Wireless performance prediction method and device, electronic device and storage medium

Abstract

The disclosed embodiments provide a wireless performance prediction method and device, an electronic device, and a storage medium. The method includes: obtaining a predicted signal-to-noise ratio and a predicted signal-to-interference-and-noise ratio using a target machine learning model; obtaining a signal-to-noise ratio parameter and a signal-to-interference ratio parameter based on the predicted signal-to-noise ratio and the predicted signal-to-interference-and-noise ratio; obtaining a current wireless resource allocation vector; obtaining a current signal-to-interference-and-noise ratio based on the current wireless resource allocation vector, the signal-to-noise ratio parameter, and the signal-to-interference ratio parameter. The dimension of the current wireless resource allocation vector is equal to the number of target interfering user devices, and in the current wireless resource allocation vector, the position corresponding to the target interfering user device that shares the current resource block with the target user device in the current transmission time interval is taken as a first value, and the position corresponding to the target interfering user device that does not share the current resource block with the target user device in the current transmission time interval is taken as a second value. Machine learning technology in the field of artificial intelligence can be used.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a wireless performance prediction method and device, an electronic device, and a computer-readable storage medium. Background Art

[0002] Interference is a core issue in wireless communication systems. The interference matrix is ​​established to predict the signal-to-interference-noise ratio between users to reflect the interference intensity relationship between users. Therefore, the interference matrix plays a vital role in resource allocation.

[0003] In related technologies, the methods for obtaining the interference matrix are mainly as follows:

[0004] One is to build an interference matrix based on frequency sweep data. However, the interference matrix generated based on frequency sweep data cannot reflect the interference situation at unknown locations, especially under dense network conditions, where a small location change may bring about a large interference change. In addition, building an interference matrix based on frequency sweep data requires physical equipment deployment, and the measurement cost is too high, making it inconvenient to implement.

[0005] The other is to establish an interference matrix based on the measurement report message of the mobile phone. However, the interference matrix established based on the measurement report message of the mobile phone only contains the information of the surrounding strong interference base stations, that is, the interference information only contains a few neighboring cells with strong signals. Therefore, the interference information is incomplete and the established interference matrix has certain errors. This construction method affects the integrity of the interference matrix, and therefore cannot effectively avoid interference, and cannot be well adapted to the actual scenario. The interference situation is even worse and more complicated in dense networks. Because when in a dense network with a large number of users, the agility and accuracy of the establishment of the interference matrix have higher requirements, but the non-real-time, inefficient, coarse-grained information and rough prediction accuracy of the current method cannot adapt to the current network situation.

[0006] In the traditional 4G (the fourth generation mobile communication technology) cellular network, the wireless resource allocation function is completed by the base station, and each cell basically manages and allocates wireless resources independently. In order to deal with inter-cell interference, the traditional 4G network negotiates and interacts with signaling between network units, and uses enhanced technology to compensate to a certain extent. For example, by using the X2 interface exchange information between base stations, ICIC (Inter Cell Interference Coordination) or eICIC (enhancedICIC) technology is used to interact with the signaling of interference indication messages between base stations to coordinate the interference between cells and cooperatively allocate resources to solve the inter-cell interference problem; or by using CoMP (Coordinated Multiple Points) technology, different base stations can collaboratively handle interference, avoid interference, or convert interference into useful signals to provide users with higher rates, thereby improving network utilization.

[0007] For ICIC and eICIC technologies, the interference control method based on service quality measurement and signaling interaction lacks accuracy, agility and flexibility due to the large latency, and is difficult to deal with highly dynamic interference. At the same time, a large number of adjacent cells in UDN (Ultra-dense network) will cause considerable signaling exchange overhead.

[0008] However, CoMP technology requires a lot of channel measurements and consumes a lot of pilot resources. At the same time, it is difficult to guarantee the measurement accuracy in a strong interference environment. The resulting channel estimation error will seriously restrict the improvement of system performance, and it requires a lot of computing resources to process and calculate the signal. Therefore, this is not a suitable solution.

[0009] In addition, in some studies, it is usually assumed that the wireless communication system can obtain the geographic location information of all terminals in real time and calculate the wireless interference based on the propagation loss model formula. Although it is theoretically feasible to calculate wireless interference based on the terminal's geographic location information and the propagation loss model, there are obviously two problems in reality: first, it is actually difficult for the wireless communication system to obtain the geographic location information of the mobile terminal at any time; second, the propagation loss model is generally only used in simulation evaluation, network planning and other scenarios, and cannot truly and accurately reflect the radio wave propagation situation in real situations. Summary of the invention

[0010] The embodiments of the present disclosure provide a wireless performance prediction method and device, an electronic device, and a computer-readable storage medium, which can improve the accuracy of the prediction of the interference intensity relationship between user devices.

[0011] The embodiment of the present disclosure provides a wireless performance prediction method, which includes: using a target machine learning model to obtain a predicted signal-to-noise ratio of a target user device and a predicted signal-to-interference-and-noise ratio between the target user device and a target interfering user device of the target user device; obtaining a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user device according to the predicted signal-to-noise ratio of the target user device and the predicted signal-to-interference-and-noise ratio between the target user device and the target interfering user device of the target user device; obtaining a current wireless resource allocation vector of the target user device, the current wireless resource allocation vector corresponding to a current resource block of the target user device in a current transmission time interval; obtaining a current signal-to-interference-and-noise ratio of the target user device according to the current wireless resource allocation vector of the target user device, the signal-to-noise ratio parameter of the target user device and the signal-to-interference ratio parameter. Wherein, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user devices of the target user device, and in the current wireless resource allocation vector, the position corresponding to the target interfering user device that shares the current resource block with the target user device in the current transmission time interval is a first value, and the position corresponding to the target interfering user device that does not share the current resource block with the target user device in the current transmission time interval is a second value.

[0012] An embodiment of the present disclosure provides a wireless performance prediction device, which includes: a signal-to-noise ratio (SNR) and a signal-to-interference-noise ratio (SINR) prediction and output unit, used to obtain a predicted SNR of a target user equipment and a predicted SNR between the target user equipment and a target interfering user equipment of the target user equipment by using a target machine learning model; a signal-to-noise ratio (SNR) and a signal-to-interference-noise ratio (SINR) parameter calculation unit, used to obtain a SNR parameter and a SINR parameter of the target user equipment according to the predicted SNR of the target user equipment and the predicted SNR between the target user equipment and a target interfering user equipment of the target user equipment; a current wireless resource allocation vector acquisition unit, used to obtain a current wireless resource allocation vector of the target user equipment, the current wireless resource allocation vector corresponding to a current resource block of the target user equipment in a current transmission time interval; and a current SINR acquisition unit, used to obtain a current SINR of the target user equipment according to the current wireless resource allocation vector of the target user equipment, the SNR parameter of the target user equipment and the SINR parameter. Among them, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and in the current wireless resource allocation vector, the position value corresponding to the target interfering user equipment that shares the current resource block with the target user equipment in the current transmission time interval is a first value, and the position value corresponding to the target interfering user equipment that does not share the current resource block with the target user equipment in the current transmission time interval is a second value.

[0013] An embodiment of the present disclosure provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the wireless performance prediction method in the above embodiment is implemented.

[0014] An embodiment of the present disclosure provides an electronic device, comprising: at least one processor; and a storage device configured to store at least one program, so that when the at least one program is executed by the at least one processor, the at least one processor implements the wireless performance prediction method as in the above embodiment.

[0015] According to one aspect of the present disclosure, a computer program product or a computer program is provided, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the wireless performance prediction method provided in various optional implementations of the above-mentioned embodiments.

[0016] In the technical solutions provided by some embodiments of the present disclosure, a target machine learning model is first used to predict a predicted signal-to-noise ratio of a target user device and a predicted signal-to-interference-plus-noise ratio between the target user device and each target interfering user device. Then, the signal-to-noise ratio parameter and the signal-to-interference ratio parameter of the target user device are calculated based on the predicted signal-to-noise ratio and the predicted signal-to-interference-plus-noise ratio output by using the target machine learning model. When a current wireless resource allocation vector corresponding to a current resource block of the target user device in a current transmission time interval is obtained, the current signal-to-interference-plus-noise ratio of the target user device is obtained based on the current wireless resource allocation vector, the signal-to-noise ratio parameter and the signal-to-interference ratio parameter of the target user device. On the one hand, using the target machine learning model to predict the predicted signal-to-noise ratio and the predicted signal-to-interference-plus-noise ratio of the target user device can achieve a better prediction effect and improve the accuracy and real-time performance of the prediction. On the other hand, the prediction of the current signal-to-interference-plus-noise ratio of the target user device is not completely handed over to the target machine learning model, but also makes full use of the generation mechanism of interference between user devices in the wireless network. The target machine learning model and the generation mechanism of interference are combined to perform interference prediction, and its physical meaning is clear, so the prediction performance is better. In addition, the method provided by the embodiment of the present disclosure does not require additional hardware or pilot resource occupation, thereby reducing the prediction cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The flowchart of the wireless performance prediction method according to an embodiment of the present disclosure is schematically shown.

[0018] Figure 2 The following is a schematic diagram of an application scenario of a wireless performance prediction method according to an embodiment of the present disclosure.

[0019] Figure 3 The flowchart of the wireless performance prediction method according to an embodiment of the present disclosure is schematically shown.

[0020] Figure 4 A schematic diagram of the training process of an integrated tree model according to an embodiment of the present disclosure is schematically shown.

[0021] Figure 5 The figure schematically shows a prediction diagram of an integrated tree model according to an embodiment of the present disclosure.

[0022] Figure 6 The diagram schematically shows a single leaf node in an integrated tree model before splitting according to an embodiment of the present disclosure.

[0023] Figure 7 The schematic diagram schematically shows a single leaf node after splitting in an integrated tree model according to an embodiment of the present disclosure.

[0024] Figure 8A schematic diagram of a simulation of interference source identification performance according to an embodiment of the present disclosure is schematically shown.

[0025] Fig. 9 The figure schematically shows a simulation diagram of SINR prediction performance according to an embodiment of the present disclosure.

[0026] Fig.10 A simulation diagram schematically illustrates the time consumption of algorithm training according to an embodiment of the present disclosure.

[0027] Fig.11 A schematic diagram of a simulation showing the time consumption for training to achieve the same performance according to an embodiment of the present disclosure is shown.

[0028] Fig.12 The block diagram of a wireless performance prediction device according to an embodiment of the present disclosure is schematically shown.

[0029] Fig.13 A schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0030] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0031] The features, structures or characteristics described in the present disclosure may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.

