CN117915481B

CN117915481B - Resource allocation method and system of ultra-dense industrial Internet of things

Info

Publication number: CN117915481B
Application number: CN202410075321.1A
Authority: CN
Inventors: 黄杰; 喻涛; 杨凡; 张仕龙; 赵梓杰; 李幸星; 姚凤航
Original assignee: Chongqing University of Technology
Current assignee: Chongqing University of Technology
Priority date: 2024-01-18
Filing date: 2024-01-18
Publication date: 2024-06-18
Anticipated expiration: 2044-01-18
Also published as: CN117915481A

Abstract

The invention relates to the technical field of telecommunications, in particular to a resource allocation method and a system of an ultra-dense industrial Internet of things, and firstly, a system model of the ultra-dense industrial Internet of things is established. Then, a hypergraph-based interference hypergraph model is built to avoid overlapping interference in UDIIoT networks. Then, considering that in UDIIoT networks, many real-time applications need to ensure timeliness of data transmission, real-time performance is quantified by using AOI indexes, a robust optimization model of incomplete CSI considering power control and resource conflict constraint is established, and the aim of maximizing network throughput is achieved. Finally, in order to solve the optimization model, a LRRA-IHAoI algorithm based on a learning theory is provided, and a more accurate robust optimization model solution is obtained by reducing the influence of imperfect CSI. Simulation results show that the algorithm has good network throughput, interference Efficiency (IE) and Energy Efficiency (EE) in UDIIoT networks, which demonstrates the effectiveness of the proposed scheme.

Description

Resource allocation method and system of ultra-dense industrial Internet of things

Technical Field

The invention relates to the technical field of telecommunications, in particular to a resource allocation method and system of ultra-dense industrial Internet of things.

Background

Along with the continuous expansion of industrial Internet of things (IIoT) application, the number of industrial Internet of things devices (IIoTD) is also remarkably increased. The devices interact, process data and communicate through the Internet of things to form an ultra-dense IIoT (UDIIoT) network. In order to meet the high reliability and real-time communication requirements in UDIIoT applications, frequent data transmission is critical, resulting in IIoTD generating large amounts of data in short time intervals. Therefore, efficient management of spectrum resources in UDIIoT networks has become a critical research hotspot. Notably, device-to-device (D2D) communication techniques hold great promise for efficient data transmission. D2D communication allows devices to communicate directly with each other, reducing the need to rely solely on Base Stations (BS). This feature distinguishes D2D from small cellular technologies such as femto cells and pico cells. D2D communication enhances spectrum resource utilization in UDIIoT networks by minimizing reliance on BSs. However, when D2D communication techniques are integrated into UDIIoT networks to enhance overall network performance, the overlapping interference caused by the numerous D2D connections sharing resources significantly compromises the performance of UDIIoT networks. It is therefore important to address this interference problem and to optimize the utilization of spectrum resources in UDIIoT networks. In recent studies, researchers have focused on resource allocation problems caused by interference in IIoT networks using D2D communication technology. As the number of D2D pairs increases, interference collisions with IIoTD also increase, resulting in reduced communication quality.

There are documents that consider the problem of mutual interference between cellular users and D2D users in a shared subchannel. They propose a quantitative feedback scheme to ensure quality of service (QoS) and improve spectral efficiency. Literature has been made to implement interference-avoiding resource allocation schemes in D2D-enabled 5G narrowband IoT networks that maximize spectrum utilization. There is literature that proposes an interference reduction scheme for a group of D2D mobiles that uses uplink non-orthogonal multiple access to maximize network throughput. There are documents that a IIoTD clustering algorithm is designed aiming at the intra-cluster transmission resources of the D2D communication so as to improve the efficiency of the intra-cluster resources and eliminate the intra-cluster interference. Meanwhile, a cooperative communication scheme for optimizing the resource utilization rate is proposed in literature by analyzing wireless service in a vehicle-mounted self-organizing network, and the intra-cluster spectrum efficiency is improved. There is a literature that proposes a distributed overlay D2D network scheme based on outdated local interference information. The scheme can effectively learn the interference mode related to time among links from outdated interference information, thereby adjusting the link power and maximizing the global sum rate. There is a literature on research on interference management in D2D supported heterogeneous cellular networks, which proposes a joint mode selection and power control scheme that makes efficient use of limited spectrum resources. In general, these studies have helped understand and implement resource management strategies to reduce interference in IIoT networks using D2D communication technologies. The existing overlapping interference scene resource management research improves the network spectrum efficiency, but the guarantee problem of the information age is not considered.

AOI is an important performance indicator for measuring information freshness and is critical for time sensitive applications in UDIIoT networks. Outdated information may lead to increased data processing time and network congestion. The real-time data of AOI quantization can effectively improve network performance. Overlapping interference can affect AOI, degrading communication quality. For example IoTDS may require more time to process the received information, which may reduce the data transmission rate. Data transmission errors or losses due to interference cannot guarantee time-sensitive service requirements, resulting in a degradation of network signal quality. However, few studies on resource management consider AOI guarantees in overlapping interference scenarios.

Disclosure of Invention

The invention provides a resource allocation method and a system of ultra-dense industrial Internet of things, which solve the technical problems that: and the effective AOI resource management under the ultra-dense industrial Internet of things overlapping interference scene is ensured.

