Bandwidth allocation method and device for computation task offloading in wireless powered network

Abstract

The application discloses a bandwidth allocation method and device for computing task offloading in a wireless energy supply network. Channel gains of all wireless devices in communication with an edge server are collected as sample data according to a preset time frame, an optimal bandwidth proportion allocation scheme corresponding to each time frame sample data is searched, and total bandwidth proportions occupied by communications of all wireless devices are calculated to form final sample data. A trained neural network model is used to train a constructed neural network model, the trained neural network model is deployed to the edge server, channel gains of all current wireless devices are obtained and input into the neural network model to obtain bandwidth proportions occupied by communications of each wireless device, and bandwidths are allocated to each wireless device according to the obtained bandwidth proportions. The trained neural network model can quickly decide a bandwidth allocation scheme in an edge computing network in a communication mode using FDMA and a partial offloading mode, and a high network computing rate is also achieved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bandwidth allocation, and particularly relates to a bandwidth allocation method and device for computing task offloading in a wireless energy supply network. BACKGROUND

[0002] With the rapid development of Internet of Things (IoT) technology, the efficiency of industry and the quality of people's life are greatly improved. The increase of various Internet of Things wireless devices also brings many challenges. Traditional Internet of Things wireless devices usually use small batteries or wired power supply, such as various sensors and health monitoring devices. However, in many cases, the cost of replacing the battery or continuously powering the wireless device is very high. Therefore, in order to solve the problem of limited life of Internet of Things devices and prolong the use time of the network, using radio frequency energy to power the Internet of Things wireless device becomes a good solution, which can provide stable energy supply for the Internet of Things wireless device. At the same time, computing-intensive and delay-sensitive Internet of Things applications such as augmented reality, real-time monitoring and target detection are becoming more and more widely used, but due to the limitation of the limited computing resources of the Internet of Things wireless device itself, such Internet of Things applications are difficult to be widely used. Mobile edge computing (MEC) is a good solution to provide additional computing resources for Internet of Things wireless devices. Compared with traditional cloud computing, edge server (ECS) is closer to Internet of Things wireless devices, and the communication delay is lower.

[0003] In the edge computing network, the offloading strategy of wireless devices and edge servers needs to be considered, such as total bandwidth allocation, which will directly lead to whether a network is efficient. Each Internet of Things wireless device follows a partial offloading strategy, that is, the computing task of the wireless device can be split, that is, part of the computing task is offloaded to the edge server. There are many indicators to evaluate whether a network is efficient, such as computing rate, maximum delay and minimum energy consumption.

[0004] Most of the current edge computing bandwidth allocation methods mostly use traditional optimization methods, which usually need to be iterated many times to get the optimal or approximate optimal solution, thus consuming too much time, which is unacceptable for delay-sensitive edge computing applications. SUMMARY

[0005] The purpose of the application is to provide a bandwidth allocation method and device for computing task offloading in a wireless energy supply network, which is used to overcome the above technical difficulties and achieve a high computing rate.

[0006] In order to achieve the above purpose, the technical scheme of the application is as follows:

[0007] A bandwidth allocation method for computing task offloading in a wireless energy supply network, comprising:

[0008] According to a preset time frame, channel gains of all wireless devices in communication with the edge server are collected as sample data;

[0009] An optimal bandwidth proportion allocation scheme corresponding to each time frame sample data is searched, and total bandwidth proportions occupied by communication of all wireless devices are calculated to form final sample data;

[0010] The sample data is used to train a constructed neural network model;

[0011] The trained neural network model is deployed to the edge server, channel gains of all current wireless devices are obtained, and are input into the neural network model to obtain bandwidth proportions occupied by communication of each wireless device, and bandwidth is allocated to each wireless device according to the obtained bandwidth proportions.

[0012] Further, the neural network model comprises three full connection layers and one classification layer, the first two full connection layers have 128 neurons, the third full connection layer has 10 neurons, and an activation function of each full connection layer is a Sigmoid function.

[0013] Further, the searching of the optimal bandwidth proportion allocation scheme corresponding to each time frame sample data comprises:

[0014] For any one time frame sample data, bandwidth allocation proportions of the wireless devices are set according to a preset step, and all allocation schemes are found out;

[0015] An optimal calculation rate corresponding to the allocation scheme is calculated;

[0016] An allocation scheme with the maximum optimal calculation rate is found out from all allocation schemes, and is taken as the optimal bandwidth proportion allocation scheme.

