Spatial Domain Resource Allocation Method Based on Determinantal Point Process Learning

Abstract

The present invention belongs to the field of mobile communication technologies, and relates to a spatial domain resource allocation method based on the learning of a determinantal point process. The method is applied to an NR-U / Wi-Fi coexistence network, and includes calculating a spatial domain resource allocation scheme by using a heuristic resource allocation algorithm, using the scheme as a training set to train a kernel matrix of a determinantal point process learning model to obtain optimal learning parameters, then using the optimal learning model to predict scenario parameters to be processed, and outputting a corresponding spatial domain resource allocation scheme; and in the current NR-U / Wi-Fi coexistence network, precoding a transmitted signal on the base station side. The present invention can convert the solution of a complex optimization problem into a simple calculation of the determinant of a kernel matrix, and at the same time solves the problems of high computational complexity and exponential growth of the computational complexity with the expansion of the network scale, and is applicable to large-scale and densely deployed NR-U / Wi-Fi coexistence networks.

Description

Technical Field

[0001] The present invention belongs to the technical field of mobile communications and relates to a method for allocating airspace resources in an NR-U / Wi-Fi coexistence network based on DPPL. Background Art

[0002] The Cisco report points out that the mobile data traffic of the mobile network reached 3.3 ZBs in 2017 and will grow at an annual growth rate of 65% during the period of 2017 - 2022. To meet the growth of this explosive data traffic, the capacity of the cellular network urgently needs to be greatly improved.

[0003] In particular, in the application scenarios of the fifth-generation mobile network defined by the International Telecommunication Union, there are enhanced mobile broadband (eMBB), ultra-high reliable and low-latency communication, and massive machine-type communication. Among them, enhanced mobile broadband requires the network capacity to be increased by 1000 times. However, since the data traffic in the current authorized frequency band has tended to saturate, the scarcity of spectrum resources has become a bottleneck for achieving the goal of enhanced mobile broadband. Therefore, cellular network operators have turned their attention to the 5GHz unlicensed frequency band with rich spectrum resources. However, this frequency band is currently used by Wi-Fi networks. The channel access mechanism adopted by Wi-Fi networks is the carrier sense multiple access with collision avoidance (CSMA / CA) mechanism based on the LBT strategy, while the channel access mechanism used by cellular networks is the centralized scheduling mechanism. Therefore, if a suitable coexistence mechanism is not deployed on the side of the cellular base station operating in the unlicensed frequency band, the transmission performance of the Wi-Fi network will be greatly attenuated.

[0004] Based on the above analysis, how to ensure the friendly coexistence of the cellular network and the Wi-Fi network in the unlicensed frequency band is a very crucial problem that urgently needs to be solved. Recently, a new coexistence mechanism based on airspace resources has been proposed in the academic community to reduce the interference of cellular base stations in the coexistence network on Wi-Fi devices operating on the same unlicensed channel, that is, the airspace coexistence mechanism. However, the heuristic algorithm designed based on the airspace coexistence mechanism has a very high computational complexity and is not suitable for large-scale and densely deployed coexistence networks. Summary of the Invention

[0005] The objective of the present invention is to provide a spatial domain resource allocation scheme for NR-U / Wi-Fi coexistence networks based on DPPL. In this scheme, it is assumed that the cellular network can obtain information about all other Wi-Fi devices within its coverage area through a control channel, and the base station can precode the transmitted signal through the spatial domain allocation scheme to change the reception mode of the signal at the receiving side. The present invention proposes a low-complexity spatial domain resource allocation method based on the learning of the determinantal point process. The method is applied to the NR-U / Wi-Fi coexistence network and includes:

[0006] Based on the scenario of the NR-U / Wi-Fi coexistence network, a spatial domain coexistence mechanism model with maximizing throughput as the optimization objective is constructed;

[0007] According to the first scenario parameters of the NR-U / Wi-Fi coexistence network, the optimization objective is solved based on a heuristic algorithm of convex optimization to obtain the first spatial domain resource allocation scheme;

[0008] Taking the first scenario parameters of the NR-U / Wi-Fi coexistence network as training set data and the corresponding first spatial domain resource allocation scheme as the training set label, input them into the determinantal point process learning model for training, and output the optimal learning parameters of the kernel matrix;

[0009] Input the third scenario parameters of the NR-U / Wi-Fi coexistence network into the model corresponding to the optimal learning parameters, and output the third spatial domain resource allocation scheme;

[0010] Using the result of the third spatial domain resource allocation scheme, in the current NR-U / Wi-Fi coexistence network, precode the transmitted signal at the base station side.

