Information processing apparatus, information processing method, and program

Abstract

This information processing apparatus comprises: an extraction unit that extracts, from a history of quality information pertaining to communication performed by a device, a feature value for each of management regions including a spatial region; and a management database that stores the feature value for each of the management regions.

Description

Information processing device, information processing method, and program

[0001] The present invention relates to a technique for estimating communication quality.

[0002] A conventional technique for estimating communication quality is disclosed in Non-Patent Document 1. In the technique disclosed in Non-Patent Document 1, a quality history is created based on the results of a connected car's use of a network (NW), the quality history is stored in a database (DB), and the quality is estimated by referring to the quality history in the DB in response to a request from an external function.

[0003] Study on quality estimation method based on network usage record of connected cars, B-6-18, IEICE Society Conference 2023

[0004] However, in the above-described conventional technology, the amount of quality history stored in the DB increases over time. Since there is a limit to the amount of data that can be stored in the DB, it is necessary to reduce the amount of data stored in the DB.

[0005] The present invention has been made in view of the above points, and has an object to provide a technique for reducing the amount of data stored in a database in a technique for estimating quality based on communication quality history.

[0006] According to the disclosed technology, an information processing device is provided that includes: an extraction unit that extracts features for each management category including a spatial category from a history of quality information related to communication performed by a device; and a management database that stores the features for each management category.

[0007] The disclosed technology provides a technology for reducing the amount of data stored in a database in a technology for estimating quality based on communication quality history.

[0008] 1 is a diagram for explaining the problem. FIG. 1 is a diagram showing an overview of a system configuration in an embodiment of the present invention. FIG. 2 is a diagram showing a configuration of an information processing device 100 in an embodiment of the present invention. FIG. 3 is a diagram for explaining a management method in the management phase. FIG. 4 is a diagram for explaining a management method in the management phase. FIG. 5 is a diagram showing an estimation method in a conventional method. FIG. 6 is a diagram showing an estimation method in a proposed method. FIG. 7 is a diagram showing estimation results. FIG. 8 is a diagram for explaining an effect of avoiding an increase in the amount of management information. FIG. 9 is a diagram for explaining an effect of avoiding an increase in the amount of management information. FIG. 10 is a diagram for explaining an example of management classification and management information items. FIG. 11 is a diagram for explaining an example of the contents of management classification and management information items. FIG. 12 is a diagram for explaining an example of the contents of management classification and management information. FIG. 13 is a diagram for explaining a processing procedure for quality estimation. FIG. 14 is a diagram showing operation with a single QIM-DB. FIG. 15 is a diagram showing operation with multiple QIM-DBs. FIG. 16 is a diagram showing an example of a case where a QIM-DB has a multi-layer structure. FIG. 17 is a diagram showing another example of the configuration of the information processing device 100. FIG. 18 is a diagram showing an example of the hardware configuration of the information processing device 100.

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.

[0010] Hereinafter, first, the problems with the technology according to the present embodiment will be described in more detail with reference to the drawings, and then the technology according to the present embodiment will be described in more detail. Note that while the technology disclosed in Non-Patent Document 1 is publicly known, the explanation of the following problems is not publicly known.

[0011] (Regarding the Issues) In the conventional technology disclosed in Non-Patent Document 1 and the like, a quality history is created from various pieces of information collected in real time, and the quality history is stored and managed in a database that is referenced when estimating quality. In this configuration, information is added to the database in real time. Furthermore, in response to a request from an external function, the quality is estimated by referencing the quality history in the database and a response is made to the external function. For example, the quality is estimated based on time, location, network type, etc.

[0012] The above-mentioned conventional technology will be described in more detail with reference to Fig. 1. In the example system shown in Fig. 1, a quality information acquisition function 4 acquires network quality information based on the usage record of a network by a device (e.g., a connected car), and stores the network quality information by time in DB 1. Furthermore, a time / location information acquisition function 5 acquires time / location information of a device using the network, and stores the time / location information of the device in DB 2.

[0013] In addition, the other related information acquisition function 6 acquires information such as the network operator and quality class used by the device for information transmission, and stores information such as the network operator and QoS (Quality of Service) control level in DB3.

[0014] Meanwhile, a quality history creation function 7 refers to the information stored in each DB, creates a quality history, and stores it in the quality DB 8. Examples of the quality history include network quality (throughput / packet loss rate / delay / jitter) by time / location / network operator / QoS control level, etc.

[0015] When the external function 10 transmits a quality estimation request to the quality estimation function 9, the quality estimation function 9 performs quality estimation by referring to the quality history in the quality DB 8, and returns the estimation result to the external function 10.

[0016] Continuous information management is necessary to achieve highly accurate quality estimation. In other words, since the quality of the estimation results depends on the management information referenced, it is necessary to improve the quality of the information that serves as the basis for estimation, such as quality history, which enables highly accurate estimation. However, quality history continues to grow on the spatiotemporal axis, and there is a limit to the amount of information that can be managed in a database, so it is necessary to carefully select and discard management information.

[0017] That is, it is necessary to improve / maintain the quality of information, which is directly linked to the accuracy of the quality estimation results, while keeping the total amount of information that is successively added / updated to the DB below a certain level.

[0018] In this embodiment, the information processing device 100 extracts features included in the quality history and manages a predefined set of features / parameters (quality indexes) as information that serves as the basis for quality estimation. In addition, in order to reduce the amount of information, the quality history itself is not retained for a long period of time.

[0019] When estimating quality based on requirements from external functions, etc., management information such as quality indicators is referenced, and quality is estimated based on logic appropriate to the situation and application.

[0020] An overview of the system configuration in this embodiment is shown in Figure 2. The method of this system may also be called QIM (Quality Index Map). As shown in Figure 2, the information processing device 100 has a quality index management unit 110, a QIM-DB 120, and a quality estimation unit 210. In the example of Figure 2, external to the information processing device 100 are shown a cooperative control unit 300, external systems such as a 4D digital platform / CPS, a remote monitoring and control system, and the like.

