011] Preferably, a large number of APs and UEs are distributed in the CF-mMIMO network. The AP set is M = {1,..., m,..., M}, and the UE set is U = {1,..., i,..., U}. All APs are connected to a central processing unit (CPU). The bandwidth resource is sliced into equal-bandwidth physical resource blocks (PRBs), and the total number of PRBs is K. UEs are assigned to different slices. The slice set is S = {1,..., s,..., S}. Each PRB can only be used by all UEs within one slice. All APs transmit data to the connected UEs downlink, and interference will occur between UEs in the same slice. UEs communicate with APs through ultra-reliable low-latency communication (URLLC) data packets, and the upper limit of the transmission delay that UEi can tolerate is t i,b 。

[0012] Preferably, in the CF-mMIMO network, UEs select the set of APs serving them by measuring the large-scale channel gain, select the APs that satisfy the channel gain not less than the preset threshold, and if none exist, select the AP with the maximum gain. The construction rule of the service set of UEs is:

[0013] C i ={m|α i,m ≥β i,m (d 0 )}∪{argmax m∈M α i,m}

[0014] Among them, C i represents the set of APs serving UE i, and α i,m , β i,m respectively represent the large-scale fading parameter and path loss between UE i and AP m. β i,m is a function of the distance d, and d 0 is the distance threshold of the preset path loss. Each AP determines the connection relationship with the UE based on the service set of the UE, and uses the 0-1 variable c i,m to represent the connection relationship between UE i and AP m. If there is a connection, then c i,m = 1, otherwise c i,m = 0.

[0015] Preferably, the optimization variables for the constructed resource allocation optimization problem are: the association variable ω i,s between UE i and slice s, the number of PRBs δ s allocated to slice s, and the power allocation coefficient ξ i,m associated with UE i-AP m. The signal-to-interference-plus-noise ratio Γ i at UE i on a single PRB is defined as:

[0016]

[0017] where ρ 0 and P respectively represent the total transmit power and the number of antennas of a single AP, and γ i,m represents the independent channel gain on a single antenna between UE i and AP m. Based on the formula of Γ i , the achievable short-packet data rate of UE i for URLLC services on a single PRB can be obtained as:

[0018]

[0019] where W represents the bandwidth of a single PRB, τ c and τ respectively represent the period of the coherence interval and the length of the pilot sequence, n represents the length of the URLLC data packet block, Q -1 (·) represents the inverse function of the Gaussian Q function, and ε i represents the decoding error probability of UE i. Therefore, the total data rate at UE i can be obtained as:

[0020]

[0021] The sum of the data rates of all UEs is:

[0022]

[0023] Based on the above formulas, a resource allocation problem can be constructed:

[0024]

[0025] Among them, represents a positive integer, N 0 represents the total data volume sent from the AP to the UE in the URLLC service. Formulas (a) and (b) restrict that each UE can only be associated with one slice; formula (c) restricts that each slice is non-empty; formulas (d) and (e) restrict that the number of PRBs obtained by each slice is an integer and at least one PRB is obtained, and at the same time all PRBs are to be allocated without being idle; formulas (f) and (g) specify the value range of the power control coefficient, restricting that each AP cannot exceed the total power upper limit; formula (h) is a hard constraint on the transmission delay of the UE.

[0026] Preferably, the successive rate lower bound maximization algorithm is used to eliminate the non-convexity of the short packet rate formula. For the short packet data rate R of UEi i , there is a lower bound:

[0027]

[0028] where λ i is a rate lower bound expansion point not less than 0.28. An auxiliary variable φ i is introduced as the lower bound of Γ i , and φ i is used as the rate lower bound expansion point. For each update of the lower bound, given a set of {φ′ i} updated in the previous round of iteration and the obtained optimization variables {ω i,s}, {δ s}, {ξ i,m}, let λ i = φ′ i , substitute the data rate lower bound into the resource allocation problem, and we can get:

[0029]

[0030] This problem is a convex optimization problem, and the optimal {φ i} of this round of iteration can be solved by the SDPT3 solver in the MATLAB CVX toolbox, and then the convex lower bound of the short packet data rate is updated.

