JP6836936B2 - Estimating method and estimation device using it - Google Patents

Description

The present invention relates to an estimation technique, and relates to an estimation method for estimating the position of an unknown source and an estimation device using the estimation method.

TOA (Time Of Arrival), RSS (Receive Signal Strengh), etc. are used to estimate the position of the radio, but RSS is used in an indoor environment where there is no line of sight and a multipath environment. Is more suitable. Hyperplane equations and factor graphs are also used to simplify processing by RSS (see, for example, Non-Patent Document 1).

Chin-Tseng Huang, Cheng-Hsuan Wu, Yao-Nan Lee and Jiunn-Tsair Chen, "A Novel Indoor RSS-Based Position Location Algorithm Using Factor Graphs", IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, JUNE 2009, VOL. 8, NO. 6, p. 3050-3058

When estimating the position of the wireless device that is the source of the signal based on RSS, information on the transmission power in the wireless device is required. However, when the wireless device to be estimated is an illegal wireless station or an unknown transmission source, it is difficult to acquire information on the transmission power. Therefore, the estimation is made based on the DRSS (Differential RSS) derived by subtracting the RSS in each of the two sensors. The DRSS requires at least four observation spots that surround the radio.

The present invention has been made in view of such a situation, and an object of the present invention is to provide a technique for estimating the position of a wireless device as a transmission source even if there is no information on transmission power.

In order to solve the above problems, in the estimation device of a certain aspect of the present invention, the signal from the target wireless device is received by each of the plurality of sensors, and the TOA (Time Of Arrival) of the sample in each sensor is received. The first acquisition unit that acquires RSS (Receive Signal Strength), the TDOA sample measurement unit that derives the TDOA (Time Difference Of Arrival) of the sample between the sensors from the TOA of the sample acquired in the first acquisition unit, and the TDOA sample. With the first factor graph processing unit that derives the first estimated value of the position coordinates of the radio device by executing the TDOA-based factor graph processing of Pitagolas using the TDOA of the sample between the sensors derived in the measurement unit. , A specific unit that identifies at least four sensors located near the first estimated value derived in the first factor graph processing unit, and training signals from known transmitters are received by each of the plurality of sensors. , A second acquisition unit that acquires the RSS of the sample in each sensor, a selection unit that selects the RSS for at least four sensors specified in the specific unit among the RSS acquired in the second acquisition unit, and a selection unit in the selection unit. Based on the RSS, the coefficient calculation unit that calculates the coefficient and the DRSS (Differential RSS) of the sample between the sensors are derived from the RSS of the sample acquired in the first acquisition unit, and the derived DRSS and the coefficient calculation unit calculate. It is provided with a second factor graph processing unit for deriving a second estimated value of the position coordinates of the radio device by executing the DRSS-based factor graph processing while using the obtained coefficient.

Another aspect of the present invention is an estimation method. In this method, a signal from a target wireless device is received by each of a plurality of sensors, and a step of acquiring a sample TOA (Time Of Arrival) and RSS (Receive Signal Strength) at each sensor and acquisition are performed. By performing a TDOA-based factor graphing process of Pitagolas using the steps of deriving the sample TDOA (Time Difference Of Arrival) between the sensors from the TOA of the sample and the TDOA of the sample between the derived sensors. A step of deriving a first estimate of the position coordinates of the radio device, a step of identifying at least four sensors located near the derived first estimate, and multiple sensors with training signals from known transmitters. The step of acquiring the sample RSS for the training signal at each sensor, the step of selecting the RSS for at least four identified sensors among the sample RSS for the training signal, and the selected RSS. Based on the step of calculating the coefficient, the DRSS (Differential RSS) of the sample between the sensors is derived from the sample RSS for the signal from the target radio device, and the derived DRSS and the calculated coefficient are used. At the same time, it includes a step of deriving a second estimate of the position coordinates of the radio device by performing DRSS-based factor graph processing.

It should be noted that any combination of the above components and the conversion of the expression of the present invention between methods, devices, systems, recording media, computer programs and the like are also effective as aspects of the present invention.

According to the present invention, the position of the wireless device as the transmission source can be estimated without the information of the transmission power.