[0011] Advantages of the present invention:

[0012] In summary, the present invention proposes an optimal resource allocation scheme for NR-U / Wi-Fi coexistence networks based on DPPL. With the help of this scheme, not only can the throughput of the coexistence network be improved while ensuring a certain Wi-Fi throughput, but also compared with the traditional heuristic algorithm based on convex optimization, the present invention can greatly reduce the computational complexity. Therefore, the present invention can be better applied to large-scale coexistence NR-U / Wi-Fi networks with dense deployment. Description of the Drawings

[0013] Figure 1 It is a coexistence NR-U / Wi-Fi network model diagram in an embodiment of the present invention;

[0014] Figure 2 It is a flowchart of the spatial domain resource allocation method based on the learning of the determinantal point process in an embodiment of the present invention;

[0015] Figure 3 This is the schematic diagram of the beam-based transmission mechanism - the spatial domain coexistence mechanism in the embodiments of the present invention. Specific embodiments

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0017] See Figure 1 , Figure 1 The coexisting NR-U / Wi-Fi network model in [reference] consists of a cellular network and a Wi-Fi network operating in the same unlicensed band, specifically including 1 base station, N UEs, and M Wi-Fi clusters.

[0018] In the embodiments of the present invention, the cellular network includes one base station (Base station, BS) and N cellular users (User equipments, UEs), denoted by u BS and , and the N UEs are uniformly distributed within the coverage area of the base station. It is assumed that the BS is equipped with an array antenna including N A antenna elements, and other UEs within its coverage area are only equipped with a single antenna. The base station can simultaneously serve multiple UEs through spatial division multiplexing, and the transmission power of the base station is denoted as P bs . For the Wi-Fi network, the present invention assumes that all APs and STAs are only equipped with a single antenna, each AP serves only one STA, and the transmission power of the AP is denoted as P ap . In addition, the present invention considers a path loss model based on logarithmic distance: l(d) = α θ log 10 (d) + β θ + 20log 10 (f c ), where when the subscript θ = l represents the cellular network and θ = w represents the Wi-Fi network, d represents the Euclidean distance between the transmitter and the receiver, and f c represents the carrier frequency in GHz. Finally, the present invention assumes that there is a small-scale fading (i.e., Rayleigh fading) in the coexisting network. Specifically, the present invention uses to represent the small-scale fading in the cellular network and to represent the small-scale fading in the Wi-Fi network.

[0019] In the embodiments of the present invention, the Wi-Fi network includes M Wi-Fi Access Points (Aps) and Wi-Fi Stations (STAs), denoted by and respectively, and the STA is only served by the AP. The positions of all UEs, APs, and STAs are uniformly distributed within the coverage of the BS. Meanwhile, the number M of APs ~ poisson(λ m ) follows a Poisson distribution with parameter λ m , where λ m = 20 and the case of M = 0 is ignored. Therefore, the positions of the APs follow a Poisson point process (PPP). Since all Wi-Fi APs adopt a channel access mechanism based on LBT, any two nearby APs cannot access the channel for transmission simultaneously. Therefore, it is assumed that the distance between any two APs is large enough to ensure that any Wi-Fi AP does not detect other Wi-Fi signals greater than the LBT threshold of -82 dBm. Also, the present invention assumes that there is sufficient data at the base station and all Wi-Fi APs. Therefore, the BS and Wi-Fi APs always have data to send.

