An Optimization Method for Heterogeneous Aerial Computing Networks Based on Smart Reflectors

Abstract

This invention belongs to the field of the Internet of Things (IoT) and relates to an optimization method for an airborne computing heterogeneous network based on a smart reflector. The method includes: constructing a RIS-assisted airborne computing heterogeneous network model M; calculating the mean square error of model M; constructing an objective optimization problem based on the device transmit power vector b, the base station receive beamforming vector v, and the smart reflector phase shift matrix Θ, with the goal of minimizing the mean square error of M; and optimizing the objective optimization problem using an alternating optimization algorithm to obtain b. * v * and Θ * Set b, v, and Θ of model M to b respectively. * v * and Θ * The optimized model is obtained. This invention aims to minimize the mean square error of model M by using an alternating optimization method to optimize the transceiver end and the RIS phase shift matrix of the RIS-assisted heterogeneous network for air computing. This reduces and converges the mean square error, which can suppress inter-cell interference to a large extent, reduce the complexity of the algorithm, and improve the system performance.

Description

Technical Field

[0001] This invention belongs to the field of the Internet of Things and relates to an optimization method for an aerial computing heterogeneous network based on a smart reflective surface. Background Technology

[0002] In recent years, with the development of wireless communication technology, the Internet of Things (IoT), as a new paradigm, holds the promise of enabling billions of devices to communicate wirelessly, thereby providing various intelligent services such as smart transportation, smart homes, and healthcare. However, IoT networks also face significant challenges in data aggregation and communication, such as high latency, high energy consumption, and limited bandwidth. One existing solution is Wireless Data Aggregation (WDA), which aims to reduce the amount of data transmitted by applying compression, encryption, or fusion technologies at intermediate nodes. Although the speed and capacity of wireless communication have greatly improved in the 5G era, according to the GSMA's "Mobile Economy 2020" report, the total number of global IoT connections reached 12 billion in 2019, and is projected to reach 24.6 billion by 2025. In 5G mobile communication networks, the number of IoT nodes will reach hundreds of millions, with a deployment density of millions per square kilometer. Faced with massive WDA with a huge number of nodes, the low-bandwidth efficiency and high latency problems inherent in the traditional method of communicating first and then computing are unavoidable.

[0003] Since fusion centers typically focus on computing a nomographic function (such as weighted sum, arithmetic mean weighted sum, arithmetic mean, geometric mean, or geometric mean), over-the-air computation (AirComp) is considered a novel technology that can effectively overcome the limitations of WDA (Wide-based Data Acquisition). Unlike traditional methods that first acquire local data from wireless devices (WDs) and then perform nomographic function computation at the fusion node, AirComp utilizes interference in multiple access channels to assist computation and directly computes the nomographic function based on the characteristics of wireless signal superposition in space. AirComp enables IoT devices to transmit data simultaneously without coordination, and receivers can directly obtain the aggregation result by sampling the received signal. Compared to traditional methods, AirComp has advantages such as simplicity, efficiency, and robustness, and supports various aggregation functions such as averaging, maximizing, minimizing, and bitwise operations. Its effectiveness has been proven by information theory and numerous papers, and it can be applied in various scenarios such as distributed sensing, federated learning, and distributed consensus control. However, AirComp also faces some challenges. As the number of connected wireless devices increases, poor communication environments significantly impact AirComp performance. To address this issue, researchers have proposed numerous methods for signal enhancement in AirComp, such as massively multi-input multi-output (MIMO) and power control. However, these methods are only applicable to single-cell scenarios. When the perspective expands to multi-cell scenarios, inter-cell interference from external sources can also greatly affect AirComp performance. Some research has proposed cooperative interference management mechanisms suitable for multi-cell scenarios and interference management mechanisms for AirComp coexistence networks. While these techniques improve AirComp performance, they are typically achieved at the cost of excessive power consumption and complexity. Therefore, how to effectively and cost-efficiently reduce AirComp computational errors remains an unsolved problem.