[0017] Further, the calculation of the optimal calculation rate corresponding to the allocation scheme comprises:

[0018] For the allocation scheme X j ={x j,1 ,x j,2 ,…,x j,M} in the jth time frame, a calculation rate formula is:

[0019]

[0020] Wherein a is a wireless energy supply duration, B is total bandwidth, v is a ratio of data offloading capacity converted into calculation rate, μ is energy absorption efficiency, N0 is noise power, h j,i is channel gain of the ith wireless device, P is transmission power of radio frequency energy, and M is the number of wireless devices.

[0021] Given a bandwidth allocation scheme X, Q is transformed into a convex problem with respect to a. Existing methods for solving convex optimization problems, such as the point method, can be used to obtain the bandwidth allocation X. j The optimal computational speed Q * .

[0022] This application also proposes a bandwidth allocation device for offloading computing tasks in a wireless power supply network, including a processor and a memory storing a plurality of computer instructions, wherein the computer instructions, when executed by the processor, implement the steps of the bandwidth allocation method for offloading computing tasks in the wireless power supply network.

[0023] This application proposes a bandwidth allocation method and apparatus for offloading computing tasks in a wireless power supply network. Compared with traditional optimization methods, it utilizes a trained neural network model to quickly determine the bandwidth allocation scheme in edge computing networks using FDMA communication and partial offloading modes, while also achieving a higher network computing speed. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of an edge computing network;

[0025] Figure 2 This is a flowchart of the bandwidth allocation method for offloading computing tasks in the wireless power supply network of this application;

[0026] Figure 3 This is a schematic diagram of the neural network model structure in an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] This application provides a bandwidth allocation method for offloading computing tasks in a wireless power supply network, which can be applied to, for example... Figure 1 In the application environment shown, edge servers can provide additional computing resources and a stable power supply for IoT wireless devices.

[0029] The application trains a neural network through a large number of samples, finally obtains a neural network model about bandwidth allocation decision, and deploys the neural network model in an edge server, and compared with a traditional optimization method, has lower delay to adapt to an edge computing application sensitive to delay, and meanwhile obtains higher network computing rate. The application is suitable for quickly deciding a total bandwidth allocation scheme of a wireless energy transmission (WPT) edge computing network in a communication mode using orthogonal frequency division multiple access (OFDMA) and a partial offloading mode, so that higher computing rate is achieved.

[0030] In one embodiment, as shown in Figure 2 A bandwidth allocation method for computing task offloading in a wireless energy supply network is provided, comprising:

[0031] Step S1, channel gains of all wireless devices communicating with the edge server are collected as sample data according to a preset time frame.

[0032] For a wireless energy transmission edge computing network composed of one edge server and M wireless devices, the following is described by taking M equal to 10 as an example. All devices in the network have only one antenna, and communicate through OFDMA, and the total bandwidth in the network is B.

[0033] Channel gain samples in 10000 time frames are collected, and the jth channel gain sample data is the channel gain of 10 wireless devices in the jth time frame, denoted as H j = [h j,1 ,h j,2 ,…,h j,10 ], h j , i denotes the channel gain of the ith wireless device in the jth time frame, wherein j = 1, 2, …, 10000, i = 1, 2, …, 10. The number of collected samples is not limited by the application, and in order to better train the neural network model, the number of samples is greater than 5000.

[0034] In this embodiment, the length of the time frame can be 1 second, or 0.9 seconds, etc. The length of the time frame is set according to actual needs.

[0035] Step S2, the optimal bandwidth proportion allocation scheme corresponding to each time frame sample data is searched, and the total bandwidth proportion occupied by the communication of all wireless devices is calculated to form the final sample data.

[0036] In one specific embodiment, one-dimensional exhaustive search is used to find the optimal bandwidth allocation proportion, that is, in the jth time frame, the communication bandwidth allocation proportion x j of the ith wireless device is calculated according to the channel gain H j,iwhere j = 1, 2, …, 10000, i = 1, 2, …, 10, and must satisfy x j,1 +x j,2 +…+x j,10 ≤ 1, including the following steps:

[0037] Step S2.1, for any one time frame sample data, set the bandwidth allocation ratio for the wireless device according to the preset step size, and find all allocation schemes.