[0032] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and steps, nor must they be executed in the order described. For example, some steps can be decomposed, and some steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0033] In this specification, the terms "a", "an", "the", "said" and "at least one" are used to indicate the presence of at least one element / component / etc.; the terms "comprising", "including" and "having" are used to express an open-ended inclusion and mean that additional elements / components / etc. may exist in addition to the listed elements / components / etc.; the terms "first", "second" and "third" etc. are used only as labels and are not intended to limit the quantity of their objects.

[0034] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0035] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0036] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0037] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. I believe that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0038] The solution provided in the embodiments of the present application involves technologies such as machine learning of artificial intelligence, which is specifically illustrated by the following embodiments.

[0039] The wireless performance prediction method provided in each embodiment of the present disclosure can be executed by any electronic device, which can be a server or a user equipment (UE), or can be implemented through interaction between a server and a user equipment.

[0040] The server in the embodiments of the present disclosure may be an independent server, a server cluster or a distributed system composed of multiple servers, or a cloud server providing cloud computing services. The UE may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The UE and the server may be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0041] The exemplary embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0042] Figure 1 The flowchart of the wireless performance prediction method according to an embodiment of the present disclosure is schematically shown. Figure 1 As shown, the method provided by the embodiment of the present disclosure may include the following steps.

[0043] In step S110, a predicted signal-to-noise ratio of a target user equipment and a predicted signal-to-interference-plus-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment are obtained using a target machine learning model.

[0044] In the disclosed embodiment, the target machine learning model can be obtained by training using any suitable machine learning algorithm. In the following examples, the XGBoost (eXtreme GradientBoosting) algorithm is selected as an example of a machine learning algorithm, and the target machine learning model obtained by training with XGBoost is referred to as an integrated tree model.

[0045] Among them, XGBoost is derived from the gradient boosting framework, but is more efficient. The algorithm can parallelize calculations, approximate tree construction, effectively process sparse data, and optimize memory usage. It has strong prediction performance and fast training speed. Therefore, this application uses the XGBoost algorithm for interference relationship mining and prediction.

[0046] In the embodiment of the present disclosure, the target UE may be any UE in the wireless communication system. Assume that there are N UEs in the wireless communication system, where N is a positive integer greater than or equal to 1. In the following example, assume that UE iAs the target UE, where i is a positive integer greater than or equal to 1 and less than or equal to N.

[0047] Among them, the target interfering UE of the target UE may refer to any UE that interferes with the target UE in the wireless communication system, or may refer to any UE that remains interfering with the target UE after being processed in the cleaning step. The target interfering UE may be one or more.

[0048] Depending on the different technologies adopted by the wireless communication system, the target interference UE may have certain differences: in a system where orthogonal resources are shared between different UEs in a cell, such as OFDMA (Orthogonal Frequency Division Multiple Access) and TDMA (Time division multiple access) systems, there is no interference between UEs in the cell, and only the interference between UEs in the cell needs to be mined. At this time, the target interference UE does not include UEs in the same cell as the target UE, but only UEs in different cells from the target UE. For systems where there is still interference with UEs in the cell, such as CDMA (Code Division Multiple Access) systems, the target interference UE includes UEs in the same cell as the target UE and UEs that are not in the same cell. By using the method provided in the embodiments of the present disclosure, the interference between different UEs including those in the cell and between cells can be accurately mined and identified.

[0049] In the embodiments of the present disclosure, the signal to noise ratio (SNR) refers to the ratio of the signal to the noise in the system. The signal refers to the electronic signal from the outside of the device that needs to be processed by the device. The noise refers to the irregular additional signal that does not exist in the original signal after passing through the device. This signal is related to the environment and does not change with the change of the original signal. In the following, the UE i The predicted signal-to-noise ratio is expressed as where σ 2 Indicates the effective power of noise, measured in dB.

[0050] In the disclosed embodiment, the signal to interference plus noise ratio (SINR) refers to the ratio of the signal to the sum of the interference and noise in the system. Interference refers to the interference caused by the system itself and the heterogeneous system, such as co-channel interference and multipath interference. In the following, the target machine learning model is used to obtain the UE i With UE i Target interference UEn The predicted signal-to-interference-to-noise ratio is expressed as SINR i,n , the unit of measurement is dB, where n is the subscript of the target interfering UE, and n is a positive integer not equal to i, greater than or equal to 1 and less than or equal to N. In a system where orthogonal resource sharing is performed between different UEs in a cell, assuming that UE i There are q UEs in the same cell, and assuming that the cleaning process of the original data set is not considered, the initial interfering user equipment in the original data set is directly used as the target interfering user equipment, then the target interfering UE n The number can be Nq; if the following cleaning process of the original data set is considered, it is assumed that y initial interfering user equipments among the Nq initial interfering user equipments in the original data set are removed, and the remaining Nqy initial interfering user equipments are used as target interfering user equipments, then the target interfering UE n The number Nqy of y is greater than or equal to 0 and less than or equal to Nq. Where q is a positive integer greater than or equal to 1 and less than or equal to N. For a system where there is still interference with UEs in the cell, and assuming that the following cleaning process of the original data set is not considered, the initial interfering user equipment in the original data set is directly used as the target interfering user equipment, then the target interfering UE n The number can be N-1; if the following cleaning process of the original data set is considered, it is assumed that y initial interfering user equipments among the N-1 initial interfering user equipments in the original data set are removed, and the remaining N-1-y initial interfering user equipments are used as target interfering user equipments, then the target interfering UE n The number can be N-1-y, where y is an integer greater than or equal to 0 and less than or equal to N-1.

[0051] In the following examples, mining the interference relationship between any two wireless links in the uplink direction of a wireless communication system is used as an example for explanation, and at this time, SINR uses the uplink signal-to-interference-plus-noise ratio (UL-SINR). However, the embodiments of the present disclosure are not limited to this, and the method provided in the embodiments of the present disclosure can also be applied to mining the interference relationship between any two wireless links in the downlink direction.

[0052] In an exemplary embodiment, before using the target machine learning model to obtain a predicted signal-to-noise ratio of a target user equipment and a predicted signal-to-interference-plus-noise ratio between the target user equipment and an interfering user equipment of the target user equipment, the method may further include: obtaining a target historical wireless resource allocation vector of the target user equipment and its target historical signal-to-interference-plus-noise ratio; and training the target machine learning model using the target historical wireless resource allocation vector of the target user equipment and its target historical signal-to-interference-plus-noise ratio.

[0053] In an embodiment of the present disclosure, a target machine learning algorithm such as XGBoost training is used to obtain a target machine learning model, and the input data is historical wireless resource allocation data (such as the usage of each RB in each TTI, that is, the target historical wireless resource allocation vector) and network measurement data (such as UL-SINR, that is, the target historical signal to interference and noise ratio).

[0054] Specifically, obtain UE i The target dataset is i The target data set includes the UE i The target historical wireless resource allocation vector and its target historical wireless signal to interference and noise ratio are used as samples to input to the UE i The target machine learning model is used to train the UE i The target machine learning model can be used to train the UE i The target machine learning model is used to obtain UE i The predicted signal-to-noise ratio SNR i,σ2 and UE i With UE n The predicted signal-to-interference-and-noise ratio SINR i,n .

[0055] The dimension of the target historical radio resource allocation vector is equal to UE i The number of target interfering UEs, in a system where orthogonal resource sharing is performed between different UEs in a cell, assuming that UE i There are q UEs in the same cell, and assuming that the following cleaning process of the original data set is not considered, and the original data set is directly used as the target data set, the dimension of the target historical wireless resource allocation vector can be equal to Nq; if it is assumed that the following cleaning process of the original data set is considered, and it is assumed that the target data set has cleaned up y initial interfering UEs in the original data set, the dimension of the target historical wireless resource allocation vector can be equal to Nqy, and y is an integer greater than or equal to 0 and less than or equal to Nq. For systems where there is still interference with UEs in the cell, and assuming that the following cleaning process of the original data set is not considered, and the original data set is directly used as the target data set, the dimension of the target historical wireless resource allocation vector can be equal to N-1; if it is assumed that the following cleaning process of the original data set is considered, and it is assumed that the target data set has cleaned up y initial interfering UEs in the original data set, the dimension of the target historical wireless resource allocation vector can be equal to N-1-y, and y is an integer greater than or equal to 0 and less than or equal to N-1.

[0056] In the target historical radio resource allocation vector, iThe position corresponding to the target interfering UE sharing the corresponding resource block (RB) within the historical transmission time interval (Transmission time interval, TTI) is taken as the first value, which is the same as the UE i The position value corresponding to the target interfering UE that does not share the corresponding RB in the historical transmission time interval is taken as the second value.

[0057] In the embodiment of the present disclosure, the first value and the second value can be set according to actual needs. In the following example, the first value is 1, indicating that the target interference UE and the corresponding target interference UE jointly allocate corresponding RBs within the historical TTI; the second value is 0, indicating that the target interference UE and the corresponding target interference UE do not jointly allocate corresponding RBs within the historical TTI.

[0058] In an exemplary embodiment, obtaining the target historical wireless resource allocation vector and the target historical signal to interference and noise ratio of the target user equipment may include: obtaining an initial historical wireless resource allocation vector and the initial historical signal to interference and noise ratio of the target user equipment, wherein each initial historical wireless resource allocation vector corresponds to each resource block of the target user equipment in each historical transmission time interval, the dimension of each initial historical wireless resource allocation vector is equal to the number of initial interfering user equipment of the target user equipment, and in each initial historical wireless resource allocation vector, the position corresponding to the initial interfering user equipment that shares the corresponding resource block with the target user equipment in the corresponding historical transmission time interval is taken as the first value, and the position corresponding to the initial interfering user equipment that does not share the corresponding resource block with the target user equipment in the corresponding historical transmission time interval is taken as the first value. The position corresponding to the initial interfering user equipment is taken as the second value; the first value of the corresponding position of the initial historical wireless resource allocation vector of the target user equipment is counted to obtain the number of occurrences of each initial interfering user equipment in the initial historical wireless resource allocation vector of the target user equipment; the initial interfering user equipment whose number of occurrences is less than the number of occurrences threshold is eliminated, and the remaining initial interfering user equipment is used as the target interfering user equipment; the initial historical wireless resource allocation vector whose corresponding position value is the first value is eliminated, and the remaining initial historical wireless resource allocation vector is used as the target historical wireless resource allocation vector, and the initial historical signal to interference and noise ratio corresponding to the target historical wireless resource allocation vector is used as the target historical signal to interference and noise ratio.