[0004] Reconfigurable Intelligent Surfaces (RIS) are a novel technology that can effectively and cost-effectively assist AirComp in interference suppression. An RIS is a plane composed of numerous passive reflective elements. By controlling these reflective elements, this technology can adjust the phase of the incident signal, thereby reconstructing the wireless channel and achieving signal enhancement and interference suppression. Deploying an RIS at a suitable location within the wireless channel can generate a line-of-sight path, thus bypassing obstacles and expanding coverage. Furthermore, RIS is characterized by low cost, low energy consumption, and flexible deployment. Based on these excellent characteristics, RIS has potential application value in improving wireless communication, enhancing energy efficiency, and strengthening privacy and security. Similarly, applying RIS to heterogeneous over-the-air computing networks can also achieve interference suppression, thereby improving AirComp performance.

[0005] However, existing research lacks studies on the mean square error optimization of heterogeneous networks for aerial computing assisted by intelligent reflectors, resulting in insufficient WDA and low MSE of such networks, thus leading to poor system performance. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention employs an optimization method for heterogeneous aerial computing networks based on intelligent reflectors, comprising:

[0007] S1. Construct a heterogeneous network model for airborne computing assisted by a smart reflector; the heterogeneous network model for airborne computing assisted by a smart reflector includes: a cell and a smart reflector RIS, wherein the cell includes: a base station and a device;

[0008] S2. Calculate the mean square error of the heterogeneous network model of air computing assisted by the intelligent reflector, and construct an objective optimization problem based on the device transmit power vector b, the base station receive beamforming vector v and the phase shift matrix Θ of the intelligent reflector with the goal of minimizing the mean square error.

[0009] S3. Utilize the alternating optimization algorithm to optimize the objective problem and obtain the optimal device transmit power vector b. * Base station received beamforming vector v * and the phase shift matrix Θ of the intelligent reflective surface;

[0010] S4. Set the device transmit power vector, base station receive beamforming vector, and smart reflector phase shift matrix of the smart reflector-assisted heterogeneous over-the-air computing network to b respectively. * v * and Θ * This results in an optimized, intelligent reflective surface-assisted heterogeneous aerial computing network.

[0011] Cellular structures include: AirComp cells and traditional cells; an AirComp cell consists of one base station (BS) and K single-antenna devices; a traditional cell consists of one base station (BS) and one single-antenna device; an AirComp cell in a smart reflector-assisted heterogeneous network model for airborne computing is used as the target cell, and the devices in the target cell... The target cell is the set of devices; the remaining cells are considered interfering cells. For the collection of AirComp cells, AirComp cellular devices are a collection of traditional cellular n-type devices. AirComp is a collection of devices in a cellular system, compared to devices in a traditional cellular system. It is a collection of devices in a traditional cellular n; the smart reflective surface has M reflective elements.

[0012] The objective optimization problem is:

[0013]

[0014] Where C2 is the constraint on the transmit power of the target cell's devices, C1 is the constraint on the phase shift matrix of the smart reflector, MSE is the mean square error of the smart reflector-assisted heterogeneous network model for over-the-air computing, b is the transmit power vector of device k, v is the receive beamforming vector of the base station in the target cell, Θ is the phase shift matrix of the smart reflector, and θ is the transmit power vector of device k. m For the phase shift of the reflective element m of the intelligent reflective surface, b k Let P be the transmission power of device k. max This represents the maximum transmit power of device k.

[0015] The mean square error of the heterogeneous network model for aerial computing assisted by intelligent reflectors is:

[0016]

[0017] Among them, b k Let represent the transmit power of device k, and v be the receive beamforming vector of the base station in the target cell. H Indicates conjugate transpose, h k , G and θ represent the channel vector between device k and the base station of the target cell, the channel vector between device k and RIS, and the channel matrix between RIS and the base station of the target cell, respectively. Θ represents the phase shift matrix of RIS, and σ represents the phase shift matrix of RIS. 2 h represents the power of additive white Gaussian noise. i , Let b represent the channel vector between device i and the base station of the target cell, and the channel vector between device i and RIS, respectively.i h represents the transmit power of device i. j , These represent the channel vector between device j and the base station of the target cell, and the channel vector between device j and RIS, respectively.