It is a figure which shows the structure of the estimation apparatus which concerns on embodiment of this invention. It is a flowchart which shows the procedure of the PTDOA-based factor graph processing by the estimation apparatus of FIG. It is a flowchart which shows the procedure of the downlink processing by the estimation apparatus of FIG. It is a flowchart which shows the procedure of the uplink processing by the estimation apparatus of FIG. It is a flowchart which shows the procedure of the calculation process by the estimation apparatus of FIG. It is a flowchart which shows the procedure of the DRSS-based factor graph processing by the estimation apparatus of FIG.

Before explaining the present invention in detail, first, an outline will be given. An embodiment of the present invention relates to an estimation device that estimates the position of an illegal radio station or an unknown source (hereinafter referred to as “radio device”). A recent trend in geoposition is to establish low cost, high accuracy, and simplicity, but this geoposition is also being used to estimate the location of unknown radios. Here, "unknown" indicates that the TOA and transmission power of the target wireless device are unknown. Radio geoposition is defined as the process of calculating the geographic coordinates of a radio device by observing the nature of electromagnetic waves. In order to estimate the geographical location of the radio device, the properties of electromagnetic waves are observed in TOA, RSS, TDOA (Time Difference Of Arrival), and combinations thereof.

So far, a combination of RSS and TOA measurements and a combination of RSS and DOA (Direction Of Arrival) measurements have been used for geopositioning. The combination of these improves the accuracy as compared with the case of RSS alone, TOA alone, and DOA alone. However, additional configuration is required to obtain the azimuth. Further, in the combination of RSS and TOA, the position of the unknown wireless device cannot be estimated without the information of the absolute transmission power and the time stamp for the unknown target.

In this embodiment, a PTDOA (Pythagorean Time Difference Of Arrival) factor graph and a DRSS factor graph are combined to estimate the location of an unknown radio device. In the algorithmic structure of the factor graph, global functions are decomposed into multiple local functions. As a local function, the factor graph contains two types of nodes. The first is the factor node, which calculates the updated software information (SI) message. The second is the variable node, which executes the thumb product algorithm and passes the value to the factor node. The mean value and the variance value of the measurement data are exchanged between these factor nodes and the variable nodes.

In geolocation technology, the factor graph runs a thumb product algorithm to estimate the position of the target using TOA. Only the mean and variance of the TOA measurements, which are assumed to be Gaussian, are exchanged between the nodes. Each sensor measures TOA and the mean and variance values are calculated. However, factor graph-based TOA requires perfect time synchronization of the target, which is not possible if the target is unknown. Therefore, Pythagoras TDOA based on the factor graph is introduced in order to eliminate the need for knowledge of time stamps. The Pythagoras TDOA, which is based on the factor graph, is a modification of the TOA, which is based on the factor graph, and adds nodes to the factor graph. At the added node, subtraction of TOA data between the two sensors is performed.

RSS based on the factor graph is also used for geolocation technology. Transmission signal sequences are transmitted from several observation spots, and RSS is acquired in watts. RSS is converted to dB and input to the factor graph. The RSS information reference data obtained by the training is used to generate a PDP (Power Decay Profile) by the interpolation method. Here, local linearity is used to estimate the position of the target. This is highly accurate indoors.

A factor graph-based DRSS is proposed to eliminate the need for knowledge of absolute transmit power in unknown radios. DRSS is a technique that modifies RSS by subtracting RSS in each of the two sensors. This subtraction requires the addition of a new node to the factor graph. However, DRSS requires RSS information from at least four suitable observation spots. This is because the DRSS algorithm operates in the difference region, and the linear equations are degenerated and the position cannot be determined unless there are four or more observation spots. It can be said that the degeneracy of the linear equations lowers the rank of the matrix by one.

This observation spot must be close to the target and placed around the target. In the DRSS factor graph, such suitable observation spots are known and the target should be in the grid of observation spots. In reality, many observation spots are arranged and require another algorithm to select an appropriate observation spot from them. As such another algorithm, the RADAR algorithm is used, but the RADAR algorithm requires knowledge of the transmission power of the target. Therefore, here we use the PTDOA factor graph instead of the RADAR algorithm. The PTDOA factor graph is used to select observation spots for the DRSS factor graph in an outdoor environment.