[0021] 2, information (video, time / location information, etc.) from the connected car is transmitted to the remote monitoring control system, whereby remote monitoring is performed. In parallel with this, the quality index management unit 110 collects quality histories, extracts feature quantities contained in the quality histories, and stores the feature quantities in the QIM-DB 120.

[0022] The information processing device 100 cooperates with an external system, and for example, based on a request from the external system, the quality estimation unit 210 performs quality estimation using information acquired from the QIM-DB 120 and notifies the cooperative control unit 300 of the estimation result. Based on this, the cooperative control unit 300 can issue control instructions to the connected car.

[0023] As described above, by using features, it becomes unnecessary to manage the quality history itself over a long period of time, which is created and added sequentially. Therefore, it is possible to avoid an increase in the amount of information to be managed on the time-space axis.

[0024] Furthermore, by making estimations based on management information that reflects quality changes on the time-space axis and logic according to the situation and application, it is possible to maintain accuracy that is equal to or greater than that of estimations based only on quality history.

[0025] (Configuration and Operation of Information Processing Device 100) Next, a more detailed configuration and operation of the information processing device 100 will be described with reference to Fig. 3. As shown in Fig. 3, the information processing device 100 has a quality index management unit 110, a QIM-DB 120, a management information monitoring unit 130, a quality history distribution unit 140, an estimation determination DB 150, a quality estimation method determination unit 160, and a quality estimation unit 210.

[0026] The "quality index management unit 110, QIM-DB 120, management information monitoring unit 130, quality history distribution unit 140, estimation judgment DB 150, and quality estimation method judgment unit 160" are used in the management phase. The "estimation judgment DB 150 and quality estimation unit 210" are used in the estimation phase.

[0027] Note that the device used only in the management phase and the device used only in the estimation phase may be separate devices. "Device used only in the management phase," "Device used only in the estimation phase," and "Both devices used only in the management phase and the estimation phase" may all be referred to as information processing device 100. Furthermore, the quality index management unit 110 and the QIM-DB 120 may be referred to as an extraction unit and a management database, respectively. Furthermore, the QIM-DB 120 may include an estimation determination DB 150.

[0028] The quality history distribution unit 140 receives quality history (e.g., quality information by time / location / network type) transmitted from a cooperative control unit (e.g., cooperative control gateway). The quality history distribution unit 140 can also receive a request for quality history from an external function and transmit the quality history to the external function.

[0029] The quality index management unit 110 acquires the quality history from the quality history distribution unit 140 and calculates (usually recalculates) quality indexes such as feature quantities based on the quality history by referring to the management information in the QIM-DB 120. If there is no existing management information, a management category is set and a new quality index is created.

[0030] Furthermore, the quality index management unit 110 sets the same management classification as the QIM-DB 120 and stores / updates management information for estimation method determination in the estimation determination DB 150. The management information monitoring unit 130 references the management information for each DB and deletes the management information as appropriate based on the set rules. Note that deletion of management information (quality history, etc.) in a DB may be performed by the DB itself, or may be performed by a functional unit other than the management information monitoring unit 130.

[0031] The quality estimation method determination unit 160 performs quality estimation using multiple quality estimation methods (logic, parameters, etc.) based on the management information used in quality estimation by the quality estimation unit 210, compares and verifies the actual quality history with the quality estimation results, and then determines the quality estimation method (logic, parameters, etc.) for each management category. The quality estimation method determination unit 160 stores the quality estimation methods determined for each management category in the estimation determination DB 150.

[0032] The quality estimation unit 210 estimates quality using management information such as feature quantities stored in the QIM-DB 120, based on the method (logic, parameters, etc.) determined by the quality estimation method determination unit 160. Specifically, the quality estimation unit 210 estimates quality by referring to the management information in the QIM-DB 120 and referring to the quality estimation method determined for each management category in the estimation determination DB 150. The quality estimation unit 210 returns a quality estimation result to the external function based on a request from the external function.

[0033] As described above, the information processing device 100 according to this embodiment manages quality indicators (feature amounts, etc.) by time / location / network type (and also temporarily manages quality history as needed) in the QIM-DB 120 based on quality history related to the network distributed from the cooperative control unit. Furthermore, the information processing device 100 provides the quality history related to the network distributed from the cooperative control unit to an external function as needed.

[0034] Furthermore, in response to a quality estimation request from an external function, the information processing apparatus 100 estimates quality based on the information it manages and provides the estimation result to the external function.

[0035] Furthermore, the information processing device 100 can compare and verify its own estimation results with quality history related to the network distributed from the cooperative control unit or the like, and control the quality estimation method based on a closed loop.

[0036] (Details of the Management Phase) Next, the management phase will be described in detail. In this embodiment, the main object of management is not the quality history itself but the feature amount. This point will be described using an example.

[0037] In the conventional method of managing information using a quality DB, the quality history itself, which is quality information by time, space, and medium, is managed, and the quality history is added sequentially. The image of a quality DB in the conventional method is as shown in Figures 4(a) and 5(a). In other words, in the conventional method, management information corresponding to each management category is managed as quality history for each management category. As a result, the amount of information managed increases monotonically with the passage of time.

[0038] In contrast, in the information management of the proposed method (QIM), the features extracted from the quality history itself are managed by classification based on spatiotemporal information and media information, and are updated sequentially.

[0039] The proposed method (QIM) is conceptually shown in Figures 4(b) and 5(b). In this example, region A is divided into four regions (regional meshes), and feature values ​​for each region are calculated from quality history data. These feature values ​​are then stored as management information for each regional mesh. Note that there is no need to store management information for regions for which quality history has not been acquired, such as region 2 shown in Figure 4(b). In other words, as soon as quality history is acquired, feature values ​​for that region can be added and managed.