[0018] The optimization of objective optimization problems using the alternating optimization algorithm includes:

[0019] S31. Set the initial values ​​of b, v, and Θ as b0, v0, and Θ0, and the iteration stopping condition threshold ∈;

[0020] S32. Fix v and Θ of the objective optimization problem to v from the previous iteration. t-1 and Θ t-1 To obtain subproblems Optimize the subproblem using the logarithmic barrier function method. The optimized b t ;

[0021] S33. Fix Θ and b of the objective optimization problem to the Θ of the previous iteration. t-1 and the optimized b t To obtain subproblems Optimize the subproblem using gradient descent. Get the optimized v t ;

[0022] S34. Fix b and v of the objective optimization problem to the optimized b. t and v t To obtain subproblems Use semidefinite programming to solve the subproblems Transform the problem into a convex problem, and use the convex optimization tool CVX to optimize the subproblems that have been transformed into convex problems. Optimize to obtain the optimized Θ t ;

[0023] S35, Based on the optimized b t v t and Θ t Calculate the ratio q t If q t If the value is less than the iteration stopping condition threshold ∈, then the currently optimized b will be... t v t and Θ t As the optimal device transmission power b * Base station received beamforming vector v * and the phase shift matrix Θ of the intelligent reflector * Otherwise, return to step S2.

[0024] Subproblems for:

[0025]

[0026] in,

[0027]

[0028] A k B1 is an intermediate parameter, b k Let represent the transmit power of device k, and v represent the receive beamforming vector of the base station of the target cell. H P represents the conjugate transpose. max h is the maximum transmission power of device k. k , G and θ represent the channel vector between device k and the base station of the target cell, the channel vector between device k and RIS, and the channel matrix between RIS and the base station of the target cell, respectively. Θ represents the phase shift matrix of RIS, and σ represents the phase shift matrix of RIS. 2 h represents the power of additive white Gaussian noise. i , Let b represent the channel vector between device i and the base station of the target cell, and the channel vector between device i and RIS, respectively. i h represents the transmit power of device i. j , Let represent the channel vector between device j and the base station of the target cell, and the channel vector between device j and RIS, respectively. It is a complex field.

[0029] Optimize the subproblem using the logarithmic barrier function method. include:

[0030] Construct the penalty function:

[0031] Solving the penalty function yields the optimal device transmit power b. k :

[0032]

[0033] Where μ is the penalty factor.

[0034] Subproblems for:

[0035]

[0036]

[0037] in,

[0038]

[0039] For a heterogeneous network model of airborne computing assisted by a smart reflector, p represents the set of all devices, and b represents the index of all devices. p Let p represent the transmit power of device p, and v represent the receive beamforming vector of the base station in the target cell. H Indicates conjugate transpose, h p , G and b represent the channel vector between device p and the base station in the target cell, the channel vector between device p and RIS, and the channel matrix between RIS and the base station in the target cell, respectively. k h represents the transmit power of device k. k , Let h represent the channel vector between device k and the base station in the target cell, and the channel vector between device k and the RIS, respectively. Let Θ represent the phase shift matrix of the RIS, and h represent the phase shift matrix of the RIS. i , Let b represent the channel vector between device i and the base station of the target cell, and the channel vector between device i and RIS, respectively. i h represents the transmit power of device i. j , Let C and D represent the channel vector between device j and the base station of the target cell, and the channel vector between device j and RIS, respectively. C and D are intermediate parameters, θ is the transpose of the phase shift matrix of the smart reflector, and m is the reflective element of the smart reflector.

[0040] Use semidefinite programming to solve the subproblems Transforming into convex problems includes:

[0041] Subproblems Transform into a problem format suitable for SDR

[0042]

[0043] Using semidefinite programming method to By relaxing the constraints, we obtain a subproblem that is transformed into a convex problem.