In order to solve the technical problems, the invention provides a resource allocation method of ultra-dense industrial Internet of things, comprising the following steps:

s1, constructing a system model of the ultra-dense industrial Internet of things;

the system model comprises a plurality of base stations BS and a plurality of industrial Internet of things devices IIoTD, wherein the plurality of industrial Internet of things devices comprise M cellular devices CD and N D2D devices DD; the set of M CDs is represented as CD _m represents the mth CD, m=1, 2,; the set of N DDs is represented asDD _n denotes the nth DD, which, n=1, 2,. -%, N; in the network, DD repeatedly uses the frequency spectrum of CD to communicate; the set of subchannels is defined as/>LK _j denotes the J-th subchannel, j=1, 2,..j;

S2, constructing a hypergraph interference model of the system model;

The hypergraph interference model is expressed as h= { V, E }, v= { V ₁,v₂,...,v_N+M } is a vertex set, e= { E ₁,e₂,...,e_D } is a hyperedge set, the vertices express IIoTD, the hyperedges express IIoTD sets of mutual interference, V _i express the ith vertex, i=1, 2,..;

S3, constructing a robust optimization model based on the hypergraph interference model so as to meet the requirements of real-time service;

and S4, solving the robust optimization model.

Further, in the step S3, the robust optimization model is constructed as follows:

C₂:H^THX＝O,

Where x _n,j represents the resource allocation factor for each DD, when x _n,j =1, which means that DD _n is allocated LK _j;R_n,j as the achievable transmission rate from DD _n to IIoTD on LK _j, aoI representing the information age; r ^th represents a minimum rate threshold; pr {. Cndot. } represents the probability of occurrence of event {. Cndot. }; ζ represents a minimum probability threshold; x represents IIoTD resource allocation matrix, X= {0,1} ^(M+N)×J, where the ith row and jth column elements in X O= {0,1} ^(M+N)×J is a zero matrix; p _n,j denotes the transmit power of DD _n on LK _j, and p ^th denotes the maximum transmit power threshold; constraint C ₁ represents a probability constraint that ensures that the AoI requirement of DD _n requires a minimum rate; constraint C ₂ represents a resource allocation constraint based on the interference hypergraph model; constraint C ₃ indicates that the transmit power of each DD cannot be the maximum transmit power threshold; max represents maximizing, s.t. represents satisfying,/>Arbitrary is represented.

Further, R ^th is obtained by solving the following formula:

Δ_M/M/1＜Δ^th

Wherein the process of DD transmitting data to IIoTD is expressed as an M/M/1 model, delta _M/M/1 represents the average AoI of the M/M/1 model, delta ^th represents the maximum AoI threshold allowed in the queue;

Δ _M/M/1 is calculated by the following formula:

ρ is the data utilization of DD, defined as Lambda is the data arrival rate.

Further, the step S4 specifically includes the steps of:

s41, constructing an ellipsoid set representing uncertainty of channel realization Expressed as:

Wherein, And C is a parameter learned from the sample dataset, H _n,j is the set of channel gains for N DDs on LK _j,/>H ₁,h₂,...,h_N denotes 1, 2..N channel gains, C denotes constraint vector,/>Representing a real number;

s42, based on the ellipsoid set Constraint C ₁ was rewritten as:

Wherein P _n,j represents the transmission power matrix of N DD pairs LK _j, Ω=σ ²I_N×1 represents the noise power matrix, σ represents the noise power, I _N×1 represents the n×1 identity matrix;

S43, eliminating the ellipsoid set Reconstructing the ellipsoidal collection/>The following are provided:

Suppose H _n,j＝[φ]_N×1 and Representing post-reconstruction/>H _n,j denotes the channel gain of DD _n on LK _j, h _m,j denotes the channel gain of CD _m on LK _j, h _n,j is a random channel parameter; [] ^T denotes matrix transposition;

S44, further rewriting the constraint C ₁ rewritten in the step S42 as:

s45, using the ellipsoid set reconstructed in the step S44 And (3) solving the robust optimization model rewritten in the step (S43) after the step (C ₁) to obtain a global optimal network throughput solution.

Further, the step S41 specifically includes the steps of:

S411, collecting I independent imperfect CSI samples Q ₁,Q₂,…,Q_I distributed in the same manner to form a sample set Q _S＝Q₁,Q₂,…,Q_I, wherein the CSI represents channel state information;

S412, dividing the sample set Q _S into two sets And/>Wherein Q _S1 includes the first I ₁ samples/>Q _S2 includes the remaining samples/>

S413, calculating parametersSum block diagonal matrix/>Gathering ellipsoids/>Represented as Is set/>Is a sample covariance of (2); Γ isIs,/>Is/>Is expressed as: /(I)

S414, defining a potential distribution 1-theta quantile Q _1-θ of a calibration value i (Q) of any sample Q through a set Q _S2, then obtaining a function value of an observed value i (Q ₁),...,i(Q_N), and arranging the observed values i (1) to … to i (N) in an ascending orderAs an upper bound for the 1- θ quantile of i (Q);

S415 according to Γ=i (Q ^*) calculates C, where Σ is/>Cholesky decomposition of (a);

S416, based on the calculated parameters And C obtaining a sample set/>

Further, in the step S45, the rewritten robust optimization model is first converted into a lagrangian dual model, a convex problem is constructed, and then a solution of the model is solved according to the KKT theorem of lagrangian.