FIG. 1 shows the configuration of the estimation device 100. The estimation device 100 includes a first sensor 10a, a second sensor 10b, a third sensor 10c, an Nth sensor 10n, an acquisition unit 12, a TDOA sample measurement unit 14, a first factor graph processing unit 16, and a specific sensor 10. A unit 18, a selection unit 20, a coefficient calculation unit 22, and a second factor graph processing unit 24 are included. The acquisition unit 12 includes a first acquisition unit 12a and a second acquisition unit 12b. The first factor graph processing unit 16 includes an Euclidean factor node processing unit 40, an Euclidean variable node processing unit 42, an Euclidean distance estimation factor node processing unit 44, a thumb product algorithm variable node processing unit 46, a Pythagoras factor node processing unit 48, and a Pythagoras variable. It includes a node processing unit 50, a conversion factor node processing unit 52, a thumb product algorithm processing unit 54, a determination unit 56, and an output unit 58. The second factor graph processing unit 24 includes an RSS factor node processing unit 80, an RSS variable node processing unit 82, a DRSS factor node processing unit 84, a DRSS variable node processing unit 86, a linear plane factor node processing unit 88, and an output unit 90. .. The thumb product algorithm variable node processing unit 46 includes a downlink processing unit 60 and an uplink processing unit 62, the Pythagoras factor node processing unit 48 includes a downlink processing unit 64 and an uplink processing unit 66, and the Pythagoras variable node processing unit 50 includes a Pythagoras variable node processing unit 50. The downlink processing unit 68 and the uplink processing unit 70 are included, and the conversion factor node processing unit 52 includes the downlink processing unit 72 and the uplink processing unit 74.

The N sensors 10 are arranged at _{coordinates (X h} , Y _h). Here, h indicates the index of the sensor 10. It has an area of 1000 x 1000 m ² and a mesh configuration grid of 100 m. Of the N sensors 10, four sensors 10 are observation spots. The four observation spots is arranged so as to surround the target, the coordinates of the observation spot are shown as (x _{m, y m).} Here, m indicates the index of the observation spot. Here, the position of the sensor 10 is known to the estimation device 100. The coordinates of the target are shown as (x, y). The target is a static, unknown radio device, which is the target radio device. The signal transmitted from the target is received by the sensor 10.

ここで、右辺の前者はセンサ１０が信号を受信した受信時間であり、後者はターゲットが信号を送信した送信時間である。未知のターゲットに対して送信時間は取得できないが、受信時間は取得できる。第１取得部１２ａは、サンプルのＴＯＡをＴＤＯＡサンプル測定部１４に出力する。 The first acquisition unit 12a acquires the TOA of the sample based on the signal received by each sensor 10 and transmitted from the target. TOA indicates the time difference between signals transmitted from the transmitting device and received by the receiving device. Here, the transmitting device corresponds to the target, and the receiving device corresponds to the sensor 10. TOA is shown as follows.

Here, the former on the right side is the reception time when the sensor 10 receives the signal, and the latter is the transmission time when the target transmits the signal. The transmission time cannot be obtained for an unknown target, but the reception time can be obtained. The first acquisition unit 12a outputs the TOA of the sample to the TDOA sample measurement unit 14.

ＴＤＯＡの導出はすべてのセンサ１０に対してなされる。 The TDOA sample measuring unit 14 receives the sample TOA from the first acquisition unit 12a and derives the sample TDOA between the sensors 10 from the sample TOA. This is to eliminate the need for knowledge of the transmission time. Specifically, the TDOA is derived by subtracting the TOA between the sensors 10 as follows.

Derivation of TDOA is made for all sensors 10.

The first factor graph processing unit 16 first estimates the position coordinates of the radio device by performing PTDOA-based factor graph processing using the TDOA of the sample between the sensors 10 derived in the TDOA sample measurement unit 14. Derived the value. Hereinafter, the first factor graph processing unit 16 will be described in detail.

The Euclidean factor node processing unit 40 in the first factor graph processing unit 16 receives the TDOA from the TDOA sample measurement unit 14. The Euclidean factor node processing unit 40 converts the sample TDOA into the difference in Euclidean distance by multiplying the TDOA by the propagation velocity of the electromagnetic wave. _{In particular, Ei and j} of the Euclidean factor node processing unit 40 convert K TDOAs into the difference in Euclidean distance, and calculate the mean value and the variance value. These are the messages exchanged in the factor graph.

ここで、「ｍ」が「平均値」を示し、「σ^２」が「分散値」を示し、矢印は、因子グラフのノード間におけるメッセージの進む方向を示す。 Only two factor nodes are directly connected to the Euclidean variable node processing unit 42. Euclidean variable node processing unit 42, E _i Euclidean factors node processing unit _40, D _i of the Euclidean distance estimation factor node processing unit 44 _j, to directly pass a message to _j. This message is presented in the form of an average value as in equation (3) and a variance value as in equation (4).