[0038] For example, in the jth time frame, the bandwidth allocation ratio x of each wireless device takes values in [0, 1] with a step size of 0.01, i.e., the value range of x is [0, 0.01, 0.02, …, 0.98, 0.99, 1], and all possible combinations of the value of the bandwidth allocation ratio of 10 wireless devices are combined to obtain 100 10 00000 kinds of bandwidth ratio allocation schemes for wireless devices. If the value of a certain bandwidth ratio allocation scheme x j,1 +x j,2 +…+x j,10 ≤ 1, then the scheme is a feasible scheme, otherwise it is an infeasible scheme.

[0039] Step S2.2, calculate the optimal calculation rate corresponding to the allocation scheme.

[0040] In the jth time frame, for a certain feasible bandwidth ratio allocation scheme X j = {x j,1 , x j,2 , …, x j,M}, the calculation rate formula of the edge computing network is:

[0041]

[0042] where a is the wireless energy supply duration, B is the total bandwidth, v is the data offloading capacity converted into the calculation rate, μ is the energy absorption efficiency, N0 is the noise power, h j,i is the channel gain of the ith wireless device, P is the transmission power of the radio frequency energy. The edge server sends radio frequency energy to the wireless device, i.e., charges the wireless device, P is the transmission power during charging, M is the number of wireless devices. Q(X j , a) is the calculation rate, also represented by Q.

[0043] When the bandwidth ratio allocation scheme X is given, Q is converted into a convex problem about a, and existing methods such as point method are used to obtain the optimal calculation rate Q j under the bandwidth ratio allocation X * .

[0044] Step S2.3, find the optimal allocation scheme with the maximum calculation rate among all allocation schemes as the optimal bandwidth proportion allocation scheme.

[0045] For all feasible schemes obtained in step S2.1, the optimal Q * is calculated according to step 2.2. * The maximum Q * corresponding to the bandwidth proportion allocation X * in the Q j of all feasible schemes is the bandwidth proportion allocation scheme under the current channel gain H j .

[0046] Then, the total bandwidth proportion X j occupied by the communication of the 10 wireless devices under the jth time frame is calculated according to the channel gain samples collected in 10,000 time frames, and X j is the label of the channel gain sample H 10 under the jth time frame, where j = 1, 2, …, 10000, and the 10,000 samples and labels are used for subsequent neural network model training.

[0047] It should be noted that, in addition to the one-dimensional exhaustive search method described above, the iterative solution method can also be used to search for the optimal bandwidth allocation proportion, which will not be described here.

[0048] Step S3, training the constructed neural network model using sample data.

[0049] The constructed neural network model of the embodiment is shown in FIG. 2, which includes three fully connected layers and one classification layer (softmax). Figure 3

[0050] The first two fully connected layers have 128 neurons, and the third fully connected layer has 10 neurons. The activation function of each fully connected layer is a Sigmoid function:

[0051]

[0052] The third fully connected layer outputs a vector Z = [z1, z2, …, z 10 ] with a length of 10, and the vector is input to a softmax layer which includes 10 neurons and has an activation function:

[0053]

[0054] After the calculation by the softmax layer, the total bandwidth proportion X = {x1, x2, …, x 10} occupied by the communication of each wireless device is obtained, and x1+x2+…+x 10 =1, where x irepresents the communication bandwidth allocation of the i-th wireless device under the current channel gain H.

[0055] During training, the input of the neural network model is the channel gain H of 10 wireless devices, and the output is the proportion of the total bandwidth occupied by the communication of each wireless device. The training of the neural network is a relatively mature technology in the technical field, and will not be described here.

[0056] It should be noted that for the collected samples and labels, 7000 in the sample set is used as the training set and 3000 is used as the validation set, that is, the network model is trained using the training set, and the trained network model is verified using the validation set.

[0057] Step S4, deploy the trained neural network model to the edge server, obtain the channel gain of all wireless devices at present, input into the neural network model, get the proportion of the bandwidth occupied by the communication of each wireless device, and allocate bandwidth to each wireless device according to the obtained bandwidth proportion.

[0058] After training the neural network model, it is deployed to the edge server. The edge server can communicate with each wireless device to obtain the channel gain of the wireless device. The channel gain of the wireless device can also be obtained by sending a data packet.

[0059] Thus, for the next time frame, the bandwidth allocation scheme to be adopted, the obtained channel gain of the wireless device is input into the neural network model to obtain the proportion of the bandwidth occupied by the communication of each wireless device, and the bandwidth is allocated to each wireless device according to the obtained bandwidth proportion.