[0031] Preferably, the SCA algorithm first introduces an auxiliary variable as the lower bound after taking the square root of the signal-to-interference-plus-noise ratio. Therefore, there is:

[0032]

[0033] Let The left side of the inequality is a non-convex function, and the right side is a convex function. Therefore, for the left function, its first-order Taylor expansion is used for approximation. The first-order Taylor expansion of

[0034]

[0035] where is the first-order Taylor expansion point, and the expansion point is updated through iterative loops. Substituting the first-order Taylor expansion into the lower bound constraint of the square root of the signal-to-interference-plus-noise ratio can eliminate the non-convexity of the signal-to-interference-plus-noise ratio.

[0036] Preferably, the alternating optimization method first fixes the power allocation coefficients {ξ i,m} of the AP-UE association, given the optimal {φ i} of this round of iteration of the successive rate lower bound maximization algorithm and the optimal first-order Taylor expansion point of the previous round of SCA algorithm iteration The original resource allocation optimization problem is transformed into:

[0037]

[0038] where and define Relaxing equation (b) to δ s > 0, this problem becomes a convex optimization problem, and the optimal continuous form of {δ s} can be solved by the SDPT3 solver of the MATLAB CVX toolbox. By rounding, the optimal discrete form of {δ s} can be obtained.

[0039] Next, fix the number of PRBs {δ s} allocated to each slice. Through variable substitution, define The original resource allocation optimization problem is transformed into:

[0040]

[0041] This problem is a convex problem of quadratic constraint quadratic programming, and the optimal {υ i} and {η i,m} can be solved by the SDPT3 solver of the MATLAB CVX toolbox, and then update the {ξ i,m} of this round of iteration of the SCA algorithm and the first-order Taylor expansion point for the next round of iteration

[0042] Preferably, the simulated annealing algorithm is performed after updating the PRB allocation and power allocation. After optimizing the user slice association variable {ω i,s} Then return to the next iteration of the successive rate lower bound maximization algorithm to start the algorithm. The loop terminates until the change value of the network sum rate between two iterations is lower than the threshold.

[0043] Finally, the optimal resource allocation scheme can be obtained, and the maximum sum rate of the network can be calculated according to the optimal resource allocation scheme.

[0044] Preferably, the method is applied to intelligent grid control services, supporting millisecond-level delay and nearly 100% transmission reliability in the wide temperature range of -40 to 60 °C.

[0045] The present invention also provides a communication system, adopting the network slice resource allocation method described above, including a CF-mMIMO network architecture, a URLLC service support module, and a resource allocation optimization module.

[0046] Compared with the prior art by adopting the above technical solutions, the network slice resource allocation method with deterministic quality of service guarantee of the present invention has the following technical effects:

[0047] The network slice resource allocation method with deterministic quality of service guarantee of the present invention constructs a network slice resource allocation problem in a multi-slice user-centric CF-mMIMO network with the goal of maximizing the sum rate of all users and considering the deterministic QoS requirements of UEs. The non-convexity is eliminated by the successive rate lower bound maximization algorithm and the SCA algorithm, and the optimization variables are solved by the alternating optimization method and the simulated annealing algorithm, and finally the optimal resource allocation scheme is obtained. Compared with the traditional network slice resource allocation method, the present invention considers the emerging network scenario of 6G, adopts a short packet data rate formula corresponding to the more practical URLLC service, and takes the QoS requirements of UEs as constraint conditions when constructing the optimization problem. Compared with the traditional network slice resource allocation algorithm, the algorithm proposed by the present invention has a faster convergence speed and a higher network sum rate, and can obtain a resource allocation scheme with higher resource utilization. Description of the Drawings

[0048] Figure 1 It is a diagram of a multi-slice user-centric CF-mMIMO network architecture.

[0049] Figure 2 It is a flowchart of the implementation of a network slice resource allocation method with deterministic quality of service guarantee.

[0050] Figure 3 It is the specific implementation process of the successive rate lower bound maximization algorithm.

[0051] Figure 4 It is the specific implementation process of the SCA algorithm.

[0052] Figure 5 This is the specific implementation process of the simulated annealing algorithm. Specific implementation manner

[0053] The following further elaborates on the specific implementation manner of the present invention in conjunction with the specification drawings and the 6G network scenario.