Further, in the interference hypergraph model, interference includes independent interference and cumulative interference; independent interference refers to comparing an interference generated by IIoTD with an interference threshold s _th, and if the interference generated by IIoTD exceeds the interference threshold s _th, then the interference generated by IIoTD is considered to be independent interference; the cumulative interference refers to the interference generated by the plurality IIoTD when the interference exceeds the interference threshold s _th, and the interference generated by the plurality IIoTD is regarded as the cumulative interference.

The invention also provides a resource distribution system of the ultra-dense industrial Internet of things, which is characterized in that: the system comprises a system model building module, a hypergraph interference model building module, a robust optimization model building module and an optimization model solving module; the system model building module, the hypergraph interference model building module, the robust optimization model building module and the optimization model solving module are respectively used for executing the steps S1, S2, S3 and S4 in the method.

According to the resource allocation method and system of the ultra-dense industrial Internet of things, a system model of the ultra-dense industrial Internet of things is established. Then, a hypergraph-based interference hypergraph model is built to avoid overlapping interference in UDIIoT networks, which can analyze the interference type and relationship between D2D Devices (DDs) and IIoTD and identify resource allocation conflicts between different DDs, thereby reducing interference and optimizing spectrum resource utilization. Then, considering that in UDIIoT networks, many real-time applications need to ensure the timeliness of data transmission, the real-time performance is quantified by using an AOI index, and in the overlapping interference scene of UDIIoT networks, complex environments exist, signal attenuation and channel uncertainty increase, so that real-time channel state information cannot be accurately acquired, and therefore, a robust optimization model of incomplete CSI considering power control and resource conflict constraint is established, and the aim of maximizing network throughput is achieved. Finally, in order to solve the optimization model, a LRRA-IHAoI algorithm based on a learning theory is provided, and a more accurate robust optimization model solution is obtained by reducing the influence of imperfect CSI. Simulation results show that the algorithm has good network throughput, interference Efficiency (IE) and Energy Efficiency (EE) in UDIIoT networks, which demonstrates the effectiveness of the proposed scheme.

The main contributions of the invention are as follows:

1) The interference type between DDs is analyzed for overlapping interference problems in UDIIoT networks. On the basis, an interference hypergraph is established, and interference relation between DD can be analyzed at the same time. Finally, a judgment method of resource allocation conflict among different DDs is designed.

2) In order to solve the challenge of ensuring timely transmission of service data in UDIIoT network overlapping interference scenes, the invention uses AoI indexes to quantify real-time performance and construct AoI models. The model may convert AoI metrics to rates to meet real-time data transmission requirements. On the basis of the interference hypergraph model and the AoI model, a robust optimization model under imperfect CSI is established, and the model combines power control constraint and resource allocation constraint and aims at maximizing network throughput.

3) Aiming at the problem that a robust optimization model with random channel parameters introduced by imperfect CSI is difficult to solve, a LRRA-IHAoI algorithm based on a learning theory is provided. The algorithm utilizes learning theory to improve the channel sample dataset of the robust optimization model, thereby generating a more accurate feasible solution, maximizing network throughput under UDIIoT networks, and reducing the impact of imperfect CSI.

Drawings

FIG. 1 is a system model diagram of an ultra-dense industrial Internet of things provided by an embodiment of the invention;

FIG. 2 is an exemplary diagram of an interference hypergraph model provided by an embodiment of the present invention;

FIG. 3 is a diagram illustrating an example AoI in UDIIoT networks provided by embodiments of the present invention;

FIG. 4 is a graph of AoI threshold Δ ^th versus DD throughput provided by an embodiment of the present invention;

FIG. 5 is a graph of the relationship between the DD power threshold p ^th and total EE provided by an embodiment of the present invention;

Fig. 6 is a graph of the relationship between the power threshold p ^th and the total IEs of DD provided by an embodiment of the present invention;

FIG. 7 is a graph of the relationship between the power threshold p ^th and the total throughput of DD provided by an embodiment of the present invention;

FIG. 8 is a graph of the relationship between the rate threshold R ^th and UDIIoT network performance provided by an embodiment of the present invention;

fig. 9 is a graph of noise power σ versus UDIIoT network performance provided by an embodiment of the present invention.

Detailed Description

The following examples are given for the purpose of illustration only and are not to be construed as limiting the invention, including the drawings for reference and description only, and are not to be construed as limiting the scope of the invention as many variations thereof are possible without departing from the spirit and scope of the invention.

The embodiment of the invention provides a resource allocation method of an ultra-dense industrial Internet of things, which comprises the following steps:

S2, building a hypergraph interference model of the system model;

S3, constructing a robust optimization model based on the hypergraph interference model to meet the requirements of real-time service;

and S4, solving the robust optimization model.

(1) Step S1: system model for constructing ultra-dense industrial Internet of things

The scenario studied by the present invention is UDIIoT scenario employing D2D technology, as shown in fig. 1. The area includes a Base Station (BS) and an industrial internet of things device (IIoTD). IIoTD includes M Cellular Devices (CDs) and N D2D Devices (DDs), a set of M Cellular Devices (CDs) being defined asWhere the mth cellular device is denoted CD _m, m=1, 2, …, M. The set of N D2D Devices (DD) is defined as/>Where the nth D2D device is denoted DD _n, n=1, 2, …, N. In the network, there are DDs that reuse the spectrum of the CD for communication. Further, let J sets of subchannels be defined as/>Where the J-th subchannel is denoted LK _j, j=1, 2, …, J.