[0059] In the embodiment of the present disclosure, the collected UE iThe common usage of each RB and other initial interfering UEs in each historical TTI is used as the initial historical wireless resource allocation vector, the corresponding UL-SINR measured is used as the initial historical signal to interference and noise ratio, and the initial historical signal to interference and noise ratio is used as the sample label to form a sample with the corresponding initial historical wireless resource allocation vector to generate UE i The original data set.

[0060] In a system where orthogonal resource sharing is performed between different UEs in a cell, assuming that UE i There are q UEs in the same cell, then the number of initial interfering UEs is Nq, that is, among the N UEs, except for the UE i All UEs other than the q UEs in the same cell are UE i For a system where there is still interference with UEs in the cell, the number of initial interfering UEs is N-1, that is, among the N UEs, UE i All others are UE i The initial interfering UE.

[0061] In some embodiments, the UE i The original data set is used to train UE i The target machine learning model.

[0062] For example, taking i=1 as an example, and in a system where orthogonal resource sharing is performed between different UEs in a cell, assuming that UE 1 There are q = 2 UEs in the same cell, that is, UE 1 With UE 2 There is no interference between them, as shown in Table 1 below. Assume that UE 1 The original dataset is shown below.

[0063] Table 1

[0064]

[0065] In the above Table 1, it is assumed that there are k historical TTIs in total, k is a positive integer greater than or equal to 1, and TTI 1 is given to UE 1 RB 1, RB 2, RB 3, and RB 4 are allocated, and UE 3 With UE 1 RB 1, RB 2, and RB 4 are jointly allocated, and UE 4 With UE 1 RB 1 and RB 4 are jointly allocated, and UE 5 With UE 1 RB 4 is jointly allocated to UE 6 With UE 1 RB 1, RB 2, and RB 3 are jointly allocated, ..., UEN With UE 1 RB 3 and RB 4 are jointly allocated; to UE in TTI 2 1 RB 3, RB 4, RB 5, RB 6, and RB 7 are allocated, and UE 3 With UE 1 RB 3, RB 4, RB 5, and RB 7 are jointly allocated, and UE 4 With UE 1 RB 5 and RB 6 are jointly allocated, and UE 5 With UE 1 RB 3 is jointly allocated to UE 6 With UE 1 RB 4, RB 5, and RB 7 are jointly allocated, ..., UE N With UE 1 RB 4 and RB 5 are jointly allocated; ..., to UE within TTIk 1 RB 9, RB 10, and RB 11 are allocated, and UE 4 With UE 1 RB 10 is allocated to UE 6 With UE 1 RB 9 and RB 10 are jointly allocated, ..., UE N With UE 1 RB 9 and RB 11 are jointly assigned.

[0066] In the original data set of Table 1, each row represents a sample, each column (except the last column) represents an initial interfering UE, and the last column represents the UL-SINR corresponding to an initial interfering UE. For example, the initial historical wireless resource allocation vector (the target historical wireless resource allocation vector at this time) corresponding to RB 1 in TTI 1 is w 1 =(w 1，3 , w 1,4 , w 1,5 , w 1,6 ,…,w 1，N )=(1,1,0,1…,0), the corresponding initial historical signal to interference and noise ratio (which is the target historical signal to interference and noise ratio at this time)=3.25dB, which is input as a sample to the target machine learning model to be trained. Other samples are processed similarly.

[0067] In the disclosed embodiment, the number of samples in the original data set is equal to the sum of the number of RBs in k TTIs = 4 + 5 + ... + 3. All samples in the original data set are directly input into the target machine learning model to be trained. Since the original data set includes the initial wireless resource allocation vectors and initial historical signal to noise ratios of all initial interfering UEs, the trained target machine learning model can predict the interference relationship between any two UEs in the wireless communication system, that is, it includes the interference information in all cells in the wireless communication system, and the interference information is complete and accurate.

[0068] In some other embodiments, the UE may first i The original data set is cleaned and processed to obtain UE i The target dataset is i The target dataset is used to train UE i The target machine learning model.

[0069] Specifically, for UE i The original data set is summed by column (except the last column). Since each row of data is assigned to the UE i The value corresponding to the initial interfering UE with the same RB is 1, otherwise it is 0. Therefore, the summation result is the value of each initial interfering UE in UE i The number of occurrences in the original dataset.

[0070] In the embodiment of the present disclosure, the occurrence threshold can be set as UE i The total number of data items (i.e., the number of samples) in the original data set is a predetermined proportion (e.g., 1%, which can be set according to actual conditions and is not limited in this disclosure). i If the predetermined ratio of the total number of data in the original data set is greater than 1, the column and all rows corresponding to the column being 1 are deleted from the original data set, thereby completing the cleaning of the data set and generating UE i The target dataset.

[0071] Taking Table 1 above as an example, assuming that UE 5 If the sum of the corresponding column is less than the occurrence threshold, the UE is deleted. 5 The corresponding column and UE 5 All rows corresponding to 1 in this column generate the target data set shown in Table 2 below.

[0072] Table 2

[0073]

[0074] In the target data set of Table 2, each row represents a sample, each column (except the last column) represents a target interfering UE, and the last column represents the UL-SINR corresponding to a target interfering UE. For example, the target historical wireless resource allocation vector corresponding to RB 1 in TTI 1 is w 1 =(w 1，3 , w 1，4 , w 1，6 ,…,w 1，N )=(1, 1, 1…, 0), the corresponding target historical signal to noise ratio = 3.25 dB, which is input as a sample to the target machine learning model to be trained. Other samples are processed similarly.

[0075] In the embodiment of the present disclosure, the number of samples in the target data set is equal to the sum of the number of RBs in k TTIs = 3+4+…+3. All samples in the target data set are directly input into the target machine learning model to be trained. Since the target data set excludes the initial interfering UEs with a small number of times of taking the value of 1, the number of samples can be reduced and the training speed of the target machine learning model can be accelerated. i The interference is small, so the impact on the accuracy of the trained target machine learning model is small.

[0076] In an exemplary embodiment, the target machine learning model may be an ensemble tree model.

[0077] Among them, using the target historical wireless resource allocation vector of the target user equipment and its target historical signal to noise ratio to train the target machine learning model can include: inputting the target historical wireless resource allocation vector into the integrated tree model; processing the target historical wireless resource allocation vector through the integrated tree model, and outputting the predicted signal to noise ratio corresponding to the target historical wireless resource allocation vector; training the integrated tree model according to the predicted signal to noise ratio corresponding to the target historical wireless resource allocation vector and the target historical signal to noise ratio. The training process of the integrated tree model can be specifically referred to below Figure 4-Figure 7 Description.

[0078] In an exemplary embodiment, obtaining a predicted signal-to-noise ratio of a target user device using a target machine learning model may include: obtaining an interference-free wireless resource allocation vector for the target user device, wherein the dimension of the interference-free wireless resource allocation vector is equal to the number of target interfering user devices of the target user device, and each bit of the interference-free wireless resource allocation vector takes the second value; inputting the interference-free wireless resource allocation vector into the target machine learning model; processing the interference-free wireless resource allocation vector through the target machine learning model, and outputting the predicted signal-to-noise ratio of the target user device.

[0079] In an exemplary embodiment, using a target machine learning model to obtain a predicted signal to interference plus noise ratio between the target user equipment and the target interfering user equipment of the target user equipment may include: respectively obtaining interference wireless resource allocation vectors of each target interfering user equipment of the target user equipment, wherein the dimension of the interference wireless resource allocation vector of each target interfering user equipment is equal to the number of target interfering user equipment of the target user equipment, and in the interference wireless resource allocation vector of each target interfering user equipment, the value of the position corresponding to the target interfering user equipment is the first value, and the value of the remaining position is the second value; respectively inputting the interference wireless resource allocation vector of each target interfering user equipment into the target machine learning model; the target machine learning model respectively processes the interference wireless resource allocation vector of each target interfering user equipment, and outputs the predicted signal to interference plus noise ratio between the target user equipment and each target interfering user equipment. Obtain UE i The predicted signal-to-noise ratio and predict SINR i，n The process can be referred to below Figure 3 Description.

[0080] In the above training process and the following online prediction process, data under different interference situations can be input, such as a single target interference UE or multiple target interference UEs, wherein a single target interference UE refers to only one target interference UE occupying the same resources (such as the same RB) as the target UE in a certain TTI (which can be a historical TTI or a current TTI) in the system; correspondingly, multiple target interference UEs refer to multiple target interference UEs occupying the same resources as the target UE in a certain TTI in the system.

[0081] In step S120, a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user equipment are obtained according to the predicted signal-to-noise ratio of the target user equipment and the predicted signal-to-interference-plus-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment.

[0082] In an exemplary embodiment, obtaining a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user equipment according to a predicted signal-to-noise ratio of the target user equipment and a predicted signal-to-interference-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment may include: obtaining the signal-to-noise ratio parameter of the target user equipment according to the predicted signal-to-noise ratio of the target user equipment; obtaining the signal-to-interference ratio parameter of the target user equipment according to the signal-to-noise ratio parameter of the target user equipment and a predicted signal-to-interference-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment. i The predicted signal-to-noise ratio and predict SINR i,n Get UEi The process of calculating the signal-to-noise ratio parameters and signal-to-interference ratio parameters can be referred to as follows Figure 3 Description.

[0083] In step S130, a current radio resource allocation vector of the target user equipment is obtained, where the current radio resource allocation vector corresponds to a current resource block of the target user equipment in a current transmission time interval.

[0084] Among them, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and in the current wireless resource allocation vector, the position value corresponding to the target interfering user equipment that shares the current resource block with the target user equipment in the current transmission time interval is a first value, and the position value corresponding to the target interfering user equipment that does not share the current resource block with the target user equipment in the current transmission time interval is a second value.

[0085] In step S140, a current signal to interference and noise ratio of the target user equipment is obtained according to a current radio resource allocation vector of the target user equipment, a signal to noise ratio parameter of the target user equipment and the signal to interference ratio parameter.

[0086] The wireless performance prediction method provided by the embodiment of the present disclosure first uses a target machine learning model to predict a predicted signal-to-noise ratio of a target user device and a predicted signal-to-interference-plus-noise ratio between the target user device and each target interfering user device, and then calculates a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user device according to the predicted signal-to-noise ratio and the predicted signal-to-interference-plus-noise ratio output by using the target machine learning model to predict. When a current wireless resource allocation vector corresponding to a current resource block of the target user device in a current transmission time interval is obtained, the current signal-to-interference-plus-noise ratio of the target user device is obtained according to the current wireless resource allocation vector, the signal-to-noise ratio parameter and the signal-to-interference ratio parameter of the target user device. On the one hand, using the target machine learning model to predict the predicted signal-to-noise ratio and the predicted signal-to-interference-plus-noise ratio of the target user device can achieve a better prediction effect, thereby improving the accuracy and real-time performance of the prediction. On the other hand, the prediction of the current signal-to-interference-plus-noise ratio of the target user device is not completely handed over to the target machine learning model to complete, but also makes full use of the generation mechanism of interference between user devices in the wireless network, and adopts a combination of the target machine learning model and the generation mechanism of interference to perform interference prediction, which has a clear physical meaning and therefore has a better prediction performance. In addition, the method provided by the embodiment of the present disclosure does not require additional hardware or pilot resource occupation, thereby reducing the prediction cost.