[0054] In the user - centered CF - mMIMO network architecture with multiple slices in this embodiment, services are provided for the URLLC services of users. A large number of APs \(M=\{1,\cdots,m,\cdots,M\}\) and UEs \(U = \{1,\cdots,i,\cdots,U\}\) are distributed in the network. All APs are connected to a CPU, and the upper limit of the transmission power of each AP is \(\rho\) 0 , the number of antennas is \(P\), and \(\xi\) i,m is used to represent the power allocation coefficient of AP \(m\) assigned to UE \(i\). \(\xi\) i,m is a continuous variable between 0 and 1. UEs are assigned to different slices, and the slice set is \(S=\{1,\cdots,s,\cdots,S\}\). Each UE can only be associated with one slice, and all slices are non - empty. The 0 - 1 variable \(\omega\) i,s is used to represent the association between the UE and the slice. If UE \(i\) is in slice \(s\), then \(\omega\) i,s = 1; otherwise, \(\omega\) i,s = 0. The bandwidth resource is divided into \(K\) PRBs with equal bandwidth, and the bandwidth of each PRB is \(W\). Each PRB can only be used by all UEs within one slice, and each slice is assigned at least one PRB. \(\delta\) s is used to represent the number of PRBs assigned to slice \(s\); all APs transmit data to the connected UEs in the downlink, and interference occurs between UEs in the same slice. UEs and APs communicate through URLLC data packets. The upper limit of the transmission delay that UE \(i\) can tolerate is \(t\) i,b . To meet the QoS requirements of UEs, the transmission delay \(t\) i of UE \(i\) implemented by the network cannot exceed \(t\) i,b . To simultaneously meet the QoS requirements of all UEs under the condition of limited spectrum resources and power resources and maximize the sum rate of all users on this basis, it specifically needs to be achieved through the following steps:

[0055] Step 1: Initialize the CF - mMIMO network, and UEs construct the set of APs that serve them.

[0056] (1) The channels between each AP and UEs are independent of each other. The channel parameter \(h\) i,m is used to characterize the channel fading between UE \(i\) and AP \(m\). The channel parameter is defined as:

[0057]

[0058] where \(\alpha\)i,m , β i,m and d i,m represent the large-scale fading parameter, path loss, and distance between UE i and AP m, respectively. β i,m is a function of d i,m . ψ i,m is a random variable that follows a log-normal distribution. α i,m is defined as the product of ψ i,m and β i,m . g i,m is the small-scale fading parameter. h i,m and g i,m are P×1 vectors. Each element of g i,m is a random variable that follows a complex Gaussian distribution with zero mean and unit variance.

[0059] (2) In the CF-mMIMO network, the UE selects the set of APs serving it by measuring the large-scale channel gain, selects the APs that satisfy the channel gain not less than the preset threshold, and if none exist, selects the AP with the maximum gain. The construction rule of the UE's serving set is as follows:

[0060] C i = {m | α i,m ≥ β i,m (d 0 )} ∪ {arg max m∈M α i,m}

[0061] where C i represents the set of APs serving UE i, and d 0 is the distance threshold of the preset path loss. Each AP determines the connection relationship with the UE based on the UE's serving set, and uses the 0-1 variable c i,m to represent the connection relationship between UE i and AP m. If there is a connection, then c i,m = 1, otherwise c i,m = 0.

[0062] Step 2: Based on the QoS requirements of the UE, construct a resource allocation optimization problem with the goal of maximizing the sum rate of all users.

[0063] (1) In the CF-mMIMO network, the AP periodically sends signals to the UE. Each period is called a coherent time slot, and its period is τ c . Each coherent time slot is divided into two stages: pilot training and downlink data transmission. The length of the pilot sequence is τ, the number of pilot training symbols in one coherent time slot is also τ, and the number of downlink data transmission symbols in one coherent time slot is τ c - τ. There are τ pairwise orthogonal pilot sequences.

[0064] The transmission power of each UE is p 0 , different pilot sequences are assigned to UEs of the same slice, and UEs of different slices can reuse the same pilot sequence. The UE transmits a pilot signal to the AP, and the AP processes the received pilot signal to be able to predict the channel parameters between the connected UEs. Assuming that the white noise at all APs follows a Gaussian distribution with zero mean and unit variance, the independent channel gain on a single antenna between UE i and AP m can be obtained as:

[0065]

[0066] (2) For the downlink data transmission phase of the coherent time slot, all APs send data signals to the connected UEs, process and send the symbol data through conjugate beamforming technology. Since different slices use different PRBs, different slices are isolated from each other in frequency. Therefore, only the signals sent to users within the same slice will interfere with each other. Assuming that the white noise at all UEs follows a Gaussian distribution with zero mean and unit variance, the signal-to-interference-plus-noise ratio on a single PRB at UE i can be calculated as:

[0067]

[0068] Based on the formula of Γ i , the achievable short-packet data rate of UE i for URLLC services on a single PRB can be obtained as:

[0069]

[0070] where n represents the length of the URLLC data packet block, Q -1 (·) represents the inverse function of the Gaussian Q function, and ε i represents the decoding error probability of UE i. Therefore, the total data rate at UE i can be obtained as:

[0071]

[0072] The sum of the data rates of all UEs is:

[0073]

[0074] (3) Based on the above content, with maximizing the system capacity as the objective function, a resource allocation problem can be constructed:

[0075]

[0076] where, represents a positive integer, N 0Denotes the total data volume sent from the AP to the UE in the URLLC service. Equations (a) and (b) constrain that each UE can only be associated with one slice; Equation (c) constrains that each slice is non-empty; Equations (d) and (e) constrain that the number of PRBs obtained by each slice is an integer and at least one PRB is obtained, and at the same time all PRBs are to be allocated and cannot be idle; Equations (f) and (g) specify the value range of the power control coefficient, constraining that each AP cannot exceed the total power upper limit; Equation (h) is a hard constraint on the transmission delay of the UE.

[0077] Step 3: Eliminate the non-convexity of the short-packet rate formula through the successive rate lower bound maximization algorithm.

[0078] (1) If this is the first iteration of the successive rate lower bound maximization algorithm, jump to (2); otherwise, given {ω i,s}, {δ s}, {ξ i,m} optimized in the previous iteration of the successive rate lower bound maximization algorithm, and the signal-to-interference-plus-noise ratio lower bound {φ′ i}, then jump to (3).

[0079] (2) This is the first iteration of the successive rate lower bound maximization algorithm. Initialize a set of {ω i,s}, {δ s}, {ξ i,m} and the signal-to-interference-plus-noise ratio lower bound {φ′ i} that satisfy the constraint conditions as the optimized variables updated in the previous iteration (the zeroth iteration) of the successive rate lower bound maximization algorithm. Initialize the maximum number of iterations I 1 of the successive rate lower bound maximization algorithm, the sum rate change value threshold and the sum rate C′ sum of the previous iteration (the zeroth iteration) of the successive rate lower bound maximization algorithm = 0.

[0080] (3) For the short-packet data rate R i of UE i, there exists a lower bound:

[0081]

[0082] where λ i is a rate lower bound expansion point not less than 0.28. Take φ i as the rate lower bound expansion point.

[0083] (4) Based on {ω i,s}, {δ s}, {ξ i,m} and the signal-to-interference-plus-noise ratio lower bound {φ′ i} optimized in the previous iteration, let λ i = φ′i Substituting the lower bound of the data rate into the resource allocation problem, we can obtain:

[0084]

[0085] This problem is a convex optimization problem, and the optimal {φ i} of this round of iteration can be solved by the SDPT3 solver in the MATLAB CVX toolbox. Assign {φ i} to the lower bound of the signal-to-interference-plus-noise ratio {φ′ i} in the previous round of iteration of the successive rate lower bound maximization algorithm. Update the convex lower bound of the short packet data rate.

[0086] Step 4: Eliminate the non-convexity of the signal-to-interference-plus-noise ratio formula through the SCA algorithm.

[0087] (1) If this is the first iteration of the SCA algorithm, jump to (2). Otherwise, given the updated {ω i,s}, {ξ i,m} in the previous round of iteration of the successive rate lower bound maximization algorithm and the updated {φi} in this round of iteration of the successive rate lower bound maximization algorithm, as well as the first-order Taylor expansion point optimized in the previous round of iteration of the SCA algorithm Then jump to (3).

[0088] (2) This is the first iteration of the SCA algorithm. Given the updated {ω i,s}, {ξ i,m} in the previous round of iteration of the successive rate lower bound maximization algorithm and the updated {φi} in this round of iteration of the successive rate lower bound maximization algorithm, initialize the first-order Taylor expansion point optimized in the previous round of iteration (the zeroth iteration) of the SCA algorithm The initialization operation is:

[0089]

[0090] Initialize the maximum number of iterations I of the SCA algorithm 2 , the threshold of the change value of the objective function and the objective function value V′ = 0 in the previous round of iteration (the zeroth iteration) of the SCA algorithm.