Here, "m" indicates the "mean value", "σ ² " indicates the "dispersion value", and the arrows indicate the direction in which the message travels between the nodes of the factor graph.

The Euclidean distance estimation factor node processing unit 44 _{Di, j} converts the difference in Euclidean distance into an equivalent Euclidean distance as follows.

R _h of sum product algorithm variable node processing unit 46, the sum product algorithm, D _i of the Euclidean distance estimation factor node processing unit _44, and updates the message from the C _h of _j and Pythagoras factor node processing unit 48. The subscript h indicates an index that is not related to either the first sensor 10 or the second sensor 10. Further, the thumb product algorithm variable node processing unit 46 advances the updated message to the destination node.

ここで、ｈは、ｈ＝ｉであるかｈ＝ｊである。また、ｈ＜ｎに対してａ＝Ｄ_ｈ，ｎ→ｒ_ｈであり、ｈ＞ｎに対してａ＝Ｄ_ｎ，ｈ→ｒ_ｈである。ノードｒ_ｈとＤ_ｉ，ｊとの間のメッセージの交換は、タイムスタンプの知識の必要性を除去するための重要なステップである。 Sum product algorithm variable node update message from r _h in the downlink processing unit 60 of the processor 46 to C _h Pythagorean factors node processing unit 48 is shown as formula (9) and (10). On the other hand, D _i of the sum product algorithm variable node Euclidean distance from r _h in the uplink processing part 62 of the processing unit 46 estimates factor node processing unit _44, update message to _j is represented by the equation (11) and (12) Is done. However, the first iteration of this step requires an initial value of the average value and variance value indicating a message from the C _h Pythagorean factors node processing unit 48.

Here, h is either h = i or h = j. In addition, h <a to the n _{= D h,} is a _n → _{r h, h>} is _{a = D n, h → r} h to the n. Exchange of messages between the node r _h and D _{i, j} is an important step to eliminate the need for knowledge of the time stamp.

ここで、ｍ^２ _{Δｘｈ→Ｃｈ}、ｍ^２ _{Δｙｈ→Ｃｈ}、δ^２ _{Δｘｈ→Ｃｈ}、δ^２ _{Δｙｈ→Ｃｈ}の初期値は、最初の繰り返しにおいて「０」に設定される。 C _h Pythagorean factors node processing unit 48 executes a Pythagorean calculation to the message. The downlink processing unit 64 of the Pythagoras factor node processing unit 48 derives a message of the relative distance between the _{variable nodes Δx h} and Δy _{h, as shown by the equations (13) to (16).} Up processing unit Pythagorean factors node processing unit 48 66, replaces the message _{r h} as shown in equation (17) and equation (18).

Here, ^{the initial values of m 2} _{Δxh → Ch} , m ² _{Δyh → Ch} , δ ² _{Δxh → Ch} , and δ ² _{Δyh → Ch} are set to “0” in the first iteration.

ここで、εは限定目的のパラメータであり、経験的に選択される。また、上り処理部６６は、次の２つの制限を加える。

ここで、δも限定目的のパラメータであり、経験的に選択される。 Furthermore, in order to maintain the stability of the messages exchanged in C _h Pythagorean factors node processing unit 48, a downlink processing part 64 adds the following restrictions.

Here, ε is a parameter for limited purposes and is empirically selected. Further, the uplink processing unit 66 imposes the following two restrictions.

Here, δ is also a parameter for limited purpose and is selected empirically.

ここで、平均値と分散値は、最初の繰り返しにおけて、初期値「０」にされる。 Since the Pythagoras variable node processing unit 50, like the Euclidean variable node processing unit 42, has _{only two nodes directly connected to the nodes Δx i} and Δy _i at relative distances, messages are exchanged directly between them. Node. That is exchanged between the C _h Pythagorean factors node processing unit 48 and A _h and B _h of the conversion factors node processing unit 52. The processing of the downlink processing unit 68 of the Pythagoras variable node processing unit 50 is shown in equations (19) and (20), and the processing of the uplink processing unit 70 is shown in equations (21) and (22).

Here, the mean value and the variance value are set to the initial values "0" in the first iteration.

_{A h} and B _h of the transforming factor node processing unit 52 transform the message of the relative distance in order to estimate the position of the target calculated by the variable nodes x _init and y _init. The processing of the downlink processing unit 72 of the conversion factor node processing unit 52 is shown in the equations (23) and (24), and the processing of the uplink processing unit 74 is shown in the equations (25) and (26).