[0044] Among them, intermediate parameters Tr is the trace of the matrix, and rank is the rank of the matrix. For matrix The elements in the m rows and m columns are positive semidefinite if they are greater than or equal to 1.

[0045] Based on the optimized b t v t and Θ t Calculate the ratio q t Including: based on the optimized b t v t and Θt Calculate the MSE of the current iteration t, and calculate the ratio of the absolute value of the difference between the MSE of the current iteration t and the MSE of the previous iteration to the MSE of the previous iteration.

[0046] Beneficial effects:

[0047] 1. This invention constructs a RIS-assisted heterogeneous AirComp network model and optimizes the design of the transceiver and RIS phase shift matrix of the RIS-assisted heterogeneous AirComp network with the goal of minimizing the system MSE. This reduces and converges the system MSE, suppresses inter-cell interference in the RIS-assisted heterogeneous AirComp network, and significantly improves system performance. 2. This invention uses an alternating optimization method to optimize the RIS-assisted heterogeneous AirComp network, which significantly reduces the complexity of the algorithm and improves system performance. Attached Figure Description

[0048] Figure 1 A flowchart illustrating an optimization method for an airborne heterogeneous computing network based on a smart reflector, provided as an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of an airborne computing heterogeneous network system model based on a smart reflector provided in an embodiment of the present invention;

[0050] Figure 3 The flowchart illustrates the optimization of the objective optimization problem using the alternating optimization algorithm provided in this embodiment of the invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] like Figure 1 As shown, this invention employs an optimization method for heterogeneous aerial computing networks based on intelligent reflectors, comprising:

[0053] S1. Construct an airborne computing heterogeneous network model assisted by intelligent reflective surfaces;

[0054] like Figure 2As shown, consider a heterogeneous cellular network with RIS assistance, consisting of multiple AirComp cells and multiple traditional cells. In this network, the target cell is surrounded by multiple other cells, thus receiving interference from the uplink transmissions of other cells. Each AirComp cell in the network consists of a base station with Nr antennas and K single-antenna devices, while each traditional cell consists of a single-antenna base station and a single-antenna device. Furthermore, the system is equipped with a RIS with M reflectors. For model generality, the base station locations are determined using the Matern hard core point process (MHCPP). The AirComp cell density is λ. a The density of traditional cellular cells is λ. b Finally, each cell has a coverage radius of R, and the locations of devices within the coverage area follow a uniform distribution.

[0055] For convenience, Let L represent the set of L interfering AirComp cells, and let Let N be a set of traditional honeycombs, and let Represents the set of K devices in the target cell. This represents the set of K devices in the AirComp cell that caused the interference. Represents a set of devices in a traditional cellular network; Represents the set of all devices in the system, that is

[0056] Let x p Representative equipment The measured data, where it is assumed that x p They are independent and identically distributed. For ease of power control, let... This represents the data transmitted by device p, where... Represents the normalization function, and s p The mean is 0 and the variance is 1, that is Therefore, the signal received by the base station of the target cell is represented as follows:

[0057]

[0058] Where n represents additive white Gaussian noise;

[0059] The post-processed signal recovered by the target base station can be represented as:

[0060]

[0061] Furthermore, it is assumed that the interfering AirComp cell employs an optimal wireless power control strategy, i.e., transmit power. Among them, f i Indicates device and Channel vectors between Representing BS The optimal value of the receiving beamforming vector can be obtained through CVX.

[0062] The channel model of this invention is the Rician model (Rician fading channel model). The channel vector between device k and the base station T of the target cell, the channel vector between device k and RIS, and the channel matrix between RIS and the base station T of the target cell are defined as follows:

[0063]

[0064] in, Let represent the path loss of the links between device k and the target cell's base station T, between device k and RIS, and between RIS and the target cell's base station, respectively, denoted by . This represents the reference path loss and path loss exponent at d0 = 1 meter. k,T d k,R and d R,T Let represent the distances between device k and the target cell's base station T, between device k and RIS, and between RIS and the target cell's base station, respectively; β represents the Rician factor. Represents the real number field. Represents the complex field; This represents the non-line-of-sight (NLoS) component, whose elements are independent and identically distributed and follow a complex Gaussian distribution. The line-of-sight (LoS) component is represented by a uniform linear array (ULA), whose array response is: in, These represent the antenna spacing and wavelength, respectively, and ω represents the angle of departure (AoD) or the angle of arrival (AoA). Where ω AOA ,ω AOD These represent AOA and AOD, respectively.

