JP6032469B2 - Source estimation method and source estimation apparatus using the same - Google Patents

Description

  The present invention relates to a transmission source estimation technique, and relates to a transmission source estimation method for estimating an unknown transmission source and a transmission source estimation apparatus using the same.

  In recent years, with the downsizing and high performance of mobile terminals, there are growing concerns about unauthorized use of mobile terminals in places where fairness such as entrance examinations is expected. First, a method of blocking radio waves can be considered as the simplest method for making a mobile terminal unusable. In this method, it is conceivable to block the radio wave by laying a radio wave absorber around the test site, and it is already possible with the current technology. However, since the price of the radio wave absorber is relatively high and it is necessary to cover all the test venues with the radio wave absorber for each test, it is not practical in terms of cost and labor. Subsequently, a method of making it impossible to perform communication itself by issuing jamming radio waves can be considered. However, this is also not practical in terms of labor, as is the case with radio wave absorbers, and there may be operational and legal problems such as obstructing communication of unrelated portable terminals.

  Such a method of directly obstructing communication is not practical because various problems are assumed in practice. Thus, as a method for solving these problems, a passive approach is considered in which communication is performed by detecting a transmission radio wave of a mobile terminal, which is useful for preventing fraud. As a simple implementation method of this approach, there is a method for detecting whether or not communication by a portable terminal is performed in a test venue. In this system, received power is monitored for each communication band of a mobile terminal using a sensor, and it is determined whether or not the mobile terminal is used in a test venue. With this method, the cost and the trouble of operation that are problematic due to the above-described communication interruption can be eliminated. However, it is highly possible that other examinees who have not committed fraud will suffer a disadvantage simply by determining whether or not communication is performed by a portable terminal in the test venue, and this is insufficient for preventing fraud.

  In view of this, a system for identifying the position of a mobile terminal that has performed an illegal act in a test venue can be considered. In this system, since estimation targets are arranged at high density in an indoor environment such as a test venue, position estimation accuracy of about 1 m is required. In recent years, high-accuracy position estimation in an indoor environment has attracted a great deal of attention and has been actively studied. In the existing position estimation method, as a representative, received signal power strength (Received Signal Strength Indicator: RSSI) (for example, see Non-Patent Document 1), radio wave arrival time difference (Time Difference of Arrival: TDOA) (for example, non-patent) There is a position estimation method using a radio wave arrival angle (Direction of Arrival: DOA) (see, for example, Non-Patent Document 3) and a position fingerprint.

  RSSI is the most widely used indoor position estimation method because it is easy to measure and can be used without existing hardware even with existing hardware. In this method, the distance between the portable terminal sensors is obtained using the distance propagation loss between the portable terminal and the receiving sensor, and the position is estimated geometrically by a combination of estimated distances from at least three sensors. However, since the received power is greatly affected by noise and fading, it is known that position estimation by RSSI is relatively inaccurate. In TDOA, the distance between mobile terminal sensors is calculated from the radio wave arrival time difference between sensors in the same manner as RSSI, and the position is estimated geometrically by a combination of estimated distances from at least three sensors.

N. Patwari etc, “Relative Location Estimation in Wireless Sensor Networks”, IEEE Trans. Signal Processing, Aug. 2003, 51, 8, p. 2137-2148 Z. Merhi, M. Elgamel and M. Bayoumi, “A Lightweight Collaborative Fault Tolerant Target Localization System for Wireless Sensor Networks”, IEEE Trans. Mobile Computing, Dec 2009, 8, 12, p. 1690-1704 C.-C. Lim, Y. Wan, B.-P. Ng, and C.-MS See, `` A realtime indoor WiFi localization system using smart antenna '', IEEE Trans. Consum. Electron., May. 2007, 53 , 2, p. 618-622

  TDOA is said to have higher noise immunity than RSSI and is capable of more accurate position estimation, but it is difficult to accurately capture the arrival time of radio waves due to the effects of indoor multipath environments. This also degrades the position estimation accuracy. DOA uses an array antenna with multiple antennas mounted on a single sensor, estimates the radio wave arrival angle from the received signal phase difference between the antennas, and estimates the position based on the intersection of the estimated directions from at least two sensors. It is a technique. DOA is known to have strong noise immunity due to clever calculation methods such as the MUSIC method, but it is affected by the indoor multipath environment as well as TDOA, and in order to improve angular resolution, it has a large scale with many antennas. This is unsuitable for indoor applications due to the need for such sensors. As described above, in the existing position estimation method, since the multipath environment of the indoor environment deteriorates the estimation accuracy, the accuracy is insufficient as the position estimation for detecting the fraud.

  On the other hand, a position fingerprinting method using the fact that the propagation path response including the multipath is a function of the position is the position fingerprint method. Literally, the position fingerprinting method is a technique for collecting a multipath environment unique to each estimated position as a position fingerprint and estimating the position based on the similarity between the fingerprint and the signal to be estimated. There is a fingerprint using multi-pass DOA information as the fingerprint, and high-precision estimation is possible when the estimation target terminal exists at the position where the fingerprint is collected. However, with this location estimation method, it is difficult to accurately estimate if there is no terminal at the location where the location fingerprint was obtained. Have difficulty.

  This invention is made | formed in view of such a condition, The objective is to provide the technique which suppresses the deterioration of estimation precision, suppressing the increase in a processing amount.