ｘ_ｉｎｉｔに対する式（２７）と式（２８）は、ｙ_ｉｎｉｔにも同様に適用される。 In the iterative processing, the thumb product algorithm processing unit 54 _{adds a message to the variable nodes x init} and y _init according to the thumb product algorithm shown in the equations (27) and (28).

Equations (27) and (28) for x _init apply similarly to _{y init.}

The determination unit 56 determines whether or not the processing results of the conversion factor node processing unit 52 have converged. Here, the amount of change between the calculation result of the formula (27) and the calculation result of the formula (28) and the calculation result of the formula (27) and the calculation result of the formula (28) derived so far is the threshold value. When it becomes smaller than, it is judged to be convergent. The convergence test is _{also made for init} , and may be made based on either the mean value or the variance value. If it has not converged, the determination unit 56 returns _{the processing result of the thumb product algorithm processing unit 54 to A h} and B _h of the uplink processing unit 74 of the conversion factor node processing unit 52.

ｙ_ｉｎｉｔに対する演算は、式（２９）と式（３０）と同様である。（ｍ_{ｘｉｎｉｔ}，ｍ_{ｙｉｎｉｔ}）の最終的な値は、ＰＴＤＯＡ因子グラフで提供されるターゲットの荒い位置推定値を示し、前述の第１推定値に相当する。 In the output unit 58, when the determination unit 56 determines that the iterations have converged, the variable nodes x _init and y _init add all the incoming messages in the same manner as in equations (27) and (28).

The _{operation for y init} is the same as in equations (29) and (30). The final value of (m _xinit , _myinit ) indicates the rough position estimate of the target provided by the PTDOA factor graph and corresponds to the first estimate described above.

The identification unit 18 identifies four sensors 10 arranged near the first estimated value derived by the first factor graph processing unit 16 as observation spots. That is, the results of the PTDOA factor graph estimation algorithm at the variable nodes (x _init , y _init ) are used as the initial position of the DRSS factor graph and a rough estimate of the target position to select the four most appropriate observation spots. become.

The second acquisition unit 12b receives the training signal from the known transmission device in each of the plurality of sensors 10, and acquires the RSS of the sample in each sensor 10. Known transmitters are placed at known position coordinates. The second acquisition unit 12b executes the averaging process on the RSS of the sample. The averaging period is set to be long enough and to have a sufficient sample volume. This is because by reducing the effects of shadowing and momentary fading, only path loss fading remains and is error-free. The second acquisition unit 12b outputs the RSS of the averaged sample to the selection unit 20. The selection unit 20 selects the four sensors 10 specified by the specific unit 18, that is, the RSS for the observation spot, among the RSS acquired by the second acquisition unit 12b.

Following the processing of the RTDOA factor graph so far, the processing of the DRSS factor graph is executed. In the DRSS factor graph, the position estimation of the target (x, y) is obtained through the relationship between the power of the received signal and the PDP. Here, the power of the received signal is obtained by the sensor 10 and the first acquisition unit 12a, and the PDP is obtained from the above-mentioned training signal. Further, in the second acquisition unit 12b, the training signals received at the observation spots are averaged. All RSS samples, including the RSS of the training signal and the signal from the target, follow Gaussian assumptions. Therefore, this is modeled by a path loss model with only a sufficiently long mean. As mentioned above, a sufficiently long average excludes instantaneous fading variability and shadowing. The result is shown as follows.

Here, P indicates a power index of path loss, and dB is used as a unit. d, d ₀ , f _c , f, n are the Euclidean distance (in meters) from the target or observation spot to the hth sensor 10, the reference distance of the path loss index model (in meters), and the carrier frequency (in Hz). , The speed of light (m / s), and the coefficient of the power index of the path loss are shown.

The first acquisition unit 12a acquires the RSS of the sample based on the signal received by each sensor 10 and transmitted from the target. The first acquisition unit 12a averages the RSS of the sample over a sufficiently long period in order to eliminate the momentary fading fluctuation and shadowing. Therefore, the RSS of the sample is expressed by the path loss exponential function model because the measurement error is modeled by Gaussian noise that does not include "0" by collecting many independent factors.