According to the rayleigh fading channel model, the signal-to-interference-and-noise ratio (SINR) is expressed as:

Where γ _m,j is the SIR of CD _m on LK _j, γ _n,j is the SIR of DD _n on LK _i, p _m,j is the transmit power of CD _m on LK _j, p _n,j is the transmit power of DD _n on LK _j, h _m,j is the channel gain of CD _m on LK _j, h _m is the CD _m to BS channel gain, h _n is the DD _n to BS channel gain, and h _n,j is the DD _n channel gain on LK _j. Meanwhile, the time-varying nature of the wireless channel between DD and non-real time channel estimation will result in imperfect CSI, h _n,j being a random channel parameter, σ ² representing the sum of the noise powers.

Based on equation (1), the achievable transfer rate R _m,j from CD _m to IIoTD on LK _j and the achievable transfer rate R _n,j from DD _n to IIoTD on LK _j are as follows:

where B represents the bandwidth of each subchannel.

(2) Step S2: construction of hypergraph interference model

Hypergraphs are general graphs of graphs in which a hyperedge may contain any number of vertices. In UDIIoT, a plurality IIoTD accumulates to form a cumulative interference that affects the communication quality of IIoTD. Thus, hypergraphs may provide more accurate interference relationships between the plurality IIoTD than graphs. To solve the interference problem in UDIIoT, the present example proposes an interference hypergraph model based on hypergraph theory to analyze IIoTD the interference relationship. For the system model of the ultra-dense industrial internet of things, the hypergraph is expressed as h= { V, E }, where v= { V ₁,v₂,...,v_N+M } is a vertex set, e= { E ₁,e₂,...,e_D } is a hyperedge set, and D represents the number of hyperedges. The relationship between the vertex and the superside is recorded by an association matrix H= {0,1} ^(M+N)×D, wherein the element H _(i,d) of the ith row and the ith column takes the following values:

wherein vertices represent IIoTD and supersides represent sets of IIoTD that interfere with each other. In the interference hypergraph model, interference mainly includes independent interference and accumulated interference. Fig. 2 shows an example of an interference hypergraph model comprising 8 vertices and 4 hyperedges, denoted (v ₁,v₂,...,v₈) and (e ₁,e₂,...,e₄), respectively.

Independent interference: the interference generated by one IIoTD is compared to an interference threshold, which is _th determines the severity of the interference at IIoTD. When interference is not present, the interference is considered to be independent interference. IIoTD exceeds the interference threshold, namely:

p _i,j represents the transmit power of the i IIoTD on LK _j, i=1, 2, …, n+m. This means that the superside includes only two vertices, such as supersides e ₃ and e ₄ in FIG. 2.

Cumulative interference: it is generated from a plurality IIoTD. By comparing the superposition of multiple IIoTDS interferers to the interference threshold of IIoTDS, it can be determined whether they are cumulative interferers. When the interference generated by the plurality IIoTD exceeds the interference threshold, namely:

This means that the superside includes more than two vertices, such as supersides e ₁ and e ₂ in FIG. 2.

To record the resources allocated to IIoTD in the interference hypergraph model, a resource allocation matrix x= {0,1} ^(M+N)×J of IIoTD is introduced, where the meaning of the element X _(i,j) in X is as follows:

Then, in order to avoid interference between IIoTD based on the association matrix H and the resource allocation matrix X, the resources of IIoTD in the same super edge must be different, i.e. the condition is satisfied as follows:

H^THX＝O (7)

where the upper right symbol T represents the matrix transpose, o= {0,1} ^(M+N)×J is a zero matrix.

In the above-mentioned interference model, the present example analyzes the interference relationship and the resource conflict relationship between IIoTD, and obtains the resource relationship matrix between devices. This matrix allows this example to observe the resource allocation between nodes. When the value in the matrix is 0, no conflict exists between the nodes; otherwise, there is a conflict. This approach helps to reduce interference and optimize spectral resource utilization. However, in this network, there are a large number of AoI applications that require timeliness of the data, and ensuring freshness of the information is indeed a challenging problem. In the context of overlapping interference, it becomes more critical to efficiently manage resources to meet the needs of these time-sensitive applications. This requires advanced algorithms and strategies to optimize resource allocation, minimize interference effects, and maintain AoI of the required information.

(3) Step S3: construction of robust optimization model

To address this challenge, the present example can utilize AoI indexes to quantify real-time data and effectively improve network performance of UDIIoT networks. In this step, the example considers that there are a large number of time-efficient services in UDIIoT networks, so the example uses AoI metrics to construct the average AoI model. In this model, the example represents the process of DD transferring data to IIoTD as an M/M/1 model and converts it to a rate constraint. By combining the resource collision matrix and the power control matrix, a robust optimization model is established, and the aim is to improve the network throughput to the maximum extent. The model aims to coordinate resource allocation and power control among nodes to maximize network throughput while meeting the freshness requirements of AoI applications.

In UDIIoT networks, the information transmitted changes over time, and the validity of the data depends largely on its timeliness. To evaluate the performance of the system in terms of IIoTDS timely receiving data, the present example introduces the concept of AOI. AOI measures IIoTDS the freshness or timeliness of the received data and provides insight into the performance of the system.