[0040] With the proposed method, features are updated each time quality history is acquired, so the number of data rows does not increase beyond the number of management categories. In other words, the amount of information to be managed can be kept below the upper limit of the number of management categories. This solves the problems with the conventional technology.

[0041] Any method may be used to divide an area into multiple regions. For example, the division method used in the regional mesh statistics disclosed on the website of the Statistics Bureau of the Ministry of Internal Affairs and Communications may be used.

[0042] (Comparison of quality estimation results with conventional method) A comparison example of quality estimation results between the conventional method and the proposed method will be described using Figs. 6 to 8. Fig. 6 is a diagram showing estimation using the conventional method. In the conventional method shown in Fig. 6, the Z% (10% in this example) tile value is calculated from the quality history of the reference object itself. In the proposed method shown in Fig. 7, the Z% (10% in this case) tile value is calculated from statistics (e.g., normal distribution) by regional mesh / division that are updated each time quality history is acquired.

[0043] In a verification example at a certain location at a certain time, we confirmed that the proposed method (using estimation based on a normal distribution) can estimate the quality of the network as well as the conventional method. Specifically, as shown in Figure 8, the proposed method can obtain good estimation results.

[0044] (Specific example of the effect of avoiding an increase in the amount of management information) Next, a specific example of the effect of avoiding an increase in the amount of management information will be described with reference to Figs. 9 and 10. In this example, an area around an actual station with sides of 2 km is targeted for management, and a regional mesh with sides of 50 m is used. In this case, the number of regional meshes is 1,600. It is assumed that vehicles travel within the managed area and quality information is acquired from the vehicles.

[0045] FIG. 9(a) shows the items of information managed in the proposed method, and FIG. 9(b) shows the items of information managed in the conventional method.

[0046] In the example shown in Figure 9(a), the amount of information stored / managed by the proposed method is 1600 x 24 x 3 x 2 = 230,400 (rows) (fixed). In reality, instead of having a regional mesh / division that covers the entire area from the beginning, a "row" is created (and subsequently updated as needed) as soon as the quality history / quality index of that regional mesh / division is acquired. Therefore, when targeting vehicles that travel along a set route, such as scheduled buses, the amount of information stored / managed by the proposed method is expected to be even less.

[0047] On the other hand, in the conventional method, the amount of information stored / managed in the quality DB increases with continued operation, and the rate of increase depends on the frequency of creating quality history.

[0048] Figure 10 shows the relationship between elapsed time and the amount of information (in rows) for the proposed method and the conventional method. In Figure 10, the conventional method shows the cases of one vehicle and 20 vehicles. Note that in the proposed method (the present invention), the amount of information does not depend on the number of vehicles.

[0049] As shown in Figure 10, in both the one-vehicle and 20-vehicle environments, the conventional method initially had less information than the proposed method, but in the one-vehicle environment, the amount of information was reversed after about three days, and in the 20-vehicle environment, the amount of information was reversed in just under four hours.

[0050] Below, more detailed examples of information stored in the DB in the management phase, a method for determining the quality estimation method, and a quality estimation method will be described.

[0051] (Management Phase: Management Classification and Management Information in QIM-DB 120) The management classification and management information in the QIM-DB 210 will be described in more detail below. First, an example of the management classification and management information will be described with reference to FIG.

[0052] As shown in Fig. 11, the management classification includes a spatial classification, a time classification, and a media classification. The spatial classification is a classification based on a position in physical space (defined by latitude, longitude, altitude, etc.). The time classification is a classification based on time (UTC, etc.). The media classification is a classification based on the target of the quality index (network, application, type and level of service, etc.).

[0053] The management information includes quality indexes, total number of quality histories, update records, and quality histories. The quality index is a feature (which can also be called a parameter) that expresses the quality of the corresponding section. The total number of quality histories is the number of quality histories created in the corresponding section. The update record is the time when the management information for the corresponding section (quality index, total number of quality histories) was updated. The quality history is the most recent quality history created in the corresponding section.

[0054] A specific example of the management classification and management information will be described with reference to Fig. 12. In the example shown in Fig. 12, the spatial classification is a regional mesh code. The regional mesh code is an identifier based on the concept of regional mesh defined by the Statistics Bureau of the Ministry of Internal Affairs and Communications. As an example, a mesh (e.g., a mesh with a side length of 50 m) obtained by subdividing a standard regional mesh with a side length of 1 km is defined, and the mesh code is used as the identifier.

[0055] The time segments are segments based on one-hour units of UTC (from midnight to 11:00 p.m.), and are assumed to include 25 segments, including all hours, for example.

[0056] The media classification includes a network type and a QoS control level. In this example, the network type is the name of the network provider, and the QoS control level is QCI.

[0057] Regarding the control categories, it is assumed that all control categories will not be prepared in the DB from the beginning, but that the relevant categories will be added one by one according to the created quality history. This is because there is no need to store in the DB categories related to locations where quality history cannot be created, such as no-entry / prohibited areas, or categories related to time.

[0058] In this example, the quality indexes are the average values ​​and standard deviations of the throughput, packet loss rate, communication delay, and jitter. Average values ​​include A (arithmetic mean), B (weighted mean), C (geometric mean), and D (harmonic mean). Regarding the packet loss rate, it is assumed that the total number of quality histories with a packet loss rate > 0 is also included as a quality index.

[0059] In addition, it is assumed that the degree of weighting at the time of recalculation in B and the number by which the numerical value of the quality history in D is divided can be changed by setting, and multiple types of average values ​​can be managed as necessary.

[0060] The value (feature amount) stored as the quality index is a statistical value required to calculate a desired percentile value of a desired distribution during quality estimation.

[0061] The update record is the Unix time when the managed feature (statistical value) was updated. Regarding the quality history, the quality history itself is also retained within a certain period / within a specified number, and the number of retained records is variable by setting, for example, a maximum of about 10.