The second factor graph processing unit 24 derives the DRSS of the sample between the sensors 10 from the RSS of the sample acquired in the first acquisition unit 12a, and executes the DRSS-based factor graph processing to obtain the position coordinates of the wireless device. Derivation of the second estimate. When executing the DRSS-based factor graph processing, the DRSS and the coefficient calculated by the coefficient calculation unit 22 are used, and the coefficient calculated by the coefficient calculation unit 22 will be described later. Hereinafter, the second factor graph processing unit 24 will be described in detail.

_{The pH} of the RSS factor node processing unit 80 of the second factor graph processing unit 24 converts the RSS sample from watt units to dB units. _{N ph} of RSS variable node processing unit 82 sends a message from the _{H ph} of RSS factors node processing section 80 _{G pi} of DRSS factors node processing unit _84, to _j.

これは、次のように示してもよい。

ＤＲＳＳ因子ノード処理部８４は、ＤＲＳＳ変数ノード処理部８６のＦ_ｐｉ，ｊでの繰り返し処理へのＤＲＳＳ入力メッセージとして使用するために次のような平均値と分散値を計算する。ここで、平均値と分散値は、測定値誤差がガウス分布にしたがうと想定されることによって導出される。

_{The GPi and j} of the DRSS factor node processing unit 84 subtract the RSS sample between the two sensors 10 to obtain the DRSS sample as follows.

This may be indicated as follows.

The DRSS factor node processing unit 84 calculates the following mean value and variance value for use as a DRSS input message for the iterative processing of the DRSS variable node processing unit 86 at F _{pi, j.} Here, the mean value and the variance value are derived by assuming that the measured value error follows a Gaussian distribution.

_{The N pi, j} of the DRSS variable node processing unit 86 directly passes _{the message from the G pi, j} of the DRSS factor node processing unit 84 _{to the F pi, j} of the linear plane factor node processing unit 88 as follows.

係数計算部２２は、４つの適切な観測スポットから送られたトレーニング信号をもとに、線形平面最小２乗（ＬＳ）方程式を解くことによってＰＤＰを生成する。線形平面最小２乗（ＬＳ）方程式は次のように示される。

ここで、ａ_ｘｉ，ｊ、ａ_ｙｉ，ｊ、ａ_Ｐｉ，ｊは、平面方程式の係数である。ｘとｙはターゲットの位置を示す。 The coefficient calculation unit 22 calculates the coefficient based on the RSS selected by the selection unit 20. Specifically, the coefficient calculation unit 22 derives the DRSS of the sample between the sensors 10 from the RSS of the training signals sent from the four appropriate observation spots as follows.

The coefficient calculation unit 22 generates a PDP by solving a linear plane least squares (LS) equation based on training signals sent from four appropriate observation spots. The linear plane least squares (LS) equation is shown as follows.

Here, a _{xi, j} , a _{yi, j} , and a _{Pi, j} are coefficients of the plane equation. x and y indicate the position of the target.

その結果、ＬＳ解は次のように示される。

その結果、係数計算部２２において導出される係数は次の通りである。

Further, the coefficient vector a, the coordinates of the observation spot and the DRSS matrix B, and the constant matrix C are shown as follows.

As a result, the LS solution is shown as follows.

As a result, the coefficients derived in the coefficient calculation unit 22 are as follows.

_{The Fpi, j} of the linear plane factor node processing unit 88 generates a linear plane in order to approximate the position of the target. The linear plane factor node processing unit 88 inputs the mean value and the variance value from the DRSS variable node processing unit 86, and also inputs the coefficient from the coefficient calculation unit 22. The linear plane factor node processing unit 88 derives the mean value and the variance value of the position coordinates x and y of the radio device based on the input mean value, variance value, and coefficient as follows.

式（４６）、式（４７）は、ｘに対するサムプロダクトアルゴリズムを示すが、ｙに対するサムプロダクトアルゴリズムも同様に示される。線形平面因子ノード処理部８８は、サムプロダクトアルゴリズムの実行結果のメッセージが収束するまで、処理を繰り返し実行する。収束についての条件は前述の通りであるので、ここでは説明を省略する。 Further, the linear plane factor node processing unit 88 executes the sum product algorithm on the derived mean value and variance value.

Equations (46) and (47) show a thumb product algorithm for x, but a thumb product algorithm for y is also shown. The linear plane factor node processing unit 88 repeatedly executes the processing until the message of the execution result of the thumb product algorithm converges. Since the conditions for convergence are as described above, the description thereof will be omitted here.