[0020] Based on the above scenario of the NR-U / Wi-Fi coexistence network, Figure 2 is a flowchart of the spatial domain resource allocation method based on the learning of the determinantal point process in the embodiments of the present invention. As Figure 2 shown, the method includes:

[0021] 101. Based on the scenario of the NR-U / Wi-Fi coexistence network, construct a spatial domain coexistence mechanism model with maximizing throughput as the optimization goal;

[0022] In the embodiments of the present invention, it is considered that there is perfect Channel State Information (CSI) at the base station. And a base station equipped with N A antennas can provide N A-1 degree of spatial freedom (DOFs). Therefore, through beamforming of the signals transmitted by the base station, the base station can not only reduce the interference to Wi-Fi devices, but also bring beamforming gain to the scheduled UEs. When the base station transmits on the unlicensed channel, the present invention assumes that all Wi-Fi APs detecting the base station signals are interference-suppressed. Therefore, the base station and all APs can transmit data on the unlicensed channel simultaneously. Therefore, this embodiment focuses on the total throughput gain of the coexistence network, which includes the throughput gain of the cellular network and the throughput gain of the Wi-Fi network. First, the signal-to-interference plus noise ratio (SINR) at the UE is expressed as:

[0023]

[0024] Where, represents the Euclidean distance between the base station and . represents to . Therefore, the present invention can calculate the throughput at through the following formula:

[0025]

[0026] Where, B represents the channel bandwidth.

[0027] In addition, the SINR at is expressed as:

[0028]

[0029] Where, represents to . represents the Euclidean distance from the BS to . Therefore, the throughput at can be expressed as:

[0030]

[0031] Therefore, the total throughput of the coexistence network is expressed as:

[0032]

[0033] Therefore, with the maximization of the throughput of the coexistence network as the optimization objective, the optimization objective model constructed in this embodiment is expressed as:

[0034]

[0035] s.t.

[0036]

[0037] C2: x ∈ {0, 1}, y ∈ {0, 1}

[0038] wherein, {x i}, {y j}, are binary variables, x i indicates whether a cellular user is scheduled, y j indicates whether a Wi-Fi station user is interference suppressed, x i = 1 indicates that the cellular user is scheduled, y j = 1 indicates that the Wi-Fi station user is interference suppressed; a l and a w respectively represent the priority factors of the cellular network and the Wi-Fi network, and a l + a w = 1; Constraint C1 means that the number of applicable airspace resources is less than the number of spatial degrees of freedom that the base station can provide, N represents the number of cellular users, M represents the number of Wi-Fi users, i.e., including the number of Wi-Fi station users and the number of Wi-Fi access point users, N A represents the number of antennas of the base station; Constraint C2 restricts x and y to be binary variables, x represents the binary variable of the cellular user, y represents the binary variable of the Wi-Fi station user, ω k represents the binary variable of the Wi-Fi access point user being interference suppressed, ω k = 1 indicates that the Wi-Fi access point user is interference suppressed.

[0039] In a preferred embodiment of the present invention, in order to avoid waste of spatial degrees of freedom, this embodiment needs to determine which Wi-Fi APs need to be interference suppressed, rather than blanking all Wi-Fi Aps in the above embodiment. Therefore, the present invention defines ω k to represent whether the AP detects a base station signal higher than the energy detection threshold. Therefore, the present invention has:

[0040]

[0041] wherein, P bs represents the transmission power of the base station for the cellular network; l() represents the path loss calculation function; Denote the Euclidean distance between the base station and the Wi-Fi access point user ; Denote the small-scale fading of the base station to the Wi-Fi access point user ; σ 2 Denote the noise power; γ denotes the energy detection threshold.

[0042] 102. According to the first scenario parameters of the NR-U / Wi-Fi coexistence network, use a heuristic algorithm based on convex optimization to solve the optimization objective, and obtain the first airspace resource allocation scheme;

[0043] In the embodiments of the present invention, since only the optimization objective is constructed in the above implementation and the specific network scenario has not been involved yet, the present invention needs to collect the scenario parameters of the large-scale NR-U / Wi-Fi coexistence network. For the sake of distinction, the scenario parameters here are called the first scenario parameters. The first scenario parameters include the transmission powers of the base station and the AP, the channel gains between each transceiver device, and the LBT threshold of the Wi-Fi network, etc.; input the first scenario parameters into the airspace coexistence mechanism model with maximizing throughput as the optimization objective, and use a heuristic algorithm based on convex optimization to solve it. Considering that due to the existence of constraint C2, the above optimization problem is a non-convex optimization problem, and the above optimization problem is a mixed integer programming problem (Mixed integer quadratic programming, MIQP), so the present invention readjusts the optimization problem and uses a heuristic algorithm based on convex optimization to solve it, specifically including:

