JP2021124999A - Communication situation estimation device, communication situation estimation method, and program - Google Patents

Abstract

To provide a communication situation estimation device that accurately predicts (estimates) the situation of communication quality data even when a communication environment situation changes in a complicated and dynamic manner like a narrow space such as a factory.SOLUTION: Since a communication situation estimation device 100 uses a multi-bandwidth kernel density distribution, in which single-bandwidth kernel density distributions acquired by a plurality of different bandwidths are superimposed by various weighting coefficients, to acquire a probability density distribution of data related to communication quality, the situation of communication quality can be accurately estimated whatever the probability density distribution of data related to communication quality is.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for estimating communication status and communication status such as wireless communication.

In a narrow space such as a factory, the wireless environment situation changes complicatedly and dynamically due to changes in the propagation environment due to the movement of machine tools, products, workers, etc., and interference between multiple wireless applications. In a narrow space such as a factory, it is difficult to grasp the position information of moving objects such as machine tools, and since the radio wave propagation is a multipath-rich environment, unexpected effects may occur in unexpected places. Occurs. In this way, if an unexpected effect occurs in an unexpected place in a narrow space, stable communication cannot be maintained in the worst case, and the manufacturing site may take measures such as stopping the manufacturing line. Productivity is reduced. That is, in a narrow space such as a factory, unexpected deterioration in communication quality (Quality of Service) may occur, resulting in a decrease in productivity.

Therefore, it is necessary to take evasive action before the extreme deterioration of QoS (communication failure) occurs, and in order to realize it, a technique for predicting the occurrence of communication failure with high accuracy is required. In order to realize such a technique, for example, a technique for acquiring a distribution of values indicating communication quality in time series and predicting a future communication quality status based on the distribution (for example, disclosed in Patent Document 1). It is conceivable to use the technology that is used.

Japanese Unexamined Patent Publication No. 2005-18304

For example, Patent Document 1 describes (1) a prediction method from each time-series data value of the prediction execution time and its vicinity in the past (for example, a prediction method using an AR model, an ARMA model, a Kalman filter, etc.) and (2). Generates a set of time-series data that shows the same tendency as the prediction target time in the same season or the same day as the prediction target time, and predicts and outputs some statistical values such as the average value and the most frequent value of the set. There is a disclosure of technology regarding predictive methods that incorporate the strengths of both methods. In the technique of Patent Document 1, the time-series data can be predicted by the above-mentioned prediction method even when the time-series data has a large time fluctuation or is unsteady.

However, Patent Document 1 is based on the premise that time-series data showing the same tendency as the prediction target time can be acquired in the same season or the same day as the prediction target time, and is complicated and dynamic like a narrow space such as a factory. When the wireless environment condition (communication environment) fluctuates (when the fluctuation of time series data rarely has the same tendency and it is difficult to obtain a set of time series data with the same fluctuation tendency), It is difficult to accurately predict (estimate) the status of communication quality data (for example, the probability density distribution of communication quality data).

Therefore, in view of the above problems, the present invention raises the status of communication quality data (for example, the probability density distribution of communication quality data) even when the communication environment status fluctuates in a complicated and dynamic manner such as in a narrow space such as a factory. The purpose is to realize a communication status estimation device, a communication status estimation method, and a program that accurately predict (estimate).

In order to solve the above problems, the first invention is a communication status estimation device including an optimum parameter acquisition processing unit and a communication quality data probability density distribution estimation unit.

The optimum parameter acquisition processing unit multiplies the single-bandwidth kernel density distribution defined by the kernel functions of multiple bandwidths based on the parameters in order to estimate the probability density distribution for the data related to communication quality. When acquiring the bandwidth kernel density distribution and estimating the probability density distribution of communication quality data from the multiband kernel density distribution, the multiband kernel density distribution is applied to multiple samples of communication quality data. Acquires the optimum parameter, which is the parameter with the maximum probability when is applied.

Communication quality data The probability density distribution estimation unit applies the probability density distribution obtained by applying the multi-bandwidth kernel density distribution set by the optimum parameters to multiple samples of data related to communication quality, and determines the probability density distribution related to communication quality. Obtained as an estimated probability density distribution for the data.

In this communication status estimator, data on communication quality (for example, QoS) is used by superimposing a single bandwidth kernel density distribution acquired by a plurality of different bandwidths with various weighting coefficients. Since the probability density distribution of data) is acquired (estimated), it is possible to estimate with high accuracy regardless of the probability density distribution of data related to communication quality (for example, QoS data). For example, QoS in which communication quality data (for example, QoS data) is concentrated on a specific value (a specific narrow range) that tends to occur in a narrow space such as a factory, and the QoS data is widely distributed. The probability density distribution of data can also be estimated with high accuracy by this communication status estimation device.

Further, in this communication status estimation device, the optimum parameter is acquired in consideration of the probability calculated from the sample of QoS data, and the probability of the QoS data is obtained by using the multi-bandwidth kernel density distribution set by the optimum parameter. Since the density distribution is acquired (estimated), the probability density distribution of the QoS data can be estimated with high accuracy regardless of the probability density distribution of the QoS data.

In this way, this communication status estimation device acquires the probability density distribution of highly accurate communication quality data (QoS data) even when the communication environment status fluctuates in a complicated and dynamic manner such as in a narrow space such as a factory (QoS data). Can be estimated). That is, this communication status estimation device can accurately determine the status of communication quality data (for example, the probability density distribution of communication quality data) even when the communication environment status fluctuates dynamically and in a complicated manner such as in a narrow space such as a factory. Can be predicted (estimated).

The second invention is the first invention, in which the optimum parameter acquisition processing unit holds a plurality of bandwidth candidates for specifying the single bandwidth kernel density distribution, and the single bandwidth kernel density It holds multiple candidates for weighting factors when acquiring a multi-bandwidth kernel density distribution by superimposing distributions. Then, the optimum parameter acquisition processing unit calculates the likelihood for all combinations of the plurality of bandwidth candidates and the plurality of weighting coefficient candidates, and determines the bandwidth and weighting coefficient having the maximum calculated likelihood. Get the set as the optimal parameter.

As a result, in this communication status estimation device, the optimum bandwidth combination and the optimum weighting coefficient set that maximize the likelihood are specified, and the probability density distribution of the QoS data is acquired (estimated). Regardless of the probability density distribution of the data, the probability density distribution of the QoS data can be estimated with high accuracy.

As a result, this communication status estimation device acquires (estimates) the probability density distribution of highly accurate communication quality data (QoS data) even when the communication environment status fluctuates in a complicated and dynamic manner such as in a narrow space such as a factory. )can.

The third invention is the first or second invention, and the optimum parameter acquisition processing unit is defined by a kernel function having an integral value of "1" in all sections of the domain and having different bandwidths. Obtain the multi-bandwidth kernel density distribution using the single-bandwidth kernel density distribution that is obtained.

As a result, this communication status estimation device can acquire the multi-bandwidth kernel density distribution using the single-bandwidth kernel density distribution defined by the kernel function satisfying the above properties.

The fourth invention is the third invention, in which the kernel function is at least one of a Gaussian function, a function that realizes a triangular shape distribution, and a function that realizes a rectangular shape distribution.

As a result, in this communication status estimation device, it is defined by at least one kernel function of a Gaussian function (a function that realizes a Gaussian distribution (normal distribution)), a function that realizes a triangular shape distribution, and a function that realizes a rectangular shape distribution. The single-bandwidth kernel density distribution that is obtained can be used to obtain the multi-bandwidth kernel density distribution.

In the communication status estimation device, the single bandwidth kernel density distribution realized by combining any two or more of the Gaussian function, the function for realizing the triangular shape distribution, and the function for realizing the rectangular shape distribution is weighted. The multi-bandwidth kernel density distribution may be obtained by weighting and adding according to the coefficient set.