  In order to solve the above-described problem, a transmission source estimation apparatus according to an aspect of the present invention is a known transmission signal transmitted from a transmission source arranged in the vicinity of any one of a plurality of discrete candidate positions. The sensor is received by each of the plurality of sensors arranged at the position, and obtains a reception signal from each of the plurality of sensors, and derives a cross-correlation between the sensors with respect to the reception signal acquired by the acquisition unit. A first deriving unit for deriving an expected value of the absolute value of the cross-correlation, a replica for the expected value derived by the first deriving unit and a replica for the candidate position, and an expected value derived by the first deriving unit A second derivation unit for deriving the error of the second derivation unit, and the second derivation unit by repeatedly executing the process while changing candidate positions in the second derivation unit. A plurality of error, comprising a selector for selecting one of the error, and an output unit for the candidate position corresponding to the selected error in the selection unit, and outputs the source is installed position. The replica used in the second derivation unit is acquired in advance for each candidate position.

  According to this aspect, since the position of the transmission source is estimated based on replicas for each of a plurality of discrete candidate positions, it is possible to suppress deterioration in estimation accuracy while suppressing an increase in processing amount.

  Another aspect of the present invention is a source estimation method. In this method, a transmission signal transmitted from a source arranged near any one of a plurality of discrete candidate positions is received by each of a plurality of sensors arranged at different known positions, Obtaining a received signal from each of a plurality of sensors, deriving a cross-correlation between the sensors for the obtained received signal, deriving an expected value of an absolute value of the cross-correlation, and A step of deriving an error between the replica for the candidate position and the derived expected value, and the step of deriving the error while changing the candidate position, and being derived by repeatedly executing the process The step of selecting one error from multiple errors, and the candidate position corresponding to the selected error, the position where the transmission source is installed And a step of then output. The replica used in the step of deriving the error is acquired in advance for each candidate position.

  It should be noted that any combination of the above-described constituent elements and a conversion of the expression of the present invention between a method, an apparatus, a system, a recording medium, a computer program, etc. are also effective as an aspect of the present invention.

  According to the present invention, it is possible to suppress an increase in estimation accuracy while suppressing an increase in processing amount.

It is a figure which shows the structure of the DSM system used as the premise technique of this invention. It is a figure which shows the process sequence of the DSM system of FIG. It is a figure which shows the frequency utilization condition mapping by the DSM system of FIG. It is a figure which shows the position estimation in the indoor environment which concerns on the Example of this invention, and the image of a feature-value. It is a figure which shows the process sequence of the detection system which concerns on the Example of this invention. It is a figure which shows the origin of the indoor experiment with respect to the detection system of FIG. It is a figure which shows the experimental environment with respect to the detection system of FIG. It is a figure which shows the resource block of LTE used in the detection system of FIG. It is a figure which shows the cross correlation between antennas by the reference signal of LTE used in the detection system of FIG. It is a figure which shows the comparison with the detection system of FIG. 5, and the conventional method. It is a figure which shows the average error range of the detection system of FIG. It is a figure which shows the structure of the transmission source estimation apparatus which concerns on the Example of this invention. FIGS. 13A to 13B are diagrams showing the configuration of a press used in the modification of the present invention.

  Before describing the present invention specifically, an outline will be given first. An embodiment of the present invention relates to a transmission source estimation apparatus that performs simple and highly accurate position estimation in an indoor environment using a plurality of sensors. Since the cross-correlation between antennas has information on the channel response at each antenna, the source estimation device uses all the information of received power intensity (RSSI), radio wave arrival time difference (TDOA), and radio wave arrival angle (DOA) The position estimation based on the obtained position fingerprint method is executed. However, since the position fingerprint method uses the locality of spatial correlation, it is necessary to collect a large number of position fingerprints in order to perform highly accurate position estimation. Therefore, in the present embodiment, the position fingerprint is statistically learned using the transmission signal of the portable terminal used on a daily basis, and the position estimation using the cross-correlation is performed by using the learned statistic. Furthermore, it will be verified through experiments that it has sufficient accuracy to detect and identify fraudulent acts using mobile devices in places where fairness is expected, such as entrance examinations. Therefore, indoor experiments show that position estimation accuracy that can be easily introduced but can withstand fraud detection can be realized.

  As a prerequisite technology of the present embodiment, not only the position estimation of illegal radio stations and unknown transmission sources but also a distributed spectrum monitor (DSM) will be described. Using a distributed spectrum monitor, including existing systems, the frequency and its area (white space) where radio waves are sparse and the area (white space) that can be used for energy harvesting can be identified efficiently. ) Is specified at the same time. As a result, more efficient use of frequency resources is realized.

  In the following, as a prerequisite technology, 1. 1. Outline of DSM 2. Unknown source estimation method using multiple sensors in DSM; Transmission power estimation for realizing DSM will be described. Following these, 4. 4. detection system; Experimental environment, 6. Statistical learning, 7. 7. Estimation method, Experimental results, 9. Device configuration, 10. The present embodiment will be described in the order of modifications. In addition, 2. Then, (i) a single antenna and (ii) a case where a plurality of antennas are used will be described separately. In statistical learning, an autocorrelation of 6.1 reference signal sequence and a 6.2 LTE reference signal sequence will be described.