式（４８）、式（４９）は、ｘに対するサムプロダクトアルゴリズムを示すが、ｙに対するサムプロダクトアルゴリズムも同様に示される。繰り返し回数が終了しなければ、線形平面因子ノード処理部８８に戻される。繰り返し回収が終了した場合、平均値が、ターゲットの位置（ｘ，ｙ）とされる。このターゲットの位置が第２推定値に相当する。 The output unit 90 inputs the converged average value and the variance value in the linear plane factor node processing unit 88. The output unit 90 executes the sum product algorithm on the input average value and variance value.

Equations (48) and (49) show the thumb product algorithm for x, but the thumb product algorithm for y is also shown. When the number of repetitions is not completed, the process is returned to the linear plane factor node processing unit 88. When the repeated collection is completed, the average value is taken as the target position (x, y). The position of this target corresponds to the second estimate.

This configuration can be realized by the CPU, memory, or other LSI of any computer in terms of hardware, and by programs loaded in memory in terms of software, but here it is realized by cooperation between them. It depicts a functional block to be done. Therefore, it will be understood by those skilled in the art that these functional blocks can be realized in various ways by hardware only, software only, or a combination thereof.

The operation of the estimation device 100 based on the above configuration will be described. FIG. 2 is a flowchart showing a procedure of PTDOA-based factor graph processing by the estimation device 100. Equations (1) and (2) are executed (S10). Equations (3) and (4) are executed (S12). Equations (5) to (8) are executed (S14). The first factor graph processing unit 16 executes downlink processing (S16). If it has converged (Y in S18), Eq. (29) and Eq. (30) are executed (S20). When the repetition is completed (Y in S22), the first estimated value is output (S26). When it has not converged (N in S18) or when the repetition has not been completed (N in S22), the first factor graph processing unit 16 executes the uplink process (S24). Return to step 14.

FIG. 3 is a flowchart showing the procedure of downlink processing by the estimation device 100. Equations (9) and (10) are executed (S50). If m _{rh → Ch} <ε (N in S52), equations (13) to (16) are executed (S56). If m _{rh → Ch} <ε (Y in S52), then m _{rh → Ch} = ε (S54), and equations (13) to (16) are executed (S56). Equations (19) and (20) are executed (S58). Equations (23) and (24) are executed (S60). Equations (27) and (28) are executed (S62).

FIG. 4 is a flowchart showing the procedure of uplink processing by the estimation device 100. Equations (25) and (26) are executed (S100). Equations (21) and (22) are executed (S102). When m ² _{rh → Ch −} m _{Δxh → Ch} <0 (Y in S104), m _{Δxh → Ch} = sign (m _{Δxh → Ch} ) × (abs (m ² _{rh → Ch} ) −δ) ( S106). This is also done for y. If m ² _{rh → Ch −} m _{Δxh → Ch} <0 (N in S104), step 106 is skipped. Equations (17) and (18) are executed (S108). When σ ² _{Ch → rh} > max (σ ² _{Dh → rh} ) (Y in S110), σ ² _{Ch → rh} = max (σ ² _{Dh → rh} ) is set (S112). If σ ² _{Ch → rh} > max (σ ² _{Dh → rh} ) is not satisfied (N in S110), step 112 is skipped. Equations (11) and (12) are executed (S114). The observation spot is specified (S116).

FIG. 5 is a flowchart showing a procedure of calculation processing by the estimation device 100. Equation (37) is executed (S150). Equations (38) and (39) are executed (S152). Equation (40) is executed (S154). Equation (41) is executed (S156).

FIG. 6 is a flowchart showing a procedure of DRSS-based factor graph processing by the estimation device 100. Equation (32) is executed (S200). Equations (33) to (35) are executed (S202). Equation (36) is executed (S204). Equations (42) to (45) are executed (S206). Equations (46) and (47) are executed (S208). If it has converged (Y in S210), Eq. (48) and Eq. (49) are executed (S212). When the repetition is completed (Y in S214), the second estimated value is output (S216). If it has not converged (N in S210), or if the repetition has not been completed (N in S214), the process returns to step 206.