A fifth invention is a communication status estimation method including an optimum parameter acquisition processing step and a communication quality data probability density distribution estimation step.

The optimal parameter acquisition processing step is multi-layered by superimposing a single bandwidth kernel density distribution defined by multiple bandwidth kernel functions based on the parameters in order to estimate the probability density distribution for data related to communication quality. When acquiring the bandwidth kernel density distribution and estimating the probability density distribution of communication quality data from the multiband kernel density distribution, the multiband kernel density distribution is applied to multiple samples of communication quality data. Acquires the optimum parameter, which is the parameter with the maximum probability when is applied.

The communication quality data probability density distribution estimation step relates to the communication quality by applying the probability density distribution obtained by applying the multi-bandwidth kernel density distribution set by the optimum parameters to a plurality of samples of communication quality data. Obtained as an estimated probability density distribution for the data.

As a result, it is possible to realize a communication status estimation method having the same effect as that of the first invention.

The sixth invention is a program for causing a computer to execute the communication status estimation method according to the fifth invention.

As a result, it is possible to realize a program for causing the computer to execute the communication status estimation method having the same effect as that of the first invention.

According to the present invention, even when the communication environment condition fluctuates dynamically and complicatedly such as in a narrow space such as a factory, the condition of communication quality data (for example, the probability density distribution of communication quality data) is predicted with high accuracy (for example). It is possible to realize a communication status estimation device (estimation), a communication status estimation method, and a program.

The schematic block diagram of the communication condition estimation apparatus 100 which concerns on 1st Embodiment. It is a flowchart of the process executed by the communication condition estimation apparatus 100. The figure for demonstrating the process executed by the communication condition estimation apparatus 100. The figure which shows the distribution acquired by the Gaussian kernel function using the sample q1 to q5 of QoS data, and the single bandwidth kernel density distribution acquired by superimposing the distribution. The figure which shows the distribution acquired by the Gaussian kernel function using the sample q1 to q5 of QoS data, and the single bandwidth kernel density distribution acquired by superimposing the distribution. The figure which shows the distribution acquired by the Gaussian kernel function using the sample q1 to q5 of QoS data, and the single bandwidth kernel density distribution acquired by superimposing the distribution. The figure for demonstrating the process of acquiring the _multi- bandwidth kernel density distribution p multi (q) from the single-bandwidth kernel density distribution acquired using the sample q1 to q5 of QoS data. The figure for demonstrating the likelihood calculation process. The graph which shows the probability density distribution of the QoS data of the simulation example. A graph showing the relationship between the number of samples and the error between the measured data in the estimated data based on the single-bandwidth kernel density distribution and the estimated data based on the multi-bandwidth kernel density distribution. The figure which shows the example of the function which the integral value in all sections becomes "1". The figure which shows the CPU bus configuration.

[First Embodiment]
The first embodiment will be described below with reference to the drawings.

<1.1: Configuration of communication status estimation device>
FIG. 1 is a schematic configuration diagram of the communication status estimation device 100 according to the first embodiment.

As shown in FIG. 1, the communication status estimation device 100 includes an optimum parameter acquisition processing unit 1 and a communication quality data probability density distribution estimation unit 2. For convenience of explanation, the communication status estimation device 100 inputs data (QoS data (for example, throughput)) indicating communication quality when a plurality of devices perform wireless communication in a narrow space such as a factory. This will be described below.

As shown in FIG. 1, the optimum parameter acquisition processing unit 1 includes a sample division processing unit 11, a bandwidth set selection processing unit 12, a single bandwidth kernel density distribution acquisition processing unit 13, and a weighting coefficient set selection processing unit 14. A multi-bandwidth kernel density distribution acquisition unit 15, a likelihood calculation unit 16, and a storage retention unit Mem1 are provided. Further, the optimum parameter acquisition processing unit 1 includes an average likelihood calculation unit 17, a storage unit Mem2, and an optimum parameter acquisition unit 18.

The sample division processing unit 11 inputs the data Din and the control signal CTL1 from a control unit (not shown) that performs various controls of the communication status estimation device 100. For convenience of explanation, the data Din is QoS data (for example, throughput) and includes a plurality of samples (QoS sample data). The sample division processing unit 11 divides the input QoS data sample into L subsets. The set of all samples of QoS data is expressed as ε, and the i-th subset (sample set) of the divided subset is expressed as ε _i (i: natural number, 1 ≦ i ≦ L).

The sample division processing unit 11 sets one sample set for likelihood calculation, and sets a set of sample sets other than the sample set for likelihood calculation in the sample set ε _data for acquiring the multi-bandwidth kernel density distribution. All the samples of QoS data, the sample set for likelihood calculation, and the sample set for acquiring the multi-bandwidth kernel density distribution are described as follows.
All samples: ε = {ε ₁ , ε ₂ , ..., ε _L }
Likelihood calculation sample set: ε _m (m: natural number, 1 ≦ m ≦ L)
Sample set for multi-bandwidth kernel density distribution acquisition: ε _data = {ε ₁ , ε ₂ , ..., ε _m-1 , ε _{m + 1} , ..., ε _L }
The sample division processing unit 11 sets one of the subsets of the samples obtained by division in the likelihood calculation sample set ε _m according to the control signal CTL 1, and sets a set of sample sets other than the likelihood calculation sample set ε _{m. Is} set in the sample set ε data for acquiring the multi-bandwidth kernel density distribution. Then, the sample division processing unit 11 outputs the sample set ε _data for acquiring the multi-bandwidth kernel density distribution to the single bandwidth kernel density distribution acquisition unit 13, and outputs the sample set ε _m for calculating the likelihood to the likelihood calculation unit 16. Output to.

The bandwidth set selection processing unit 12 inputs the control signal CTL2 from a control unit (not shown) that performs various controls of the communication status estimation device 100. The bandwidth set selection processing unit 12 holds a plurality of bandwidth sets (sets) for defining a kernel density distribution with a single bandwidth.

Based on the control signal CTL2, the bandwidth set selection processing unit 12 sets the bandwidth (set) for defining the kernel density distribution by a single bandwidth from the set (set) of the retained bandwidth. ) Is selected. Then, the bandwidth set selection processing unit 12 outputs the selected bandwidth set as data b_selected to the single bandwidth kernel density distribution acquisition unit 13 and the average likelihood calculation unit 17.

_{The single bandwidth kernel density distribution acquisition unit 13 obtains a subset ε data} of a sample of QoS data output from the sample division processing unit 11 and a bandwidth set data b_selected output from the bandwidth set selection processing unit 12. input. The single bandwidth kernel density distribution acquisition unit 13 calculates (acquires) the kernel density distribution using each _{bandwidth included in the bandwidth set data b_selected and each sample included in the subset ε data.} Then, the single-bandwidth kernel density distribution acquisition unit 13 outputs the data including the kernel density distribution acquired for each bandwidth to the multi-bandwidth kernel density distribution acquisition unit 15 as _{data D_uni (ε data).}

The weighting coefficient set selection processing unit 14 inputs the control signal CTL3 from a control unit (not shown) that performs various controls of the communication status estimation device 100. The weighting coefficient set selection processing unit 14 holds a plurality of weighting coefficient sets (sets) for weighting and synthesizing (superimposing) the kernel density distribution with a single bandwidth.

The weighting coefficient set selection processing unit 14 selects one weighting coefficient set from the holding weighting coefficient sets (sets) based on the control signal CTL3. Then, the weight coefficient set selection processing unit 14 outputs the selected weight coefficient set as data w_selected to the multi-bandwidth kernel density distribution acquisition unit 15 and the average likelihood calculation unit 17.

