WO2019016853A1

WO2019016853A1 - Wireless machine identification device and wireless machine identification method

Info

Publication number: WO2019016853A1
Application number: PCT/JP2017/025914
Authority: WO
Inventors: 貴文松田; 正資大島; 網嶋　武; 信弘鈴木
Original assignee: 三菱電機株式会社
Priority date: 2017-07-18
Filing date: 2017-07-18
Publication date: 2019-01-24
Also published as: JP6567242B2; JPWO2019016853A1

Abstract

The present invention is provided with: a rise detection unit (3) which calculates an energy time waveform in a spectrogram of a Fourier transform signal output from a Fourier transform unit (2) and detects a rise time of a wireless signal from the energy time waveform; and a feature amount calculation unit (7) which extracts, from the Fourier transform signal output from the Fourier transform unit (2), a Fourier transform signal in a time period including the rise time detected by the rise detection unit (3) and calculates, from a complex time signal of a set frequency component included in the extracted Fourier transform signal, a variation of the extracted Fourier transform signal as the feature amount, wherein a wireless machine identification unit (11) identifies a wireless machine on the basis of the feature amount calculated by the feature amount calculation unit (7).

Description

RADIO IDENTIFICATION DEVICE AND RADIO IDENTIFICATION METHOD

The present invention relates to a wireless device identification device for identifying a wireless device and a wireless device identification method.

It is known that there are individual differences in transient response such as rising of a radio signal transmitted from a wireless device, and in recent years, a wireless device is identified by analyzing the wireless signal transmitted from the wireless device. Machine identification devices have been proposed.
For example, the wireless device identification device calculates the feature amount of the rising edge of the wireless signal transmitted from the wireless device, and identifies the wireless device based on the calculated feature amount.
The characteristic quantities of the rising edge of the wireless signal include the dispersion value of the amplitude or the like in the wireless signal at the rising edge, the skewness, and the kurtosis.

However, in an environment where the signal-to-noise ratio (SNR) is low, the calculation accuracy of the rising feature amount is reduced, and the identification accuracy of the wireless device is reduced. For this reason, the following Patent Document 1 discloses a wireless device identification apparatus that calculates the feature value of the rising after improving SNR by performing short-time Fourier transform on a wireless signal transmitted from the wireless device.
In this wireless device identification device, the rise time constant is calculated as the rise feature amount.

Japanese Patent Application Laid-Open No. 2002-26826

The conventional wireless device identification apparatus employs a wireless device identification method based on the rise time constant. In the radio identification method based on the rise time constant, the radio type can be identified. However, even with different types of wireless devices, there is a problem in that it is not possible to identify an individual wireless device because the difference in the rise time constants is small even if the wireless devices are different.

The present invention has been made to solve the problems as described above, and it is an object of the present invention to obtain a wireless device identification device and a wireless device identification method capable of performing individual identification of a wireless device.

The wireless device identification apparatus according to the present invention performs short-time Fourier transform of a wireless signal received by the signal receiving unit and a signal receiving unit that receives a wireless signal transmitted from a wireless device to be identified. A Fourier transform unit that outputs a Fourier transform signal indicating a result of Fourier transform, and a time waveform of energy in a spectrogram of the Fourier transform signal output from the Fourier transform unit are calculated, and a rise time of a wireless signal is detected from the time waveform of energy. The Fourier transform signal of the time zone including the rise time detected by the rise detection unit is extracted from the rise detection unit and the Fourier transform signal output from the Fourier transform unit, and is included in the extracted Fourier transform signal The variation of the Fourier transform signal extracted from the complex time signal of the set frequency component A feature quantity calculating unit that calculates Te provided, the wireless device identification unit, based on the feature amount calculated by the feature calculation unit, in which so as to identify the radio.

According to the present invention, the rise detection unit which calculates the time waveform of energy in the spectrogram of the Fourier transform signal output from the Fourier transform unit and detects the rise time of the wireless signal from the time waveform of energy, and the output from the Fourier transform unit The Fourier transform signal of the time zone including the rise time detected by the rise detection unit is extracted from the detected Fourier transform signal, and the complex time signal of the set frequency component included in the extracted Fourier transform signal is extracted Since the wireless device identification unit is configured to identify the wireless device based on the feature amount calculated by the feature amount calculation unit, the feature amount calculation unit calculates the variation of the Fourier transform signal as the feature amount. There is an effect that can identify the individual of the radio.

It is a block diagram which shows the radio | wireless machine identification apparatus by Embodiment 1 of this invention. It is a hardware block diagram which shows the radio | wireless machine identification apparatus by Embodiment 1 of this invention. It is a hardware block diagram of a computer in case the component except the signal receiving part 1 in a radio | wireless apparatus identification apparatus is implement | achieved by software or a firmware. It is a flowchart which shows the learning process in the radio | wireless machine identification method by Embodiment 1 of this invention. FIG. 6 is an explanatory view showing STFTs of a learning radio signal s _g (n) and a discrimination radio signal s _d (n) by the Fourier transform unit 2. FIG. 6 is an explanatory view showing a detection process of rising time T ₀ by the rising detection unit 3; FIG. 8 is an explanatory view showing calculation processing of a feature amount by the feature amount calculation unit 7; In FIG. 8A, since the parameter for calculating the feature amount is not appropriate, the feature amount vector 係る for the wireless device of type (A), the feature amount vector ◇ for the wireless device of type (B), and the type (C) Is an explanatory view showing an example in which the feature quantity vector Δ pertaining to the wireless device partially overlaps, and FIG. 8B is a feature quantity pertaining to the wireless device of type (A) because the parameters for feature value calculation are appropriate. FIG. 9 is an explanatory view showing an example in which a vector と, a feature amount vector ◇ for a wireless device of type (B), and a feature amount vector Δ for a wireless device of type (C) are not overlapped. FIG. 5 is an explanatory view showing a learning wireless signal to which a serial number is stored, which is stored in a first database unit 9; FIG. 6 is an explanatory view showing an example of feature amounts of a plurality of known wireless devices stored in a second database unit 10; It is a flowchart which shows the identification process in the radio | wireless machine identification method by Embodiment 1 of this invention. FIG. 6 is an explanatory view showing identification processing of a wireless device by the wireless device identification unit 11;