1. Overview of DSM FIG. 1 shows a configuration of a DSM system 100 according to an embodiment of the present invention. The DSM system 100 includes a fusion sensor 10, a first sensor 12a, a second sensor 12b, a third sensor 12c, which are collectively referred to as a sensor 12, and a first unknown transmission source 14a, a second unknown transmission source, which are collectively referred to as an unknown transmission source 14. 14b. Here, three sensors 12 are shown, but the number of sensors 12 is not limited to this. A signal transmitted from the first unknown transmission source 14 a or the second unknown transmission source 14 b is received by each of the plurality of sensors 12. The position where the unknown transmission source 14 is installed and the transmission power of the signal are unknown. Each sensor 12 outputs a reception signal to the fusion center 10. The fusion center 10 estimates the position where the unknown transmission source 14 is installed and the transmission power of the signal based on the received power from each sensor 12. Specific processing for estimation will be described later.

  FIG. 2 shows a processing procedure of the DSM system 100. The DSM system 100 includes a known information database 16 in addition to the above-described fusion center 10 and sensor 12. The DSM system 100 includes a propagation path learning process, an unknown transmission source estimation process, and a spectrum map creation process. Hereinafter, specific processing in each process will be described in order.

  First, in the propagation path learning process, learning / acquisition of propagation path information and its calculation amount between the sensors 12 from an arbitrary position in the actual environment necessary for unknown source estimation. However, it is very difficult to install a known transmission source at any arbitrary position and perform propagation path estimation. Therefore, the propagation path from an arbitrary position to the sensor 12 and its statistics are modeled by ray tracing simulation using terrain data and building data. However, in an actual environment, the simulated propagation path may be different depending on a moving object such as a person or a car or an obstacle that does not exist in advance data. Therefore, a cellular system or the like is utilized as a known transmission source, and the propagation path from the known position to the sensor 12 is actually measured and its statistics are acquired. Using this, the propagation path from the arbitrary position simulated earlier is corrected repeatedly, and the propagation path information between the arbitrary position and the sensor 12 in the actual environment and its statistics are acquired.

  Next, in the unknown transmission source estimation process, the unknown transmission source is obtained using the statistics of the propagation path information acquired in the propagation path learning process, the transmission signal pattern candidate database of the unknown transmission source 14, and the reception signal from the unknown transmission source 14. 14 information is estimated. Unknown using the feature that the cross-correlation created from the statistics of the propagation path information between each sensor 12 or antenna and the convolution integral of the autocorrelation of the transmission signal are equal to the cross-correlation of the received signal at each sensor 12 or antenna. Estimate the source.

  Finally, in the spectrum map creation process, a spectrum map is created using information on the estimated unknown transmission source 14 and information on a known transmission source possessed in advance. FIG. 3 shows frequency usage mapping by the DSM system 100. The received power at an arbitrary position and frequency is mapped onto the terrain data, and a charging space 202 that can receive sufficient power and an unused frequency domain white space 200 are found.

2. Estimation Method of Unknown Source Using Multiple Sensors in DSM Here, a broadband model is assumed, and a propagation model uses a statistical model in which a propagation delay profile is a multipath according to an exponential distribution. Each path is uncorrelated according to the Rayleigh distribution. For simplicity, it is assumed that there is one unknown transmission source 14 and that the propagation path learning process, which is the first process, has been learned.

(I) In the case of a single antenna The received signal received from an arbitrary position by the i-th (i = 1,..., M) sensor 12 is an unknown transmission source when the broadband signal and the propagation environment change from moment to moment. It is expressed as follows using the dynamic propagation channel at the position ψ ₀ of 14 and the transmission signal of the unknown transmission source 14.
Where t and τ _i represent the time and delay time on the signal sequence,
Is the time representing the fluctuation of the propagation path. g ₀ is in the form of a power spectrum (hereinafter referred to as “normalized filter”) in which the transmission power of the unknown source is 1, and depends on the transmission filter and the modulation method. p ₀ is the transmission power of the unknown source. Also,
Is the transmission signal of an unknown source,
Is the propagation path response from the unknown source position ψ ₀ to the i-th sensor 12,
Is white Gaussian noise with variance σ ² .

The cross-correlation of the reception signal of the i-th sensor 12 and the reception signal of the j-th sensor 12 expressed by the equation (1) with respect to the time t is observed by the i-th sensor 12 and the j-th sensor 12. It can be calculated as follows to express the propagation delay difference (TDOA) of the signal. However, i = j is originally referred to as autocorrelation, but is hereinafter referred to as cross-correlation for simplification. Further, for simplification of the equation, description of ψ ₀ , g ₀ , p ₀ is omitted, and Rx [A, B] is set as the correlation of functions A and B with respect to x.

Where D _ij is 1 when i = j,
Is 0.
Is a statistical process
Does not depend on. here,
(2) can be further modified as follows.

Represents the autocorrelation with respect to the time of the transmission signal of the unknown transmission source and the cross-correlation between the propagation path from the unknown transmission source to the i-th sensor 12 and the propagation path to the j-th sensor 12, respectively. That is, the cross-correlation between the reception signals from the unknown transmission source in each sensor 12 can be expressed by a convolution integral of the cross-correlation between the propagation paths from the unknown transmission source to each sensor 12 and the autocorrelation of the transmission signal. At this time, the shape of the autocorrelation of the transmission signal is determined by the transmission power of the unknown transmission source and the shape of the normalization filter. The cross-correlation between channel responses is
Since it has a peak value at this time, it becomes a function having information on the difference between the propagation delay to the i-th sensor 12 and the propagation delay to the j-th sensor 12, and its shape depends on the environment of the propagation path to each sensor. It is determined. here,
Is
It is necessary to take the expected value because it changes.