[0062] As an example of the management classification, a management information sheet (≒ mesh) may exist for each location / time period / network type / QoS control level. For example, even if the location is the same, if the time period or network type is different, they may be managed on different sheets and treated as different meshes.

[0063] Furthermore, multiple levels of regional mesh subdivision may be used. For example, it may be possible to imagine that there are both management information sheets corresponding to location divisions (regional meshes) with a side length of 100 meters and management information sheets corresponding to location divisions (regional meshes) with a side length of 10 meters. Depending on the implementation, information may be managed in the same database (identified by the number of digits, etc.), or databases may be separated by level.

[0064] (Management Phase: Management Classification and Management Information in Estimation Determination DB 150) Next, the management classification and management information in the estimation determination DB 150 in the management phase will be described.

[0065] As shown in Fig. 13, the management classification is the same as that of the QIM-DB 210. The quality index in the management information is a parameter (feature) that expresses the quality of the corresponding classification. The total number of quality histories and update records are the same as those of the QIM-DB 210. The quality history is the quality history created in the corresponding classification.

[0066] The management information of the estimation determination DB 150 includes a quality estimation method determination result. The quality estimation method determination result includes the determined quality estimation method. The quality estimation method determination method will be described later.

[0067] A specific example of the management classification and management information will be described with reference to Fig. 14. The area mesh code, time classification, NW type, and QoS control level are the same as those in the QIM-DB 120.

[0068] Regarding quality indices, in addition to the quality indices held by QIM-DB120, it also holds "multiple types of weighted averages with varying degrees of weighting" and "multiple types of harmonic averages with varying magnitudes of the numbers divided by the numerical values ​​of the quality history."

[0069] The total number of quality histories and update records are the same as those in the QIM-DB 120. Quality histories are sequentially deleted after processing required to determine the quality estimation method, such as comparative verification with quality estimation results and distribution generation. Regarding the quality estimation method, logic and parameters are determined according to the comparative verification results.

[0070] (Quality estimation method determination flow) In this embodiment, the quality estimation method determination unit 160 determines the quality estimation method according to the following procedure. The following flow is executed as needed in the closed loop described above, and the quality estimation method for each management category is updated as time passes.

[0071] S1 (Step 1): In S1, the quality estimation method determination unit 160 performs quality estimation in response to a quality estimation request transferred from the quality estimation unit 210. The logic and types of parameters used at this time are, for example, those set in the configuration.

[0072] S2: In S2, the quality estimation method determination unit 160 refers to the quality history for the time closest to the time included in the quality estimation request (referring to the quality history (correct answer) linked to that time based on the management category), and compares that quality history with the quality estimation result. However, the time difference must be less than t (e.g., 1) seconds. If there is no quality history for less than t seconds, no comparison is made. This is because there may be a situation where an estimation and its corresponding result do not exist in the same management category, for example, due to a large error in the prediction of the future position.

[0073] S3: In S3, the quality estimation method determining unit 160 determines the logic and parameters to be used for estimation based on the result of the comparison between the quality history and the quality estimation result, and stores the determination result in the estimation determination DB 150.

[0074] (Specific examples of quality estimation method determination methods) Specific examples of the quality estimation method determination methods explained in the above flow will be explained. Below, determination methods 1 to 4 will be explained as examples. The determination method to be used can be arbitrarily set (added / changed) depending on the purpose of quality estimation. For example, an implementation may be made in which determination method 1 is used as a base, but determination method 4 can also be used.

[0075] <Determination Method 1> The procedure for determination method 1 is as follows.

[0076] S1: An arbitrary threshold value x is set for each numerical value such as throughput included in the quality history, and a range above x and a range below x are defined.

[0077] S2: Determine to which region each of the quality history and quality estimation results to be compared belongs.

[0078] S3: Accuracy is calculated based on whether the quality history to be compared and the area category to which the quality estimation result belongs match, and Recall is calculated based on whether the quality history was correctly assigned to a certain area category.

[0079] S4: Accuracy (A) is A １ % or more, and Recall (R) is R １ % or more, the quality estimation method that maximizes A+R is determined.

[0080] S5: If there is no quality estimation method that corresponds to S4 above, determine the quality estimation method that maximizes R.

[0081] <Determination Method 2> The procedure for determination method 2 is as follows.

[0082] S1: Accuracy and Recall are calculated in the same manner as in Determination Method 1.

[0083] S2: A is A １ % or more and R is R １ % or more, the quality estimation method with the maximum A is determined.

[0084] <Determination Method 3> The procedure for determination method 3 is as follows.

[0085] S1: Accuracy and Recall are calculated in the same manner as in Determination Method 1.

[0086] S2: A is A １ % or more and R is R １ % or more, the quality estimation method that maximizes A+R is determined.

[0087] <Determination Method 4> The procedure for determination method 4 is as follows.

[0088] S1: Calculate the difference between each numerical value of the quality estimation result to be compared and each numerical value of the throughput etc. included in the quality history to be compared.

[0089] S2: Determine the quality estimation method that minimizes the sum of the absolute values ​​of the differences.

[0090] (Estimation Phase: Quality Estimation Flow Based on Feature Amounts / Parameter Groups) Next, the flow of quality estimation based on feature amounts / parameter groups in the estimation phase will be described with reference to FIG.

[0091] In step S1, the quality estimation unit 210 receives a quality estimation request from an external function. In step S2, the quality estimation unit 210 determines the management category to be referenced from the information included in the request.

[0092] In S3, the quality estimation unit 210 acquires management information from the QIM-DB 120 based on the determined management classification, and acquires the quality estimation method to be used from the estimation determination DB 150, and estimates the quality using these.

[0093] In S4, the quality estimation unit 210 returns the quality estimation result to the external function. Note that the number of estimation results may be multiple.

[0094] <Specific Example of Quality Estimation Flow> A more specific quality estimation flow is as follows.