In this scenario, the main task of the network is for the DD to send packets to the CD, assuming the source is the DD for purposes of IIoTDS. AOI at time t is defined as:

Δ(t)＝t-U(t) (8)

Where U (t) is the timestamp of the latest update received by the BS. Fig. 3 shows a schematic diagram of AoI over time in UDIIoT networks. The time curve in the graph represents the state age time delta (t), which increases linearly and then decreases to a smaller value when a packet is received. Let t _i be the number of times the DD sends the generated packet and t _i' be the number of times the CD receives the packet. In UDIIoT networks, this example models the DD packet transmission process as an M/M/1 queue model. In the M/M/1 model, the data arrival rate follows the Poisson distribution, and the data service time follows the exponential distribution. Lambda is the data arrival rate. The example assumes that P is an exponential distribution of data arrival time, ρ is the data utilization of DD, defined as Thus,/>And is also provided with Indicating the desire.

Average AoI is calculated as:

Where TA represents the trapezoidal area in fig. 3, S ^T is the system time, i.e., the sum of the queuing time W and the service time S, as follows:

S^T＝W+S (10)

wherein the method comprises the steps of Therefore, the formula (9) can be rewritten as:

in the M/M/1 model, when the network reaches steady state, the system time S ^T is randomly the same, and the probability density function of the system time S ^T of the M/M/1 model is:

s represents time.

Thus, the condition of the wait time W _i given P _i =p is expected to be obtainable as:

According to the formulas (12) and (13), the desired can be obtained as follows

In this example, by substituting formulas (11) and (14) into formula (9), the average AoI of M/M/1 can be obtained as follows:

From the derived average AoI, the example can observe that average AoI in the M/M/1 model is related to IIoTD process data rate and data arrival rate. In UDIIoT networks, there are a large number of time-efficient services, and it is necessary to guarantee the time-efficient of data transmission. Thus, this example builds an average AoI model for AoI services. To meet the timeliness requirements of AoI services, average AoI must be less than the maximum AoI threshold Δ ^th allowed in the queue, expressed as follows:

Δ_M/M/1＜Δ^th (16)

Thus, the rate of DD expression for AoI services is obtained by the following equation:

wherein the method comprises the steps of Is a solution of equation (16) that represents the minimum rate threshold of DD _n for AoI services. Equation (17) transforms the AoI requirement of DD _n for AoI service to a rate requirement. It achieves the minimum rate constraint required to ensure AoI requirements of the DD _n for time sensitive applications.

In UDIIoT networks, the resource allocation problem is achieved by maximizing network throughput while ensuring AoI requirements. Under the condition of incomplete channel state information, a resource allocation model based on interference hypergraph is established. Combining the transmitting power constraint and the channel selection constraint, a robust optimization model aiming at maximizing the network throughput is established, and the mathematical model is as follows:

Where x _n,j represents the resource allocation factor for each DD, when x _n,j =1, this means that DD _n is LK _j. ζ represents a minimum probability threshold. Constraint C ₁ represents a probability constraint that ensures that the AoI requirement of DD _n requires a minimum rate; constraint C ₂ represents a resource allocation constraint based on the interference hypergraph model; constraint C ₃ indicates that the transmit power of each DD cannot be the maximum transmit power threshold. max represents maximizing, s.t. represents satisfying, Arbitrary is represented. Pr {. Cndot. } represents the probability of occurrence of event {. Cndot. }. The minimum rate threshold of R ^th, i.e./>

(4) Step S4: solving of robust optimization model

From the robust optimization model, it can be seen that the channel gain h _n,j is uncertain under imperfect CSI, so that the model cannot solve for the global optimum by using the conventional optimization method. Thus, the idea of split learning theory is employed herein to train random probability constraints to obtain a more accurate set of uncertainties and apply it to the solution of a robust optimization model.

In this step, this example proposes a LRRA-IHAoI algorithm (a learning-based robust resource allocation algorithm that takes into account interference hypergraphs and AOI requirements) to solve the robust optimization model. Aiming at the problems encountered in the robust optimization model, the method firstly collects a plurality of independent and uniformly distributed incomplete CSI samples to learn a channel gain uncertainty set. Then, a more accurate channel gain uncertainty set is obtained by using a shape learning and size calibration method, so that an optimization model is solved.

In the robust optimization model, the objective function is not easily solved due to the randomness of the transmission rate under imperfect CSI. In order to learn the uncertainty model, this example requires collecting samples of the channel gain. In a robust optimization model (18), the rate constraints can be reformulated as:

wherein the method comprises the steps of Representing real number,/>Representing a set of ellipses.

In an imperfect CSI environment, H _n,j is the uncertainty that needs to be learned. The present example selects a set of ellipses to characterize the uncertainty of the channel. The expression is as follows:

wherein the method comprises the steps of And T is a parameter learned from the sample dataset. Thus, this example collects sample Q _S＝Q₁,Q₂,...,Q_I,/>For H _n,j learning. However, the uncertainty of the a priori distribution of imperfect CSI makes it difficult to learn the dataset. Thus, this example uses a confidence approach to construct a priori distribution of imperfect CSI. Formula (20) can be restated as:

Pr{Pr{H_n,j∈H_n,j(Q_S)}≥ζ}≥1-θ (21)

Where 1- θ is a prescribed confidence level, H _n,j(Q_S) represents the value of the random variable H _n,j under condition Q _S. The present example then devised a learning method to determine parameters, including shape learning and size calibration. Before this, the present example divided the sample dataset into two parts, i.e And/>

Shape learning: aggregationCan be reformulated as:

Wherein, Is set/>Is/>Is,/>And Γ > 0. /(I)Is/>Can be expressed as:

The matrix is a block diagonal matrix whose diagonal elements are the sample covariance of the dataset, expressed as:

wherein the method comprises the steps of Any diagonal element/>The calculation is as follows:

Where a=1, 2 and b=1, 2 denote the rows and columns of blocks in the diagonal matrix, Q _i is a feature vector, superscripts 2 (n-1) +a and 2 (n-1) +b denote the indices of the feature vectors, Is used as a reference value,/>Representing eigenvalues/>Mean value of/(I)Is used as a reference value,/>Representing eigenvalues/>Is a mean value of (c).