However, since the phase of the propagation path response is uniformly distributed, if the expected value of Equation (2) is taken in a dynamic environment, the cross-correlation value becomes 0, and information regarding the position cannot be extracted. Therefore, phase information is rejected by taking the absolute value of Formula (2). This process
It is limited only to the amplitude of each channel of the i-th sensor 12 and the j-th sensor 12. Therefore, take an absolute value for equation (2),
The expected value for is expressed as follows. Note that Ex [•] is set as an expected value for x.

Since the amplitude of the propagation path response varies stochastically, it shows that RSSI and TDOA information can be extracted from Equation (4). That is, the cross-correlation of the propagation path includes two pieces of information determined by the positions of TDOA and RSSI. Therefore, as shown in Expression (1), the received signal also depends on the position, transmission power, and normalization filter. Therefore, the cross-correlation expressed by Expression (3) and the expected value of the absolute value expressed by Expression (4) Is created from the received signal of each sensor, and the received signal is normalized by the maximum average received power in the sensor, it can be expressed by the following equation.

Here, a cross-correlation replica is created from the estimated candidate locations, and the replica and
The place where the difference in the values of the values is the minimum is taken as the estimated value. Specifically, the replica created from the position candidate ψ, the normalized filter candidate g, and the statistics of the propagation path information acquired in the propagation learning process is expressed as follows from Equation (3).

Is the replica transmission signal created from the estimated normalization filter candidate g
Autocorrelation
And the propagation path responses of the i th sensor 12 and the j th sensor 12 created from the estimated candidate position ψ and the statistics of the dynamic propagation path response obtained in the propagation path learning process.
Cross-correlation
After convolving and taking the absolute value
Is the expected value for.

Therefore, using the equations (5) and (8), the position of the unknown transmission source and the normalization filter can be estimated by the following maximum likelihood estimation method.
In the equation (9), the two parameters of the unknown source position and the normalization filter were changed and created from the actual received signal.
And a replica of the estimate
The sum of the squares of the differences between the two is the estimated value of these two parameters. A value obtained by subtracting the cross-correlation between true values is noise.

(Ii) In the case of a plurality of antennas A replica of a received signal received from an arbitrary position by the k-th antenna (k = 1,..., N _i ) of the i-th sensor 12 includes a wideband signal, a linear array antenna, and a propagation environment. Assuming that it changes dynamically from moment to moment, it is expressed as follows using the dynamic propagation path response at the position ψ ₀ of the unknown source and the transmission signal of the unknown source.

Here, d _k is the distance from the reference point to the position of the k-th element when the antenna of the first element is used as the reference point, λ is the wavelength of the incoming wave, and θ _i (τ _i ) is the angle of arrival. Represent. At this time, the behavior when the expected value is obtained differs between the cross-correlation between the same sensors and the cross-correlation between the antennas of different sensors. In the following, description of ψ ₀ , g ₀ , and p ₀ is omitted for the sake of brevity, as in the case of a single antenna.

(Ii-1) Cross-correlation between antennas in the same sensor The cross-correlation between the k-th antenna and the l-th antenna between the same sensors with respect to the i-th sensor 12 is expressed as follows.
However,
It is. Therefore,
Shows that a cross-correlation peak appears at
It can be said that the influence of the change on the cross-correlation is the amplitude and the arrival angle of the propagation path response.

In the cross-correlation of equation (12)
The expected value is expressed as follows.

Therefore,
When the value is 0,
Is expressed by the following formula.
From equation (14)
Indicates that RSSI and DOA information can be extracted even if the cross-correlation between different antennas of the same sensor takes an expected value for the change in the dynamic channel response.

(Ii-2) Cross-correlation between antennas in different sensors The cross-correlation between the k-th antenna of the i-th sensor 12 and the l-th antenna of the j-th sensor 12 is expressed as follows.

Therefore,
It can be said that the effect of the change in the cross-correlation is the amplitude and phase of the channel response and the arrival angle. In general, since the phase of the dynamic channel response is uniformly distributed, information regarding the position cannot be extracted if the expected value of Equation (15) is taken. Therefore, the absolute value is taken for equation (15) as in the case of a single antenna. This process
Only the amplitude of the channel of the i-th sensor 12 and the j-th sensor 12 changes with respect to. Therefore, for the absolute value of equation (15),
The expected value for is expressed as follows.

In general, in a dynamic channel response, the probability distribution of amplitude takes an arbitrary form, and therefore Equation (16) indicates that RSSI and TDOA information can be extracted. From the above, when multiple antennas are used, it has been shown that three information determined by the positions of RSSI, TDOA, and DOA are included in the cross-correlation of the propagation path. Therefore, as shown in Expression (10), the received signal also depends on the position, the transmission power, and the normalization filter. Therefore, the cross-correlation represented by Expression (12) and Expression (15) and Expression (13) and Expression If the expected value of the absolute value shown in (16) is created from the received signal of each sensor, and the received signal is normalized by the maximum average received power in the sensor, it can be written in the following equation.

A replica created from the position candidate ψ, the normalized filter candidate g, and the statistics of the propagation path information acquired by the propagation learning process is expressed as follows from the equation (3).
Is the replica transmission signal created from the estimated normalization filter candidate g
Autocorrelation
And the channel response of the k-th antenna of the i-th sensor 12 and the l-th antenna of the j-th sensor 12 created from the estimated candidate position ψ and the statistics of the dynamic channel response acquired in the channel learning process.
Convolve the cross-correlation of
Is the expected value for.