[0095] In S1, the external function requests the quality estimation unit 210 to estimate quality based on "time (UTC) / location (latitude and longitude) / network type (network operator name) / QoS control level (QCI)". The quality estimation unit 210 receives the request.

[0096] In S2-1, the quality estimation unit 210 calculates a regional mesh code from the "position" included in the request. Note that multiple meshes may be selected by reducing the number of digits in the code.

[0097] In S2-2, the quality estimation unit 210 determines the corresponding time segment from the "time" included in the request. Note that multiple time segments may be selected.

[0098] In S2-3, the quality estimation unit 210 determines the management category to be referenced based on the "network type / QoS control level" included in the request and the calculation / determination results of S2-1 and S2-2.

[0099] In S3, the quality estimation unit 210 acquires management information and a quality estimation method to be used based on the determined management classification, and estimates the quality.

[0100] As an example, the quality estimation unit 210 calculates the 10th percentile value from the average value and standard deviation of the throughput performance values, assuming a normal distribution, and sets this as the throughput estimate value.

[0101] In S4, the quality estimation unit 210 returns the quality estimation result to the external function. Note that there may be multiple estimation results.

[0102] (Estimation Phase: Quality Estimation Method) A more specific example of the quality estimation method in S3 above will be described. In principle, quality estimation is carried out in S1 and S2 below.

[0103] S1: If there is a quality history immediately before (within b seconds before the time of the estimation target) in the management category of the estimation target, an estimate is calculated based on that quality history.

[0104] S2: If there is no quality history, an estimate is calculated based on the quality index and a continuous probability distribution. In this embodiment, the quality estimation method is assumed to be a combination of logic, distribution type, and parameters. In other words, the combination is selected based on the judgment of the quality estimation method.

[0105] Examples of the above S1 and S2 will be explained below.

[0106] <Example 1 of S1> Z of quality history １ (Example: 25) Calculate the percentile value.

[0107] <Example 2 of S1 (throughput estimation only)> Extract only quality history data with a packet loss rate greater than L (e.g., 0), and １ (Example: 10) Calculate the percentile value.

[0108] If there is no extracted quality history, the quality history Z １ (Example: 50) Calculate the percentile value.

[0109] <Example 3 of S1> Not performed (only step 2 is performed).

[0110] <Example 1 of S2> Z calculated based on quality index and continuous probability distribution ２ (Example: 10) Calculate the percentile value.

[0111] <Example 2 of S2 (Packet Loss Occurrence Estimation Only)> The probability of packet loss occurrence is calculated from the "total number of quality histories with packet loss rate > 0" and the "total number of quality histories."

[0112] <Other Example 1 (Example 1 that deviates from the rule)> Both Example 1 of S1 and Example 1 of S2 are carried out, and the larger estimated value is adopted.

[0113] <Other Example 2 (Example 2 outside the rules)> Both Example 1 of S1 and Example 1 of S2 are carried out, and the smaller estimated value is adopted.

[0114] <Other Example 3 (Example 3 outside the rules)> Both Example 1 of S1 and Example 1 of S2 are implemented, and both estimated values ​​are adopted. The response format is determined depending on the source of the quality estimation request.

[0115] (Examples of Distribution Types) Examples of probability distributions used in quality estimation will be described below, along with their respective characteristics and assumptions.

[0116] <Example 1 of distribution type: normal distribution> The average value is the mode, and the probability of occurrence of values ​​with large differences from the average value is low. The degree of reduction in occurrence probability is constant regardless of the direction of increase / decrease.

[0117] <Example 2 of Distribution Type: t Distribution> When the number of quality histories is small, this is more likely to be effective than the normal distribution.

[0118] <Distribution type example 3: Gamma distribution> Gamma distribution corresponds to the probability of the time it takes for a phone to ring k times. It assumes events that occur discretely, but by adjusting the shape and scale parameters, it is possible to generate continuous data distributions with various biases. Therefore, the ability to generate distributions based on network usage records and make estimations based on those distributions is thought to be compatible with the concept of this proposed method.

[0119] <Distribution type example 4: Exponential distribution> Exponential distribution corresponds to the probability of the time until the next phone ring, and assumes that an event occurs discretely. Note that exponential distribution is an example of gamma distribution, with a specific shape and one parameter. For this reason, it is thought to have an affinity with estimating quality degradation or good quality.

[0120] <Distribution type example 5: Chi-squared distribution> The chi-squared distribution corresponds to the probability when there is a bias in the probability of occurrence of discrete events such as coins or dice. It is intended for statistical hypothesis testing rather than direct parameter estimation. Since throughput based on network usage records is biased around the set transmission rate, it is thought to have an affinity with the chi-squared distribution.

[0121] <Distribution Type Example 6: Weibull Distribution> The Weibull distribution corresponds to modeling the probability of the time until a phone breaks down, its lifespan, and its failure rate. The Weibull distribution generates a distribution based on shape parameters and scale parameters. The Weibull distribution is thought to be compatible with estimating quality deterioration or good quality.

[0122] <Distribution Type Example 7: Beta Distribution> The beta distribution corresponds to the probability when the occurrence probability of each event is estimated from the results when there is a possibility that there is a bias in the occurrence probability of discrete events such as coins or dice. The beta distribution is considered to have affinity as an estimation method based on the network usage record.

[0123] <Distribution Type Example 8: Dirichlet Distribution> The Dirichlet distribution is considered to have affinity as an estimation method based on the NW usage record.

[0124] <Distribution Type Example 9: Lognormal Distribution> The lognormal distribution corresponds to the probability of how often a phone will ring during a certain period of time. The lognormal distribution is thought to be compatible with the concept of quality degradation detection.

[0125] The quality estimation methods described above are based on the premise that the average value and standard deviation of the quality history are stored as quality indicators, but the statistics to be stored are not limited to these.