The multi-bandwidth kernel density distribution acquisition unit 15 includes data D_uni (ε _data ) output from the single-bandwidth kernel density distribution acquisition unit 13 and weight coefficient set data w_selected output from the weight coefficient set selection processing unit 14. Enter. The multi-bandwidth kernel density distribution acquisition unit 15 uses the weighting coefficient set data w_selected to weight-synthesize (superimpose) the kernel density distribution for each bandwidth included in the _{data D_uni (ε data), and the multi-bandwidth kernel density distribution.} To get. Then, the multi-bandwidth kernel density distribution acquisition unit 15 outputs the data including the acquired multi-bandwidth kernel density distribution to the likelihood calculation unit 16 as _{data D_multi (ε data).}

The likelihood calculation unit 16 inputs the likelihood calculation sample ε _{m of the QoS data output from the sample division processing unit 11 and the data D_multi (ε data} ) output from the multi-bandwidth kernel density distribution acquisition unit 15. do. The likelihood calculation unit 16 calculates the likelihood of the multi-bandwidth kernel density distribution included in the data D_multi (ε _{data ) by using the sample for calculating the likelihood of QoS data ε m.} Then, the likelihood calculation unit 16 outputs the data including the calculated likelihood as data D_Lh _m (ε _data ) to the storage holding unit Mem1.

The storage holding unit Mem1 stores and holds the likelihood data D_Lh _m (ε _data ) output from the likelihood calculation unit 16. Further, the storage holding unit Mem1 reads the stored likelihood data in accordance with the reading command from the average likelihood calculation unit 17, and outputs the read likelihood data to the average likelihood calculation unit 17. Here, when a subset of L divided samples, that is, ε ₁ , ε ₂ , ..., And ε _L are set in the likelihood calculation sample set ε _{m for the average likelihood calculation.} It is assumed that the calculated likelihood data D_Lh ₁ (ε _data ), D_Lh ₂ (ε _data ), ..., D_Lh _L (ε _data ) are used. Then, L likelihood data D_Lh ₁ (ε _data ), D_Lh ₂ (ε _data ), ..., D_Lh _L (ε _data ) are expressed as D_Lh _{1 ... L} (ε _data ).

The average likelihood calculation unit 17 reads out _{L likelihood data D_Lh 1 ... L} (ε _{data) from the storage holding unit Mem1.} Further, the average likelihood calculation unit 17 inputs data including the number of divisions of QoS data num_div (the number of subsets of the sample of QoS data) from the control unit (not shown). Further, the average likelihood calculation unit 17 inputs the data b_selected output from the bandwidth set selection processing unit 12 and the data w_selected output from the weighting coefficient set selection processing unit 14. The average likelihood calculation unit 17 calculates the average value of _{L likelihood data D_Lh 1 ... L} (ε _data). Then, the average likelihood calculation unit 17 collects data including the calculated average likelihood, the bandwidth set data b_selected and the weighting coefficient set data w_selected used when the average likelihood is calculated, as data D_ave_Lh (b_selected). , W_selected) and output to the storage unit Mem2.

The storage unit Mem2 stores the data D_ave_Lh (b_selected, w_selected) output from the average likelihood calculation unit 17. Further, the storage unit Mem2 reads the stored average likelihood data D_ave_Lh (b_selected, w_selected) in accordance with the read command from the optimum parameter acquisition unit 18, and outputs the read average likelihood data to the optimum parameter acquisition unit 18. do.

The optimum parameter acquisition unit 18 reads the average likelihood data from the storage unit Mem2, identifies the maximum average likelihood among the read average likelihood data, and acquires the maximum average likelihood. The bandwidth set b_selected and the weighting coefficient set w_selected are acquired as the optimum bandwidth set b_set_best and the optimum weighting coefficient set w_set_best, respectively. Then, the optimum parameter acquisition unit 18 outputs the data including the optimum bandwidth set b_set_best and the optimum weight coefficient set w_set_best acquired as described above to the communication quality data probability density distribution estimation unit 2 as the optimum parameter prm_opt.

The communication quality data probability density distribution estimation unit 2 inputs the data Din and the optimum parameter prm_opt output from the optimum parameter acquisition unit 18 of the optimum parameter acquisition processing unit 1. The communication quality data probability density distribution estimation unit 2 applies the probability density obtained by applying the multi-bandwidth kernel density distribution specified by the optimum parameter prm_opt to all the samples ε of the QoS data included in the data Din. The distribution is acquired as an estimated probability density distribution for the QoS data. Then, the communication quality data probability density distribution estimation unit 2 outputs the data including the acquired estimated probability density distribution as data Dout.

<1.2: Operation of communication status estimation device>
The operation of the communication status estimation device 100 configured as described above will be described below with reference to the drawings.

FIG. 2 is a flowchart of processing executed by the communication status estimation device 100.

FIG. 3 is a diagram for explaining the processing executed by the communication status estimation device 100.

(Step S1):
In step S1, the sample division processing unit 11 of the optimum parameter acquisition processing unit 1 executes the sample division processing. Specifically, the data Din and the sample of QoS data (for example, throughput) input to the communication status estimation device 100 are divided into L subsets.

(Steps S2 and S3):
In steps S2 and S3, the bandwidth set selection processing unit 12 executes a process of selecting a bandwidth set. The bandwidth set selection processing unit 12 holds a plurality of bandwidth sets (sets) for defining the kernel density distribution by a single bandwidth, and here, the bandwidth set selection processing unit 12 has a bandwidth set selection processing unit 12. As shown in FIG. 3, data of K (K: natural number) bandwidth candidate sets B (first bandwidth candidate set B ⁽¹⁾ to Kth bandwidth candidate set B ^(K) ) are held. It shall be. The i-th bandwidth candidate set ^{is expressed as B (i)} (i: natural number, 1 ≦ i ≦ K), and the number of bandwidth data included in the i-th bandwidth candidate set is expressed as N (i). Then, the N (i) bandwidth data included in the i-th bandwidth candidate set B ⁽ⁱ⁾ _{are referred to as b 1} ⁽ⁱ⁾ , ..., B _{N (i)} ⁽ⁱ⁾ . in short,
B ⁽ⁱ⁾ = {b ₁ ⁽ⁱ⁾ , ..., b _{N (i)} ⁽ⁱ⁾ }
Is.

The bandwidth set selection processing unit 12 selects bandwidth data one by one from ^{the first bandwidth candidate set B (1)} to the Kth bandwidth candidate set B ^{(K), and selects the band.} A set of width data (bandwidth set) is acquired as b_selected. Incidentally, denoted bandwidth data selected from the i bandwidth candidate set ^{B (i)} and ^{_{b j (i) (i)}} . j (i) is the j (i) th bandwidth data included in the i-th bandwidth candidate set B ⁽ⁱ⁾ (j (i): natural number, 1 ≦ j (i) ≦ N (i)). ..

That is, the bandwidth set selection processing unit 12 sets the bandwidth set b_selected = {b _{j (1)} ⁽¹⁾ , ..., b _{j (1)} ^(K) }.
And outputs the bandwidth set b_selected to the single bandwidth kernel density distribution acquisition unit 13. ^{The bandwidth data selected from the first bandwidth candidate set B (1)} to the Kth bandwidth candidate set B ^(K) by the bandwidth set selection processing unit 12 is designated by the control signal CTL2. NS.

(Steps S4, S5):
In steps S4 and S5, the weighting coefficient set selection processing unit 14 executes a process of selecting the weighting coefficient set. The weighting coefficient set selection processing unit 14 holds a plurality of weighting coefficient sets (sets) for weighting and synthesizing (superimposing) the kernel density distribution with a single bandwidth. As shown in FIG. 3, the weighting coefficient set selection processing unit 14 holds data of N (α) (N (α): natural numbers) weighting coefficient sets (the set of weighting coefficient sets is referred to as A). It is assumed that Each weighting factor set contains K weighting coefficients, and the i-th weighting factor set is expressed _{as {α i} ⁽¹⁾ , ..., α _i ^(K)}.

