JP7234803B2 - Target signal separation device, passive radar device and target signal separation method - Google Patents

Description

The present invention relates to a target signal separation device and a target signal separation method, and more particularly to technology for predicting noise signals.

It is known that when a moving target with a small RCS (Rader Cross Section) is detected by a radar, it is effective to use radio waves of a lower frequency. This is because a resonance phenomenon occurs when the wavelength of the radio wave is close to the size of the moving target. Radio waves in the VHF/UHF bands are considered effective for detecting flight targets.

Normally, when detecting a moving target, a radar using pulse waves is used for which signal processing is easy.
However, since the above-mentioned low-frequency radio waves are not permitted to be used for high-power radar using pulse waves, they must be used as continuous waves instead of pulse waves. Non-Patent Document 1 discloses a method of detecting a target with a passive radar using FM broadcast waves.

When a target is detected by radar, the received signal contains clutter, which is reflected waves from sources other than the target. Reflection from the ground surface and structures on the ground is called ground clutter, reflection from the sea surface and waves is called sea clutter, and reflection from rain clouds is called weather clutter.
Since the received signal of the passive radar is weak, it is important to remove the clutter and separate the target signal.

As a method of removing clutter, there is known a method of separating reflected signals from a moving target based on the difference between the moving speed of ground structures, waves, rain clouds, etc. and the moving speed of the moving target. In radar using pulse waves, it is possible to detect the Doppler velocity from the phase difference of the received signal with respect to each pulse signal. can be used to detect moving targets.
Further, Patent Literature 1 discloses a technique for adjusting the gain of an auxiliary antenna and removing ground clutter from a signal received by a main antenna.

Japanese Unexamined Patent Application Publication No. 2017-211238

“FM radio based bistatic radar”, IEE Proceedings - Radar, Sonar and Navigation (Vol.152, 2005)

A problem with using continuous waves to detect low RCS targets is the inability to filter out clutter.

When a continuous wave is used, signal processing using the phase difference of the received signal cannot be performed unlike the pulse wave. It is also believed that the scattering properties of clutter also change. Therefore, clutter cannot be removed like the conventional radar using pulse waves.
Further, the method described in Patent Document 1 can only remove ground clutter, and cannot remove weather clutter and sea clutter that fluctuate depending on the weather.

The present disclosure has been made in order to solve the above problems, and includes a target signal separating apparatus and a passive radar capable of separating signals from moving targets by removing clutter signals from continuous wave received signals. It is an object to provide an apparatus and target signal separation method.

The target signal separation device according to the present disclosure includes acquisition means for acquiring received signals and observed parameters, and a prediction formula obtained by machine learning using past received signals and past observed parameters when the target is absent as teacher data. analysis means for predicting a noise signal corresponding to the current observation parameter acquired by the acquisition means, and the current received signal acquired by the acquisition means and the noise signal predicted by the analysis means. and output means for performing a process of obtaining a difference and separating the target signal from the received signal when the difference exceeds a predetermined threshold.

The target signal separation method according to the present disclosure acquires the current received signal and the current observed parameter, and the prediction obtained by machine learning using the past received signal and the past observed parameter when the target is absent as teacher data Using the equation, predict the noise signal for the current observed parameter, perform processing to obtain the difference between the current received signal and the noise signal, and separate the target signal when the difference exceeds a predetermined threshold. It is a thing.

According to the present disclosure, it is possible to provide a target signal separation device, a passive radar device, and a target signal separation method capable of removing clutter signals from continuous wave received signals and separating signals from moving targets. can.

1 is a configuration diagram of a target signal separation device according to the present invention; FIG. 1 is a schematic diagram showing an overview of the present invention; FIG. FIG. 4 is a conceptual diagram showing reception directions and types of clutter used for noise signal prediction according to the present invention; FIG. 4 is a diagram showing an array D(i, j) indicating reception directions used for noise signal prediction according to the present invention; FIG. 4 is a diagram showing an example of a decision tree used for noise signal prediction according to the present invention; FIG. 4 is a diagram showing weather parameters corresponding to a decision tree used for noise signal prediction according to the present invention; FIG. 4 is a conceptual diagram showing a process of taking a difference in the target signal separation device according to the present invention; 4 is a flow chart showing an example of processing of the target signal separation device according to the present invention;

<First embodiment>
A configuration of a target signal separation device according to an embodiment of the present invention will be described with reference to FIG. This target signal separation device 10 includes an acquisition unit 11, an analysis unit 12, and an output unit 13, as shown in FIG.