[0044] Step 1: Introduce slack variables and auxiliary variables, and perform an equivalent transformation on the original optimization problem to obtain another solvable optimization model,

[0045] Step 2: Define a very small Gap threshold ε, such as ε = 0.01,

[0046] Step 3: Use the Gurobi solver to solve the transformed optimization problem to obtain the gap Gap between the optimal value,

[0047] Step 4: Determine whether Gap is less than ε. If so, execute Step 5; if not, repeat Step 3,

[0048] Step 5: Obtain the optimal solution {x *},{y *}, and the optimal solution {x *},{y *} is the first airspace resource allocation scheme, that is, the first airspace resource allocation scheme obtained according to the first scenario parameters. These first airspace resource allocation schemes are the optimal airspace resource allocation schemes belonging to the first scenario parameters. This method can obtain a series of optimal airspace resource allocation schemes, but its complexity is relatively high. Therefore, the embodiments of the present invention will adopt the method of determinantal point process learning in the subsequent process to reduce the complexity.

[0049] 103. Use the first scenario parameters of the NR-U / Wi-Fi coexistence network as training set data, and use the corresponding first airspace resource allocation scheme as training set labels to input into the determinantal point process learning model for training, and output the optimal learning parameters of the kernel matrix;

[0050] In the embodiments of the present invention, since a large number of optimal airspace resource allocation schemes have been calculated in step 102, in this embodiment, these optimal airspace resource allocation schemes are used as training set labels, and the first scenario parameters are used as training set data, and they are jointly input into the determinantal point process learning model for learning and training to optimize the learning model.

[0051] For the determinantal point process learning model, the present invention assumes that for each input X, the conditional DPPP(Y = Y|X) is a measure of each feasible subset conditional probability model, where X represents the set of first scenario parameters, Y represents the set of first airspace resource allocation schemes, and Ω represents the universal set, that is, the set of all first airspace resource allocation schemes corresponding to the set of first scenario parameters. The L-ensemble format of the model is expressed as:

[0052] P(Y = Y|X) ∝ det(L Y |X)

[0053] where L(X) ∈ R |Ω(X)|×Ω(X) is a positive semi-definite kernel matrix depending on the input X. The present invention considers the conditional DPP kernel matrix L(X; θ) parameterized by a θ, and the conditional probability model of the output Y corresponding to the given input X is expressed as:

[0054]

[0055] And for each set of scenario parameters X k , then there is

[0056] where, L Y (X k ; θ, Θ) represents the kernel matrix corresponding to the output airspace resource allocation scheme Y k under the set of scenario parameters X k ; L(X k; (θ, Θ) represents the corresponding kernel matrix under the set of scenario parameters X k under which, I k represents the interference signal of Wi-Fi access point users under the set of scenario parameters X k where K represents the first set of scenario parameters. The purpose of machine learning is to select a suitable θ value based on the data in the training set to predict the output of unknown inputs. The present invention defines the learning model for quality measurement as a log-linear model:

[0057] q j (X; θ) = exp(θ T f j (X))

[0058] where q j (X; θ) represents the measurement function of the transmission quality at the Wi-Fi station user with the first learning parameter θ as the learning objective under the scenario parameter X; f j (X) is a feature vector representing the measurement function of the transmission quality at, θ is a learning parameter vector, i.e., the first learning parameter, and the first learning parameter can include multiple learning components, and these components are actually also learning parameters. In the present invention, the present invention sets the quality measurement function as:

[0059] q j (X; θ) = exp(θ 1 + θ 2 S * + θ 3 I 1 + θ 4 I 2 + θ 5 I 3 )

[0060] where S * represents the received signal, I 1 , I 2 , I 3 represent the maximum AP interference signal, the second largest AP interference signal, and the base station interference signal, θ 1 、θ 2 、θ 3 、θ 4 and θ 5 respectively represent different learning components in the first learning parameter, and these five learning components together constitute the first learning parameter. At the same time, the present invention selects the Gaussian kernel as the learning model of the similarity measurement function:

[0061]

[0062] where S k,j(X; Θ) represents the Wi-Fi access point users under the scenario parameter X with the second learning parameter Θ as the learning target and the Wi-Fi station users is the similarity measurement function g k,j (X) = ||x(t k ) - x(r j )|| 2 +||x(t j ) - x(r k )|| 2 is the similarity measurement between the k-th and j-th items, where x(t k ) and x(r j ) represent the positions of the Wi-Fi access point users and the Wi-Fi station users respectively; j, k ∈ [1, 2, … M], and M represents the number of Wi-Fi users, that is, including the number of Wi-Fi station users and the number of Wi-Fi access point users.