[0091] (3) Introduce an auxiliary variable as the lower bound after taking the square root of the signal-to-interference-plus-noise ratio. Therefore, we have:

[0092]

[0093] Let The left side of the inequality is a non-convex function, and the right side is a convex function. Therefore, for the left function, its first-order Taylor expansion is used for approximation. The first-order Taylor expansion of

[0094]

[0095] wherein is the first-order Taylor expansion point, and the expansion point is updated through iterative loop. Substituting the first-order Taylor expansion into the lower bound constraint of the square root of the signal-to-interference-plus-noise ratio can eliminate the non-convexity of the signal-to-interference-plus-noise ratio.

[0096] Step 5: Optimize the frequency resources and power resources respectively through the alternating optimization method.

[0097] (1) Fix the power allocation coefficients {ξ i,m} of the AP-UE association, and given the optimal {φ i} of the successive rate lower bound maximization algorithm in this round of iteration and the optimal first-order Taylor expansion point of the previous round of SCA algorithm iteration The original resource allocation optimization problem is transformed into:

[0098]

[0099] wherein and define Relax equation (b) to δ s > 0, , and this problem becomes a convex optimization problem. The optimal continuous-form {δ s} can be solved through the SDPT3 solver of the MATLAB CVX toolbox, and the optimal discrete-form {δ s} can be obtained by rounding.

[0100] (2) Fix the number of PRBs {δ s} allocated to each slice, and define through variable substitution The original resource allocation optimization problem is transformed into:

[0101]

[0102] This problem is a convex problem of quadratic constraint quadratic programming. The optimal {υ i} and {η i,m} can be solved through the SDPT3 solver of the MATLAB CVX toolbox, and then update the {ξ i,m} of this round of iteration of the SCA algorithm and the first-order Taylor expansion point for the next round of iteration Assign to the first-order Taylor expansion point in the previous round of iteration of the SCA algorithm

[0103] (3) Calculate the optimal objective function value V of the quadratic constraint quadratic programming problem in this round of iteration of the SCA algorithm, and judge Whether it holds. If it holds, jump back to step 4; if not, use the {δ s}, {ξ i,m} optimized in this round of iteration of the SCA algorithm as the {δ s}, {ξ i,m} optimized in this round of iteration of the successive rate lower bound maximization algorithm, and then enter step 6.

[0104] Step 6: Optimize the user slice association variables using the simulated annealing algorithm.

[0105] (1) Given the {δ s}, {ξ i,m} optimized in this round of iteration of the successive rate lower bound maximization algorithm. Initialize the initial temperature T 0 , the termination temperature T min , the temperature reduction rate α t and the maximum number of iterations I t at each temperature. Randomly initialize {ω′ i,s}, ensuring that there is at least one UE in each slice. Calculate the sum rate C sum (ω′ i,s ), and initialize the optimal The current temperature T = T 0 . Loop I t times at one temperature.

[0106] (2) Generate a neighborhood solution by randomly generating The generation operation of the neighborhood solution is as follows: with a probability of 50%, randomly transfer a UE in a slice to another slice; with a probability of 50%, randomly swap a UE in each of two slices. It is necessary to ensure that each slice is non-empty when generating a new solution. Calculate the sum rate corresponding to the neighborhood solution

[0107] (3) If then assign to {ω′ i,s}, C sum (ω′ i,s ), and jump to (5); otherwise, enter step (4)

[0108] (4) Randomly generate a number between 0 and 1. If the number is less than then assign to {ω′ i,s}, C sum (ω′ i,s ).

[0109] (5) If the number of iterations at the current temperature has not reached It , return to (2); otherwise, judge C sum (ω′ i,s ) is greater than If it holds, then assign {ω′ i,s}, C sum (ω′ i,s ) to

[0110] (6) Update the current temperature T = T × α t . Judge whether T < T min holds. If it holds, enter (7); otherwise, return to (2).

[0111] (7) Assign to the {ω i,s} optimized in this round of iteration of the successive rate lower bound maximization algorithm.

[0112] Step 7: Calculate the maximum sum rate according to the optimal allocation result.

[0113] (1) Given the {δ s}, {ξ i,m} and {ω i,s} optimized in this round of iteration of the successive rate lower bound maximization algorithm. Calculate the sum rate C sum . Judge If it holds, then assign C sum to C′ sum , and jump back to Step 3; otherwise, enter (2).