[0087] In a standard 4G network, each base station can use all RBs, and each base station is only responsible for resource allocation for UEs connected to it, regardless of the resource allocation status of UEs connected to other base stations, or only performs simple interference coordination through a rough mechanism. Therefore, each base station basically allocates resources independently of each other. The 5G (fifth Generation mobile communication technology) network adopts a CU (Centralized Unit) / DU (Distributed Unit) separation architecture, that is, multiple base stations can be connected to the same CU, so centralized resource allocation and coordination can be achieved.

[0088] The method and apparatus proposed in the embodiments of the present disclosure may be applicable to any wireless communication system. In the following embodiments, an OFDMA wireless communication system 200 of an ultra-dense network (UDN) is taken as an example to illustrate a system scenario.

[0089] Consider a dual-strip model. In a small wireless service hotspot, in order to meet the huge demand for services and throughput, operators will deploy a large number of network devices to form UDN. Figure 2 As shown, assuming that there are 2 subspace bands on both sides of a 20m wide corridor (corridor) 201, each subspace band is composed of Nr subspaces / rooms with a side length of 10m, Nr is a positive integer greater than or equal to 1, for example, assuming Nr = 7, then one of the subspace bands includes subspaces (subareas) 2021, subspace 2022, subspace 2023, subspace 2024, subspace 2025, subspace 2026, and subspace 2027, that is, each subspace size is 10m×10m, and each subspace is deployed with a small base station (Small basestation, SBS) or FAP (Femtocell access point, femtocell access point), each small base station or FAP is assumed to access q users, that is, assuming UE 1 ,UE 2 ,…,UE q Registered with SBS 1 , UE q+1 ,UE q+2 ,…,UE 2q Registered with SBS 2There are a total of R = 4Nr SBSs or FAPs and N = 4Nr*q UEs. UEs registered under the same SBS or FAP can only be allocated orthogonal resources. Small base stations and users (corresponding to the corresponding UEs, users and UEs are corresponding) are randomly deployed in the corresponding rooms, and it is assumed that the minimum distance between small base stations is 8m. The channel model of the urban indoor scene can be used.

[0090] Figure 2 In the embodiment, assuming that q=2, subspace 2021 includes FAP 2031, UE 2041, and UE 2042, and FAP 2031 may correspond to SBS 1 UE 2041 and UE 2042 may correspond to UE 1 and UE 2 Assuming that subspace 2022 includes FAP 2032 and UE 2043, FAP 2032 may correspond to SBS 2 UE 2043 can correspond to UE 3 Core network 206, cloud 205. Each FAP is connected to the cloud 205 via Fronthaul, and the cloud 205 is connected to the core network 206 via Backhaul (also called signal tunnel).

[0091] It should be noted that Figure 2 The scenario shown is only a setting made to facilitate subsequent description and system simulation. The method and device proposed in the embodiment of the present disclosure are actually applicable to any wireless communication networking model and are not limited to the above scenario and parameter settings.

[0092] Let's take XGBoost as an example, combined with Figure 2The UDN network shown illustrates the method provided by the embodiment of the present disclosure. The method provided by the embodiment of the present disclosure uses the wireless resource allocation data (such as the target historical wireless resource allocation vector) and network measurement data (such as the target historical signal to noise ratio) generated in the scheduling process without the need for additional physical equipment deployment and user complete measurement reports, and mines the wireless network interference relationship between users through big data analysis and machine learning algorithms, thereby providing a complete and accurate uplink user interference modeling solution. At the same time, in addition to applying the XGBoost algorithm to mine interference relationships, the embodiment of the present disclosure also derives a wireless network interference model by deeply analyzing the generation mechanism and related information of interference in the wireless network, and combines it with the XGBoost algorithm to predict interference performance. Moreover, the solution provided by the embodiment of the present disclosure is simple to implement, closer to the actual network scenario, and realizes real-time, efficient, high-precision and complete interference prediction. At the same time, compared with the long learning time of the neural network algorithm, the XGBoost algorithm realizes the second-level training data volume and sub-second training time when the prediction performance meets the requirements. It is explained in detail below.

[0093] The user-to-user uplink interference modeling solution based on the XGBoost machine learning algorithm provided in the embodiment of the present disclosure includes two parts: offline training of the XGBoost model and online prediction of interference intensity. The overall solution flow chart is as follows: Figure 3 shown.

[0094] Figure 3 The flowchart of the wireless performance prediction method according to an embodiment of the present disclosure is schematically shown. Figure 3 As shown, the method provided by the embodiment of the present disclosure may include the following steps.

[0095] In step S301, i is initialized to 1.

[0096] The method provided in the embodiment of the present disclosure can be used to mine the interference relationship between any users registered in different small base stations. In the following example, UE i =UE 1 For example, predict UE 1 The interference strength between UEs and all other target interfering UEs. In the disclosed embodiment, UL-SINR is used to characterize the interference strength between UEs.

[0097] In step S302, an original data set of the i-th target user equipment is obtained.

[0098] For example, to obtain UE 1 The original data set can be found in Table 1 above.

[0099] In step S303, the original data set of the i-th target user equipment is cleaned to obtain a target data set of the i-th target user equipment.

[0100] For example, for the obtained UE 1 The original data set is cleaned to obtain the target data set shown in Table 2 above.

[0101] In step S304, the target data set of the i-th target user equipment is used to train and obtain an integrated tree model of the i-th target user equipment.

[0102] In the disclosed embodiment, the integrated tree model of each target user device is trained separately, that is, the target data set of each target user device is used to train the corresponding integrated tree model as the target machine learning model.

[0103] For example, UE 1 The training process of the integrated tree model includes: each time training, input UE 1 The samples in the target dataset (for example, UE 1 The historical wireless resource allocation vectors of each target) are input to the XGBoost model. During the training process, the XGBoost model will predict and output the corresponding That is, the predicted UL-SINR, and then the objective function is solved according to the predicted UL-SINR and the label corresponding to the sample, that is, the actual UL-SINR as the target historical signal to noise ratio, and the sample iteration is continuously input until the number of subtrees reaches the preset maximum number of subtrees K (K is a positive integer greater than 1, and K=100 is used as an example in the following example), the training process ends, and the XBGoost model has been trained.

[0104] Specifically, to model UE 1 As an example, the target data set input to the integrated tree model is the corresponding RB in the corresponding historical TTI in the system and the UE 1 The usage of the UE that may cause interference, such as the target historical radio resource allocation vector shown in Table 2, and the UE 1 The UL-SINR per RB per historical TTI is, for example, a target historical signal to interference and noise ratio as shown in Table 2.

[0105] For a system where there is still interference with the UE in the cell, and assuming that the following cleaning process of the original data set is not considered, the UE 1 Target interference UE n It can be all UEs except itself, that is, UE 2 ,…UE N , that is, the value of n is 2, …N.

[0106] In a system where orthogonal resource sharing is performed between different UEs in a cell, assuming that UE 1 There are q = 2 UEs in the same cell, and it is assumed that the following cleaning process of the original data set is not considered. 1 Target interference UE n Can be used for UE 3 ,UE 4 ,…UE N , that is, the value of n is 3, 4, …N.

[0107] These UE 1 Target interference UE n As a characteristic attribute of the data, the attribute value is {0,1}. Taking TTI t in Table 2 as an example, the last column of data represents the UE 1 UL-SINR, and other columns of data indicate possible 1 Target interfering UE n Each row of data represents the UE 1 Target interference UE n Case, 1 means the target interferes with UE in TTI t n With UE 1 Use the same resources, 0 means not using the same resources, that is, w i ={w i,n}. Record the UE in the system i The resource allocation data and network measurement data for all TTIs during data transmission are provided.

[0108] When considering a system such as OFDMA where there is no interference between users in a cell in a typical scenario, only the interference between users in the cell can be mined. i Target interference UE n The other q-1 UEs registered with the same base station can be removed, reducing the number of interfering users, the amount of training data and the complexity of the algorithm. It should be noted that if the interference between users in the cell still needs to be considered, the UE i Target interference UE n There is no need to remove UEs in the same cell.

[0109] The data in Table 1 or Table 2 are used as the input data of the XGBoost model. The maximum depth d of the algorithm input tree max Can be set to all possible UE i The number of UEs causing interference. If the interference between users in the cell is considered, then d max =N-1; if the interference between users in the cell is not considered, then d max=Nq. Each leaf node needs at least one sample to be estimated, so the minimum number of samples in the leaf node ε can be set to 1; the remaining parameters can be tried and adjusted according to the quality of RMSE (Root Mean Squared Error); the output of the model is the predicted UL-SINR, which is the predicted signal to interference and noise ratio corresponding to the target historical wireless resource allocation vector.

[0110] In step S305, the interference-free wireless resource allocation vector of the ith target user equipment and the interference wireless resource allocation vectors of each target interfering user equipment are respectively input into the trained ensemble tree model of the ith target user equipment, and the predicted signal-to-noise ratio and predicted signal-to-interference-plus-noise ratio of the ith target user equipment are output.

[0111] In the embodiments of the present disclosure, based on the aforementioned typical system models and scenarios, the generation mechanism of interference between wireless network cells is deeply analyzed, and on this basis, a SINR prediction technology based on the trained XGBoost model and the wireless network interference model obtained by analysis is proposed.

[0112] In the uplink direction of the mobile communication network, assuming that UE i The transmitted uplink signal reaches the home cell Useful signal receiving power of small base stations for:

[0113]

[0114] In the above formula, j i is a positive integer greater than or equal to 1 and less than or equal to R, Indicates the jth i SBS;P i For UE i The transmitted uplink signal transmission power; For UE i To the home cell Channel gain of small base stations.

[0115] With user UE i Target interfering UE occupying the same radio resources (e.g., the same RB) n Arrival at the community Interference signal power of small base stations for:

[0116]

[0117] In the above formula, P n For UE n The transmitted uplink signal transmission power; Interfering UE nTo the community Channel gain of small base stations.

