JP6668270B2 - Apparatus and method for extracting communication quality features - Google Patents

Description

  The present invention relates to a technique for extracting a feature quantity representing a change in communication quality from a group of quality indexes measured a plurality of times.

  Generally, when there are a plurality of samples of the quality index, feature amounts such as sample average and sample variance are used.

JP-A-2005-049586

  To cope with the deterioration of communication quality, it is necessary to quantitatively grasp the tendency of fluctuations, but it is not possible to express gradual fluctuations with the sample mean and sample standard deviation, which are general distribution features. difficult.

  In view of the above problems, an object of the present invention is to provide a technique for extracting a feature amount representing a gradual change in communication quality.

  In order to solve the above-described problem, one aspect of the present invention is to collect a measurement value of a communication quality index generated from a model, and a data collection unit that forms a column of a pair of the collected measurement value and percentile value, A regression processing unit that determines a regression line for each subsequence by performing a linear function regression on each subsequence of the configured pair of columns; The present invention relates to a communication quality feature quantity extraction device having a feature quantity extraction unit for extracting a feature quantity based on the longest partial sequence.

  According to the present invention, it is possible to extract a feature amount representing a gradual change in communication quality.

FIG. 1 is a block diagram showing a configuration of a communication quality feature quantity extraction device according to one embodiment of the present invention. FIG. 2 is a graph showing a pair of a measured value and a percentile value according to an embodiment of the present invention. FIG. 3 is a flowchart showing a communication quality feature quantity extraction process according to one embodiment of the present invention. FIG. 4 is a flowchart showing a communication quality feature value extraction process according to another embodiment of the present invention. FIG. 5 is a graph showing a simulation result in a case where the average and the standard deviation are used as the feature amounts. FIG. 6 is a graph showing a simulation result in a case using a feature according to an embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  In an embodiment described later, a communication quality feature amount extraction device is disclosed. In summary of the following embodiment, the communication quality feature quantity extraction device collects the measured values of the communication quality index generated from the model and forms a column of pairs of the collected measured values and percentile values. The communication quality feature quantity extraction device determines a regression line for each subsequence by performing a linear function regression on each subsequence of the constructed pair of columns, and determines a regression line having an error equal to or less than a predetermined threshold. , The feature amount is extracted based on the longest partial sequence. As a result, it is possible to extract a feature amount representing a gradual change in communication quality as compared with a case where a general average and a standard deviation are used as the feature amount.

  First, a communication quality feature quantity extraction device according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing a configuration of a communication quality feature quantity extraction device according to one embodiment of the present invention.

  As shown in FIG. 1, the communication quality feature quantity extraction device 100 includes a data collection unit 110, a regression processing unit 120, and a feature quantity extraction unit 130.

  The data collection unit 110 collects measured values of the communication quality index generated from the model, and forms a column of pairs of the collected measured values and percentile values. Specifically, when the number of measurements (number of samples) of the communication quality index is n, n samples generated from an arbitrary model (parameters α_1, α_2,...) Are x_k (x_1, x_2,..., X_n ) Represents the value of the ((k-1) / (n-1)) percentile point of an arbitrary model (k = 1,..., N). Y_k (y_1, y_2,..., Y_n) is an index indicating communication quality measured n times within an arbitrary time. At this time, the data collection unit 110 forms a pair (x_k, y_k) of the percentile value and the measurement value, and creates a graph when (x_k, y_k) is changed to k = 1,..., N ( (Fig. 2).

The regression processing unit 120 determines a regression line for each subsequence by performing a linear function regression on each subsequence of the configured pair of columns. Specifically, the regression processing unit 120 generates (x_i, y_i), (x_ {i + 1}, y_ {i + 1}), (x_i, y_i) in the graph of (x_k, y_k) created by the data collection unit 110. A linear function regression is performed on the points…, (x_j, y_j) (however, i <j, i = 2,…, n-2, j = 2,…, n-1). In other words, the regression processing unit 120 performs a linear function regression on the partial sequence of the point sequence of {(x_k, y_k)}. At this time, a straight line regressed so as to minimize the index s is set as y = a * x + b (a and b are parameters). The method of calculating the index s includes, for example, the following method. Here, it is assumed that y'_k = a * x_k + b.
(A) L1
Add the absolute value of the difference between y'_k and y_k up to k = i, ..., j and divide by (j-i + 1).