And (3) size calibration: the task of the size calibration is to calibrate the uncertainty set conforming to equation (21) with a confidence level of 1- θ. The main problem is to estimate the quantiles of the data sample transforms:

wherein formula (26) is from random space To/>Q represents any one sample. Then, for the purpose of calibration/>The example defines the (1- θ) quantile Q _1-θ of the potential distribution of i (Q) based on the sample dataset Q _S2 as follows:

Pr{i(Q)≤q_1-θ}＝1-θ (27)

then, the present example can obtain the function value of the observed value i (Q ₁),...,i(Q_N). After the observed values i (1) is more than or equal to … and is less than or equal to i (N) are arranged in ascending order, the values are May be used as an upper bound for the quantile i (Q). Thus, the collection/>May be set to a size of:

Pr{Γ≥q_1-θ}＝1-θ (28)

Wherein Γ=i (Q ^*). Based on the above learning, the present example can calculate C as:

Where Σ is Cholesky decomposition of (i.e./>)Then, by combining the expressions (23) and (29), the present example can obtain the related data of the channel gain.

Then, by using the established sample set, a robust version of the problem can be deduced, and the SINR can be rewritten as:

To facilitate model calculation, assume:

the method can obtain the following steps:

based on equation (31), the robust optimization problem can be rewritten as:

Wherein C is the norm of the vector C, Ω=σ ²I_N×1 denotes a noise power matrix. Wherein:

thus, the robust optimization problem can be rewritten as:

in the above optimization, the random channel is learned and trained to However, some remote points are also involved in the learning process. In order to make the random channel more accurate, improvements to the data set are needed.

In this section, the present example sets samplesThe method is divided into a plurality of independent sample sets, samples far away from the center are removed, and therefore a more accurate sample set is obtained. This example assumes that H _n,j＝[φ]_N×1 and phi= [ H _n,j,h_m,j]^T, equation (33) can be re-expressed as:

wherein the method comprises the steps of Based on the same approach, this example reconstructs the ellipse set:

wherein the method comprises the steps of The center of (c) is represented as follows:

/>

then, the sample set q= { Q ₁,Q₂,...Q_I } is divided into N samples again.

Taking Q ₁ as an example, the relationship between different channel realizations is calculated as:

wherein the method comprises the steps of Representing the length of sample Q ₁, Q _1i represents the i-th eigenvector in sample Q ₁, a=1, 2 and b=1, 2,Values representing the a-th dimension of feature vector Q _1i,/>Representing the mean value in the a-th dimension in feature space,/>Value representing the b-th dimension of feature vector Q _1i,/>Representing the mean in dimension b in the feature space.

For size calibration, this example defines the calibration function as:

Thus, the first and second substrates are bonded together, The size of (c) can be rewritten as follows:

Γ＝i_Q(Q^*) (40)

Wherein Q represents one of the values in formula (39).

C can then be recalculated as:

thus, based on equations (35) and (36), the robust optimization model can be reformulated as follows:

Through the above learning, this example can obtain a more accurate sample data set. In order to solve the robust optimization model to maximize the network throughput, LRRAIHAoI algorithm is adopted in the method, and the robust optimization model is converted into a convex optimization problem through Lagrangian dual theorem to solve. The pseudo code of LRRAIHAoI resource allocation method is shown in algorithm 1. The channel parameters are random due to imperfections in CSI. Thus, the present example uses a learned channel sample dataset to obtain more accurate channel parameters. And initializing system parameters, setting relevant parameters such as an initial point p ⁰, and solving a solution of the optimization model through iteration.

From the above pseudocode, it can be seen that in algorithm 1, each channel must be learned through a sample dataset to obtain more accurate channel sample parameters, and then solve the model using the Lagrangian pair theorem to obtain a viable solution to the model. Thus, the overall complexity of the proposed LRRA-IHAoI algorithm is O ((INJ)).

Based on the method, the embodiment of the invention also provides a resource allocation system of the ultra-dense industrial Internet of things, which comprises a system model building module, a hypergraph interference model building module, a robust optimization model building module and an optimization model solving module. The system model building module, the hypergraph interference model building module, the robust optimization model building module and the optimization model solving module are respectively used for executing steps S1, S2, S3 and S4 in the method.

Numerical values and simulation results will be provided below to demonstrate the performance of the proposed LRRAIHAoI algorithm and to verify the validity of the proposed LRRA-IHAoI algorithm by comparing the joint robust algorithm with the non-robust algorithm. The simulated scene is UDIIoT using D2D technology. The coverage of the BS is a circle with a radius of 600m, the bandwidth B of the BS is 5MHz, the DD number is 30, and the data set number is 100. The remaining simulation parameters are shown in table 1.