[0060] For example, in a completely new time frame, the channel gain H of 10 wireless devices in the time frame is first obtained, and then H is input into the trained neural network model, and the model outputs the proportion of the total bandwidth occupied by the communication of each wireless device X={x1, x2, …, x 10}x i B is the bandwidth allocated to the i-th wireless device under the channel gain H of the current time frame.

[0061] In another embodiment, the application also provides a bandwidth allocation device for computing task offloading in a wireless energy supply network, comprising a processor and a memory storing a plurality of computer instructions, which are executed by the processor to implement the steps of the bandwidth allocation method for computing task offloading in the wireless energy supply network.

[0062] The specific definitions of the bandwidth allocation apparatus for computing task offloading in a wireless powered network can refer to the definitions of the bandwidth allocation method for computing task offloading in a wireless powered network, which will not be repeated here. The bandwidth allocation apparatus for computing task offloading in a wireless powered network can be realized by software, hardware, and combinations thereof, in whole or in part. It can be embedded in the processor in the form of hardware or independent of the processor in the computer device, or stored in the memory in the computer device in the form of software, so as to be called and executed by the processor to perform the above corresponding operations.

[0063] The memory and the processor are directly or indirectly electrically connected to realize the transmission or interaction of data. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines. The memory stores a computer program executable on the processor, and the processor realizes the network topology layout method in the embodiment of the application by running the computer program stored in the memory.

[0064] The memory can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read only memory (PROM), an erasable programmable read only memory (EPROM), an electrically erasable programmable read only memory (EEPROM), etc. The memory is used to store programs, and the processor executes the programs after receiving an execution instruction.

[0065] The processor can be an integrated circuit chip with data processing capability. The processor can be a general purpose processor, including a central processing unit (CPU), a network processor (NP), etc. It can realize or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the application. The general purpose processor can be a microprocessor or any conventional processor, etc.

[0066] The above-described embodiments are merely illustrative of several embodiments of the present application, which are described in more detail and in a specific manner, but should not be construed as limiting the scope of the patent. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A bandwidth allocation method for computation task offloading in a wireless powered network, characterized in that, The bandwidth allocation method for computing task offloading in the wireless powered network comprises the following steps: According to a preset time frame, the channel gain of all wireless devices communicating with the edge server is collected as sample data; The optimal bandwidth proportion allocation scheme corresponding to each time frame sample data is searched, and the total bandwidth proportion occupied by all wireless devices is calculated to form the final sample data; The neural network model is trained by using the sample data; The trained neural network model is deployed to the edge server, the channel gain of all wireless devices is obtained, and the bandwidth proportion occupied by each wireless device is obtained by inputting the channel gain into the neural network model; The optimal bandwidth proportion allocation scheme corresponding to each time frame sample data is searched, and the total bandwidth proportion occupied by all wireless devices is calculated to form the final sample data; For any one time frame sample data, the bandwidth allocation proportion of the wireless device is set according to a preset step, and all allocation schemes are found out; The optimal computing rate corresponding to the allocation scheme is calculated; The allocation scheme with the maximum optimal computing rate is found out from all allocation schemes, and is taken as the optimal bandwidth proportion allocation scheme; The optimal computing rate corresponding to the allocation scheme comprises the following steps: Under the jth time frame, for the allocation scheme X j = {x j,1 , x j,2 , …, x j,M}, the rate formula is: ， where a is the wireless powered duration, B is the total bandwidth, v is the ratio of data offloading capability converted into computing rate, μ is the energy absorption efficiency, N0 is the noise power, h j,i is the channel gain of the ith wireless device, P is the transmit power of the radio frequency energy, and M is the number of wireless devices. When the bandwidth proportion allocation scheme X is given, Q is transformed into a convex problem about a, and the existing methods such as point method are used to obtain the optimal calculation rate Q under the bandwidth proportion allocation X j . * .

2. The method of claim 1, wherein, The neural network model comprises three fully connected layers and one classification layer, the first two fully connected layers have 128 neurons, the third fully connected layer has 10 neurons, and the activation function of each fully connected layer is a Sigmoid function.

3. A bandwidth allocation device for offloading computing tasks in a wireless power supply network, comprising a processor and a memory storing a plurality of computer instructions, characterized in that, The computer instructions are executed by the processor to realize the steps of the method in any one of claims 1 to 2.

Images