[0126] (Management Phase: Flow of Operating Quality Estimation Based on Gamma Distribution) Here, as an example, a flow of operating quality estimation based on gamma distribution will be described below.

[0127] <Preliminary Measurement> In S1, the quality index management unit 110 collects quality history in a target area for quality estimation, and calculates each quality index based on the management classification.

[0128] In S2, the quality index management unit 110 determines the shape parameters and scale parameters of the gamma distribution from the collected quality history for each management category using the maximum likelihood estimation method.

[0129] <During Whole System Operation> In S1, in real-time quality estimation method determination, the quality estimation method determination unit 160 fixes the shape parameter and the scale parameter and then optimizes Z. Note that Z is Z of the Z % tile value.

[0130] In S2, when x or more quality histories have been accumulated for each management category, the quality estimation method determination unit 160 again determines the shape parameters and scale parameters of the gamma distribution based on those quality histories using the maximum likelihood estimation method, and reflects these in the quality estimation method.

[0131] <At the time of analysis and evaluation after demonstration> In S1, the quality estimation method determination unit 160 determines the shape parameters and scale parameters of the gamma distribution from the quality history accumulated in each management category by the maximum likelihood estimation method.

[0132] In S2, the quality estimation method determination unit 160 compares and verifies the correct data with the estimation result, and performs optimization of Z.

[0133] Regarding the maximum likelihood estimation method performed in multiple steps as described above, it is theoretically possible to calculate the shape parameters and scale parameters from the mean and standard deviation, but it is necessary to consider the resources required to solve the nonlinear equations that include time and the stability of the results.

[0134] Furthermore, in order to avoid the generation of a distribution that is far from the actual situation when the number of quality histories is small, and to avoid an increase in the time required for parameter determination, a parameter determination method such as setting upper and lower limits for each parameter or selecting the combination of parameters with the highest fitting rate from among predefined candidate combinations of parameters may be adopted.

[0135] (Mechanism for Preventing Management Information from Becoming Stale) Next, an embodiment of a mechanism for preventing management information from becoming stale will be described with reference to FIGS. 16 and 17. FIG.

[0136] There is a need for diverse information management that responds to quality changes over time. Therefore, in this embodiment, the QIM-DB 120 is multi-layered, and information is managed so that features can be retained and used for quality estimation in a manner that is not affected by past performance. This makes it possible to estimate quality according to the use cases and requirements to which the proposed method (QIM) is applied. A more specific example will be described with reference to FIGS. 16 and 17.

[0137] In FIG. 16, a single QIM-DB is used, and the quality estimation unit 210 performs quality estimation by referring to the QIM-DB 120 .

[0138] However, a system that simply updates features each time quality history is acquired leaves the influence of past performance data, such as the quality history at the time information acquisition begins, unaffected. For example, when spatial changes occur, such as the construction or demolition of buildings that could affect wireless network quality, if the influence of information acquired before the changes remains in the features, there is a high possibility that the quality estimation accuracy will decrease (the features will not accurately represent the state). Therefore, in this embodiment, as shown in FIG. 17 , operation is performed using multiple QIM-DBs. Note that the multiple DBs, such as QIM-DB 120 and QIM-DB 121 described below, may be physically multiple DBs (e.g., multiple DB servers), or may physically mean that multiple data groups (e.g., the previous day, the current day) exist within a single DB.

[0139] In the example of (a), there is provided a QIM-DB 120 that holds information for the current day, and a QIM-DB 121 that holds information up to the previous day. The management information of the QIM-DB 120 that holds information for the current day is a feature extracted / updated only from the quality history for the most recent day (refreshed daily in this example), and reflects the latest situation without being influenced by past performance. Note that the latest DB period being "one day" is just an example, and it may be shorter or longer than "one day."

[0140] The management information of the QIM-DB 121, which holds information up to the previous day, makes it possible to cover areas where quality history is not frequently acquired based on past performance.

[0141] If the QIM-DB 120 does not contain information for the current day, the quality estimation unit 210 refers back to the QIM-DB 121. Alternatively, the quality estimation unit 210 may refer to both the information in the QIM-DB 120 and the information in the QIM-DB 121 and perform quality estimation from each of them. This also applies to the following examples.

[0142] The example of (b) uses a QIM-DB 120 that holds information for the current day, a QIM-DB 122 that holds information for the previous day, and a QIM-DB 123 that holds information for the past two days.

[0143] When a QIM-DB is refreshed (e.g., after one day has passed), the information before the refresh is reflected in the QIM-DB that manages older information than itself. In the example (b), the management information of QIM-DB 120 (today) is replaced with the management information of QIM-DB 122 (previous day), and the management information of QIM-DB 123 (past up to two days ago) is updated taking into account the management information of QIM-DB 122 (previous day). The same applies to the example (c) below.

[0144] The example of (c) uses QIM-DB120, which holds information for the current day, QIM-DB122, which holds information for the previous day, QIM-DB124, which holds information for two days ago, and QIM-DB125, which holds information for up to three days ago.

[0145] In the above example, the database is divided by day, but it is also possible to manage and reference information by day of the week or week, and to discard (refresh) the managed information at regular intervals to avoid the influence of past performance.

[0146] Furthermore, when recalculating / updating features based on the most recent quality history and management information in the QIM-DB, it is also possible to use a method such as tuning that emphasizes the most recent results by using a weighted average. The innovations in the calculation / tuning method can also be applied when operating with a single QIM-DB.

[0147] (Mechanism for Estimating Quality Taking into Account Not Only Past Performance but Also Latest Status) Next, an embodiment of a mechanism for estimating quality taking into account not only past performance but also the latest status will be described.

[0148] There is a problem in that quality estimation is required in an environment where there is a large change on the time axis (where it is difficult to see a certain correlation). In response to this problem, this embodiment uses both quality estimation based on the most recent situation including the latest quality history and quality estimation from a statistical perspective. This makes it possible to estimate quality according to the use case and requirements to which this proposed method (QIM) is applied. This embodiment will be described with reference to FIG. 18.