The weighting coefficient set selection processing unit 14 selects a weighting coefficient set of the j (α) th (j (α): natural number, 1 ≦ j (α) ≦ N (α)) according to the control signal CTL3, and selects the selected weight. The coefficient set is acquired as w_selected (= {α _{j (α)} ⁽¹⁾ , ..., α _{j (α)} ^(K) }). Then, the weighting coefficient set selection processing unit 14 outputs the selected weighting coefficient set w_selected to the multi-bandwidth kernel density distribution acquisition unit 15. The weighting coefficient set selected from the N (α) weighting coefficient sets by the weighting coefficient set selection processing unit 14 is designated by the control signal CTL3.

(Steps S6, S7):
In steps S6 and S7, the sample division processing unit 11 performs setting processing of the _{sample set ε m for calculating the likelihood and the sample set ε data} for acquiring the multi-bandwidth kernel density distribution. Specifically, the setting process is executed as follows.

The control unit (not shown) _{generates a control signal CTL1 indicating that the m-th sample set ε m} (m: natural number, 1 ≦ m ≦ L) is designated as the likelihood calculation sample set, and the control signal CTL1 is generated. Is output to the sample division processing unit 11. Here, m = 1.

Then, the sample division processing unit 11 sets the m-th sample set ε _m as the likelihood calculation sample set according to the control signal CTL1, and sets the set of sample sets other than the likelihood calculation sample set as the multi-bandwidth kernel density. Set in the sample set for distribution acquisition ε _data (= {ε ₁ , ε ₂ , ..., ε _m-1 , ε _{m + 1} , ..., ε _L }).

Then, the sample division processing unit 11 outputs the sample set ε _data for acquiring the multi-bandwidth kernel density distribution to the single bandwidth kernel density distribution acquisition unit 13, and outputs the sample set ε _m for calculating the likelihood to the likelihood calculation unit 16. Output to.

κ（ｘ）：ガウスカーネル関数
Ｎ_ｑ：サブセットε_ｄａｔａに含まれるサンプル数
ｂ：ハンド幅
図４〜図６に、ＱｏＳデータのサンプルｑ_１〜ｑ_５を用いてガウスカーネル関数により取得される分布と、当該分布を重ね合わせて取得される単一バンド幅カーネル密度分布とを示す。具体的には、図４の上図は、第１バンド幅候補セットからバンド幅ｂ_１ ^（１）を選択した場合において、ＱｏＳデータのサンプルｑ_１〜ｑ_５を用いてガウスカーネル関数により取得される分布を示している。図４の下図は、第１バンド幅候補セットからバンド幅ｂ_１ ^（１）を選択した場合において、ＱｏＳデータのサンプルｑ_１〜ｑ_５を用いて取得される単一バンド幅カーネル密度分布を示している。なお、図４の上図との関係を分かりやすくするために、図４の下図は、単一バンド幅カーネル密度分布ｐ^（１）（ｑ）をＮ_ｑ倍（図４の場合、Ｎ_ｑ＝５）した分布を示している。 (Step S8):
In step S8, the single bandwidth kernel density distribution acquisition unit 13 is included in the bandwidth set data b_selected (= {b _{j (1)} ⁽¹⁾ , ..., b _{j (1)} ^(K) }). The kernel density distribution is calculated (acquired) using each bandwidth obtained and each sample contained in the subset ε _data. Specifically, a single bandwidth kernel density distribution obtaining unit 13, to the value q _i of QoS data of each sample included in the subset epsilon _data, executes the processing corresponding to the following formula, a kernel density distribution Calculate (acquire).

κ (x): Gaussian kernel function N _q : Number of samples included in the subset ε _data b: Hand width Figures 4 to 6 show the distribution obtained by the Gaussian kernel function using _{samples q 1 to q 5 of QoS data.} And the single bandwidth kernel density distribution obtained by superimposing the distributions. Specifically, the upper figure of FIG. 4 is obtained by the Gauss kernel function using _{samples q 1 to} q ₅ of QoS data when the _{bandwidth b 1} ^{(1) is selected from the first bandwidth candidate set.} The distribution is shown. The lower figure of FIG. 4 shows a single bandwidth kernel density distribution obtained by using _{samples q 1 to} q ₅ of QoS data when _{bandwidth b 1} ^{(1) is selected from the first bandwidth candidate set.} ing. In order to make it easier to understand the relationship with the upper figure of FIG. 4, the lower figure of FIG. 4 shows the single bandwidth kernel density distribution p ⁽¹⁾ (q) _{multiplied by N q (N q} = in the case of FIG. 4). 5) The distribution is shown.

The upper figure of FIG. 5 shows the distribution obtained by the Gauss kernel function using _{samples q 1 to} q ₅ of QoS data when _{bandwidth b 1} ^{(2) is selected from the second bandwidth candidate set.} There is. The lower figure of FIG. 5 shows a single bandwidth kernel density distribution obtained by using _{samples q 1 to} q ₅ of QoS data when _{bandwidth b 1} ^{(2) is selected from the second bandwidth candidate set.} ing. In order to make it easier to understand the relationship with the upper figure of FIG. 5, in the lower figure of FIG. 5, the single bandwidth kernel density distribution p ⁽²⁾ (q) is _{multiplied by N q} (in the case of FIG. 5, N _q = 5) The distribution is shown.

The upper figure of FIG. 6 shows the distribution obtained by the Gauss kernel function using _{samples q 1 to} q ₅ of QoS data when _{bandwidth b 1} ^{(3) is selected from the third bandwidth candidate set.} There is. The lower figure of FIG. 6 shows a single bandwidth kernel density distribution obtained using _{samples q 1 to} q ₅ of QoS data when _{bandwidth b 1} ^{(3) is selected from the third bandwidth candidate set.} ing. In order to make it easier to understand the relationship with the upper figure of FIG. 6, in the lower figure of FIG. 6, the single bandwidth kernel density distribution p ⁽³⁾ (q) is _{multiplied by N q} (in the case of FIG. 6, N _q = 5) The distribution is shown.

The same applies to the single bandwidth kernel density distributions p ⁽⁴⁾ (q) to ^{p (K) (q).} That is, the single bandwidth kernel density distribution acquisition unit 13 uses the QoS data samples (QoS data samples q _{1 to} q _{5 in} the case of FIGS. 4 to 6) to obtain the distribution acquired by the Gaussian kernel function. ^{Single bandwidth kernel density distributions p (4)} (q) to ^{p (K)} (q) are obtained by superimposing and averaging by the number of samples.

Single bandwidth kernel density distribution obtaining unit 13, to the value q _i of QoS data of each sample included in the subset epsilon _data, by processing as described above, a single bandwidth kernel density distribution p ^{(1 )} (Q) to p ^(K) (q) is calculated (acquired). Then, the single bandwidth kernel density distribution acquisition unit 13 ^{collects data including the single bandwidth kernel density distributions p (1)} (q) to ^{p (K)} (q) acquired for each bandwidth as data D_uni (ε). _Data ) is output to the multi-bandwidth kernel density distribution acquisition unit 15.

なお、上記数式において、ｗ＿ｓｅｌｅｃｔｅｄ（＝｛α_ｊ（α） ^（１），・・・，α_ｊ（α） ^（Ｋ）｝の各要素をα^（ｋ）としている。つまり、α^（ｋ）＝α_ｊ（α） ^（ｋ）（ｋ：自然数、１≦ｋ≦Ｋ）である。 (Step S9):
In step S9, the multi-bandwidth kernel density distribution acquisition unit 15 uses the weighting coefficient set data w_selected (= {α _{j (α)} ⁽¹⁾ , ..., α _{j (α)} ^(K) }). Multi-band by weighting and synthesizing (superimposing) the kernel density distribution for each bandwidth included in the data D_uni (ε _data ) (= {p ⁽¹⁾ (q), ..., p ^{(K) (q)})} Get the width kernel density distribution. Specifically, the multi-bandwidth kernel density distribution acquisition unit 15 acquires the multi-bandwidth kernel density distribution p _multi (q) by executing a process corresponding to the following mathematical formula.