[0114] (2) Take the {δ s}, {ξ i,m} and {ω i,s} optimized in this round of iteration of the successive rate lower bound maximization algorithm as the optimal allocation result, and calculate the maximum sum rate

[0115] This embodiment targets the 6G network scenario. Based on the user-centric CF-mMIMO network, with the goal of maximizing the sum rate of all users, combined with the deterministic QoS requirements of the UE, by constructing a network slice resource allocation problem, a corresponding optimization scheme is proposed. First, the successive rate lower bound maximization algorithm and the SCA algorithm are used to eliminate the non-convexity in the problem, and the optimization variables are solved by the alternating optimization method and the simulated annealing algorithm, and finally an optimal resource allocation scheme is achieved. Compared with the traditional network slice resource allocation method, the deterministic QoS requirements of the UE are used as constraints, and the short packet data rate formula corresponding to the URLLC service that conforms to the actual scenario is adopted; through algorithm design, the resource allocation scheme has a faster convergence speed, significantly improves the network sum rate, effectively improves the resource utilization rate, and provides an efficient and reliable solution for network slice resource allocation in the 6G network scenario.

[0116] Another embodiment is a communication system that adopts the network slice resource allocation method described above, including a CF-mMIMO network architecture, a URLLC service support module, and a resource allocation optimization module.

[0117] The specific embodiments of this specification are described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A network slice resource allocation method with deterministic service quality assurance, characterized in that: The following steps are involved: Build a multi-slice user-centric CF-mMIMO network architecture and define the connection relationship between access points (APs) and user equipment (UEs) in the CF-mMIMO network architecture; Determine the QoS requirements of each UE based on the connection relationship, and construct a resource allocation optimization problem with the goal of maximizing the sum rate of all users; The non-convexity of the short packet rate formula in the resource allocation optimization problem model is eliminated by maximizing the successive rate lower bound algorithm; The continuous convex approximation (SCA) algorithm is used to eliminate the non-convexity of the signal-to-interference-noise ratio formula in the resource allocation optimization problem model; The alternating optimization method is used to optimize bandwidth resource and power resource allocation respectively to obtain the optimal resource allocation result; The user slice association variables are optimized by simulated annealing algorithm to obtain the optimal user slice association result; The maximum sum rate is calculated based on the optimal resource allocation result and the optimal user slice association result.

2. The method according to claim 1, characterized in that In the CF-mMIMO network: The bandwidth resources are divided into physical resource blocks (PRBs) of equal bandwidth, and each PRB is only used by UEs in one slice; The UE selects the service AP set by measuring the large-scale channel gain. If the channel gain does not reach the threshold, it selects the AP with the largest gain. The QoS requirement is defined as that the transmission delay of the UE must not exceed a preset tolerance upper limit.

3. The method according to claim 1, characterized in that The successive rate lower bound maximization algorithm includes: Iteratively updating a convex lower bound of a short packet rate, wherein the convex lower bound is a convex function of a signal to interference noise ratio lower bound; Each update of the convex lower bound is obtained by solving a convex optimization problem with it as the optimization variable; The convex function of the lower bound of the signal-to-interference-noise ratio is solved by the SDPT3 solver of the MATLAB CVX toolbox.

4. The method according to claim 1, characterized in that: The SCA algorithm introduces an auxiliary variable as the lower bound of the square root of the signal to interference noise ratio and uses a first-order Taylor expansion to eliminate non-convexity.

5. The method according to claim 1, characterized in that The alternating optimization method comprises: The power allocation coefficient associated with a fixed user slice is used to optimize the number of PRBs allocated to each slice; The number of PRBs allocated to each slice is fixed to optimize the power allocation coefficient; The two optimization problems are solved by the SDPT3 solver of MATLAB CVX toolbox.

6. The method according to claim 1, characterized in that The simulated annealing algorithm optimizes user slice association variables through a neighborhood solution generation strategy, including randomly transferring or exchanging UEs within a slice to ensure that each slice is non-empty.

7. The method according to claim 1, characterized in that The method is applied to smart grid control services, supporting millisecond-level delay and nearly 100% transmission reliability in a wide temperature range of -40 to 60°C.

8. A communication system, characterized in that: The network slice resource allocation method described in any one of claims 1 to 7 includes a CF-mMIMO network architecture, a URLLC service support module and a resource allocation optimization module.

Images