[0149] As shown in FIG. 18, the QIM-DB has a three-layer structure. Note that this is just one example, and any other multi-layer structure / format may be used. In the example of FIG. 18, QIM-DB 120-1 holds the most recent quality history itself. QIM-DB 120-2 holds information for the previous 10 minutes. QIM-DB 120-2 holds feature amounts as management information. QIM-DB 120-3 holds past information, including the most recent information. QIM-DB 120-3 holds feature amounts as management information.

[0150] With the configuration shown in FIG. 18, for example, the quality estimation unit 210 returns to the external function an estimation result response such as "Estimation result 1: Estimation result based on QIM1," "Estimation result 2: Estimation result based on QIM2," and "Estimation result 3: Estimation result based on QIM3."

[0151] In addition, the following configuration is also envisioned, in which the quality history acquired by the vehicle in front when vehicles are traveling in a line is immediately utilized by the vehicle in the rear.

[0152] The rear vehicle will be able to carry out various judgments, decisions, and controls based on both the perspectives of "what was the quality just before (based on the quality history acquired by the front vehicle)" and "whether quality deterioration is statistically estimated (based on the basic concept of QIM)."

[0153] (Other Configuration Examples) The information processing device 100 may have the configuration shown in Fig. 19. Note that the configuration shown in Fig. 3 is a detailed example of the configuration shown in Fig. 19.

[0154] 19 includes an extraction unit 510, a management database 520, a quality estimation unit 530, and a quality estimation method determination unit 540. Note that the quality estimation unit 530 and the quality estimation method determination unit 540 may be provided outside the information processing device 100.

[0155] The extraction unit 510 extracts feature amounts for each management category including a space category from the history of quality information related to communications performed by the device, and the management database 520 stores the feature amounts for each management category.

[0156] The quality estimation unit 530 receives a quality estimation request from an external function, references the management database, and estimates quality based on information in the management category determined based on the request. The quality estimation method determination unit 540 performs quality estimation using multiple quality estimation methods and determines the quality estimation method for each management category by comparing the quality estimation results with the actual quality history.

[0157] (Other Examples) Examples 1 to 5 will now be described as other examples related to the above embodiments.

[0158] Example 1: Regarding the scaled (distributed, multiple) form mentioned above, it is possible to incorporate not only the proposed method (QIM) but also (existing) quality history creation functions and external functions.

[0159] Example 2: Based on local specialization of distributed learning / quality estimation functions using edge computing, low-latency / high-accuracy real-time estimation may be performed in use cases such as connected cars.

[0160] Example 3: As a feature of the quality index, not only throughput (bandwidth) but also information related to network quality such as packet loss rate, packet loss occurrence probability, communication delay, and jitter, and information related to the quality of the application layer such as video quality including VMAF (Video Multimethod Assessment Fusion), bit rate, and frame rate may be used.

[0161] Example 4: As an example of the management division of this proposed method, a mesh structure based on latitude and longitude for the spatial axis is explained, but it is also possible to use a division method based on three-dimensional polygons that add altitude, or divisions that are not fixed shapes such as squares or cubes (polygons, divisions of different sizes or shapes that are divided according to the properties of objects in the space or the space, etc.).

[0162] Example 5: Closed-loop (feedback) control utilizing machine learning / AI may be applied to determine the quality estimation method.

[0163] (Scope of application of proactive cooperative control) The scope of application of proactive cooperative control is explained below. Proactive cooperative control aims to continue (avoid interruptions in) information transmission, such as video transmission for real-time remote monitoring.

[0164] To achieve this goal, the system controls the transmission information and the transmission method. Specifically, it selects networks with good quality that can be used for information transmission (excluding networks with deteriorating quality that cannot be used for information transmission), and distributes the transmission information according to the network quality.

[0165] The following predictions 1 and 2 are necessary for controlling the transmission method.

[0166] Prediction 1: Detection of network quality degradation that makes it unusable for information transmission. For example, when the throughput is x １ Less than Mbps, packet loss rate / occurrence probability is x ２ % or more, communication delay is x ３ ms or more, or jitter is x ４ ms or more is detected in advance.

[0167] Prediction 2: Estimation of the network quality used for information transmission. For example, if the throughput is x １ Mbps or more (and x １" Mbps or less), communication delay is x ３' Less than ms and jitter is x ４' A priori estimation is made that the time is less than ms.

[0168] The scope of application assumed for this proposed method (QIM) is as follows (1) and (2).

[0169] (1) Detecting network quality degradation based on network usage records. Throughput is x. １ Less than Mbps, packet loss probability is x ２ % or more, communication delay is x ３ ms or more, or jitter is x ４ ms or more is detected in advance.

[0170] (2) Estimation of the lower limit of quality of the network used for information transmission. The throughput is x. １' Mbps or more, communication delay x ３'Less than ms and jitter is x ４' A priori estimate is made that it will be less than ms.

[0171] (Hardware Configuration Example) The information processing device 100 described in this embodiment can be realized, for example, by causing a computer to execute a program. This computer may be a physical computer or a virtual machine on a cloud. Furthermore, the information processing device 100 may be a server on a network, a base station, an edge server, or a terminal held by a user.

[0172] That is, the information processing device 100 can be realized by using hardware resources such as a CPU and memory built into a computer to execute a program corresponding to the processing performed by the information processing device 100. The program can be recorded on a computer-readable recording medium (such as a portable memory) and can be saved or distributed. The program can also be provided via a network such as the Internet or email.

[0173] Fig. 20 is a diagram showing an example of the hardware configuration of the computer. The computer in Fig. 20 includes a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, and the like, all of which are interconnected by a bus BS. The computer may further include a GPU.

[0174] The program that realizes the processing on the computer is provided by a recording medium 1001, such as a CD-ROM or a memory card. When the recording medium 1001 storing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001, but may be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files, data, etc.