TABLE 1

Fig. 4 shows the relationship between the maximum AoI threshold Δ ^th of the DD and the throughput of the DD in UDIIoT networks. It can be observed that as the maximum AoI threshold Δ ^th increases, the throughput of DD also decreases. This is because the real-time traffic demand of DDs in UDIIoT networks increases and the timeliness of data transmission between DDs will decrease, thereby reducing the communication quality of UDIIoT networks. In addition, it is observed that as the data arrival rate Δ increases, the throughput of UDIIoT networks also increases. This is because as the data transfer rate between DDs increases, packet congestion may result, resulting in a decrease in throughput of UDIIoT networks.

Fig. 5 shows the relationship between DD power threshold p ^th and EE. Figure 5 clearly shows that the proposed LRRA-IHAoI algorithm achieves a higher EE than the joint robust algorithm. This shows that in UDIIoT networks with imperfect CSI, the LRRA-IHAoI algorithm is superior to the joint robust algorithm in implementation EE. This improvement may be attributed to the ability of the LRRA-IHAoI algorithm to reduce channel uncertainty by enhancing the learning sample dataset. Therefore, the algorithm effectively reduces the power consumption of the DD and comprehensively improves the network EE.

Fig. 6 and 7 show the relationship between the power threshold p ^th of the DD and the IE and throughput. As shown in fig. 6, the IE of the proposed LRRAIHAoI algorithm is comparable to the IE of the comparative algorithm. However, FIG. 7 shows that the network throughput of the LRRA-IHAoI algorithm exceeds that of the comparative algorithm. This shows that the LRRA-IHAoI algorithm achieves better quality of service in UDIIoT networks with imperfect CSI than the comparative algorithm.

Meanwhile, from fig. 5, 6 and 7, it is also observed that when the power threshold p ^th increases, the IEs and EE of DD decrease and the throughput increases. This may be due to the fact that as the real-time service requirements of the DD increase, the power consumption of the DD also increases. Thus, this increase in power consumption affects IE and EE, resulting in an overall increase in UDIIoT network throughput.

Fig. 8 shows the relationship between the rate threshold R ^th and UDIIoT network performance, fig. 8 (a) shows the relationship between the rate threshold R ^th and the total EE of the DD, fig. 8 (b) shows the relationship between the rate threshold R ^th and the total IE of the DD, and fig. 8 (c) shows the relationship between the rate threshold R ^th and the total throughput of the DD. As can be seen from fig. 8 (a), (b) and (c), when the rate threshold R ^th increases, the IE, EE and throughput of the DD increase, and the IE, EE and throughput of the proposed LRRA-IHAoI algorithm are greater than those of the comparative algorithm, proving that the performance of the proposed LRRA-IHAoI algorithm is superior to that of the comparative algorithm and the non-robust algorithm. This shows that in UDIIoT networks with imperfect CSI, the LRRA-IHAoI algorithm can obtain better quality of service in differentiated services than the comparative algorithm. At the same time, comparing fig. 8 (a) and (b), it can be observed that the DD IE is much larger than EE, since the proposed LRRA-IHAoI algorithm has good power efficiency, which has an impact on both IE and EE.

Fig. 9 shows the relationship between the noise power σ and UDIIoT network performance, where fig. 9 (a) shows the relationship between the noise power σ of DD and total EE, and fig. 9 (b) shows the relationship between the noise power σ and total throughput of DD. As can be seen from fig. 9 (a) and (b), when the noise power σ increases, EE and throughput of DD decrease. Furthermore, the LRRA-IHAoI algorithm has higher EE and throughput than the comparative algorithm at different noise powers σ. This is because the proposed LRRA-IHAoI algorithm has good robustness and can effectively reduce the impact of imperfect CSI for UDIIoT networks. The results show that the performance of the proposed LRRA-IHAoI algorithm is superior to the comparative algorithm and the non-robust algorithm. This shows that the LRRAIHAoI algorithm presented herein exhibits better performance than the comparison algorithm, and can effectively maximize network throughput in the UDIIoT scenario.

In summary, the method and the system for resource allocation of the ultra-dense industrial internet of things provided by the embodiment of the invention first establish a system model of the ultra-dense industrial internet of things. Then, a hypergraph-based interference hypergraph model is built to avoid overlapping interference in UDIIoT networks, which can analyze the interference type and relationship between D2D Devices (DDs) and IIoTD and identify resource allocation conflicts between different DDs, thereby reducing interference and optimizing spectrum resource utilization. Then, considering that in UDIIoT networks, many real-time applications need to ensure the timeliness of data transmission, the real-time performance is quantified by using an AOI index, and in the overlapping interference scene of UDIIoT networks, complex environments exist, signal attenuation and channel uncertainty increase, so that real-time channel state information cannot be accurately acquired, and therefore, a robust optimization model of incomplete CSI considering power control and resource conflict constraint is established, and the aim of maximizing network throughput is achieved. Finally, in order to solve the optimization model, a LRRA-IHAoI algorithm based on a learning theory is provided, and a more accurate robust optimization model solution is obtained by reducing the influence of imperfect CSI. Simulation results show that the algorithm has good network throughput, interference Efficiency (IE) and Energy Efficiency (EE) in UDIIoT networks, which demonstrates the effectiveness of the proposed scheme.