In the above formula, each element of _{w_selected (= {α j (α)} ⁽¹⁾ , ..., α _{j (α)} ^{(K) } is α (k)} . That is, α ^(k) = α _{j (α)} ^(k) (k: natural number, 1 ≦ k ≦ K).

FIG. 7 is a diagram for explaining a process of acquiring a _multi- bandwidth kernel density distribution p multi (q) from a single-bandwidth kernel density distribution acquired using _{samples q 1 to} q _{5 of QoS data.} As shown in FIG. 7, the multi-bandwidth kernel density distribution acquisition unit 15 has a single-bandwidth kernel density distribution p ⁽¹⁾ (q), p ⁽²⁾ (q), p ⁽³⁾ (q), ... By weighting and adding ^{p (K) and} (q) with weighting coefficients α ₁ ⁽¹⁾ , α ₁ ⁽²⁾ , α ₁ ⁽³⁾ , ..., α ₁ ^{(K), respectively.} _{Obtain the multi-} bandwidth kernel density distribution p multi (q). As a result, for example, the multi-bandwidth kernel density distribution p _multi (q) as shown in the lower figure of FIG. 7 is acquired.

Multi bandwidth kernel density distribution obtaining unit 15, to the value _{q i} of QoS data of each sample included in the subset epsilon _data, by processing as described above, the multi-bandwidth kernel density distribution _{p multi} (q) Is calculated (acquired). Then, the multi-bandwidth kernel density distribution acquisition unit 15 _{outputs the data including the acquired multi-bandwidth kernel density distribution p multi} (q) to the likelihood calculation unit 16 as _{data D_multi (ε data).}

なお、尤度Ｊ_{ｊ（１），...,ｊ（Ｋ），ｊ（α），ｍ}は、（１）バンド幅のセットデータｂ＿ｓｅｌｅｃｔｅｄ（＝｛ｂ_ｊ（１） ^（１），・・・，ｂ_ｊ（１） ^（Ｋ）｝）と、（２）重み係数セットデータｗ＿ｓｅｌｅｃｔｅｄ（＝｛α_ｊ（α） ^（１），・・・，α_ｊ（α） ^（Ｋ）｝）とから導出したマルチバンド幅カーネル密度分布ｐ_{ｍｕｌｔｉ}（ｑ）に対して、尤度算出用サンプルε_ｍを用いて算出した尤度Ｊであることを示している。 (Step S10):
In step S10, the likelihood calculation unit 16 uses the QoS data likelihood calculation sample ε _m to determine the likelihood of the _multi- bandwidth kernel density distribution p multi (q) contained in the data D_multi (ε _data). calculate. Specifically, the likelihood calculation unit 16 executes a process corresponding to the following mathematical formula on the QoS data (= q) included in the _{sample ε m} for calculating the likelihood of the QoS data to obtain the likelihood J (likelihood J). Calculate the log-likelihood J).

The likelihoods J _{j (1), ..., j (K), j (α), m} are (1) bandwidth set data b_selected (= {b _{j (1)} ⁽¹⁾ , ...・, B _{j (1)} ^(K) }) and (2) Weight coefficient set data w_selected (= {α _{j (α)} ⁽¹⁾ , ..., α _{j (α)} ^(K) }) It is shown that the likelihood J is calculated by using the likelihood calculation sample ε _m with respect to the derived multi-bandwidth kernel density distribution p _{multi (q).}

FIG. 8 is a diagram for explaining the likelihood calculation process.

The likelihood calculation unit 16 is, for example, as shown in FIG. 8, a sample of QoS data ε _data (for example, in the case of FIG. 7, ε _data = {q ₁ , q ₂ , q ₃ , q ₄ , q ₅ }). (1) Bandwidth set data b_selected (= {b ₁ ⁽¹⁾ , ..., b ₁ ^(K) }) and (2) Weight coefficient set data w_selected (= {α ₁ ^{(1)) )} , ..., α ₁ ^(K) }) for the multi-bandwidth kernel density distribution p _multi (q), the sample for calculating the likelihood ε _m (for example, in the case of FIG. 8, ε _m = The likelihood J is calculated using {q _m1 , q _m2 , q _m3 , q _m4 , q _m5}). That is, in the case of FIG. 8, the likelihood calculation unit 16 is
J = log (P _multi (q _m1 )) + log (P _multi (q _m2 )) + log (P _multi (q _m3 )) + log (P _multi (q _m4 )) + log (P _multi (q _m5 ))
To calculate the likelihood J.

The likelihood calculation unit 16 calculates (acquires) the likelihood J by processing the QoS data of each sample included in the _{likelihood calculation sample ε m as described above.} Then, the likelihood calculation unit 16 outputs the acquired data including the likelihood J as data D_Lh _m (ε _data ) to the storage holding unit Mem1. Then, the storage holding unit Mem1 stores the data D_Lh _m (ε _data ).

(Step S11):
In step S11, when all of the L sample sets are set to the likelihood calculation sample set ε _m and it is determined that the processes of steps S7 to S10 have not been executed (when m <L). The process returns to step S6. On the other hand, when all of the L sample sets are set to the likelihood calculation sample set ε _m and it is determined that the processes of steps S7 to S10 have been executed (when m = L), the process is performed in step S12. Proceed to.

That is, in the loop processing of steps S6 to S11, the likelihood calculation sample set ε _m and the multi-bandwidth kernel density distribution acquisition sample set ε _data are used with m = 1, 2, ..., L. , The processing of steps S7 to S10 is executed L times. As a result, in the communication status estimation device 100, L likelihood Js acquired as m = 1, 2, ..., L are acquired, and the L likelihoods J are stored in the storage holding unit Mem1. NS. The change of m is executed by the sample division processing unit 11 according to the control signal CTL1.

なお、
なお、平均尤度Ｊａｖｅ_{ｊ（１），...,ｊ（Ｋ），ｊ（α）}は、（１）バンド幅のセットデータｂ＿ｓｅｌｅｃｔｅｄを｛ｂ_ｊ（１） ^（１），・・・，ｂ_ｊ（１） ^（Ｋ）｝）とし、（２）重み係数セットデータｗ＿ｓｅｌｅｃｔｅｄを｛α_ｊ（α） ^（１），・・・，α_ｊ（α） ^（Ｋ）｝としたときに、算出される平均尤度Ｊａｖｅであることを示している。 (Step S12):
In step S12, the average likelihood calculation unit 17 reads out _{L likelihood data D_Lh 1 ... L} (ε _data ) (data including L likelihood J) from the storage holding unit Mem1. Further, the average likelihood calculation unit 17 inputs data including the number of divisions of QoS data num_div (the number of subsets of the sample of QoS data) from the control unit (not shown). Further, the average likelihood calculation unit 17 inputs the data b_selected output from the bandwidth set selection processing unit 12 and the data w_selected output from the weighting coefficient set selection processing unit 14. The average likelihood calculation unit 17 calculates the average value of _{L likelihood data D_Lh 1 ... L} (ε _data). That is, the average likelihood calculation unit 17 calculates the average value Jav (average likelihood _{Jav) of L likelihood data D_Lh 1 ... L} (ε _data ) by executing the process corresponding to the following mathematical formula. do.

note that,
In addition, the average likelihood Jav _{j (1), ..., j (K), j (α)} sets (1) bandwidth set data b_selected to {b _{j (1)} ⁽¹⁾ , ..., b _{j (1)} ^(K) }), and (2) Calculated when the weighting coefficient set data w_selected is {α _{j (α)} ⁽¹⁾ , ..., α _{j (α)} ^(K) }. It shows that the average likelihood is Jave.

Then, the average likelihood calculation unit 17 collects data including the calculated average likelihood, the bandwidth set data b_selected and the weighting coefficient set data w_selected used when the average likelihood is calculated, as data D_ave_Lh (b_selected). , W_selected) and output to the storage unit Mem2.