[0118] UE i The signal is in the home cell The signal-to-interference-to-noise ratio at the small base station is:

[0119]

[0120] In the above formula, N represents the set of users in the system, N i Indicates that it is not compatible with UE i The set of users that cause interference, for example, in a system where orthogonal resource sharing is performed between different UEs in a cell, N i Indicates that UE i The collection of UEs in the same cell, i.e., UE n is a UE in the system, and UE n Not with UE i In the same community; i,n ∈{0,1} represents UE n Whether to use the corresponding radio resources such as RB to i Causes interference, which can be called the wireless resource allocation vector (including the target historical wireless resource allocation vector and the current wireless resource allocation vector); σ 2 is the noise power.

[0121] The above formula (3) can be further transformed to obtain:

[0122]

[0123] In the above formula, For UE i The received signal signal-to-interference ratio (SIR, considering interference but not noise); For UE i The signal-to-noise ratio (SNR) of the received signal (without considering interference and considering noise).

[0124] make:

[0125]

[0126]

[0127]

[0128] Then substitute formula (5)-(7) into the above formula (4) and simplify formula (4) to:

[0129]

[0130] Then, considering that UL-SINR is usually expressed in dB in actual usage scenarios, γ i The value is converted to dB, that is, the logarithm of both ends of the equation is taken to obtain the following formula:

[0131]

[0132] At this time, γ i In dB, after taking the logarithm of both ends of the equation, the multiplicative small-scale fading becomes additive, which can be well captured by the error term; at the same time, the optimization target of the model also becomes the dB value of UL-SINR, which is more in line with the actual usage scenario. In addition, the model can help ICIC and eICIC improve performance by mining the signal-to-interference ratio SIR and signal-to-noise ratio SNR; and SINR performance prediction can assist subsequent resource management.

[0133] According to the above formula, we can deduce:

[0134]

[0135]

[0136] In the above formula, Called UE i A signal-to-noise ratio parameter, where the signal-to-noise ratio parameter is related to the ratio of the average power of the received signal to the average power of the noise in the target UE; Called UE i The signal-to-interference ratio parameter is related to the ratio of the energy of the received signal of the target UE to the interference energy (such as co-channel interference, multipath, etc.), excluding noise energy.

[0137] The disclosed embodiment mines interference relationships with the help of the XGBoost model, and uses the network interference relationship modeling function shown in formula (9) of the wireless network interference model to perform interference prediction. After the XGBoost model training is completed, the interference intensity prediction is performed online, and the input data corresponding to different interference situations (such as a single interfering user, multiple interfering users, etc.) are input into the XGBoost model, and the corresponding interference prediction results are output.

[0138] In the network, the performance of a user can be expressed by the transmission rate, and the transmission rate is directly affected by the UL-SINR. Therefore, the performance of a user depends not only on the interference received, but also on the received signal strength of the user. The higher the received signal strength, the higher the tolerance to interference, that is, under the same interference condition, the higher the signal strength, the higher the UL-SINR, the higher the transmission rate, and the better the performance. Therefore, the embodiment of the present disclosure comprehensively considers the received signal strength and the interference signal.

[0139] Specifically, the trained XBGoost model is used to predict and output the corresponding and SINR i,n At this time, the format of the data input into the trained XBGoost model is shown in Table 3 below to perform interference relationship Mining. It should be noted that the dimension of the data input into the trained XBGoost model in Table 3 is illustrated by taking Table 2 as an example, but the present disclosure is not limited thereto. If the input data used in the training process is Table 1, Table 3 can also be changed to the dimension of the data in Table 1 accordingly.

[0140] Table 3

[0141]

[0142] In Table 3 above, the (0,0,0,…,0) vector input to the trained XGBoost model is called the interference-free wireless resource allocation vector. The interference-free wireless resource allocation vector does not consider interference but considers noise. Therefore, after the XGBoost model processes the interference-free wireless resource allocation vector, the predicted output is UE i The predicted signal-to-noise ratio (Table 3 takes i=1 as an example for illustration) The vector (1,0,0,…,0), (0,1,0,…,0),…(0,0,0,…,1) input to the trained XGBoost model is called the interference wireless resource allocation vector. The interference wireless resource allocation vector takes interference into account. Therefore, after the XGBoost model processes the corresponding interference wireless resource allocation vectors respectively, the predicted output is UE i and the corresponding target interference UE n The predicted signal-to-interference-and-noise ratio SINR i,n (Table 3 is illustrated by taking i=1, n=3 or 4 or 6 or ... or N as an example).

[0143] The data format input to the trained XGBoost model is shown in Table 3, which respectively inputs the interference-free wireless resource allocation vector of the non-interfering UE and all the different single target interfering UEs. n The corresponding interference wireless resource allocation vectors in the case are input into the trained XGBoost model respectively, and the corresponding and SINR i,n , in dB; finally, the dB value is converted into a real value and calculated using the above formulas (10) and (11) to obtain and Thereby the corresponding interference relationship is extracted.

[0144] It should be noted that, unless otherwise specified, in the embodiments of the present disclosure, SINR refers to uplink SINR, i.e., UL-SINR, and the two are interchangeable. Similarly, when considering systems such as OFDMA where there is no interference between users in a cell, only the interference between users in the cell needs to be mined. In this case, the UE i The target interfering UE does not include other q-1 users registered in the same small base station, which reduces the number of target interfering UEs and reduces the number of input data from N to N-q+1.

[0145] In Table 3, each characteristic attribute represents the possible i Other users or UEs that cause interference (i.e., target interfering UEs) n ), each row of data represents a certain RB in a certain TTI for UE i Target interference UE n situation, 1 means that this target interferes with UE n With UE i The RB is reused, and 0 means that the RB is not reused, so each data sample represents the UE in a certain TTI. i Sharing a RB with a certain user, inputting this data sample into the trained XGBoost model can get UE i SINR when interfering with any other single UE.

[0146] Among them, in the analysis of UE i The UL-SINR for single UE or single user interference does not distinguish between RBs and TTIs, which means that no matter the UE i Interfering with target UE n The UL-SINR of the RB in the system is the same in which TTI (without considering the influence of fast fading), so when constructing UE i Interfering with target UE n RB and TTI are not considered when analyzing the interference relationship of UE. That is, when considering large-scale fading, the interference effects caused by different RBs and TTIs are the same and do not need to be distinguished; while small-scale fading is unpredictable and is therefore not considered. i When calculating the UL-SINR for single-user interference, it is assumed that different RBs and TTIs have the same impact on the interference relationship between users, so no distinction is made between RBs and TTIs.

[0147] In step S306, a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the i-th target user equipment are calculated according to the predicted signal-to-noise ratio and the predicted signal-to-interference-plus-noise ratio of the i-th target user equipment.

[0148] In the disclosed embodiment, the corresponding interference relationship is mined through the trained XGBoost model. and Then, using the formula (9) derived above, we get As the network interference relationship modeling function, interference prediction is performed. Through the network interference relationship modeling function, the corresponding w is input i,n The current wireless resource allocation vector w is 0 or 1 i , the UL-SINR can be predicted under any multi-user interference condition. i M t In this paper, the interference vector data corresponding to different multi-user interferences are randomly input into the XGBoost model, the corresponding interference intensity is predicted, and compared with the theoretical SINR value, and the prediction performance is represented by RMSE.

[0149] In step S307, an interference vector of the i-th target user equipment is generated according to the signal-to-noise ratio parameter and the signal-to-interference ratio parameter of the i-th target user equipment.

[0150] In the embodiment of the present disclosure, the training of Table 3 is respectively input into the trained XGBoost model, and the prediction output is and SINR i,n , then, through formula (10) and formula (11) get When each target interfering UE is obtained n of Afterwards, they can be combined to form UE i The corresponding interference vector, such as The value of n depends on the target interference UE n Depends on the value of .

[0151] Then, Substitute into formula (9) to obtain the predicted output γ i , as the final predicted UL-SINR value, that is, UE i That is, in the embodiment of the present disclosure, the trained XGBoost and network interference relationship modeling function are jointly used to predict the UL-SINR value.

[0152] In step S308, i=i+1.

[0153] In step S309, determine whether i is greater than N; if i is greater than N, execute step S310; if i is less than or equal to N, jump back to the above step S302 and repeat the above steps S302-S308.

[0154] In step S310, interference matrices of N user equipments are generated according to interference vectors of the first to Nth target user equipments.

[0155] Repeat the above operation for N UEs in the system, and finally obtain the interference intensity relationship between any two UEs in the system, and perform SINR performance prediction. For example, assume that the interference matrix finally obtained is as shown in Table 4 below.

[0156] Table 4

[0157]

[0158]

[0159] In step S311, the current radio resource allocation vector of the j-th target user equipment is obtained.

[0160] In the embodiment of the present disclosure, j may be a positive integer greater than or equal to 1 and less than or equal to N, that is, the jth target user equipment may be any UE in the system. For example, assuming that the current radio resource allocation vector of the jth target user equipment is input: j =(w j,1 , w j,2 ,…,w j,n ,…,w j,N )=(1,0,…,1,…,1).

[0161] In step S312, the current signal to interference and noise ratio of the jth target user equipment is predicted according to the current radio resource allocation vector of the jth target user equipment and the interference vector corresponding to the jth target user equipment in the interference matrix.

[0162] According to the current radio resource allocation vector w of the jth target user equipment j , obtain the interference vector corresponding to the jth target user equipment in the interference matrix. At this time, i in the above formula (9) can be expressed as j, that is, The current signal to interference and noise ratio of the j-th target user equipment is predicted and obtained.

[0163] Combine the following Figures 4 to 7 This paper gives an example to illustrate the training process of the ensemble tree model.

[0164] Figure 4 The following schematically shows a training process diagram of an integrated tree model according to an embodiment of the present disclosure. Figure 4 As shown in Figure 1, XGBoost uses a greedy method to learn tree by tree, and each tree fits the residual of the previous model. Here, it is assumed that the cleaning process is not considered, and the original data set D of the target UE is directly used to train the weak learner f 1 , calculate the weak learner f 1 The residual of the residual E 1 , using the residual E 1 Train weak learner f 2, calculate the weak learner f 2 The residual E 2 , using the residual E 2 Train weak learner f 3 , ...compute the weak learner f K-1 The residual E K-1 , using the residual E K-1 Train weak learner f K , get the weak learner f K , the weak learner f 1 , weak learner f 2 , weak learner f 3 ,…, weak learner f K The strong learner f is obtained by summing up. Each weak learner is a tree or subtree in the ensemble tree model. It is assumed that there are K subtrees in total, that is, the number of trees or subtrees created after training. K is a positive integer greater than 1. In the following example, K=100 is used as an example, but the present disclosure is not limited to this and can be set according to the actual scenario.