Also,
After taking the absolute value for the convolution
Is the expected value for. Therefore, using the equations (17), (18), and (21), the position of the unknown transmission source and the normalized filter can be estimated by the following maximum likelihood estimation method.
The two parameters of the unknown source position and the normalization filter were changed in the equation (22), and created from the actual received signal.
And created from replica
The sum of the squares of the differences is the estimated value of these two parameters. A value obtained by subtracting the cross-correlation between true values is noise.

3. Transmission power estimation for realizing DSM The position of an unknown transmission source can be estimated by using Equations (9) and (22). To create a DSM spectrum map, the transmission power of an unknown transmission source can be estimated. You also need to know. Therefore, the transmission power P _tx of the unknown source is estimated from the following maximum likelihood estimation method by using the propagation path information and RSSI learned after estimating the position of the unknown source.

However,
And represents the average received power. P _{tx, 0} is the true transmission power of the unknown source,
Is the estimated average received power created from the estimated transmit power of the unknown source and the learned propagation path information. In Equation (23), the transmission power of the unknown transmission source is changed as a parameter, and the sum of the difference from the actual average reception power is minimized to be an estimated value of the transmission power parameter.

4). FIG. 4 shows an image of position estimation and feature amounts in an indoor environment according to an embodiment of the present invention. A classroom is assumed as the indoor environment, and the first sensor 12a, the second sensor 12b, and the third sensor 12c are installed in the classroom. In the classroom, there is a person who carries the mobile terminal 302. FIG. 4 also shows a propagation path response and a delay time due to a signal output from the mobile terminal 302 in each sensor 12.

  FIG. 5 shows a processing procedure of the detection system 300 according to the embodiment of the present invention. The detection system 300 is, for example, a fraud detection system, and FIG. 5 shows a flowchart from the learning of the position fingerprint to the position estimation from the overall image aimed by the fraud detection system. As shown in FIG. 5, the detection system 300 includes a mobile terminal 302 that is a signal transmission source, a sensor group that includes a plurality of sensors 12 that receive the signal, a camera that maps the position fingerprint to the real space, and the position fingerprint and the real space. Are stored in association with each other, and are composed of four elements of the transmission source estimation apparatus 50 including a database for performing statistics and feature quantity extraction.

  In the learning process, signals transmitted from the portable terminals 302 that are used on a daily basis are received by the plurality of sensors 12, and a position fingerprint is created. The position fingerprint is a feature value for each seat that is discretely arranged, but this alone has no relation to the position in the real space, so the position fingerprint that has already been learned in association with the real space can be used. It is necessary to use it as a function of position by associating it with information on the position in real space. The transmission source estimation apparatus 50 records the position fingerprint associated with the position information in this way in the database as a function of the position. Furthermore, the database performs not only the function of recording the position fingerprint but also more efficient learning such as enhancing the feature amount by statistically processing the accumulated position fingerprint.

  In the estimation process, first, it is determined from the received power whether or not the received signal is transmitted from within the test site. If the received power is sufficiently large, it is determined that communication has been performed from within the test site, and position estimation is performed. In the estimation process, similar to the learning process, the transmission source estimation device 50 receives a signal transmitted from the mobile terminal 302 by a plurality of sensors, creates a position fingerprint, and compares it with the position fingerprint learned as a function of the position. And the seat with the closest position fingerprint is output as the position estimation result.

5. Experimental environment In order to verify the detection system 300, an experiment of position estimation using an LTE (Long Term Evolution) terminal as the portable terminal 302 was performed. FIG. 6 shows the specifications of the indoor experiment for the detection system 300, and FIG. 7 shows the experimental environment for the detection system 300. The experimental environment was a classroom of 10.8 m x 16.85 m, and sensors with one antenna were placed at the four corners of the room. These are indicated as the first antenna 60a to the fourth antenna 60d. In the center of the room, sensors with three circular array antennas (interval 0.45λ) were arranged. These are denoted as the fifth antenna 60e to the seventh antenna 60g.

  The measurement interval was assumed to be a test, and the measurement was performed at a distance of 2 seats in the horizontal direction of 1.2 m and in the vertical direction of 0.78 m, for a total of 114 seats. Also, in order to assume a quasi-static environment in which the classroom environment does not change, but the terminal that performs the fraudulent act itself is not fixed, the measurement of each seat is performed by moving the transmitting terminal every 2 cm. Point measurements were taken. At this time, two points are measured at each measurement point, and a total of 20 samples are measured for each seat. In such an environment, in this experiment, a signal transmitted from the LTE terminal is received and propagation path information at each seat is obtained as position fingerprint information. Note that this time, position information is not linked using a camera, and the seat to be learned is known. In addition, the time synchronization between the sensors has been calibrated this time, and there is one LTE terminal for the demonstration experiment. Furthermore, for future explanation, the seat numbers are defined as 1, 2, 3 from the upper left seat to the right.

6). Statistical learning The statistical learning used in this experiment utilizes the property that the cross-correlation between received signals of multiple sensors includes information on the channel response. This is because position estimation using a part of information is common among RSSI, TDOA, and DOA included in the propagation path response in the existing position fingerprint method, but here, all the information of the propagation path response is used. be able to. In the present embodiment, a method using the received signal cross-correlation between the antennas as a position fingerprint and a known reference signal sequence is proposed from the nature of the received signal cross-correlation between the antennas.