[0065] S2. Calculate the mean square error of the heterogeneous network model of air computing assisted by the intelligent reflector, and construct an objective optimization problem based on the device transmit power vector b, the base station receive beamforming vector v and the phase shift matrix Θ of the intelligent reflector with the goal of minimizing the mean square error.

[0066] The performance of AirComp can be estimated by function g and the corresponding estimation function. express

[0067]

[0068] Denotes the diagonal phase shift matrix of RIS, where The phase shift of the reflecting element is represented by the function g, which represents the ideal received signal, i.e., the signal transmitted by the device.

[0069] To minimize the MSE under the constraints of device power and the unit mode of phase shift of the RIS reflector, it is necessary to jointly optimize the transmit vector b, the receive beamforming vector v, and the phase shift matrix Θ. Therefore, the optimization problem can be obtained as follows:

[0070]

[0071] in, It is the device's transmit power constraint and It is a smart reflective surface phase shift matrix constraint.

[0072] S3. Utilize the alternating optimization algorithm to optimize the objective problem and obtain the optimal device transmit power vector b. * Base station received beamforming vector v * and the phase shift matrix Θ of the intelligent reflector * ;

[0073] like Figure 3 As shown, the specific steps are as follows:

[0074] S31. Set the initial values ​​of b, v, and Θ as b0, v0, and Θ0, and the iteration stopping condition threshold ∈;

[0075] S32. Fix v and Θ of the objective optimization problem to v of the previous iteration t-1. t-1 and Θ t-1 To obtain subproblems Optimize the subproblem using the logarithmic barrier function method. The optimized b t ;

[0076] S33. Fix Θ and b of the objective optimization problem to the Θ of the previous iteration. t-1 and the optimized b t To obtain subproblems Optimize the subproblem using gradient descent. Get the optimized v t ;

[0077] S34. Fix b and v of the objective optimization problem to the optimized b. t and v t To obtain subproblems Use semidefinite programming to solve the subproblems Transform the problem into a convex problem, and use the convex optimization tool CVX to optimize the subproblems that have been transformed into convex problems. Optimize to obtain the optimized Θ t ;

[0078] S35, Based on the optimized b t v t and Θ t Calculate the ratio q t If q t If the iteration stopping threshold ∈ is greater than or equal to the threshold value, then return to step S2; otherwise, set the current optimized device transmit power b. t Base station received beamforming vector v t and the phase shift matrix Θ of the intelligent reflector t As the optimal b * v * and Θ * .

[0079] Optimizing the emission coefficient includes:

[0080] Let v and Θ be fixed values, and we can obtain the subproblem P3 as follows:

[0081]

[0082] in,

[0083]

[0084]

[0085] A k B1 is an intermediate parameter.

[0086] question It consists of an objective function composed of the sum of squares of multiple norms and a quadratic constraint, belonging to Quadratically Constrained Quadratic Programming (QCQP), which can be solved using the logarithmic barrier function method, as shown below.

[0087] First, the penalty function is constructed as follows:

[0088]

[0089] Here, μ is the penalty factor, which is a very small positive number, i.e., μ→0.

[0090] Secondly, the penalty function with respect to the optimization variable b k The first derivative and the Hessian matrix are as follows:

[0091]

[0092]

[0093] in, b k The conjugate of the first squared. This indicates a conjugate operation.