[0063] The present invention represents the training set sum as T := {(X 1 , Y 1 ), …, (X K , Y K )}, where X K represents the input, that is, the first scenario parameter, and Y K represents the output, that is, the first spatial domain resource allocation scheme corresponding to the first scenario parameter. Then, the optimization problem of machine learning is expressed as maximizing the likelihood probability of the training set sum:

[0064]

[0065] where, θ * represents the first optimal learning parameter, Θ * represents the second optimal learning parameter, and P θ,Θ is P parameterized by θ, Θ L . In addition, the objective function of this optimization problem is a concave function, so this optimization problem can be directly solved.

[0066] The solution process is as follows:

[0067] Step 1: Initialize θ = [1, 1, 1, 1, 1], Θ = 1, ε, the training data set, and the scenario parameter

[0068] Step 2: Calculate L = q(θ)S(Θ)q T (θ)

[0069] Step 3: Solve the optimization problem to obtain the optimal parameters θ * , Θ *

[0070] Step 4: Calculate L = q(θ * )S(Θ * )q T (θ * )

[0071] Step 5: Perform eigenvalue decomposition on L

[0072] Step 6: Define J = φ

[0073] Step 7: Traverse each eigenvalue and eigenvector, and with probability take n, and J = J ∪ {n}

[0074] Step 8: V = {v n}_(n∈J), define Z = φ

[0075] Step 9: With probability sample i from Ω, and Z = Z ∪ i

[0076] Step 10: Output Z to represent the spatial domain scheme.

[0077] To further illustrate the training process of the determinant point process learning model, the embodiments of the present invention give the following steps, including:

[0078] Calculate the kernel matrix according to the scenario parameters of the NR-U / Wi-Fi coexistence network;

[0079] According to the determinant point process learning model, calculate the optimal learning parameters of the kernel matrix according to the first scenario parameters and the corresponding first spatial domain resource allocation scheme;

[0080] Substitute the optimal learning parameters into the kernel matrix and perform eigenvalue decomposition on the kernel matrix;

[0081] Traverse each eigenvalue and eigenvector, and generate a second spatial domain resource allocation scheme by means of probability sampling.

[0082] 104. Input the third scenario parameters of the NR-U / Wi-Fi coexistence network into the model corresponding to the optimal learning parameters, and output the third spatial domain resource allocation scheme;

[0083] In an embodiment of the present invention, the third scenario parameters of the NR-U / Wi-Fi coexistence network to be processed are input into the learned determinantal point process learning model. Similarly, the third scenario parameters here are similar to the first scenario parameters, except that they belong to different times or scales. By inputting the third scenario parameters into the learned determinantal point process learning model, the third airspace resource allocation scheme can be directly predicted, which can reduce the problems of high computational complexity in the airspace resource allocation process and the exponential growth of the computational complexity with the expansion of the network scale.

[0084] 105. Using the result of the third airspace resource allocation scheme, in the current NR-U / Wi-Fi coexistence network, precoding is performed on the transmitted signal at the base station side.

[0085] In an embodiment of the present invention, the result of the third airspace resource allocation scheme is the optimal solution {x *}, {y *} corresponding to the third scenario parameters. Through this optimal solution, it can be known which cellular users are scheduled and which Wi-Fi station users are interference-suppressed. After determining the corresponding Wi-Fi station users, the base station can perform precoding on the signal, that is, the base station can change the reception mode of the base station signal at the Wi-Fi device through precoding to reduce the interference at the Wi-Fi device.