s = (1 / (j-i + 1)) * Σ_ {k = i} ^ {j} | y'_k-y_k |
(B) L2
The sum of the square of the absolute value of the difference between y′_k and y_k up to k = i,..., j is raised to the half power, and divided by (j−i + 1).

s = (1 / (j-i + 1)) * Σ_ {k = i} ^ {j} | y'_k-y_k | ^ 2) ^ (1/2)
The feature amount extraction unit 130 extracts a feature amount based on the longest partial sequence among the partial sequences of the regression line having an error equal to or less than a predetermined threshold. Specifically, the feature amount extraction unit 130 calculates an index d representing a distance between y_k and a * x_k + b at k = i,..., J. Note that the index d can be obtained by the methods (A) and (A) described above, but may be independent of the method of calculating the index s. For example, the index d and the index s may be calculated by the same method, or may be calculated by different methods. The feature amount extraction unit 130 calculates d for all pairs of (i, j) satisfying 2 ≦ i <j ≦ n−1. When d becomes equal to or less than a predetermined threshold value d *, a set of (i, j) that maximizes ji is selected and output. That is, the feature amount extraction unit 130 determines that the length ji of the subsequence among the subsequences in which the error between the measured value y_k and the value of the derived regression line x_k (y′_k) is equal to or smaller than the threshold value The maximum (i, j) pair is determined.

  After that, the data collection unit 110 changes the parameters α_1, α_2,... Of the model, and assigns a percentile value to new x′_1, x′_2,. A pair (x'_k, y_k) is constructed with the measured values and a graph is created when (x'_k, y_k) is changed to k = 1,..., N. Thereafter, in the same manner, the regression processing unit 120 determines (x'_i, y_i), (x '_ {i + 1}, y_) in the graph of (x'_k, y_k) created by the data collection unit 110. {i + 1}),..., (x′_j, y_j) are subjected to linear function regression, and the feature quantity extraction unit 130 calculates the distance between y_k and a * x′_k + b Is selected and output when the index d representing becomes equal to or less than a predetermined threshold d *.

  After acquiring the set of (i, j) for the values of the parameters α_1, α_2,... By the above-described procedure, the feature amount extraction unit 130 maximizes ji among the set of (i, j). .. And i, j of the model are determined, and α_1, α_2,... And i / n, j / n are extracted as features of the communication quality index.

  The communication quality feature quantity extraction device 100 includes, for example, a drive device, an auxiliary storage device, a memory device, a processor, an interface device, and a communication device interconnected via a bus. Various computer programs including programs for realizing various functions and processes in the communication quality feature quantity extraction device 100 are provided by a recording medium such as a CD-ROM (Compact Disk-Read Only Memory), a DVD (Digital Versatile Disk), and a flash memory. May be done. When the recording medium storing the program is set in the drive device, the program is installed from the recording medium to the auxiliary storage device via the drive device. However, the installation of the program is not necessarily performed by a recording medium, and may be downloaded from any external device via a network or the like. The auxiliary storage device stores installed programs and also stores necessary files and data. The memory device reads out a program or data from the auxiliary storage device and stores it when there is an instruction to start the program. The processor executes various functions and processes of the communication quality feature quantity extraction device 100 according to various data such as a program stored in the memory device and parameters necessary for executing the program. The interface device is used as a communication interface for connecting to a network or an external device. The communication device executes various communication processes for communicating with a network such as the Internet. However, the communication quality feature quantity extraction device 100 is not limited to the hardware configuration described above, and may be implemented by any other appropriate hardware configuration.

  Next, with reference to FIG. 3, the processing in the communication quality feature quantity extraction device according to one embodiment of the present invention will be described. FIG. 3 is a flowchart showing a communication quality feature quantity extraction process according to one embodiment of the present invention.

  As shown in FIG. 3, in step S101, the data collection unit 110 pairs a pair (x_k, n) of n measurement values y_k and percentile values x_k collected from an arbitrary model based on the set parameters α_1, α_2,. y_k), and create a graph when (x_k, y_k) is changed to k = 1,..., n.

  In step S102, the regression processing unit 120 sets an index indicating the distance between y′_k and y_k for the subsequence (x_i, y_i),..., (X_j, y_j) A linear function regression is performed to derive a regression line y = a * x + b that minimizes s.

  In step S103, the feature amount extraction unit 130 calculates an index d representing a distance between y_k and a * x + b for each set of (i, j), and the index d is equal to or less than the threshold d *. Among the (i, j) pairs that satisfy (d ≦ d *), the (i, j) pair that maximizes ji is determined.