Acquisition unit 11 acquires the current received signal.
The acquisition unit 11 also acquires the current observation parameters. Observation parameters are parameters related to observation conditions.

The analysis unit 12 predicts the noise signal in the current observation parameters acquired by the acquisition unit 11 based on the prediction formula. The prediction formula is calculated by machine learning using past received signals and past observed parameters obtained when no target exists as teacher data.
As the data acquired when the target does not exist and which is used as the teacher data, the data for which the output unit 13 determines that the target does not exist may be used.
The analysis unit 12 outputs the predicted noise signal to the output unit 13 .

The output unit 13 performs processing to find the difference between the current received signal acquired by the acquisition unit 11 and the noise signal output by the analysis unit 12, and determines that the target exists when the difference exceeds a threshold. Then, when it is determined that the target exists, the difference result is separated as the target signal.
The received signal and observed parameters when the threshold is not exceeded may be used for machine learning of the prediction formula used by the analysis unit 12 .

The target signal separation device according to this embodiment predicts a noise signal using a prediction formula obtained by machine learning. Therefore, it is possible to cope with changes in clutter accompanying changes in observation parameters. Therefore, the target signal separating apparatus according to the present embodiment can separate the target signal from the received continuous wave signal.

<Second embodiment>
An overview of the target signal separation device 10 according to the embodiment of the present invention will be described with reference to FIG. The target signal separation device 10 separates the target signal from the signal received by the passive radar 23 . A target signal is a signal in which a continuous wave transmitted by the transmitting station 21 is reflected by the moving target 22 . The radio waves transmitted by the transmitting station 21 are, for example, FM broadcast waves, television broadcast waves, or radio waves used in mobile phones, but may be other VHF/UHF band radio waves. The moving targets 22 are, for example, airplanes, ships, automobiles, and may be low RCS targets.

A configuration of a target signal separation device 10 according to an embodiment of the present invention will be described with reference to FIG. This target signal separation device 10 includes an acquisition unit 11, an analysis unit 12, and an output unit 13, as shown in FIG.

Acquisition unit 11 acquires a current received signal from radar, which is observation equipment. The radar is a passive radar that only receives.
The received signal includes the target signal, which is the radio wave reflected by the moving target 22, and the noise signal, which is the radio wave reflected by the clutter.

Acquisition unit 11 acquires measurement parameters, weather parameters, and reception parameters.
A measurement parameter is a parameter related to the time and position of the measurement. The measurement parameters are, for example, measurement date and time or measurement location (latitude, longitude, altitude).
A weather parameter is a parameter related to the weather. Weather parameters are, for example, weather (sunny, cloudy, rainy, snowy, foggy), amount of rain/precipitation, amount of clouds, atmospheric pressure, wind speed/direction, temperature, and humidity.
A reception parameter is a parameter related to a reception condition. The reception parameters are, for example, reception time, reception frequency, reception direction (elevation angle θ, horizontal direction φ), and reception resolution.
Meteorological data such as temperature and wind speed are obtained from meteorological equipment or the Internet.

Acquisition unit 11 associates the received signal with each parameter and stores them in a database provided in acquisition unit 11 . The received signal is Fourier transformed and stored.
At this time, the reception direction (elevation angle .theta., horizontal direction .phi.) may be set as a fixed parameter, and weather/reception parameters other than the reception direction may be stored in a separate table in an optimum table format. When obtaining a prediction formula for each reception direction (elevation angle direction θ, horizontal direction φ), machine learning can be performed using training data in the corresponding reception direction without using training data in other reception directions. In this case, the amount of teacher data used for machine learning can be reduced, and learning can be performed at high speed, which is advantageous when the number of times of learning is large.
Acquisition unit 11 outputs the current received signal to output unit 13 . The acquisition unit 11 also outputs the current weather parameters and the current reception parameters to the analysis unit 12 .