(Step S13):
If all the weighting coefficient sets are selected in step S13 and it is determined that the processing of steps S5 to S12 has not been executed, the processing is returned to step S4. On the other hand, when it is determined that all the weight coefficient sets have been selected and the processes of steps S5 to S12 have been executed, the process proceeds to step S14.

That is, in the loop processing of steps S5 to S13, the processing of steps S5 to S12 is executed for each of the N (α) weight coefficient sets included in the set A of the weighting coefficient sets, and the average likelihood is acquired. NS. As a result, in the communication status estimation device 100, the average likelihood Jave (N (α)) average likelihood when each of the N (α) weight coefficient sets included in the set A of the weight coefficient sets A is selected. Jav) is acquired, and the N (α) average likelihoods J are stored in the storage unit Mem2. Which of the N (α) weight coefficient sets included in the set of weight coefficient sets A is selected is determined by the weighting coefficient set selection processing unit 14 according to the control signal CTL3.

(Step S14):
If all the bandwidth sets are selected in step S14 and it is determined that the processes of steps S3 to S13 have not been executed, the process is returned to step S2. On the other hand, when all the bandwidth sets are selected and it is determined that the processes of steps S3 to S13 have been executed, the process proceeds to step S15.

That is, in the loop processing of steps S3 to S13, for N (1) to N (K) ^{bandwidths included in the sets B (1) to} B ^{(K) of K bandwidth sets, respectively.} Then, the processes of steps S3 to S13 are executed, respectively, and the average likelihood is acquired. in short,
(1) One bandwidth is selected from N (1) bandwidths included in the ^{set B (1), and one bandwidth is selected.}
(2) Select one bandwidth from the N (2) bandwidths included in the ^{set B (2), and}
・ ・ ・
(K) Select one bandwidth from the N (K) bandwidths included in the ^{set B (K), and select one.}
The processes of steps S3 to S13 are executed and the average likelihood is acquired. That is, the processes of steps S3 to S13 are executed N (1) × N (2) × ・ ・ ・ × N (K) times, and N (1) × N (2) × ・ ・ ・ × N (K) × The average likelihood of N (α) is obtained. Note that N (α) is the number of weighting coefficient sets included in the set of weighting coefficient sets A.

In the communication status estimation device 100, the average likelihoods of N (1) × N (2) × ... × N (K) × N (α) acquired by the above processing are stored in the storage unit Mem2. The bandwidth set selection processing unit 12 determines which bandwidth is selected from the sets B ^{(1) to} B ^{(K) of the K bandwidth sets according to the control signal CTL2.}

つまり、最適パラメータ取得部１８は、上記数式に相当する処理を実行し、平均尤度Ｊａｖｅが最大となる、最適なバンド幅の組み合わせ（最適バンド幅セットｂ＿ｓｅｔ＿ｂｅｓｔ）、および、最適な重み係数セット（最適重み係数セットｗ＿ｓｅｔ＿ｂｅｓｔ）を
ｂ＿ｓｅｔ＿ｂｅｓｔ＝｛ｂ_{ｊｏｐｔ（１）} ^（１），・・・，ｂ_{ｊｏｐｔ（Ｋ）} ^（Ｋ）｝
ｗ＿ｓｅｔ＿ｂｅｓｔ＝｛α_{ｊｏｐｔ（α）} ^（１），・・・，α_{ｊｏｐｔ（α）} ^（Ｋ）｝
として、取得する。 (Step S15):
In step S15, the optimum parameter acquisition unit 18 receives the average likelihood data (N (1) × N (2) × ・ ・ ・ × N (K) ×” acquired from the storage unit Mem2 by the processing of steps S2 to S13. N (α) average likelihoods) are read out, and the maximum average likelihood is specified in the read average likelihood data. Then, the optimum parameter acquisition unit 18 acquires the bandwidth set b_selected and the weighting coefficient set w_selected when the maximum average likelihood is acquired as the optimum bandwidth set b_set_best and the optimum weighting coefficient set w_set_best, respectively. .. That is, the optimum parameter acquisition unit 18 executes the process corresponding to the following mathematical formula, and specifies the optimum bandwidth combination and the optimum weighting coefficient set that maximizes the average likelihood Jav.

That is, the optimum parameter acquisition unit 18 executes the process corresponding to the above formula, and the optimum bandwidth combination (optimal bandwidth set b_set_best) and the optimum weighting coefficient set (optimal bandwidth set b_set_best) that maximizes the average likelihood Jav. Optimal weight coefficient set w_set_best) is set to b_set_best = {b _{fit (1)} ⁽¹⁾ , ..., b _{fit (K)} ^(K) }
w_set_best = {α _{Jopt (α)} ⁽¹⁾ , ..., α _{Jopt (α)} ^(K) }
To get as.

Then, the optimum parameter acquisition unit 18 outputs the data including the optimum bandwidth set b_set_best and the optimum weight coefficient set w_set_best acquired as described above to the communication quality data probability density distribution estimation unit 2 as the optimum parameter prm_opt.

(Step S16):
In step S16, the communication quality data probability density distribution estimation unit 2 inputs the data Din and the optimum parameter prm_opt output from the optimum parameter acquisition unit 18 of the optimum parameter acquisition processing unit 1. The communication quality data probability density distribution estimation unit 2 applies the probability density obtained by applying the multi-bandwidth kernel density distribution specified by the optimum parameter prm_opt to all the samples ε of the QoS data included in the data Din. The distribution is acquired as an estimated probability density distribution for the QoS data. That is, the communication quality data probability density distribution estimation unit 2 sets the optimum bandwidth set b_set_best (= {b _{jopt (1)} ⁽¹⁾ , ..., b _{jopt (K)} ^{(K) for all the samples ε of the QoS data.} Applying the multi-bandwidth kernel density distribution specified by ^K) }) and the optimal weighting coefficient set w_set_best (= {α _{Jopt (α)} ⁽¹⁾ , ..., α _{Jopt (α)} ^(K)}). The probability density distribution acquired in is acquired as an estimated probability density distribution for QoS data.

Then, the communication quality data probability density distribution estimation unit 2 outputs the data including the acquired estimated probability density distribution as data Dout.

By processing as described above, the communication status estimation device 100 probabilities of highly accurate communication quality data (QoS data) even when the communication environment status fluctuates in a complicated and dynamic manner such as in a narrow space such as a factory. The density distribution can be obtained (estimated).

In the communication status estimation device 100, the sample of QoS data is divided into L, and the sample set for acquiring the multi-bandwidth kernel density distribution ε _data (= {ε ₁ , ε ₂ , ..., Ε _m-1 , ε _{m + 1} , ...・ ・, Ε _L }) and the sample set ε _m for calculating the likelihood are set.

Then, in the communication status estimation device 100, a sample set ε _data for acquiring a multi-bandwidth kernel density distribution is applied, and a plurality of bandwidth candidate sets (sets of bandwidths) B ⁽¹⁾ to each include a plurality of bandwidths. Multiple single bandwidth kernel density distributions specified by the bandwidth set b_selected (= {b _{j (1)} ⁽¹⁾ , ..., b _{j (1)} ^(K) }) selected from B ^(K). Weight coefficient set (set of weight coefficient sets) The weight coefficient set data w_selected selected from A is superimposed by {α _{j (α)} ⁽¹⁾ , ..., α _{j (α)} ^(K) } (weight addition). By doing so, the multi-bandwidth kernel density distribution p _multi (q) is obtained.

Then, the communication status estimation device 100 calculates the likelihood of the multi-bandwidth kernel density distribution p _multi (q) _{by using the likelihood calculation sample set ε m.} Further, in the communication status estimation device 100, the process of calculating the likelihood is _{performed for L likelihood calculation sample sets by changing the likelihood calculation sample set ε m} , and L likelihood Js are obtained. calculate. Then, the communication status estimation device 100 calculates the average likelihood Jav of the L likelihoods.