According to the embodiment of the present invention, since PTDOA is used, knowledge of time stamps can be eliminated. Moreover, since PTDOA is executed, the position of the target can be roughly estimated. Moreover, since the position of the target is roughly estimated, four observation spots can be selected from a plurality of sensors. Also, since four observation spots surrounding the target are selected, DRSS can be executed. Further, since the DRSS of the sample between the sensors is derived and the process is executed based on the RSS of the sample in each sensor, a relative value can be used. Moreover, since it is used relative to each other, it is possible to eliminate the need for information on absolute transmission power. Further, since the absolute transmission power information is not required, the position of the target that becomes the transmission source can be estimated based on the RSS even if there is no transmission power information. Moreover, since the position of the target that becomes the transmission source is estimated, it is possible to search for illegal radio stations and unknown sources.

The present invention has been described above based on examples. This embodiment is an example, and it will be understood by those skilled in the art that various modifications are possible for the combination of each of these components, and that such modifications are also within the scope of the present invention.

In this embodiment, the identification unit 18 specifies four sensors 10 as observation spots. However, the present invention is not limited to this, and for example, the specific unit 18 may specify four or more sensors 10 as observation spots. According to this modification, the estimation accuracy can be improved.

10 Sensor, 12 Acquisition part, 14 TDOA sample measurement part, 16 First factor graph processing part, 18 Specific part, 20 Selection part, 22 Coefficient calculation part, 24 Second factor graph processing part, 40 Euclidean factor node processing part, 42 Euclidean variable node processing unit, 44 Euclidean distance estimation factor node processing unit, 46 thumb product algorithm variable node processing unit, 48 Pythagoras factor node processing unit, 50 Pythagoras variable node processing unit, 52 conversion factor node processing unit, 54 thumb product algorithm processing unit. Unit, 56 Judgment unit, 58 Output unit, 60 Down processing unit, 62 Up processing unit, 64 Down processing unit, 66 Up processing unit, 68 Down processing unit, 70 Up processing unit, 72 Down processing unit, 74 Up processing unit, 80 RSS factor node processing unit, 82 RSS variable node processing unit, 84 DRSS factor node processing unit, 86 DRSS variable node processing unit, 88 linear plane factor node processing unit, 90 output unit, 100 estimator.

Claims (2)

A signal from a target wireless device is received by each of a plurality of sensors, and a first acquisition unit that acquires TOA (Time Of Arrival) and RSS (Receive Signal Strength) of a sample at each sensor, and a first acquisition unit.
A TDOA sample measuring unit for deriving a sample TDOA (Time Difference Of Arrival) between sensors from the sample TOA acquired in the first acquisition unit, and a TDOA sample measuring unit.
A first factor graph for deriving a first estimate of the position coordinates of a radio device by performing a Pythagoras TDOA-based factor graph process using the TDOA of the sample between the sensors derived in the TDOA sample measurement unit. Processing unit and
A specific unit that identifies at least four sensors arranged near the first estimated value derived in the first factor graph processing unit, and a specific unit.
A training signal from a known transmitter is received by each of the plurality of sensors, and a second acquisition unit that acquires RSS of a sample at each sensor, and
Among the RSS acquired in the second acquisition unit, a selection unit that selects RSS for at least four sensors specified in the specific unit, and
A coefficient calculation unit that calculates a coefficient based on the RSS selected in the selection unit, and a coefficient calculation unit.
DRSS (Differential RSS) of the sample between the sensors is derived from the RSS of the sample acquired in the first acquisition unit, and the DRSS-based factor graph processing is performed using the derived DRSS and the coefficient calculated in the coefficient calculation unit. The second factor graph processing unit that derives the second estimated value of the position coordinates of the wireless device by executing
An estimation device comprising. A signal from the target wireless device is received by each of the plurality of sensors, and a step of acquiring TOA (Time Of Arrival) and RSS (Receive Signal Strength) of a sample at each sensor, and
The step of deriving the sample TDOA (Time Difference Of Arrival) between the sensors from the acquired sample TOA, and
The step of deriving the first estimate of the position coordinates of the radio device by performing the Pythagoras TDOA-based factor graph processing using the sample TDOA between the derived sensors.
Steps to identify at least four sensors located near the derived first estimate, and
Training signals from known transmitters are received at each of the sensors, and the steps to obtain a sample RSS for the training signal at each sensor, and
Of the sample RSS for the training signal, the step of selecting the RSS for at least four identified sensors, and
Steps to calculate the coefficients based on the selected RSS,
A sample DRSS (Differential RSS) between sensors is derived from the sample RSS for the signal from the target radio device, and the DRSS-based factor graph processing is executed using the derived DRSS and the calculated coefficient. By doing so, the step of deriving the second estimated value of the position coordinates of the wireless device,
An estimation method characterized by comprising.

Images