The analysis unit 12 predicts the noise signal in the current weather/reception parameters acquired by the acquisition unit 11 based on the prediction formula. The prediction formula is created by machine learning using past received signals, past weather and reception parameters obtained when no target exists as teacher data.
As the data acquired when the target used for supervised learning does not exist, the data for which the output unit 13 determines that the target does not exist can be used.

ωは周波数を表す。ｋは、観測の時期、つまり観測日時を表すパラメータである。a_0、a_kは回帰係数、つまり重み付け係数である。ｆ_ｋ（ω）は受信強度関数である。ｋは１からｎの整数とする。 The prediction formula will be described below.
As will be described later, the prediction formula is basically obtained by combining an array D(i, j) corresponding to the frequency and the reception direction (elevation angle direction θ _{i ,} horizontal direction φ _j ) as variables with regression coefficients as weighting coefficients. configuration.
The prediction formula is a function representing the reception intensity of clutter. First, considering that the prediction formula is represented by partial sums of received signals at past observation dates and times observed under various weather conditions, the prediction formula F(ω) is expressed as in formula (1).

ω represents frequency. k is a parameter representing the time of observation, that is, the date and time of observation. a _0, a _k are regression coefficients, ie weighting coefficients. f _k (ω) is the received strength function. Let k be an integer from 1 to n.

式（１）と同様に、ωは周波数を表す。ｋは、観測の時期、つまり観測日時を表すパラメータである。a_0|p、a_k|pは回帰係数、つまり重み付け係数である。ｆ_ｋ（ω）は受信強度関数である。ｋは１からｎの整数とする。パラメータｐは、後述する機械学習によって決定される。 Since the occurrence of clutter is affected by weather conditions, the regression coefficients in Equation (1) vary with weather. Therefore, using a parameter p that indicates the type of prediction formula, the prediction formula F _p (ω) is expressed as the following formula (2).

As in equation (1), ω represents frequency. k is a parameter representing the time of observation, that is, the date and time of observation. a _0|p , a _k|p are regression coefficients, ie weighting coefficients. f _k (ω) is the received strength function. Let k be an integer from 1 to n. The parameter p is determined by machine learning, which will be described later.

Furthermore, the regression coefficients of the prediction formula are different for each direction. For example, if the receiving antenna is installed on a mountain, the ratio of sea clutter, ground clutter, and weather clutter changes depending on the receiving direction (elevation angle θ, horizontal direction φ). FIG. 3 shows the relationship between reception directions (elevation angle θ, horizontal direction φ) and clutter. When the surface of the earth is used as a reference, θ represents the angle in the vertical plane and φ represents the angle in the horizontal plane. Ground clutter is waves reflected by the ground surface and structures on the ground surface. Sea clutter is reflected waves from the sea surface and waves. Weather clutter is waves reflected by rain clouds.

式（１）と同様に、ωは周波数を表す。ｋは、観測の時期、つまり観測日時を表すパラメータである。a_0|p、a_k|pは回帰係数、つまり重み付け係数である。ｆ_ｋ（ω）は受信強度関数である。ｋは１からｎの整数である。
Ｄ（ｉ，ｊ）は受信方向（仰角方向θ_i、水平方向φ_j）に対応する配列である。Ｄ（ｉ，ｊ）は、仰角方向θを１～Iに分割し、水平方向φを１～Jに分割することにより、受信方向をI×Jの要素に分割した配列である。Ｄ（ｉ，ｊ）と受信方向（仰角方向θ_i、水平方向φ_j）の関係を図４に示す。図４において、Ｄ_ijは、仰角方向がθ_iで水平方向がφ_jである受信方向を表している。
式（３）を用いることで、後述する決定木が受信方向ごとに作成されることになる。
クラッタは、受信方向並びに、気温、湿度、風向などの観測データを与えれば再現できるため、式（３）を予測式として用いた機械学習によりクラッタの受信強度を計算することができる。 Therefore, F _p,D (ω) shown in Equation (3) obtained by adding information about direction D to Equation (2) is used as a prediction equation.