[0165] Figure 5 The following schematically shows a prediction diagram of an integrated tree model according to an embodiment of the present disclosure. Figure 5 As shown, for the target historical wireless resource allocation vector w, the subtree f 1 Assume that it includes nodes 501, 503, 504, 505, and 506, wherein nodes 501 and 503 are non-leaf nodes in tree f1, and nodes 504, 505, and 506 are leaf nodes in tree f1. Node 501 is the parent node of nodes 503 and 504, and node 503 is the parent node of leaf nodes 505 and 506.

[0166] Subtree f 2 Assume that the tree includes node 502, node 507 and node 508, wherein node 502 is a non-leaf node in tree f2, and node 507 and node 508 are leaf nodes in tree f2. Node 502 is the parent node of leaf node 507 and leaf node 508.

[0167] ∑ sum is performed on leaf nodes 504 and 507 to obtain As the predicted signal to interference and noise ratio corresponding to the target historical wireless resource allocation vector w.

[0168] The XGBoost model uses the cumulative output values ​​of K subtrees to predict the output. The residual of the model before each tree is fitted can be expressed as follows:

[0169]

[0170] Where: fk (w) is the kth subtree f k The weight of the leaf node corresponding to the target historical wireless resource allocation vector w.

[0171] Figure 5 The following is a diagram of a simple XGBoost prediction model consisting of two trees, which is used for illustration only.

[0172] In the embodiment of the present disclosure, the objective function L used by XGBoost can be:

[0173]

[0174] Where i is a positive integer greater than or equal to 1 and less than or equal to I, I is a positive integer greater than 1, assuming that there are I samples in the target dataset or original dataset used to train the XGBoost model, i represents the i-th sample, γ (i) represents the target value of the i-th sample, i.e., the true UL-SINR as the label, Represents the predicted value of the i-th sample, that is, the predicted signal-to-interference-noise ratio of the i-th sample; the first part is the loss function, part 2 is the regularization term.

[0175] In the above formula (13), the loss function measures the difference between the predicted value and the target value of all I samples in the target data set or the original data set. The embodiment of the present disclosure uses the root mean square error RMSE:

[0176]

[0177] The regular term is used to control the complexity of the tree and prevent overfitting. For any tree / subtree, it can be expressed as:

[0178]

[0179] In the above formula, T is the total number of leaf nodes of the tree / subtree f, and T is a positive integer greater than or equal to 1; s t is the weight of the tth leaf node of the tree / subtree f, and t is a positive integer greater than or equal to 1; α and β are hyperparameters for determining the L0 and L2 regularization strengths of the tree f, respectively. The larger these hyperparameters are, the more we hope to obtain a tree with a simpler structure. They can be set according to actual needs, and the present disclosure does not limit this.

[0180] Then, determine the kth subtree f k The weight of each leaf node. Since the model is trained in the form of addition, formally, f k (w (i)) as the predicted value of the i-th sample in the k-th iteration, add f k (w (i) ) to minimize the following objective function, that is, the objective function when training the kth tree is L (k) :

[0181]

[0182] Greedily add f through the objective function k (w (i) ) to improve the model to the greatest extent. For formula (16), a second-order approximate solution can be used to quickly optimize the objective.

[0183] The loss function is expanded using the second-order Taylor approximation, so the loss function can be customized as long as it satisfies the second-order differentiability:

[0184]

[0185] In the above formula, constant is a constant term, g i and h i are the first-order and second-order derivatives of the loss function of the i-th sample, respectively, which can be expressed as:

[0186]

[0187]

[0188] Then substitute the objective function L (k) , let the sample set in leaf node t be D t , where the model is trained for each UE i Interfering with other target UEs n The interference relationship model between them is, for example, the XGBoost model, so I samples are the UE i The I target historical wireless resource allocation vectors w, and the sample set D of leaf node t t In the process of tree creation, according to the splitting rules of leaf nodes, sample I is continuously divided into different subsets on different leaf nodes t, thus forming a sample set.

[0189] For any i∈D t ,have:

[0190]

[0191] Then we have:

[0192]

[0193] Since the optimization objective function is to find the minimum value, the constant term can be removed to simplify the objective function.

[0194] The above formula (21) is the weight s of leaf node t t The quadratic function of is:

[0195]

[0196]

[0197] The optimal weight can be obtained as:

[0198]

[0199] To avoid overfitting and leave room for subsequent subtree optimization, after the subtree is created, the weights of all leaf nodes are multiplied by the learning rate η∈(0,1), so that the final node weight is:

[0200]

[0201] The goal of XGBoost model training is to minimize the objective function L. For each subtree, the leaf nodes will be continuously split and the node weights will be updated.

[0202] Figure 6 The schematic diagram of a single leaf node before splitting in an integrated tree model according to an embodiment of the present disclosure is schematically shown. Figure 6 As shown, it is assumed that it includes node 601, node 602, node 603, node 604, node 605, node 606 and node 607. Node 601 is the parent node of node 602 and node 603, node 602 is the parent node of node 604 and node 605, and node 603 is the parent node of node 606 and node 607.

[0203] Figure 6 Node 601, node 602, and node 603 are non-leaf nodes, and node 604, node 605, node 606, and node 607 are leaf nodes.

[0204] Figure 7 The schematic diagram schematically shows a single leaf node after splitting in the integrated tree model according to an embodiment of the present disclosure. Figure 7 As shown, assuming that Figure 6 Node 605 in is split into node 701 and node 702. Figure 6 Node 605 in the Figure 7 Node 701 and node 702 are leaf nodes.

[0205] The original leaf node t (here Figure 6 For example, the leaf node 605 in Figure 7 The leaf node 701 (labeled as t L ) and leaf node 702 (labeled as t R ) as an example, substitute formula (24) into the objective function, that is, formula (21), and we can get Therefore, the change ΔL of the objective function can be obtained as follows:

[0206]

[0207] Among them, the first is the score value of the parent node 605 before segmentation. The second item and the third The sum of the scores of the left and right subtrees after segmentation, and the last item The complexity is caused by introducing additional leaf nodes.

[0208] The original leaf node t may contain multiple samples, so there are multiple splitting schemes, corresponding to different ΔL; therefore, the scheme with the smallest ΔL and less than 0 is selected for node splitting and the weights are updated until at least one of the following conditions is met for each leaf node:

[0209] The number of samples contained in the leaf node is not greater than the minimum number of samples ∈;

[0210] The depth of the leaf node reaches the maximum depth d of the tree max ;

[0211] For all splitting schemes of the current leaf node, ΔL ≥ 0.

[0212] Once the leaf nodes of the subtree are split, the subtree is created.

[0213] Create a new subtree and split it. The previously created subtree will not change. When the number of subtrees reaches the maximum number of subtrees K, the subtree creation stops and the training process ends.

[0214] The method provided by the embodiment of the present disclosure can be used for modeling uplink interference between users, and can predict uplink SINR between users. XGBoost is used to mine the wireless network interference relationship between users through training of the target data set, and the derived wireless network interference model is used to predict the uplink interference situation of users at future moments: For example, the maximum depth d of the tree in the parameter of XGBoost is maxCorresponding to the number of target interfering UEs; each leaf node needs at least 1 interference sample to be estimated, so the minimum number of samples ∈ in the leaf node is set to 1; the objective function is set to the mean square error function, corresponding to the root mean square error of the interference performance evaluation index. The disclosed embodiment also proposes to use historical wireless resource allocation data (such as RB usage) and network measurement data (such as UL-SINR) as input data of the machine learning model for model training. A method for cleaning the original data set is also proposed. On the one hand, the disclosed embodiment can more accurately mine the interference relationship between users with fewer training samples, and can cope with wireless networks with highly dynamic interference; at the same time, since the learning time and training overhead required for machine learning model training are shortened, it can better adapt to the real-time requirements of the actual network. For example, when the method provided in the disclosed embodiment is trained using 5000 samples (second-level data), the prediction performance requirement of SINR prediction performance error less than 0.5dB can be met.

[0215] Table 5 below summarizes the parameters used in the XGBst model simulation:

[0216] Table 5

[0217]

[0218] According to the method proposed in the embodiment of the present disclosure, taking the XGBoost model applied to ultra-dense network interference prediction as an example, the performance of interference source identification performance, SINR prediction performance, algorithm training time and time consumption to achieve the same performance will be evaluated and displayed respectively, and compared with the performance of NN-MLP (Neural network-multilayer perceptron) and linear regression algorithm (Linear regression algorithm, LRA). XGB-PF represents the method proposed in the embodiment of the present disclosure, which first uses the XGBoost model to mine interference relationships, and then uses the network interference relationship modeling function to predict interference.

[0219] Figure 8 The following schematically shows a simulation diagram of interference source identification performance according to an embodiment of the present disclosure. Figure 8 As shown in the figure, the horizontal axis is the size of the training data set for each user, and the vertical axis is the average RMSE (dB).

[0220] Depend on Figure 8It can be seen that when the data volume per user is not less than 2000 samples, the interference source identification performance of the XGB-PF algorithm is better than that of the linear regression (LRA) algorithm; when the data volume per user is not less than 5000 samples, the performance of the XGB-PF algorithm is better than that of the neural network (NN-MLP) algorithm; and both the XGB-PF algorithm and the neural network algorithm are far superior to the linear regression algorithm.

[0221] When the data volume per user is 10,000 samples, the interference source identification performance prediction error of the XGB-PF algorithm is less than 0.5dB, meeting the performance requirements; when the data volume per user reaches 200,000 samples, the prediction error of the XGB-PF algorithm has reached 0.12dB, which is much better than the 0.25dB of the neural network algorithm and the 2.59dB of the linear regression algorithm.

[0222] At the same time, the linear regression algorithm also has the following problems: the predicted value is not always positive, which makes it impossible to convert to dB; the linear regression algorithm has poor processing ability for small training variables; there is a gap between the optimization goal of the linear regression algorithm and the target problem. Based on this, although the linear regression algorithm has a faster training time, its prediction performance is worse than that of the neural network algorithm.

[0223] Since accurate interference source identification can improve the performance of traditional technologies such as ICIC, eICIC and CoMP, and the interference source identification performance of the XGB-PF algorithm is much better than that of the neural network algorithm and the linear regression algorithm, the XGB-PF algorithm can be used to assist traditional technologies such as ICIC, eICIC and CoMP to improve their performance.