6.1 Autocorrelation of reference signal sequence The reception signal cross-correlation between antennas convolves information on the autocorrelation of the transmission signal in addition to the propagation path response. Since the autocorrelation of the transmission signal does not include information related to position estimation, it is desirable that there is little fluctuation regardless of the position. However, since the modulation scheme of the transmission signal is not always constant, the shape of the autocorrelation of the transmission signal changes every moment. Therefore, consider using a reference signal sequence of a wireless communication system. This sequence is a sequence for carrying out symbol synchronization and channel quality estimation without carrying the amount of information, and the sequence pattern is generally known.

  A pseudo-noise matrix such as a normal PN sequence is usually used as the reference signal sequence, and it is known that the autocorrelation shows a high value only when the delay time τ = 0. Therefore, by using the cross-correlation signal between antennas of a known reference signal sequence as a position fingerprint, it is possible to obtain a position fingerprint that has a constant autocorrelation of the transmission signal and that depends only on the channel response, with higher accuracy. It is thought that accurate position estimation can be performed. In the present embodiment, the position fingerprint was actually acquired by the received signal cross-correlation using a known reference signal sequence.

6.2 LTE Reference Signal Sequence As described in the experimental environment section, since the experiment was performed with the mobile terminal 302 using LTE as a communication system, the LTE uplink specifications are briefly described here, and the LTE used in this embodiment is used. The reference signal sequence will be described. The LTE uplink employs a communication method called OFDM (Orthogonal Frequency Division Multiplexing) called SC-FDMA, and communication by many subcarriers is performed.

  FIG. 8 shows LTE resource blocks used in the detection system 300. This shows the LTE frequency and symbol time design. The horizontal direction represents time, and the vertical direction represents frequency resources. In LTE, a 1 ms time slot is composed of 14 or 12 symbols as a minimum resource unit, and a 180 kHz resource block is composed of 12 15 kHz subcarriers. In addition to such communication channels, there are DRS (Demodulate Reference Signal) for symbol synchronization and SRS (Sounding Reference Signal) for estimating channel quality.

  In the present embodiment, SRS for propagation path estimation is used. SRS is a reference signal sequence for propagation path estimation, which is a wideband signal (6-110RB) for the purpose, and is periodically (2, 5, 10, 20, 40, 80, 160, 320 ms) during communication. Every). Therefore, it can be said that a broadband reference signal sequence can be obtained relatively frequently. Although the SRS bandwidth and the transmission interval depend on the operator, in this embodiment, 3.6 MHz = 20 RB is used as the SRS bandwidth.

The SRS time signal is a Zadoff-Chu sequence (ZC sequence) DFT defined by the following equation.
Here, s _SRS (kΔt) is an SRS discrete time signal, k is a time index, Δt is a sample time, n is a subcarrier index, u is a ZC sequence group number, v is a group sequence number, and α is between users. Orthogonalization parameter, _NZC is the sequence length of the ZC sequence,
Is the number of subcarriers.

  The ZC sequence DFT is also known to be a ZC sequence, and the autocorrelation of the ZC sequence DFT, that is, the SRS time signal, has a high peak value only when the delay time τ = 0. FIG. 9 shows cross-correlation between antennas based on LTE reference signals used in the detection system 300. This is an image related to the cross-correlation between received signals of SRS antennas using a ZC sequence. In this way, it is possible to obtain a position fingerprint including band-limited channel response information from which the autocorrelation of the transmission signal is eliminated. In an indoor experiment, by receiving this SRS, a position fingerprint of a seat at each discrete position is acquired.

  It is possible to statistically learn the position fingerprint by performing statistical processing on the position fingerprint acquired in this way, and the problem of the huge number of fingerprint acquisition points that was a problem of the existing position fingerprint method Can be solved. In the indoor experiment, the learning amount was processed by using 19 for learning and 1 for estimation for 20 data accumulated for each seat. For 19 data for learning, KL conversion was performed on the value of the cross-correlation in the delay time direction, thereby extracting the feature amount and appropriately reducing the time limit.

7). Estimation method In the position estimation, it becomes a problem to search for the seat where the position fingerprint for estimation is the closest to the database of the position fingerprint obtained in the previous section. When the shape of the position fingerprint is viewed as a pattern, various pattern matching methods can be applied. In the indoor experiment, since the KL conversion base and the converted position fingerprint feature amount are acquired from the learning position fingerprint in advance, the estimation position fingerprint is converted by the KL conversion base for each seat, and the converted vector is obtained. An approach to maximum likelihood estimation was taken.

8). Experimental results The experimental specifications were as described above, and TDOA, RSSI, and all of the present examples performed position estimation 20 times using 20 samples, and evaluated the average distance error of the position estimation result for each seat. In this example, the estimated samples were taken out one by one from the 20 samples, and the position was estimated 20 times for each seat. FIG. 10 shows a comparison between the detection system 300 and the traditional method. Here, the horizontal axis is the seat number shown in FIG. 7, and the vertical axis is the average distance error for each seat, and the average estimated distance error of all seats in TDOA, RSSI, and this embodiment is 3.94 m, 3. 02m and 0.42m. From these, it can be said that the present embodiment greatly improves the estimation accuracy.