As in equation (1), ω represents frequency. k is a parameter representing the time of observation, that is, the date and time of observation. a _0|p , a _k|p are regression coefficients, ie weighting coefficients. f _k (ω) is the received strength function. k is an integer from 1 to n.
D(i, j) is an array corresponding to the reception direction (elevation angle θ _{i ,} horizontal direction φ _j ). D(i, j) is an array obtained by dividing the reception direction into I×J elements by dividing the elevation direction θ into 1 to I and by dividing the horizontal direction φ into 1 to J. FIG. 4 shows the relationship between D(i, j) and the reception direction (elevation angle θ _{i ,} horizontal direction φ _j ). In FIG. 4, D _ij represents the receiving direction whose elevation direction is θ _i and whose horizontal direction is φ _j .
By using Equation (3), a decision tree, which will be described later, is created for each receiving direction.
Clutter can be reproduced by giving observation data such as reception direction, temperature, humidity, wind direction, etc. Therefore, the clutter reception intensity can be calculated by machine learning using equation (3) as a prediction equation.

The analysis unit 12 creates a decision tree that classifies the regression coefficients of Equation (3) for data acquired when no target exists. In order to learn the branching conditions, each parameter is set at the branching point, and calculation is repeated using the prediction formula. Entropy used in information theory can be used for the arrangement order and branching conditions in the decision tree. Entropy is an index representing the disorder of information. The degree of learning effect is reflected in the regression coefficient.

FIG. 5 shows an analysis example using a decision tree. This is an example of using numerical values of temperature, humidity, and wind speed, which are weather attribute classifications, as branching conditions. Entropy can be used to determine the end of learning.
An example of weather parameters corresponding to the decision tree of FIG. 5 is shown in FIG. FIG. 6 shows data in a state where the moving target 22 shown in FIG. 2 does not exist, and the numerical condition (greater than or less than) of each item corresponds to the branch point of the decision tree. Here, the state in which the moving target 22 does not exist is a state in which the signal from the moving target 22 could not be separated when the process of obtaining the difference was performed in the past.
In the case of the analysis example shown in the decision tree of FIG. 5, when the receiving antenna is oriented at a certain angle, the temperature parameter is 26° C. and branches to “wind speed”. Next, the wind speed is classified into 8 [m/s] or more and less than 8 [m/s].
For example, when the wind is strong, the waves are high, so the waves rise and the received intensity of the reflected waves due to sea clutter increases. This is the same characteristic for other clutter, and the clutter reception intensity depends on the reception direction (θ, φ).
As a result of machine learning, even after "wind speed", it is classified according to the numerical conditions of the parameters given to the branch points, and a prediction formula is obtained.

In FIG. 5, the branch destination _Fp of the decision tree represents a prediction formula. Prediction formulas F ₁ , F ₂ , F ₃ , and F ₄ are No. 1 shown in FIG. 1, 2, 3 and 4 weather parameters are supported.
The decision tree shown in FIG. 5 is created by using machine learning to determine the presence/absence of branching at each branching point and determination values that serve as branching references. Regarding the presence or absence of branching, for example, it is learned that there is no further branching when the temperature is less than 5°C. The prediction formula at this time is _F1 . The regression coefficient a _k|p of the prediction formula F _p is obtained by machine learning.
Further, as a result of machine learning, the judgment value that serves as a reference for branching may differ depending on the branching point even if the weather parameter is the same. For example, when the prediction formula is _F2 , the humidity is 80% or more, so the judgment value is 80. On the other hand, when the prediction formula is _F3 , the humidity is less than 70%, so The judgment value is 70.
Therefore, the presence/absence of branching at each branching point of the decision tree, the judgment value that serves as a reference for branching, and the regression coefficient a _k|p are obtained as a result of machine learning.

The analysis unit 12 obtains the parameter p of the prediction formula corresponding to the weather/reception parameter received from the acquisition unit 11 . Then, the noise signal is predicted by the prediction formula (3) corresponding to p and output to the output unit 13 .