The main contributions of the invention are as follows:

2) In order to solve the challenge of ensuring timely transmission of service data in UDIIoT network overlapping interference scenarios, the present example uses AoI index to quantify real-time performance and construct AoI model. The model may convert AoI metrics to rates to meet real-time data transmission requirements. On the basis of the interference hypergraph model and the AoI model, a robust optimization model under imperfect CSI is established, and the model combines power control constraint and resource allocation constraint and aims at maximizing network throughput.

Experimental results show that LRRA-IHAoI algorithm can effectively improve network performance, and the effectiveness of the method and the system is verified.

The above examples are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above examples, and any other changes, modifications, substitutions, combinations, and simplifications that do not depart from the spirit and principle of the present invention should be made in the equivalent manner, and the embodiments are included in the protection scope of the present invention.

Claims

1. The resource allocation method of the ultra-dense industrial Internet of things is characterized by comprising the following steps:

the system model comprises a plurality of base stations BS and a plurality of industrial Internet of things devices IIoTD, wherein the plurality of industrial Internet of things devices comprise M cellular devices CD and N D2D devices DD; the set of M CDs is represented as CD _m represents the mth CD, m=1, 2,; the set of N DDs is denoted/>DD _n denotes the nth DD, which, n=1, 2,. -%, N; in the network, DD repeatedly uses the frequency spectrum of CD to communicate; the set of subchannels is defined as/>LK _j denotes the J-th subchannel, j=1, 2,..j;

S2, constructing a hypergraph interference model of the system model;

In the step S3, the robust optimization model is constructed as follows:

C₂:H^THX＝O,

Where x _n,j represents the resource allocation factor for each DD, when x _n,j =1, which means that DD _n is allocated LK _j;R_n,j as the achievable transmission rate from DD _n to IIoTD on LK _j, aoI representing the information age; r ^th represents a minimum rate threshold; pr {. Cndot. } represents the probability of occurrence of event {. Cndot. }; ζ represents a minimum probability threshold; x represents IIoTD resource allocation matrix, X= {0,1} ^(M+N)×J, where the ith row and jth column elements in X O= {0,1} ^(M+N)×J is a zero matrix; p _n,j denotes the transmit power of DD _n on LK _j, and p ^th denotes the maximum transmit power threshold; constraint C ₁ represents a probability constraint that ensures that the AoI requirement of DD _n requires a minimum rate; constraint C ₂ represents a resource allocation constraint based on the interference hypergraph model; constraint C ₃ indicates that the transmit power of each DD cannot be the maximum transmit power threshold; max represents maximizing, s.t. represents satisfying,/>Represents arbitrary;

r ^th is obtained by solving the following formula:

Δ_M/M/1＜Δ^th

Δ _M/M/1 is calculated by the following formula:

ρ is the data utilization of DD, defined as Lambda is the data arrival rate;

and S4, solving the robust optimization model.

2. The method for resource allocation of the ultra-dense industrial internet of things according to claim 1, wherein the step S4 specifically comprises the steps of:

s42, based on the ellipsoid set Constraint C ₁ was rewritten as:

Assuming H _n,j＝[φ]_N×1 and phi= [ H _n,j,h_m,j]^T, Representing post-reconstruction/>H _n,j denotes the channel gain of DD _n on LK _j, h _m,j denotes the channel gain of CD _m on LK _j, h _n,j is a random channel parameter; [] ^T denotes matrix transposition;

S44, further rewriting the constraint C ₁ rewritten in the step S42 as:

3. The method for resource allocation of ultra-dense industrial internet of things according to claim 2, wherein the step S41 specifically comprises the steps of:

S411, collecting I independent imperfect CSI samples Q ₁,Q₂,...,Q_I distributed in the same manner to form a sample set Q _S＝Q₁,Q₂,...,Q_I, wherein the CSI represents channel state information;

S414, defining a potential distribution 1-theta quantile Q _1-θ of a calibration value i (Q) of any sample Q through a set Q _S2, then obtaining a function value of an observed value i (Q ₁),...,i(Q_N), and arranging the observed values i (1) to be less than or equal to i (N) in ascending order, wherein the values are equal to or less than or equal to i (N)As an upper bound for the 1- θ quantile of i (Q);

S416, based on the calculated parameters And C obtaining a sample set/>

4. The resource allocation method of the ultra-dense industrial internet of things according to claim 3, wherein the method comprises the steps of: in the step S45, the rewritten robust optimization model is first converted into a lagrangian dual model, a convex problem is constructed, and then a solution of the model is solved according to the KKT theorem of lagrangian.

5. The resource allocation method of the ultra-dense industrial internet of things according to claim 1, wherein the method comprises the following steps: in the interference hypergraph model, the interference comprises independent interference and accumulated interference; independent interference refers to comparing an interference generated by IIoTD with an interference threshold s _th, and if the interference generated by IIoTD exceeds the interference threshold s _th, then the interference generated by IIoTD is considered to be independent interference; the cumulative interference refers to the interference generated by the plurality IIoTD when the interference exceeds the interference threshold s _th, and the interference generated by the plurality IIoTD is regarded as the cumulative interference.

6. The resource allocation system of the ultra-dense industrial Internet of things is characterized in that: the system comprises a system model building module, a hypergraph interference model building module, a robust optimization model building module and an optimization model solving module; the system model building module, the hypergraph interference model building module, the robust optimization model building module and the optimization model solving module are respectively used for executing the steps S1, S2, S3 and S4 of any one of claims 1 to 5.