The communication status estimation device 100 performs the above processing on all the bandwidths included in the bandwidth candidate set and all the weighting coefficient sets included in the set of weighting coefficient sets, and calculates the average likelihood J for each. Then, in the communication status estimation device 100, the optimum parameter acquisition unit 18 executes a process corresponding to the following mathematical formula to obtain an optimum bandwidth combination and an optimum weighting coefficient set that maximizes the average likelihood Jav. Identify and get the optimal parameters.

Then, in the communication status estimation device 100, the optimum bandwidth set b_set_best (= {b _{jopt (1)} ⁽¹⁾ , ..., b _{jopt (K)} ^(K) }) is applied to all the samples ε of the QoS data. And obtained by applying the multi-bandwidth kernel density distribution specified by the optimal weighting coefficient set w_set_best (= {α _{Jopt (α)} ⁽¹⁾ , ..., α _{Jopt (α)} ^(K)}). The probability density distribution is acquired as an estimated probability density distribution for QoS data.

In the communication status estimation device 100, as described above, the single bandwidth kernel density distribution acquired by a plurality of different bandwidths is superposed by various weighting coefficients, and the QoS data is obtained by using the multi-bandwidth kernel density distribution. Since the probability density distribution is acquired (estimated), it is possible to estimate with high accuracy regardless of the probability density distribution of the QoS data. For example, regarding the probability density distribution of QoS data, which tends to occur in a narrow space such as a factory, where the QoS data is concentrated on a specific value (a specific narrow range) and the QoS data is widely dispersed. The communication status estimation device 100 can estimate with high accuracy.

Further, the communication status estimation device 100 identifies the optimum bandwidth combination and the optimum weighting coefficient set in consideration of the likelihood calculated from the QoS data sample, and acquires the probability density distribution of the QoS data (the probability density distribution of the QoS data). Therefore, regardless of the probability density distribution of the QoS data, the probability density distribution of the QoS data can be estimated with high accuracy.

In this way, the communication status estimation device 100 acquires the probability density distribution of highly accurate communication quality data (QoS data) even when the communication environment status fluctuates in a complicated and dynamic manner such as in a narrow space such as a factory (QoS data). Can be estimated). That is, the communication status estimation device 100 can accurately determine the status of communication quality data (for example, the probability density distribution of communication quality data) even when the communication environment status fluctuates in a complicated and dynamic manner such as in a narrow space such as a factory. Can be predicted (estimated).

Here, as an example of one simulation, a case where the probability density distribution of the communication quality data (QoS data) is acquired (estimated) by the communication status estimation device 100 will be described for the wireless communication status in the automobile factory. The wireless communication status in the automobile factory of this simulation example is as follows.
(1) A production line is located in the automobile factory, one camera for inspection is installed, three stations that generate periodic traffic are installed, and traffic is randomly generated. One station is installed.
(2) On the production line, the vehicle body (production target) arrives at random and the production work is carried out.
(3) One inspection camera, three stations that generate periodic traffic, and one station that randomly generates traffic use the same channel for wireless LAN communication (wireless by IEEE802.11g). Communication).
(4) The arrival interval of the vehicle body is 40 seconds on average, and the inspection time of one vehicle body is set to 50 seconds.
(5) The inspection camera captures a 3M pixel grayscale image and captures 30 images of one vehicle body.

FIG. 9 shows actual measurement data on the wireless communication status in the automobile factory, estimation result data based on the single bandwidth kernel density distribution, and estimation result data based on the multi-bandwidth kernel density distribution (communication status estimation device) in the above simulation example. It is a graph which shows (data acquired by 100). In FIG. 9, the horizontal axis represents QoS data (throughput), and the vertical axis represents the probability density on a logarithmic scale.

In this simulation example, in the communication status estimation device 100, the probability density distribution of the communication quality data (QoS data) was acquired (estimated) by the following settings.
≪Settings for multi-bandwidth kernel density distribution acquisition≫
(1) Kernel function: Gaussian function (2) Number of sample divisions L = 10
(3) Number of bandwidths K = 2
(4) First bandwidth candidate set B ⁽¹⁾ = {1000, 1274, 1624, 2069, 2637, 3360, 4281, 5456, 6952, 8859} (unit is bps)
(5) Second bandwidth candidate set B ⁽²⁾ = {11288, 14384, 18330, 23357, 29764, 37927, 48329, 61585, 78476,100,000} (unit is bps)
(6) Weight coefficient set A = {(0.05, 0.95), (0.15, 0.85), (0.25, 0.75), (0.35, 0.65), ( 0.45,0.55), (0.55,0.45), (0.65,0.35), (0.75,0.25), (0.85,0.15), ( 0.95,0.05)}
In addition, as a comparative example, the kernel density distribution with a single bandwidth was obtained with the following settings.
≪Settings for single bandwidth kernel density distribution acquisition≫
(1) Kernel function: Gaussian function (2) Number of sample divisions L = 10
(3) Bandwidth candidate set = {1000, 1274, 1624, 2069, 2637, 3360, 4281, 5456, 6952, 8859, 11288, 14384, 18330, 23357, 29764, 37927, 48329, 61585, 78476, 100,000} ( Unit is bps)
As can be seen from FIG. 9, the probability density distribution (multi-bandwidth kernel density distribution) (data shown by the solid line) of the QoS data (throughput) acquired by the communication status estimation device 100 is the measured data (data shown by the dotted line). ), And the probability density distribution (multi-bandwidth kernel density distribution) of QoS data (throughput) can be estimated with high accuracy by the communication status estimation device 100. In particular, the probability density distribution can be estimated with high accuracy in the region where the peaks of regions R1 to R3 in FIG. 9 exist (regions showing spike-like characteristics). On the other hand, in the single bandwidth kernel density distribution (data shown by the alternate long and short dash line), the error from the measured data is large as a whole, and in particular, the peaks (spike-like characteristics) of the regions R1 to R3 in FIG. 9 are large. The error is large in the region where the region indicating) exists.

As described above, in the communication status estimation device 100, the probability density of the QoS data is high even in a communication environment in which the QoS data (throughput) has peaks at a plurality of locations and the QoS data is distributed in a large probability density distribution. The distribution can be estimated with high accuracy.

Further, in the communication status estimation device 100, the probability density distribution of the QoS data is estimated by the multi-bandwidth kernel density distribution, so that the probability density distribution of the QoS data can be highly accurate even when the number of samples of the QoS data is small. Can be estimated to. Since the communication status estimation device 100 uses a multi-bandwidth kernel density distribution in which single bandwidth kernel density distributions acquired with different bandwidths (for example, narrow bandwidth and wide bandwidth) are superimposed by weighting coefficients, QoS is used. Even when the data is distributed over a wide area (a wide definition area of QoS data), the probability density distribution of QoS data can be appropriately estimated.

Ｐ（ｍ）：実測値
Ｐ_ｅｓｔ（ｍ）：推定値
図１０のグラフから分かるように、マルチバンド幅カーネル密度分布による推定データ（通信状況推定装置１００で取得される推定データ）は、サンプル数が少ない領域においても二乗平均平方根誤差（ＲＭＳＥ）が小さい。つまり、通信状況推定装置１００では、サンプル数が少ない場合であっても、高精度に、ＱｏＳデータの確率密度分布を取得することができる。 FIG. 10 shows an error between the number of samples and the measured data in the estimation data based on the single bandwidth kernel density distribution and the estimation data based on the multi-bandwidth kernel density distribution (estimated data acquired by the communication status estimation device 100). It is a graph which shows the relationship of. In the graph of FIG. 10, the horizontal axis is the number of samples, and the vertical axis is the root mean square error (RMSE). The root mean square error RMSE is obtained by the following formula.