[0094] Clearly, the Hessian matrix is ​​positive definite, therefore the point where the first derivative is zero is the optimal solution, i.e. Among them, P max ≠|b k | 2 Furthermore, due to the power constraints of the equipment |b k | 2 ≤P max Therefore, the optimal transmitter coefficient can be obtained as:

[0095]

[0096] Optimization of the received beamforming vector includes:

[0097] By fixing b and Θ, we can obtain subproblems. as follows:

[0098]

[0099] in, Let represent the cascaded equivalent channel coefficients between device k in the target cell, device i in the AirComp cell of the interfering cell, device j in the traditional cell of the interfering cell, and the BS of the target cell, respectively.

[0100] This is an unconstrained quadratic programming (QP) problem. Since the gradient of the objective function is solvable, its first derivative with respect to v can be obtained as follows:

[0101]

[0102] Optimization of the RIS phase shift matrix includes:

[0103] By setting b and v to fixed values, we can obtain subproblems. as follows

[0104]

[0105]

[0106] in,

[0107]

[0108]

[0109]

[0110] C and D are intermediate parameters, and θ is the transpose of the phase shift matrix of the intelligent reflector.

[0111] question This is a non-convex problem. Semidefinite relaxation (SDR) is generally considered an efficient method for solving this problem. Before proceeding, the target problem needs to be transformed into a form suitable for SDR, as shown below:

[0112]

[0113]

[0114] in, Tr is the trace of the matrix, and rank is the rank of the matrix. For matrix The elements in the m rows and m columns are positive semidefinite if they are greater than or equal to 1.

[0115] Using the SDR method By relaxing the constraints and then using CVX for optimization, the optimal RIS phase shift matrix can be obtained.

[0116] Each round optimizes three variables sequentially. After a round of optimization, if the absolute value of the difference between the new round and the previous round's MSE is greater than the ratio of the previous round's MSE to a predetermined threshold ∈, then optimization continues; otherwise, the current device's transmit power {b} is degraded. k The optimal values ​​for the base station receiving beamforming vector v and the smart reflector phase shift matrix Θ are denoted as follows: v * and Θ * .