[0224] Fig. 9 The following schematically shows a simulation diagram of SINR prediction performance according to an embodiment of the present disclosure. Fig. 9 As shown in the figure, the horizontal axis is the size of the training data set for each user, and the vertical axis is the average training time (s). Fig. 9 The XGB-old method in the paper refers to directly mining interference relationships and predicting SINR performance through the XGBoost algorithm, that is, the XGBoost algorithm directly identifies interference sources and predicts SINR performance after training. After completing the XGBoost model training, the interference intensity prediction is performed online, and the input data corresponding to different interference situations (such as a single interfering user, multiple interfering users, etc.) are input into the XGBoost model, and the corresponding interference prediction results are output.

[0225] Since the XGBoost algorithm can also be used directly to predict SINR and obtain results, Fig. 9In the above, the XGB-old method is added. It can be seen that although the performance of XGB-old is slightly better than the neural network (NN-MLP) algorithm, it is worse than the XGB-PF algorithm proposed in the embodiment of the present disclosure. XGB-PF performs SINR prediction based on XGB-old and combines the network interference relationship modeling function, so the prediction performance is better and the performance is improved.

[0226] For each user, when 5,000 samples (second-level data) are used for training, the XGB-PF algorithm can meet the prediction performance requirement of SINR prediction performance error less than 0.5dB, reaching 0.49dB, which is much better than the 2.78dB of the linear regression (LRA) algorithm and the 0.81dB of the neural network (NN-MLP) algorithm. When the SINR prediction performance of the neural network algorithm reaches 0.5dB, 50,000 samples are required, which is 10 times that of the XGB-PF algorithm; the linear regression algorithm converges too slowly, and the prediction performance is still greater than 1dB even at 200,000 samples, which is too poor.

[0227] When the data volume per user reaches 200,000 samples, the SINR prediction performance error of the XGB-PF algorithm has reached 0.21dB, which is still much better than the 0.31dB of the XGB-old method, the 0.38dB of the neural network algorithm, and the 1.16dB of the linear regression algorithm.

[0228] Due to the excellent SINR prediction performance of the XGB-PF algorithm, the XGB-PF algorithm can be used to quickly and accurately identify the interference sources and interference strengths in the wireless network, and construct interference vectors based on the interference identification results to form an interference matrix, thereby serving the subsequent wireless resource management and allocation.

[0229] Fig.10 The following schematic diagram shows the simulation time consumption of algorithm training according to an embodiment of the present disclosure. Fig.10 As shown, the horizontal axis is the average RMSE (dB), and the vertical axis is the average training time (s), XGB-old, NN-MLP, XGB-PF, and LRA.

[0230] As shown in the figure above, the XGB-PF algorithm takes one order of magnitude less time than the neural network algorithm (NN-MLP) under the same training data set scale, and is basically the same as the XGB-old method. When the SINR prediction accuracy is less than 0.5dB, that is, the training samples are 5000, the XGB-PF algorithm requires less than 1s to train each user, reaching 0.34s, and each user can achieve sub-second training; while the neural network algorithm requires about 4.26s of training time for each user when the training data set is 5000, which cannot meet the needs well. Although the linear regression (LRA) algorithm has the shortest training time, its SINR prediction performance is too poor.

[0231] Fig.11 A schematic diagram of a simulation showing the time consumption for training to achieve the same performance according to an embodiment of the present disclosure is shown.

[0232] like Fig.11 As shown, the horizontal axis is the average RMSE (dB), and the vertical axis is the average training time (s), XGB-old, NN-MLP, XGB-PF, and LRA.

[0233] As shown in the figure above, the neural network algorithm (NN-MLP) takes two orders of magnitude more time than the XGB-PF algorithm to achieve the same prediction accuracy of 0.5dB; when achieving an SINR prediction accuracy of less than 0.5dB, the training time required for each user of the XGB-PF algorithm is less than 1s, reaching 0.34s; while the neural network algorithm needs about 25.3s to achieve the same accuracy, which is nearly 100 times that of XGB-PF. At this time, the XGB-old method takes about 1.6s, which is about 5 times that of the XGB-PF algorithm; and compared with the linear regression (LRA) algorithm, the XGB-PF algorithm can reduce the average prediction error by one order of magnitude under the same time consumption, and the improvement is also very obvious.

[0234] The wireless performance prediction method provided by the embodiment of the present disclosure has broad application prospects because interference is the core problem in wireless systems. The interference relationship mining and wireless performance identification technology of this patent has the following different application possibilities in the wireless resource management alone: ​​directly used for wireless resource allocation of wireless systems; used to further enhance ICIC / eICIC technology.

[0235] Fig.12 The block diagram of the wireless performance prediction device according to an embodiment of the present disclosure is schematically shown. Fig.12As shown, the wireless performance prediction device 1200 provided by the embodiment of the present disclosure may include a signal-to-noise ratio and signal-to-interference-plus-noise ratio prediction output unit 1210, a signal-to-noise ratio and signal-to-interference-plus-noise ratio parameter calculation unit 1220, a current wireless resource allocation vector acquisition unit 1230, and a current signal-to-interference-plus-noise ratio acquisition unit 1240.

[0236] In the embodiment of the present disclosure, the signal-to-noise ratio and signal-to-interference-plus-noise ratio prediction output unit 1210 may be configured to obtain a predicted signal-to-noise ratio of a target user equipment and a predicted signal-to-interference-plus-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment using a target machine learning model.

[0237] The signal-to-noise ratio and signal-to-interference ratio parameter calculation unit 1220 may be configured to obtain a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user equipment according to the predicted signal-to-noise ratio of the target user equipment and the predicted signal-to-interference-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment.

[0238] The current radio resource allocation vector obtaining unit 1230 may be configured to obtain a current radio resource allocation vector of the target user equipment, where the current radio resource allocation vector corresponds to a current resource block of the target user equipment in a current transmission time interval.

[0239] The current signal to interference plus noise ratio obtaining unit 1240 may be configured to obtain the current signal to interference plus noise ratio of the target user equipment according to the current radio resource allocation vector of the target user equipment, the signal to interference plus noise ratio parameter of the target user equipment and the signal to interference ratio parameter.

[0240] Among them, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and in the current wireless resource allocation vector, the position value corresponding to the target interfering user equipment that shares the current resource block with the target user equipment in the current transmission time interval is a first value, and the position value corresponding to the target interfering user equipment that does not share the current resource block with the target user equipment in the current transmission time interval is a second value.

[0241] In an exemplary embodiment, the signal-to-noise ratio (SNR) and signal-to-interference ratio (SIR) parameter calculation unit 1220 may include: a signal-to-noise ratio (SNR) parameter calculation unit, which may be used to obtain a SNR parameter of the target user equipment according to a predicted SNR of the target user equipment; and a signal-to-interference ratio (SIR) parameter calculation unit, which may be used to obtain a SIR parameter of the target user equipment according to a predicted SNR between the target user equipment and a target interfering user equipment of the target user equipment.

[0242] In an exemplary embodiment, the wireless performance prediction device 1200 may further include: a target historical wireless resource allocation vector acquisition unit, which can be used to obtain the target historical wireless resource allocation vector and its target historical signal to noise ratio of the target user equipment before using the target machine learning model to obtain the predicted signal to noise ratio of the target user equipment and the predicted signal to interference and noise ratio between the target user equipment and the interfering user equipment of the target user equipment; a target machine learning model training unit, which can be used to train the target machine learning model using the target historical wireless resource allocation vector and its target historical signal to noise ratio of the target user equipment.

[0243] In an exemplary embodiment, the target historical wireless resource allocation vector obtaining unit may include: an initial historical wireless resource allocation vector obtaining unit, which may be used to obtain the initial historical wireless resource allocation vector and the initial historical signal to interference noise ratio of the target user equipment, wherein each initial historical wireless resource allocation vector corresponds to each resource block of the target user equipment in each historical transmission time interval, the dimension of each initial historical wireless resource allocation vector is equal to the number of initial interfering user equipment of the target user equipment, and in each initial historical wireless resource allocation vector, the position corresponding to the initial interfering user equipment that shares the corresponding resource block with the target user equipment in the corresponding historical transmission time interval is taken as the first value, and the position corresponding to the initial interfering user equipment that does not share the corresponding resource block with the target user equipment in the corresponding historical transmission time interval is taken as the second value; the initial interfering user equipment The device occurrence count counting unit may be used to count the first value of the corresponding position of the initial historical wireless resource allocation vector of the target user equipment, and obtain the occurrence count of each initial interfering user equipment in the initial historical wireless resource allocation vector of the target user equipment; the target interfering user equipment determination unit may be used to remove the initial interfering user equipment whose occurrence count is less than the occurrence count threshold, and use the remaining initial interfering user equipment as the target interfering user equipment; the target historical wireless resource allocation vector determination unit may be used to remove the initial historical wireless resource allocation vector whose corresponding position of the removed initial interfering user equipment takes the first value, and use the remaining initial historical wireless resource allocation vector as the target historical wireless resource allocation vector, and use the initial historical signal to interference plus noise ratio corresponding to the target historical wireless resource allocation vector as the target historical signal to interference plus noise ratio.

[0244] In an exemplary embodiment, the target machine learning model may be an integrated tree model. The target machine learning model training unit may include: a target historical wireless resource allocation vector input unit, which may be used to input the target historical wireless resource allocation vector into the integrated tree model; a predicted signal to interference and noise ratio output unit, which may be used to process the target historical wireless resource allocation vector through the integrated tree model and output the predicted signal to interference and noise ratio corresponding to the target historical wireless resource allocation vector; and an integrated tree model training unit, which may be used to train the integrated tree model according to the predicted signal to interference and noise ratio corresponding to the target historical wireless resource allocation vector and the target historical signal to interference and noise ratio.

[0245] In an exemplary embodiment, the signal-to-noise ratio and signal-to-interference-plus-noise ratio prediction output unit 1210 may include: an interference-free wireless resource allocation vector acquisition unit, which can be used to obtain the interference-free wireless resource allocation vector of the target user equipment, the dimension of the interference-free wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and each bit of the interference-free wireless resource allocation vector takes the second value; an interference-free wireless resource allocation vector input unit, which can be used to input the interference-free wireless resource allocation vector into the target machine learning model; and a predicted signal-to-noise ratio output unit, which can be used to process the interference-free wireless resource allocation vector through the target machine learning model and output the predicted signal-to-noise ratio of the target user equipment.

[0246] In an exemplary embodiment, the signal-to-noise ratio and signal-to-interference-plus-noise ratio prediction output unit 1210 may include: an interference wireless resource allocation vector obtaining unit, which may be used to respectively obtain the interference wireless resource allocation vectors of each target interfering user equipment of the target user equipment, wherein the dimension of the interference wireless resource allocation vector of each target interfering user equipment is equal to the number of target interfering user equipment of the target user equipment, and in the interference wireless resource allocation vector of each target interfering user equipment, the value of the position corresponding to the target interfering user equipment is the first value, and the value of the remaining position is the second value; an interference wireless resource allocation vector input unit, which may be used to respectively input the interference wireless resource allocation vector of each target interfering user equipment into the target machine learning model; and a predicted signal-to-interference-plus-noise ratio output unit, which may be used for the target machine learning model to respectively process the interference wireless resource allocation vector of each target interfering user equipment, and output the predicted signal-to-interference-plus-noise ratio between the target user equipment and each target interfering user equipment.