The output unit 13 performs processing to obtain the difference between the current received signal obtained by the obtaining unit 11 and the noise signal output by the analyzing unit 12, and determines that the target exists when the difference exceeds a threshold. In such a case, the output unit 13 separates the difference result as the target signal. The threshold may be determined based on an average value or the like by performing statistical processing. FIG. 7 is a conceptual diagram showing the process of obtaining the difference. A predicted value is calculated by a prediction formula that is a linear sum of reception intensity functions, and a difference from the current observed value is calculated to calculate an estimated value.
Since the present invention uses a continuous wave instead of a pulse wave, it is easy to perform processing for obtaining a difference. The reason is as follows. In the pulse wave method, since radio waves are emitted intermittently on the time axis, there is no modulation in the time obtained by subtracting the pulse width from the pulse repetition period. On the other hand, in the continuous wave method, since radio waves are emitted continuously on the time axis, the reception intensity of the reflected waves from the target and clutter can be continuously obtained.

The output unit 13 sets the reception direction in which the target exists as the target direction in which the target exists. Also, the distance to the target can be obtained from the difference in reception time between the signal transmitted from the transmitting station 21 and the signal reflected by the target. The position of the target can be predicted from the position of the transmitting station 21, the position of the passive radar 23, the target direction, and the distance.
The transmission signal may be received by a dedicated antenna, or may be received using an optical fiber or the like.
The received signal and parameters when the threshold is not exceeded can be used to create a prediction formula used by the analysis unit 12 .
The output unit 13 transmits the determination result to the acquisition unit 11 . This is for classifying the received signals stored in the acquisition unit 11 according to the presence or absence of the moving target 22 .

Next, an example of processing in the target signal separation device according to the second embodiment of the present invention will be described with reference to FIG.

The acquisition unit 11 saves past received signals and past measurement/weather/reception parameters in a database (S101). The data is sorted based on whether the target signal is present or not. Observation is actually performed regardless of the presence or absence of the moving target 22 .

The acquisition unit 11 Fourier-transforms the received signal acquired by the observation equipment at the present time, and stores it in the database. The acquisition unit 11 saves the current measurement/weather/reception parameters in the database (S102). After determining the presence or absence of the target signal (S104), the data are classified according to the presence or absence of the target signal.

The analysis unit 12 performs machine learning to calculate a prediction formula based on past received signals and past weather/reception parameters, which are classified when no target exists on the database. Note that machine learning need not be performed for each observation, and a prediction formula calculated in advance may be used.
Then, the analysis unit 12 calculates a prediction value of the reception strength of the clutter signal from a prediction formula corresponding to the current measurement/weather/reception parameters (S103).

The output unit 13 performs processing to obtain the difference between the predicted noise signal and the current received signal. The output unit 13 separates the target signal if it is equal to or greater than the threshold. If it is equal to or less than the threshold, the current received signal is classified as data with no target in the acquisition unit 11, and can be used as learning data (S104). In addition, the following observations can be made continuously.

The target signal separation device according to the present embodiment manages past observation data associated with a large number of parameters as a database, so that highly accurate prediction can be performed by machine learning. Also, by increasing the number of times of learning, it is possible to improve the prediction accuracy of target separation.
Further, the target signal separating apparatus according to the present embodiment can separate the target signal from the received continuous wave signal. This is because when a continuous wave is used, the result of the Fourier transform becomes a stable waveform, and it is easy to perform the process of obtaining the difference.

Note that the content of the present embodiment is not limited to the above description. In the above description, the clutter signal is considered as noise generated by radio wave irradiation, but similar processing can be performed even if other noise is included. Other noise is, for example, fading due to wavelength interference.

In the above explanation, the explanation was given as a radar for ground equipment, but it can also be used for airplanes and artificial satellites that detect targets in the air.
Also, when detecting an unmanned aircraft or a future flying car with a passive radar, in addition to FM broadcast waves, television broadcast waves and radio waves used for mobile phones can also be used.

Some or all of the above-described embodiments can also be described in the following supplementary remarks, but are not limited to the following.

(Appendix 1)
acquisition means for acquiring the received signal and the observed parameters;
Analysis of predicting the noise signal corresponding to the current observation parameters acquired by the acquisition means using a prediction formula obtained by machine learning using past received signals and past observation parameters as teacher data when the target is absent. means and
Performing a process of obtaining a difference between the current received signal obtained by the obtaining means and the noise signal predicted by the analyzing means, and separating the target signal from the received signal when the difference exceeds a predetermined threshold. an output means;
A target signal separator, comprising:

(Appendix 2)
2. The target signal separating apparatus of claim 1, wherein the observed parameters include weather parameters.