  FIG. 11 shows the average error range of the detection system 300. In this embodiment, the error is visualized by a circle. From this, it can be seen that the error of this embodiment is at most up to the adjacent seat, which is sufficient as the position estimation accuracy required by the detection system 300.

9. Device Configuration FIG. 12 is a diagram showing a configuration of a transmission source estimation device 50 according to an embodiment of the present invention. The fusion center 10 includes an acquisition unit 70, a first derivation unit 72, a storage unit 74, a second derivation unit 76, a selection unit 78, and an output unit 80. First, (i) the case of a single antenna will be described. In the acquisition unit 70, a transmission signal transmitted from the mobile terminal 302 disposed in the vicinity of any one of the plurality of discrete candidate positions is received by each of the plurality of sensors 12 disposed at different known positions. The reception signal is acquired from each of the plurality of sensors 12. Here, the plurality of discrete candidate positions correspond to the respective seats shown in FIG. The received signal is expressed as shown in Equation (1).

  The first deriving unit 72 derives a cross-correlation between the sensors 12 with respect to the reception signal acquired by the acquiring unit 70 and derives an expected value of the absolute value of the cross-correlation. The processing in the first deriving unit 72 corresponds to Expression (5) to Expression (7). In particular, the expected value of the absolute value of the cross-correlation corresponds to Equation (5).

  The storage unit 74 stores in advance a replica for the expected value derived by the first deriving unit 72. Such a replica is acquired in advance for each candidate position. As described above, since two points are measured at each measurement point included in one candidate position in order to derive a replica, a total of 20 samples are measured for each candidate position. That is, one replica is generated by reflecting received signals from each of a plurality of measurement points in the vicinity of one candidate position. More specifically, the replica is expressed as in equation (8), and one candidate is obtained by averaging the replica derived based on the received signal from the measurement point over a plurality of measurement points. A replica for the position is derived.

  The second deriving unit 76 derives an error between the replica stored in the storage unit 74 and the expected value derived by the first deriving unit 72. The selecting unit 78 selects one error from a plurality of errors derived by causing the second deriving unit 76 to repeatedly execute the process while using the replica whose candidate position is changed in the second deriving unit 76. To do. This process corresponds to selecting the minimum value in Equation (9). The output unit 80 outputs the candidate position corresponding to the error selected by the selection unit 78 as the position of the seat where the mobile terminal 302 exists.

  Next, (ii) the case of multiple antennas will be described. When the sensor 12 includes a plurality of antennas 60, the acquisition unit 70 acquires a reception signal for each antenna 60. The received signal is expressed as in equations (10) and (11). The first deriving unit 72 derives the expected value of the absolute value of the cross-correlation between the antennas 60 provided in different sensors 12. The expected value of the absolute value of the cross correlation corresponds to Expression (18). The first deriving unit 72 also derives an expected value of the cross-correlation of received signals between the antennas 60 provided in the same sensor 12. The expected value of cross-correlation corresponds to equation (17).

  The second deriving unit 76 derives an error from the expected value and the replica between the antennas 60 provided in different sensors 12 and also calculates the error from the expected value and the replica between the antennas 60 provided in the same sensor 12. To derive. The former corresponds to the second term on the right side of Equation (22), and the latter corresponds to the first term on the right side of Equation (22).

  This configuration can be realized in terms of hardware by a CPU, memory, or other LSI of any computer, and in terms of software, it can be realized by a program loaded in the memory, but here it is realized by their cooperation. Draw functional blocks. Accordingly, those skilled in the art will understand that these functional blocks can be realized in various forms by hardware only, software only, or a combination thereof.

10. In the embodiment, a situation where a test is performed in a classroom is assumed, and the detection system is applied to such a situation. At that time, the detection system identifies the mobile terminal brought into the classroom, thereby suppressing fraud in the test. On the other hand, in the modified example, a situation is assumed in which a press machine or the like is operating in a factory, and the detection system is applied to such a situation. In that case, the worker who exists in a factory is recognized by specifying a portable terminal in a factory by a detection system. In the classroom where the test is being conducted, the propagation environment fluctuates due to human movement. On the other hand, in a factory, the propagation environment varies as a person operates, and the propagation environment also varies depending on the operation of a press or the like. Therefore, the fluctuation of the propagation environment in the factory tends to be larger than the fluctuation of the propagation environment in the classroom.

  FIGS. 13A to 13B are diagrams showing the configuration of the press machine 230 used in the modification of the present invention. This is an example of a machine installed in a factory, and the machine is not limited to this. The press machine 230 includes a frame 210, a bolster 212, a slide 214, a connecting rod 216, a crankshaft 218, a slide give 220, and a flywheel / clutch 222.

  The frame 210 receives pressure. A mold is fixed to the bolster 212. A mold is attached to the slide 214 to operate. The connecting rod 216 connects the crankshaft 218 and the slide 214 to transmit energy. The crankshaft 218 converts the rotational movement into a linear movement via the connecting rod 216. The slide give 220 restrains the movement of the slide 214. The flywheel / clutch 222 stores energy with the flywheel and operates the slide 214 by the engagement / disengagement of the clutch.

  With such a configuration, the press machine 230 transmits the rotation of the motor to the crankshaft 218 and moves the slide 214 constrained to the slide give 220 via the connecting rod 216 up and down. The slide 214 reciprocates (generally up and down) and has a bolster 212 fixed thereto. The bolster 212 is fixed on the bed and is connected to the slide drive mechanism by a frame 210, a tie rod or the like, and receives a pressure.