[0086] Specifically, referring to Figure 3 , Figure 3 describes the beam-based transmission mechanism, that is, the schematic diagram of the airspace coexistence mechanism mentioned in the present invention. The airspace coexistence mechanism refers to the base station performing precoding on the signal to reduce the interference of the base station signal to the Wi-Fi device and increase the beamforming gain of the base station signal to the UE. When the base station transmits on the unlicensed channel, if the Wi-Fi APs that could originally detect the base station signal are interference-suppressed by the base station, then these APs can also access the unlicensed channel to transmit data to the downlink channel simultaneously with the base station.

[0087] In the embodiments of the present invention, a coexistence network model of NR-U and Wi-Fi operating in the millimeter-wave band is constructed. By analyzing the interference model therein, a coexistence mechanism based on a beam-based transmission mechanism is proposed by the present invention. Deploying this coexistence mechanism on the cellular network base station side can ensure the friendly coexistence of the cellular network and the Wi-Fi network in the unlicensed band. Meanwhile, in the embodiments of the present invention, a coexistence network model and a spatial domain coexistence mechanism are also provided. The throughput of the cellular network and the throughput of the Wi-Fi network in the coexistence network are analyzed for performance. An optimization model is constructed to maximize the total throughput of the coexistence network on the premise of ensuring a certain throughput of the Wi-Fi network. And a heuristic algorithm based on convex optimization is proposed to solve the throughput performance optimization model in the coexistence network. Further still, in order to solve the shortcomings that the computational complexity of the traditional heuristic algorithm based on convex optimization is very high and increases exponentially with the increase of the network scale, the present invention also proposes a brand-new estimation algorithm based on DPPL to solve the spatial domain allocation problem in the coexistence network, and compared with the computational time of the heuristic algorithm based on convex optimization, the present invention can greatly reduce the computational time.

[0088] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by terms such as "coaxial", "bottom", "one end", "top", "middle", "the other end", "upper", "one side", "top", "inner", "outer", "front", "center", "both ends", etc. are based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be construed as a limitation of the present invention.