P (m): Measured value _Best (m): Estimated value As can be seen from the graph in FIG. 10, the estimated data based on the multi-bandwidth kernel density distribution (estimated data acquired by the communication status estimation device 100) is the number of samples. The root mean square error (RMSE) is small even in the region where there is little. That is, the communication status estimation device 100 can acquire the probability density distribution of QoS data with high accuracy even when the number of samples is small.

[Other Embodiments]
In the above embodiment, the case where the multi-bandwidth kernel density distribution is acquired by using the Gaussian kernel function in the communication status estimation device 100 has been described, but the present invention is not limited to this. For example, in the communication status estimation device 100, the single bandwidth kernel density distribution is calculated by using another function whose integral value in all sections is "1", and the calculated single bandwidth kernel density distribution is weighted. The multi-bandwidth kernel density distribution may be obtained by superimposing the above. FIG. 11 shows an example of a function in which the integral value in the entire interval (assuming that the horizontal axis is x, the region of x (domain) [−∞: ∞] (all intervals)) is “1”. FIG. 11A shows three Gaussian kernel functions (functions func _{21 to func 23} ) (functions that realize a Gaussian distribution) having different bandwidths (standard deviation σ). FIG. 11B is a function example (functions fun _{21 to func 23} ) showing a polygonal line (triangular shape) distribution having a maximum value at x = 0, where the horizontal axis is the x-axis. _{FIG. 11C is a function example (functions func 31 to func 33} ) showing a distribution of a line-symmetrical shape (rectangular shape) with respect to x = 0, where the horizontal axis is the x-axis.

In the communication status estimation device 100, the single bandwidth kernel density is used by using the functions (functions having different hand widths) as shown in FIGS. 11 (b) and 11 (c) in which the integrated value in all sections is "1". The multi-bandwidth kernel density distribution may be obtained by calculating the distribution and superimposing the calculated single-bandwidth kernel density distribution by the weighting coefficient.

Further, in the above embodiment, the method of setting the bandwidth candidate is not particularly mentioned, but when the bandwidth candidate can predict the tendency of the probability density distribution of the QoS data. , The tendency (property of probability density distribution) may be taken into consideration when setting. For example, when the probability density of QoS data increases exponentially in a predetermined region (a predetermined range of a domain), bandwidth candidates may be set to increase exponentially.

Further, in the above embodiment, the method of setting the candidate of the weighting coefficient set is not particularly mentioned, but the degree of randomness becomes high so that the multi-bandwidth kernel density distribution realized by the weighting coefficient set becomes diverse. Candidates for the weighting factor set may be set as such (eg, close to a uniform distribution).

Further, in the above embodiment, the communication status estimation device 100 divides the sample of QoS data into L pieces, sets all the divided samples in the sample set for likelihood calculation, and calculates the likelihood, that is, The case where the L-likelihood is calculated and the average likelihood is calculated has been described, but the present invention is not limited to this. For example, in the communication status estimation device 100, in order to reduce the amount of calculation, the likelihood calculation process is thinned out, the likelihood calculation process is performed several times less than L, and the average value of the likelihoods acquired by the process is averaged. As the likelihood, the process may be executed.

Further, in the communication status estimation device 100 described in the above embodiment, each block may be individually integrated into one chip by a semiconductor device such as an LSI, or may be integrated into one chip so as to include a part or all of the blocks. good.

Although it is referred to as LSI here, it may be referred to as IC, system LSI, super LSI, or ultra LSI depending on the degree of integration.

Further, the method of making an integrated circuit is not limited to the LSI, and may be realized by a dedicated circuit or a general-purpose processor. An FPGA (Field Programmable Gate Array) that can be programmed after the LSI is manufactured, or a reconfigurable processor that can reconfigure the connection and settings of circuit cells inside the LSI may be used.

In addition, a part or all of the processing of each functional block of each of the above embodiments may be realized by a program. Then, a part or all of the processing of each functional block of each of the above embodiments is performed by the central processing unit (CPU) in the computer. Further, the program for performing each process is stored in a storage device such as a hard disk or a ROM, and is read and executed in the ROM or the RAM.

Further, each process of the above embodiment may be realized by hardware, or may be realized by software (including a case where it is realized together with an OS (operating system), middleware, or a predetermined library). Further, it may be realized by mixed processing of software and hardware.

For example, when each functional unit of the above embodiment is realized by software, the hardware configuration shown in FIG. 12 (for example, a hardware configuration in which a CPU, ROM, RAM, input unit, output unit, etc. are connected by a bus Bus). May be used to realize each functional unit by software processing.

Further, the execution order of the processing methods in the above-described embodiment is not necessarily limited to the description of the above-described embodiment, and the execution order can be changed without departing from the gist of the invention.

A computer program that causes a computer to perform the above-mentioned method and a computer-readable recording medium on which the program is recorded are included in the scope of the present invention. Here, examples of computer-readable recording media include flexible disks, hard disks, CD-ROMs, MOs, DVDs, DVD-ROMs, DVD-RAMs, large-capacity DVDs, next-generation DVDs, and semiconductor memories. ..

The computer program is not limited to the one recorded on the recording medium, and may be transmitted via a telecommunication line, a wireless or wired communication line, a network represented by the Internet, or the like.

The specific configuration of the present invention is not limited to the above-described embodiment, and various changes and modifications can be made without departing from the gist of the invention.

100 Communication status estimation device 1 Optimal parameter acquisition processing unit 2 Communication quality probability density distribution estimation unit

Claims (6)

Obtain a multi-bandwidth kernel density distribution by superimposing a single-bandwidth kernel density distribution defined by multiple bandwidth kernel functions based on parameters in order to estimate the probability density distribution for data related to communication quality. In addition, when estimating the probability density distribution of the data related to the communication quality from the multi-bandwidth kernel density distribution, the multi-bandwidth kernel density distribution was applied to a plurality of samples of the data related to the communication quality. The optimum parameter acquisition processing unit that acquires the optimum parameter, which is the parameter with the maximum probability at times,
The probability density distribution obtained by applying the multi-bandwidth kernel density distribution set by the optimum parameters to a plurality of samples of the communication quality data is the estimated probability density distribution for the communication quality data. Communication quality data to be acquired as Probability density distribution estimation unit and
A communication status estimation device including. The optimum parameter acquisition processing unit
It holds multiple bandwidth candidates to identify the single bandwidth kernel density distribution.
It holds a plurality of candidates for weighting factors when acquiring the multi-bandwidth kernel density distribution by superimposing the single-bandwidth kernel density distribution.
Likelihoods are calculated for all combinations of the plurality of bandwidth candidates and the plurality of weighting coefficient candidates, and the set of the bandwidth and the weighting coefficients that maximizes the calculated likelihood is used as the optimum parameter. get,
The communication status estimation device according to claim 1. The optimum parameter acquisition processing unit
The multi-bandwidth kernel density distribution is acquired using the single-bandwidth kernel density distribution defined by kernel functions having different bandwidths and the integral value in all sections of the domain is "1". ,
The communication status estimation device according to claim 1 or 2. The kernel function is at least one of a Gaussian function, a function that realizes a triangular shape distribution, and a function that realizes a rectangular shape distribution.
The communication status estimation device according to claim 3. Obtain a multi-bandwidth kernel density distribution by superimposing a single-bandwidth kernel density distribution defined by multiple bandwidth kernel functions based on parameters in order to estimate the probability density distribution for data related to communication quality. In addition, when estimating the probability density distribution of the data related to the communication quality from the multi-bandwidth kernel density distribution, the multi-bandwidth kernel density distribution was applied to a plurality of samples of the data related to the communication quality. The optimum parameter acquisition processing step to acquire the optimum parameter, which is the parameter with the maximum probability at times, and
The probability density distribution obtained by applying the multi-bandwidth kernel density distribution set by the optimum parameters to a plurality of samples of the communication quality data is the estimated probability density distribution for the communication quality data. Communication quality data to acquire as the probability density distribution estimation step and
Communication status estimation method. A program for causing a computer to execute the communication status estimation method according to claim 5.

Images