(Appendix 3)
3. A target signal separation apparatus according to any one of claims 1 or 2, wherein said observation parameter includes reception direction.

(Appendix 4)
The analysis means performs machine learning using past received signals and past observation parameters as teacher data to predict a noise signal corresponding to the current observation parameters acquired by the acquisition means. A target signal separation device according to any one of the preceding claims.

(Appendix 5)
5. Any one of Appendices 1 to 4, wherein the analysis means uses a prediction formula that is a linear sum of past received signals and predicts the noise signal using the prediction formula obtained by creating a decision tree through machine learning. A target signal separation device as described in .

(Appendix 6)
6. The target signal separating apparatus according to any one of appendices 1 to 5, wherein the obtaining means associates the received signal with the observation parameter and stores the same in a database.

(Appendix 7)
7. The target signal separating apparatus according to appendix 6, wherein the acquiring means manages a part of the observation parameters other than the reception direction in a table different from that for the reception direction.

(Appendix 8)
8. The target signal separation device according to any one of appendices 1 to 7, wherein the received signal is a VHF/UHF band radio wave.

(Appendix 9)
9. The target signal separation device according to any one of appendices 1 to 8, wherein the received signal is an FM broadcast wave, a television broadcast wave, or a radio wave used in a mobile phone.

(Appendix 10)
A passive radar apparatus comprising a passive radar and a target signal separation apparatus according to any one of appendices 1 to 9.

(Appendix 11)
Get the current received signal and the current observed parameters,
Predicting the noise signal for the current observation parameter using a prediction formula obtained by machine learning using the past received signal when the target is absent and the past observation parameter as teacher data,
performing a process of obtaining a difference between the current received signal and the noise signal, and separating the target signal from the received signal when the difference exceeds a predetermined threshold;
Target signal separation method.

REFERENCE SIGNS LIST 10 target signal separation device 11 acquisition unit 12 analysis unit 13 output unit 21 transmission station 22 moving target 23 passive radar

Claims (10)

acquisition means for acquiring the received signal and the observed parameters;
Analysis of predicting the noise signal corresponding to the current observation parameters acquired by the acquisition means using a prediction formula obtained by machine learning using past received signals and past observation parameters as teacher data when the target is absent. means and
Performing a process of obtaining a difference between the current received signal obtained by the obtaining means and the noise signal predicted by the analyzing means, and separating the target signal from the received signal when the difference exceeds a predetermined threshold. an output means;
A target signal separator, comprising: 2. The target signal separating apparatus of claim 1, wherein said observed parameters include weather parameters. 3. A target signal separating apparatus according to claim 1 or 2, wherein said observation parameter comprises reception direction. 4. The analyzing means performs machine learning using past received signals and past observed parameters as teacher data to predict a noise signal corresponding to the current observed parameters acquired by the acquiring means. The target signal separation device according to any one of Claims 1 to 3. 5. Any one of claims 1 to 4, wherein the analysis means uses a prediction formula that is a linear sum of past received signals, and predicts the noise signal using the prediction formula obtained by creating a decision tree through machine learning. A target signal separation device according to any one of the preceding paragraphs. 6. The target signal separating apparatus according to any one of claims 1 to 5, wherein said acquiring means associates a received signal with an observed parameter and stores them in a database. 7. The target signal separation device according to claim 1, wherein said received signal is a VHF/UHF band radio wave. The target signal separation device according to any one of claims 1 to 7, wherein the received signal is an FM broadcast wave, a television broadcast wave, or a radio wave used in mobile phones. A passive radar device comprising a passive radar and a target signal separation device according to any one of claims 1 to 8. Get the current received signal and the current observed parameters,
Predicting the noise signal for the current observation parameter using a prediction formula obtained by machine learning using the past received signal when the target is absent and the past observation parameter as teacher data,
performing a process of obtaining a difference between the current received signal and the noise signal, and separating the target signal from the received signal when the difference exceeds a predetermined threshold;
Target signal separation method.

Images