  In step S104, the data collection unit 110 determines whether or not to change to the next parameter value. If the parameter value needs to be changed, for example, if there is another parameter value, in step S105, the data collection unit 110 changes the parameters α_1, α_2,. Then, steps S101 to S103 described above are executed for the changed parameters.

  On the other hand, when the change is unnecessary, for example, when the above-described steps S101 to S103 are performed for all the predetermined parameter values, in step S106, the feature amount extraction unit 130 acquires The parameters α_1, α_2,... And i / n, j / n that maximize ji are extracted from the set of (i, j) as the feature amount of the communication quality index.

  Next, with reference to FIG. 4, a process in the communication quality feature quantity extraction device according to another embodiment of the present invention will be described. In the embodiment illustrated in FIG. 3, the feature amount is extracted from the communication quality index of an arbitrary model. However, in the present embodiment, the communication quality feature amount extracting device 100 determines the feature amount from the communication quality index according to the standard normal distribution. Is extracted. FIG. 4 is a flowchart showing a communication quality feature value extraction process according to another embodiment of the present invention.

  As shown in FIG. 4, in step S201, the data collection unit 110 pairs a pair (x_k) of n measurement values y_k and percentile values x_k collected from a standard normal distribution with an average μ = 0 and a standard deviation σ = 1. , y_k), and create a graph when (x_k, y_k) is changed to k = 1,..., n.

  In step S202, the regression processing unit 120 determines an index representing the distance between y′_k and y_k for the subsequence (x_i, y_i),..., (X_j, y_j) A linear function regression is performed to derive a regression line y = a * x + b that minimizes s.

  In step S203, the feature amount extraction unit 130 calculates an index d representing a distance between y_k and a * x + b for each set of (i, j), and the index d is equal to or less than the threshold d *. Among the (i, j) pairs that satisfy (d ≦ d *), the (i, j) pair that maximizes ji is determined.

  In step S204, the feature amount extraction unit 130 extracts the parameters a and b and i / n and j / n of the regression line that maximizes j-i as feature amounts of the communication quality index.

  Next, with reference to FIGS. 5 and 6, a simulation result regarding a comparison between a feature value according to the present invention and a general average and standard deviation will be described. In this simulation, feature amounts are plotted using RTT (Round-Trip Time) measured every second for 5 minutes.

  FIG. 5 shows a case where the average and the standard deviation are used as the feature amounts, and FIG. 6 shows a case where the above-mentioned feature amounts i / n, j / n, a and b are used. As can be observed from the illustrated graph, the feature amounts i / n, j / n, a, and b show a gradual change as compared with the average and the standard deviation, which are general indices.

  As described above, the embodiments of the present invention have been described in detail. However, the present invention is not limited to the specific embodiments described above, and various modifications may be made within the scope of the present invention described in the appended claims.・ Change is possible.

Reference Signs List 100 communication quality feature extraction device 110 data collection unit 120 regression processing unit 130 feature extraction unit

Claims (4)

A data collection unit that collects the measurement values of the communication quality index generated from the model, and configures a column of a pair of the collected measurement value and the percentile value.
A regression processing unit that determines a regression line for each subsequence by performing a linear function regression on each subsequence of the configured pair of columns;
Among the substrings of the regression line having an error equal to or less than a predetermined threshold, a feature amount extraction unit that extracts a feature amount based on the longest subsequence,
A communication quality feature quantity extraction device having: The model is any model that follows variable parameters,
The feature amount extraction unit determines the longest subsequence for each parameter value, and calculates the parameter value that is the longest subsequence of the longest subsequences and the index at both ends of the longest subsequence. The communication quality feature quantity extraction device according to claim 1, wherein a value obtained by dividing by a number of measurement values is extracted as the feature quantity. The model is a standard normal distribution,
The feature value extraction unit extracts, as the feature value, a parameter value that defines a regression line of the longest subsequence and a value obtained by dividing an index at both ends of the longest subsequence by the number of the measurement values. The communication quality feature quantity extraction device according to claim 1. Collecting a measurement value of a communication quality index generated from the model, and configuring a column of a pair of the collected measurement value and percentile value,
Determining a regression line for each subsequence by performing a linear function regression on each subsequence of the constructed pair of columns;
Extracting a feature amount based on the longest subsequence among the subsequences of the regression line having an error equal to or less than a predetermined threshold value;
A communication quality feature amount extraction method having the following.

Images