[0175] The memory device 1003 reads and stores a program from the auxiliary storage device 1002 when an instruction to start the program is received. The CPU 1004 realizes functions related to the information processing device 100 in accordance with the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a network, etc. The display device 1006 displays a GUI (Graphical User Interface) or the like according to the program. The input device 1007 is composed of a keyboard, mouse, buttons, a touch panel, etc., and is used to input various operation instructions. The output device 1008 outputs the results of calculations.

[0176] (Summary of the embodiment, effects, etc.) As explained above, continuous information management is necessary to achieve highly accurate quality estimation, but conventional techniques have had the problem that the amount of data in the DB continues to increase.

[0177] In contrast, the technology according to the present embodiment extracts features included in the quality history, manages a predefined set of features / parameters (quality indexes) as information that serves as the basis for quality estimation, and does not retain the quality history itself for a long period of time. Furthermore, when estimating quality, the management information such as the quality index is referenced, and the quality is estimated based on logic appropriate for the situation and application.

[0178] The technology according to the present embodiment as described above eliminates the need for long-term management of the quality history itself, which is created / added sequentially, and therefore prevents the amount of management information from expanding on the time-space axis. As a result, the total amount of management information is kept below a certain level, enabling continuous automatic operation of the entire system based on real-time information collection.

[0179] Furthermore, estimation based on management information that reflects changing quality and logic according to the situation and application maintains accuracy that is equal to or greater than that of estimation based solely on quality history. By maintaining and improving the quantity and quality of the information that forms the basis of quality estimation, it becomes possible to maintain and improve the availability and accuracy of quality estimation.

[0180] Furthermore, mesh-based management, rather than pinpointing locations, ensures the availability of quality estimation (the "quantity" of information that affects quality estimation). Also, real-time quality / changes in the situation are reflected / feature values / parameters that are tuned as needed ensure the accuracy of quality estimation (the "quality" of information).

[0181] Furthermore, by managing quality information and quality estimation methods based on the concept of edge computing tailored to regional characteristics, such as using information from a certain region only in that region (where it makes sense to use it), and the local specialization / distribution of input learning data and output algorithms, it is possible to avoid the expansion of the amount of information that would result from centralized management, while also being able to accommodate use cases that require regional distribution and low latency.

[0182] The following additional notes are provided regarding the above-described embodiments.

[0183] <Additional Notes> (Additional Item 1) An information processing device comprising: an extraction unit that extracts a feature for each management category including a spatial category from a history of quality information related to communication performed by a device; and a management database that stores the feature for each management category. (Additional Item 2) The information processing device according to Additional Item 1, further comprising: a quality estimation unit that receives a quality estimation request from an external function, refers to the management database, and estimates quality based on information in the management category determined based on the request. (Additional Item 3) The information processing device according to Additional Item 2, wherein the quality estimation unit estimates quality using the feature and the history. (Additional Item 4) The information processing device according to Additional Item 1, further comprising: a quality estimation method determination unit that performs quality estimation using a plurality of quality estimation methods and determines the quality estimation method for each management category by comparing the results of the quality estimation with a history of actual quality. (Additional Item 5) The information processing device according to Additional Item 1, wherein the spatial categories are regions divided into sections of a map, and the extraction unit extracts the feature for each region. (Supplementary Item 6) The information processing device according to Supplementary Item 1, wherein the management database includes a database for storing first feature amounts extracted only from history for a most recent predetermined period, and a database for storing second feature amounts extracted from history older than the predetermined period. (Supplementary Item 7) An information processing method executed by an information processing device, comprising: extracting feature amounts for each management category including a spatial category from a history of quality information related to communications performed by a device; and storing the feature amounts for each management category in the management database. (Supplementary Item 8) A non-transitory storage medium storing a program for causing a computer to function as each unit in the information processing device according to any one of Supplementary Item 1 to 6.

[0184] Although the present embodiment has been described above, the present invention is not limited to such a specific embodiment, and various modifications and changes are possible within the scope of the gist of the present invention described in the claims.

[0185] 1-3 DB 4 Quality information acquisition function 5 Time / location information acquisition function 6 Other related information acquisition function 7 Quality history creation function 8 Quality DB 9 Quality estimation function 10 External function 100 Information processing device 110 Quality index management unit 120 QIM-DB 130 Management information monitoring unit 140 Quality history distribution unit 150 Estimation judgment DB 160 Quality estimation method judgment unit 210 Quality estimation unit 510 Extraction unit 520 Management database 530 Quality estimation unit 540 Quality estimation method judgment unit 1000 Drive device 1001 Recording medium 1002 Auxiliary storage device 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device

Claims

1. An information processing device comprising: an extraction unit that extracts features for each management category including a spatial category from a history of quality information related to communication performed by a device; and a management database that stores the features for each management category.

2. The information processing device according to claim 1, further comprising a quality estimation unit that receives a request for quality estimation from an external function, references the management database, and estimates quality based on information in a management category determined based on the request.

3. The information processing device according to claim 2, wherein the quality estimation unit estimates quality using the feature amount and the history.

4. The information processing device according to claim 1, further comprising a quality estimation method determination unit that performs quality estimation using a plurality of quality estimation methods and determines the quality estimation method for each management category by comparing the results of the quality estimation with the actual quality history.

5. The information processing device according to claim 1, wherein the spatial divisions are regions divided into sections of a map, and the extraction unit extracts the feature amount for each region.

6. The information processing device according to claim 1, wherein the management database comprises a database that stores first features extracted only from history for the most recent specified period, and a database that stores second features extracted from history older than the specified period.

7. An information processing method executed by an information processing device, comprising: a step of extracting features for each management category including a spatial category from a history of quality information related to communications performed by a device; and a step of storing the features for each management category in a management database.

8. A program for causing a computer to function as each unit in the information processing device according to any one of claims 1 to 6.

Images