  In FIG. 13A, the slide 214 is arranged at a position above the vertical movement, and in FIG. 13B, the slide 214 is arranged at a position below the vertical movement. The propagation environment varies depending on the position of the slide 214. By the operation of the press machine 230, the state of FIG. 13A and the state of FIG. 13B are repeated repeatedly. As a result, the propagation environment varies.

  The detection system 300 according to the modified example is the same type as in FIG. 5, and the transmission source estimating apparatus 50 according to the modified example is the same type as in FIG. Here, the difference will be mainly described. One replica stored in the storage unit 74 is generated by reflecting a received signal corresponding to each moving position of the object. That is, not only a plurality of measurement points are set in the vicinity of the candidate position and replicas at each measurement point are generated as in the past, but also at one measurement point, the state of FIG. And the replica in the state of FIG. 13B are generated. A replica at one candidate position is generated by averaging these replicas.

  According to the embodiment of the present invention, since the position of the mobile terminal is estimated based on the replicas for each of the plurality of discrete candidate positions, the position candidates to be estimated can be limited. Moreover, since the position candidates to be estimated are limited, an increase in the processing amount can be suppressed. Further, since the replica is acquired in advance, an increase in the processing amount can be suppressed. Moreover, since the position candidate to be estimated is associated with the seat, it is possible to suppress the deterioration of the estimation accuracy while suppressing an increase in the processing amount. In addition, since one replica is generated by reflecting received signals from each of a plurality of positions in the vicinity of one candidate position, it is possible to reduce the influence due to the shift of the position of the mobile terminal. Further, since one replica is generated so as to reflect the operating position of the machine, the influence due to the operation of the machine can be reduced. Moreover, since the increase in the processing amount is suppressed, the cost can be reduced.

  In the above, this invention was demonstrated based on the Example. This embodiment is an exemplification, and it will be understood by those skilled in the art that various modifications can be made to combinations of the respective constituent elements, and such modifications are also within the scope of the present invention.

  DESCRIPTION OF SYMBOLS 10 Fusion center, 12 Sensor, 14 Unknown transmission source, 16 Known information database, 50 Transmission source estimation apparatus, 60 Antenna, 70 Acquisition part, 72 1st derivation part, 74 Storage part, 76 2nd derivation part, 78 Selection part, 80 output unit, 100 DSM system, 300 detection system, 302 mobile terminal.

Claims (5)

A transmission signal transmitted from a transmission source arranged near any one of a plurality of discrete candidate positions is received by each of a plurality of sensors arranged at different known positions, and the plurality of sensors An acquisition unit for acquiring a reception signal from each of the
A first derivation unit for deriving a cross-correlation between sensors with respect to the reception signal acquired in the acquisition unit, and deriving an expected value expected as an average of absolute values of the cross-correlation;
Auto-correlation of a replica transmission signal created from a candidate for a normalization filter that is a replica to be subtracted from the expected value derived in the first deriving unit and has a transmission power of 1 at the transmission source, and a candidate position And the replica derived as an expected value expected as an average after calculating the absolute value after convolving the cross-correlation of the transmission path characteristics between the two sensors acquired in the propagation learning process, and the first derivation A second derivation unit for deriving an error from the expected value derived in the unit;
A selection unit that selects a minimum error from a plurality of errors derived by causing the second derivation unit to repeatedly execute a process while changing a candidate position in the second derivation unit;
An output unit that outputs a candidate position corresponding to the error selected in the selection unit as a position where a transmission source is installed;
The replica used in the second derivation unit is acquired in advance for each candidate position. The one replica used in the second derivation unit is generated by averaging transmission path characteristics from each of a plurality of known positions in the vicinity of one candidate position to the sensor. The transmission source estimation apparatus according to 1. The one replica used in the second derivation unit is generated by averaging transmission path characteristics from each moving position of the known transmission source to the sensor. The source estimation apparatus described. The acquisition unit acquires a reception signal for each antenna when the sensor includes a plurality of antennas;
The first deriving unit derives an expected value that is expected as an average of absolute values of cross-correlation between antennas provided in different sensors, and receives signals between antennas provided in the same sensor. Derived expected correlation value,
The second deriving unit derives an error from an expected value and a replica between antennas provided in different sensors, and derives an error from an expected value and a replica between antennas provided in the same sensor. The transmission source estimation apparatus according to any one of claims 1 to 3, wherein A transmission signal transmitted from a transmission source arranged near any one of a plurality of discrete candidate positions is received by each of a plurality of sensors arranged at different known positions, and the plurality of sensors Obtaining received signals from each of the
Deriving a cross-correlation between sensors for the acquired received signal and deriving an expected value expected as an average of absolute values of the cross-correlation;
Auto-correlation of replica transmission signal created from a candidate for a normalization filter that is a replica to be subtracted from the derived expected value and the transmission power of the transmission source is 1, and acquired by candidate position and propagation learning process After calculating the absolute value after convolving the cross-correlation of the transmission path characteristics between the two sensors , the error between the expected value expected as the average and the derived expected value is derived. Steps,
Selecting a minimum error from a plurality of errors derived by repeatedly executing a process for the step of deriving the error while changing a candidate position;
Outputting a candidate position corresponding to the selected error as a position where the transmission source is installed,
A source estimation method, wherein a replica used in the step of deriving the error is acquired in advance for each candidate position.