[0117] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An optimization method for intelligent surface-based aerial computing heterogeneous networks, characterized in that, Comprise: S1, construct an intelligent reflecting surface assisted aerial computing heterogeneous network model; The intelligent reflecting surface assisted aerial computing heterogeneous network model comprises: a cell and an intelligent reflecting surface (RIS), wherein the cell comprises a base station and a device; The cells include: AirComp cells and traditional cells; an AirComp cell is composed of one base station BS and K single-antenna devices; a traditional cell is composed of one base station BS and one single-antenna device; one AirComp cell in the intelligent reflecting surface assisted air computing heterogeneous network model is taken as a target cell, and the devices of the target cell The devices of the target cell are a set of devices in the target cell; the remaining cells are taken as interference cells The devices of the target cell are a set of devices in the target cell; the remaining cells are taken as interference cells The devices of the target cell are a set of devices in the target cell; the remaining cells are taken as interference cells The devices of the target cell are a set of devices in the target cell; the remaining cells are taken as interference cells The devices of the target cell are a set of devices in the target cell; the remaining cells are taken as interference cells S2, calculate the mean square error of the intelligent reflecting surface assisted aerial computing heterogeneous network model, and construct a target optimization problem based on a device transmit power vector b, a base station receive beamforming vector v and an intelligent reflecting surface phase shift matrix Θ, with the aim of minimizing the mean square error; S3, using an alternating optimization algorithm to optimize the target optimization problem, obtaining the optimal device transmit power vector b * , a base station receive beamforming vector v * , and an intelligent reflecting surface phase shift matrix Θ * ; Optimizing the target optimization problem by using an alternating optimization algorithm comprises: S31, setting initial values b0, v0 and Θ0 of b, v and Θ, and an iteration stop condition threshold value ∈; S32, fix v and Θ of the target optimization problem as v and Θ of the last iteration respectively t-1 and Θ t-1 , get sub-problem Optimize the sub-problem using the logarithmic barrier function method Get the optimized b t ; S33, fix Θ and b of the target optimization problem as Θ of last iteration and b respectively t-1 and optimized b t , get sub-problem Optimize the sub-problem using gradient descent method Get optimized v t ; S34, fix b and v of the target optimization problem as optimized b t and v t , to get sub-problem Convert the sub-problem into a convex problem using the semi-definite programming method, and optimize the sub-problem converted into a convex problem using the convex optimization tool CVX to get optimized Θ t ; S35、According to the optimized b t , t and Θ t , calculate the ratio q t , if q t is less than the iteration stop condition threshold ∈, take the current optimized b t , t and Θ t as the optimal device sending power b * , base station receiving beamforming vector v * and smart reflecting surface phase shift matrix Θ * ; otherwise, go back to step S2; Sub-problems are: Wherein, A k , B1is an intermediate parameter, b k denotes the transmit power of device k, (·) H denotes the conjugate transpose, P max is the maximum value of the transmit power of device k, h k , and G denote the channel vector between device k and the base station of the target cell, the channel vector between device k and the RIS, and the channel matrix between the RIS and the base station of the target cell, respectively, σ 2 denotes the power of the additive white Gaussian noise, h i , denote the channel vector between device i and the base station of the target cell, the channel vector between device i and the RIS, respectively, b i denotes the transmit power of device i, h j , denote the channel vector between device j and the base station of the target cell, the channel vector between device j and the RIS, respectively, is the complex field; S4, set the device transmit power vector, base station receive beamforming vector and intelligent reflecting surface phase shift matrix of the intelligent reflecting surface assisted aerial computing heterogeneous network as b * , v * and Θ * , to obtain an optimized intelligent reflecting surface assisted aerial computing heterogeneous network. 2.The optimization method of smart reflective surface based air computing heterogeneous network according to claim 1, characterized in that, The target optimization problem is: Wherein, C2 is the constraint of the transmission power of the device of the target cell, C1 is the constraint of the phase shift matrix of the intelligent reflecting surface, MSE is the mean square error of the intelligent reflecting surface assisted air computing heterogeneous network model, and θ m is the phase shift of the reflecting element m of the intelligent reflecting surface. 3.The optimization method of smart reflective surface based air computing heterogeneous network according to claim 2, characterized in that, The mean square error of the intelligent reflecting surface assisted aerial computing heterogeneous network model is:

4. The optimization method for smart reflective surface based air computing heterogeneous network according to claim 1, characterized in that, Optimizing sub-problems using a logarithmic barrier function method comprising: Constructing the penalty function: Solving the penalty function obtains the optimal device transmit power b k : Wherein, μ is a penalty factor.

5. The optimization method for smart reflective surface based air computing heterogeneous network according to claim 1, characterized in that, Sub-problems are: Wherein, A set of all devices for the intelligent reflecting surface-assisted aerial computing heterogeneous network model, p represents the index of all devices, b p represents the transmission power of device p, h p 、 and G represent the channel vector between device p and the base station in the target cell, the channel vector between device p and the RIS, and the channel matrix between the RIS and the base station in the target cell, respectively, C and D are intermediate parameters, θ is the transpose of the phase shift matrix of the intelligent reflecting surface, and m is the reflecting element of the intelligent reflecting surface.

6. The optimization method for smart reflective surface based air computing heterogeneous network according to claim 5, characterized in that, The subproblem is converted to a convex problem using a semidefinite programming method includes: Transforming the sub-problem into the SDR-appropriate problem form : The constraints of the problem are relaxed by using the semi-definite programming method to obtain a sub-problem which is transformed into a convex problem ; where the intermediate parameter Tr is the trace of a matrix, rank is the rank of a matrix, is the element in the mth row and mth column of the matrix is semi-positive definite.

7. The optimization method for smart reflective surface based air computing heterogeneous network according to claim 1, characterized in that, According to the optimized b t , v t and Θ t , the ratio q t is calculated. According to the optimized b t , v t and Θ t , the MSE of the current iteration t is calculated, and the ratio of the absolute value of the difference between the MSE of the current iteration t and the MSE of the last iteration and the MSE of the last iteration is calculated.

Images