[0089] In the present invention, unless otherwise clearly specified and defined, terms such as "installation", "setting", "connection", "fixation", "rotation", etc. should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or integrated; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements or the interaction relationship between two elements. Unless otherwise clearly defined, for those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0090] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An airspace resource allocation method based on determinant point process learning, characterized in that, the method is applied to an NR-U / Wi-Fi coexistence network, and the method includes: Based on the scenario of the NR-U / Wi-Fi coexistence network, construct an airspace coexistence mechanism model with maximizing throughput as the optimization goal; the airspace coexistence mechanism model with maximizing throughput as the optimization goal is expressed as: s.t. C1: C2: x ∈ {0, 1}, y ∈ {0, 1} Among them, represents the throughput at and represents the throughput at is a binary variable, x i represents whether a cellular user is scheduled, y j represents whether a Wi-Fi station user is interference-suppressed, x i = 1 indicates that the cellular user is scheduled, y j = 1 indicates that the Wi-Fi station user is interference-suppressed; a l and a w respectively represent the priority factors of the cellular network and the Wi-Fi network, and a l + a w = 1; Constraint C1 means that the number of applicable airspace resources is less than the number of spatial degrees of freedom that the base station can provide. N represents the number of cellular users, M represents the number of Wi-Fi users, that is, including the number of Wi-Fi station users and the number of Wi-Fi access point users, N A represents the number of antennas of the base station; Constraint C2 restricts x and y to be binary variables. x represents the binary variable of the cellular user, y represents the binary variable of the Wi-Fi station user, ω k represents the binary variable of whether a Wi-Fi access point user is interference-suppressed, ω k = 1 indicates that the Wi-Fi access point user is interference-suppressed; The determination process of whether a Wi-Fi access point user is interference-suppressed includes determining whether the Wi-Fi access point user detects a base station signal higher than the energy detection threshold. If it is higher, it is interference-suppressed, otherwise it is not. Among them, the calculation formula for the binary variable of whether a Wi-Fi access point user is interference-suppressed is expressed as:​​​​​​​​​​ Among them, P bs represents the transmission power of the base station; l() represents the path loss calculation function; represents the Euclidean distance between the base station and the Wi-Fi access point user ; represents the small-scale fading from the base station to the Wi-Fi access point user ; σ 2 represents the noise power; γ represents the energy detection threshold; According to the first scenario parameters of the NR-U / Wi-Fi coexistence network, solve the optimization goal based on the heuristic algorithm of convex optimization to obtain the first airspace resource allocation scheme; Use the first scenario parameters of the NR-U / Wi-Fi coexistence network as training set data, and use the corresponding first airspace resource allocation scheme as training set labels to input into the determinant point process learning model for training, and output the optimal learning parameters of the kernel matrix; the determinant point process learning model is expressed as: Among them, θ * represents the first optimal learning parameter, θ represents the first learning parameter, and Θ * represents the second optimal learning parameter, and Θ represents the second learning parameter; P θ,Θ (Y k |X k ) represents the conditional probability model for measuring each feasible airspace resource allocation scheme in the corresponding scenario under the models corresponding to θ and Θ and the input scenario parameter set X k under which; ; represents the kernel matrix corresponding to the output airspace resource allocation scheme Y k under the scenario parameter set X k ; L(X k ; θ, Θ) represents the kernel matrix corresponding to the scenario parameter set X k under which, I k represents the interference signal of Wi-Fi access point users under the scenario parameter set X k under which, K represents the first scenario parameter set, and Ω(X) represents the set of all first airspace resource allocation schemes corresponding to the first scenario parameter set; The calculation expression of the kernel matrix is: L = q(θ)S(Θ)q T (θ) where L represents the kernel matrix; q(θ) represents the quality measurement function; S(Θ) represents the learning model of the similarity measurement function; θ represents the first learning parameter, Θ represents the second learning parameter, and T represents the coordinate transpose; The calculation expression of the quality measurement function is: q j (X; θ) = exp(θ 1 + θ 2 S * + θ 3 I 1 + θ 4 I 2 + θ 5 I 3 ) where q j (X; θ) represents a measurement function of the transmission quality at the Wi-Fi site user under the scenario parameter X with the first learning parameter θ as the learning target ; S * represents the received signal, I 1 , I 2 , I 3 represents the maximum Wi-Fi access point user interference signal, the second largest Wi-Fi access point user interference signal, and the base station interference signal, θ 1 , θ 2 , θ 3 , θ 4 and θ 5 respectively represent different learning components in the first learning parameter The similarity measurement function is: Among them, S k,j (X; Θ) represents the similarity measurement function between Wi-Fi access point users and Wi-Fi station users under the scenario parameter X with the second learning parameter Θ as the learning target, and Wi-Fi station users The similarity measurement function between them is g k,j (X) = ||x(t k ) - x(r j )|| 2 + ||x(t j ) - x(r k )|| 2 is the similarity measurement between the k-th and j-th items, where x(r k ) and x(r j ) represent the positions of Wi-Fi access point users and Wi-Fi station users respectively; x(r j ) and x(r k ) represent the positions of Wi-Fi access point users and Wi-Fi station users respectively; j, k ∈ [1, 2, … M], and M represents the number of Wi-Fi users, that is, the number of Wi-Fi station users and the number of Wi-Fi access point users; where the training process of the determinant point process learning model includes: Calculate the kernel matrix according to the scenario parameters of the NR-U / Wi-Fi coexistence network; According to the determinant point process learning model, calculate the optimal learning parameters of the kernel matrix according to the first scenario parameters and the corresponding first airspace resource allocation scheme; Substitute the optimal learning parameters into the kernel matrix and perform eigenvalue decomposition on the kernel matrix; Traverse each eigenvalue and eigenvector, and generate a second airspace resource allocation scheme through probability sampling; Input the third scenario parameters of the NR-U / Wi-Fi coexistence network into the model corresponding to the optimal learning parameters, and output the third airspace resource allocation scheme; Utilize the result of the third airspace resource allocation scheme to perform precoding on the transmitted signal at the base station side in the current NR-U / Wi-Fi coexistence network.

2. The airspace resource allocation method based on determinant point process learning according to claim 1, characterized in that, the NR-U / Wi-Fi coexistence network includes a cellular network and multiple Wi-Fi networks operating in the same unlicensed band; the cellular network includes a base station and multiple cellular users, and the Wi-Fi network includes multiple APs and Wi-Fi users.

Images