[0247] Other contents of the wireless performance prediction device of the embodiment of the present disclosure may refer to the above embodiment.

[0248] It should be noted that although several units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0249] Reference below Fig.13 , which shows a schematic diagram of the structure of an electronic device suitable for implementing an embodiment of the present application. Fig.13 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0250] Reference Fig.13 The electronic device provided by the embodiment of the present disclosure may include: a processor 1301, a communication interface 1302, a memory 1303 and a communication bus 1304.

[0251] The processor 1301 , the communication interface 1302 and the memory 1303 communicate with each other via the communication bus 1304 .

[0252] Optionally, the communication interface 1302 may be an interface of a communication module, such as an interface of a GSM (Global System for Mobile communications) module. The processor 1301 is used to execute a program. The memory 1303 is used to store a program. The program may include a computer program, which includes computer operation instructions. Among them, the program may include: a game client program.

[0253] The processor 1301 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present disclosure.

[0254] The memory 1303 may include a high-speed RAM (random access memory) memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0255] Wherein, the program can be specifically used to: obtain the predicted signal-to-noise ratio of the target user equipment and the predicted signal-to-interference-noise ratio between the target user equipment and the target interfering user equipment of the target user equipment by using the target machine learning model; obtain the signal-to-noise ratio parameter and the signal-to-interference ratio parameter of the target user equipment according to the predicted signal-to-noise ratio of the target user equipment and the predicted signal-to-interference-noise ratio between the target user equipment and the target interfering user equipment of the target user equipment; obtain the current wireless resource allocation vector of the target user equipment, the current wireless resource allocation vector corresponds to the current resource block of the target user equipment in the current transmission time interval; obtain the current signal-to-interference-noise ratio of the target user equipment according to the current wireless resource allocation vector of the target user equipment, the signal-to-noise ratio parameter of the target user equipment and the signal-to-interference ratio parameter. Wherein, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and in the current wireless resource allocation vector, the position corresponding to the target interfering user equipment that shares the current resource block with the target user equipment in the current transmission time interval takes a first value, and the position corresponding to the target interfering user equipment that does not share the current resource block with the target user equipment in the current transmission time interval takes a second value.

[0256] According to one aspect of the present application, a computer program product or a computer program is provided, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in various optional implementations of the above-mentioned embodiments.

[0257] It should be understood that any number of elements in the drawings of the present disclosure is for illustration rather than limitation, and any naming is only for distinction rather than having any limiting meaning.

[0258] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0259] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A wireless performance prediction method, It is characterized in that include: Obtaining a predicted signal-to-noise ratio of a target user equipment and a predicted signal-to-interference-and-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment using a target machine learning model; Obtaining a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user equipment according to a predicted signal-to-noise ratio of the target user equipment and a predicted signal-to-interference-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment, wherein the signal-to-noise ratio parameter is related to a ratio of an average power of a received signal in the target user equipment to an average power of noise; and the signal-to-interference ratio parameter is related to a ratio of energy of a received signal of the target user equipment to interference energy; Obtaining a current radio resource allocation vector of the target user equipment, where the current radio resource allocation vector corresponds to a current resource block of the target user equipment in a current transmission time interval; Obtaining a current signal to interference and noise ratio of the target user equipment according to a current radio resource allocation vector of the target user equipment, a signal to noise ratio parameter of the target user equipment, and the signal to interference ratio parameter; Among them, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and in the current wireless resource allocation vector, the position value corresponding to the target interfering user equipment that shares the current resource block with the target user equipment in the current transmission time interval is a first value, and the position value corresponding to the target interfering user equipment that does not share the current resource block with the target user equipment in the current transmission time interval is a second value.

2. The method according to claim 1, It is characterized in that Obtaining a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user equipment according to a predicted signal-to-noise ratio of the target user equipment and a predicted signal-to-interference-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment, including: Obtaining a signal-to-noise ratio parameter of the target user equipment according to the predicted signal-to-noise ratio of the target user equipment; The signal-to-interference ratio parameter of the target user equipment is obtained according to the signal-to-noise ratio parameter of the target user equipment and the predicted signal-to-interference-and-noise ratio between the target user equipment and the target interfering user equipment of the target user equipment.

3. The method according to claim 1, It is characterized in that Before using the target machine learning model to obtain a predicted signal-to-noise ratio of a target user equipment and a predicted signal-to-interference-and-noise ratio between the target user equipment and an interfering user equipment of the target user equipment, the method further includes: Obtaining a target historical radio resource allocation vector and a target historical signal to interference and noise ratio of the target user equipment; The target machine learning model is trained using the target historical wireless resource allocation vector and the target historical signal to interference and noise ratio of the target user equipment.

4. The method according to claim 3, It is characterized in that Obtaining a target historical radio resource allocation vector and a target historical signal to interference and noise ratio of the target user equipment, including: Obtaining an initial historical wireless resource allocation vector and an initial historical signal to interference and noise ratio of the target user equipment, wherein each initial historical wireless resource allocation vector corresponds to each resource block of the target user equipment in each historical transmission time interval, the dimension of each initial historical wireless resource allocation vector is equal to the number of initial interfering user equipment of the target user equipment, and in each initial historical wireless resource allocation vector, a position value corresponding to an initial interfering user equipment that shares a corresponding resource block with the target user equipment in a corresponding historical transmission time interval is taken as the first value, and a position value corresponding to an initial interfering user equipment that does not share a corresponding resource block with the target user equipment in a corresponding historical transmission time interval is taken as the second value; Counting the first value of the corresponding position of the initial historical wireless resource allocation vector of the target user equipment to obtain the number of occurrences of each initial interfering user equipment in the initial historical wireless resource allocation vector of the target user equipment; Eliminate initial interfering user equipment whose occurrence times are less than the occurrence times threshold, and use the remaining initial interfering user equipment as the target interfering user equipment; The initial historical wireless resource allocation vectors whose corresponding positions of the eliminated initial interfering user equipment have the first value are eliminated, and the remaining initial historical wireless resource allocation vectors are used as the target historical wireless resource allocation vectors, and the initial historical signal to interference plus noise ratio corresponding to the target historical wireless resource allocation vectors are used as the target historical signal to interference plus noise ratio.

5. The method according to claim 3, It is characterized in that The target machine learning model is an integrated tree model; wherein the target machine learning model is trained using the target historical wireless resource allocation vector and the target historical signal to interference and noise ratio of the target user equipment, including: Inputting the target historical wireless resource allocation vector into the integrated tree model; Processing the target historical wireless resource allocation vector through the integrated tree model, and outputting a predicted signal to interference and noise ratio corresponding to the target historical wireless resource allocation vector; The integrated tree model is trained according to the predicted signal to interference plus noise ratio corresponding to the target historical wireless resource allocation vector and the target historical signal to interference plus noise ratio.

6. The method according to claim 1, It is characterized in that Use the target machine learning model to obtain the predicted signal-to-noise ratio of the target user device, including: Obtaining a non-interference wireless resource allocation vector of the target user equipment, where the dimension of the non-interference wireless resource allocation vector is equal to the number of target interfering user equipments of the target user equipment, and each bit of the non-interference wireless resource allocation vector is the second value; Inputting the interference-free wireless resource allocation vector into the target machine learning model; The interference-free wireless resource allocation vector is processed by the target machine learning model to output a predicted signal-to-noise ratio of the target user equipment.

7. The method according to claim 6, It is characterized in that Obtaining a predicted signal-to-interference-plus-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment by using a target machine learning model, including: Obtaining interference radio resource allocation vectors of target interfering user equipment of the target user equipment respectively, wherein the dimension of the interference radio resource allocation vector of each target interfering user equipment is equal to the number of target interfering user equipment of the target user equipment, and in the interference radio resource allocation vector of each target interfering user equipment, the value of the position corresponding to the target interfering user equipment is the first value, and the value of the remaining position is the second value; Inputting the interference wireless resource allocation vector of each target interfering user equipment into the target machine learning model respectively; The target machine learning model processes the interference wireless resource allocation vector of each target interfering user equipment respectively, and outputs the predicted signal to interference and noise ratio between the target user equipment and each target interfering user equipment.

8. A wireless performance prediction device, It is characterized in that include: A signal-to-noise ratio and signal-to-interference-plus-noise ratio prediction output unit, configured to obtain a predicted signal-to-noise ratio of a target user equipment and a predicted signal-to-interference-plus-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment by using a target machine learning model; a signal-to-noise ratio and a signal-to-interference ratio parameter calculation unit, configured to obtain a signal-to-noise ratio parameter and a signal-to-interference ratio parameter of the target user equipment according to a predicted signal-to-noise ratio of the target user equipment and a predicted signal-to-interference-noise ratio between the target user equipment and a target interfering user equipment of the target user equipment, wherein the signal-to-noise ratio parameter is related to a ratio of an average power of a received signal in the target user equipment to an average power of noise; and the signal-to-interference ratio parameter is related to a ratio of energy of a received signal of the target user equipment to interference energy; A current wireless resource allocation vector obtaining unit, configured to obtain a current wireless resource allocation vector of the target user equipment, wherein the current wireless resource allocation vector corresponds to a current resource block of the target user equipment in a current transmission time interval; a current signal to interference and noise ratio obtaining unit, configured to obtain a current signal to interference and noise ratio of the target user equipment according to a current radio resource allocation vector of the target user equipment, a signal to noise ratio parameter of the target user equipment and the signal to interference ratio parameter; Among them, the dimension of the current wireless resource allocation vector is equal to the number of target interfering user equipment of the target user equipment, and in the current wireless resource allocation vector, the position value corresponding to the target interfering user equipment that shares the current resource block with the target user equipment in the current transmission time interval is a first value, and the position value corresponding to the target interfering user equipment that does not share the current resource block with the target user equipment in the current transmission time interval is a second value.

9. A computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method according to any one of claims 1 to 7.

10. An electronic device, include: at least one processor; A storage device configured to store at least one program, when the at least one program is executed by the at least one processor, enables the at least one processor to implement the method according to any one of claims 1 to 7.

11. A computer program product, comprising computer instructions, and executing the computer instructions implements the method according to any one of claims 1 to 7.

Images