Hereinafter, in order to explain the present invention in more detail, a mode for carrying out the present invention will be described according to the attached drawings.

Embodiment 1
FIG. 1 is a block diagram showing a radio device identification apparatus according to a first embodiment of the present invention.
FIG. 2 is a hardware configuration diagram showing a radio device identification apparatus according to Embodiment 1 of the present invention.
In FIGS. 1 and 2, the signal receiving unit 1 is realized by, for example, the signal receiving circuit 21 shown in FIG.
The signal receiving unit 1 receives a wireless signal transmitted from a known wireless device as a learning wireless signal, and converts the received learning wireless signal from an analog signal to a digital signal to thereby obtain a digital learning wireless signal. Output to the Fourier transform unit 2 and the first database unit 9.
In addition, the signal reception unit 1 receives a wireless signal transmitted from a wireless device to be identified as a wireless signal for identification, and converts the received wireless signal for identification from an analog signal to a digital signal to thereby perform digital identification. The radio signal is output to the Fourier transform unit 2.

The Fourier transform unit 2 is realized by, for example, the Fourier transform circuit 22 shown in FIG.
The Fourier transform unit 2 performs short time Fourier transform (STFT: Short Time Fourier Transform) on the digital learning wireless signal output from the signal receiving unit 1, and obtains the STFT result that is the short time Fourier transform of the learning wireless signal. A process of outputting the Fourier transform signal shown below (hereinafter referred to as a learning Fourier transform signal) to the rise detection unit 3 and the feature quantity calculation unit 7 is performed.
Further, the Fourier transform unit 2 performs STFT on the digital identification radio signal output from the signal reception unit 1 and indicates a Fourier transform signal (hereinafter referred to as identification Fourier transform signal) indicating the STFT result of the identification radio signal. The processing to be output to the rising edge detection unit 3 and the feature amount calculation unit 7 is performed.
The STFT is a method of extracting radio signals of different times at fixed time intervals, and performing fast Fourier transform (FFT) on the extracted radio signals.

The rising edge detection unit 3 is realized by, for example, the rising edge detection circuit 23 shown in FIG.
The rising edge detection unit 3 calculates a time waveform of energy in the spectrogram of the learning Fourier transform signal output from the Fourier transform unit 2 (hereinafter referred to as an energy time waveform for learning), and learns from the energy time waveform for learning A process of detecting the rise time of the wireless signal is performed.
Further, the rising edge detection unit 3 calculates a time waveform of energy in the spectrogram of the identification Fourier transform signal output from the Fourier transform unit 2 (hereinafter, referred to as an energy time waveform for identification), and an energy time waveform for identification From this, the process of detecting the rise time of the identification radio signal is performed.

The parameter setting unit 4 includes a first parameter setting unit 5 and a second parameter setting unit 6, and is realized by, for example, the parameter setting circuit 24 illustrated in FIG.
The first parameter setting unit 5 sets a parameter indicating a time range Δt which is a length of a time zone including the rising time detected by the rising detection unit 3 as a parameter for calculating the feature amount, and a setting frequency component A process of setting parameters indicating frequency bins is performed.
The second parameter setting unit 6 performs a process of setting weighting parameters indicating the weights of the first to ninth feature amounts calculated by the feature amount calculation unit 7.
Further, the second parameter setting unit 6 carries out a process of setting the number of samples of the window function and the number of samples for shifting the window function used when the STFT is performed by the Fourier transform unit 2.

The feature amount calculation unit 7 is realized by, for example, the feature amount calculation circuit 25 shown in FIG.
The feature amount calculation unit 7 recognizes the time zone including the rise time and the set frequency bin from the parameters for feature amount calculation set by the parameter setting unit 4.
Among the learning Fourier transform signals output from the Fourier transform unit 2, the feature amount calculation unit 7 performs a learning Fourier transform signal of a time zone including the rise time in the learning wireless signal detected by the rise detection unit 3. Implement the process to extract.
Then, the feature amount calculation unit 7 performs processing of calculating, as a feature amount, the fluctuation of the extracted learning Fourier transform signal from the complex time signal of the set frequency bin included in the extracted learning Fourier transform signal.

In addition, the feature quantity calculation unit 7 performs Fourier transform for identification of a time zone including a rise time in the identification wireless signal calculated by the rise detection unit 3 among the identification Fourier transform signals output from the Fourier transform unit 2. Implement the process of extracting the signal.
Then, the feature amount calculation unit 7 performs processing of calculating, as a feature amount, the fluctuation of the extracted identification Fourier transform signal from the complex time signal of the set frequency component included in the extracted identification Fourier transform signal.

The database unit 8 includes a first database unit 9 and a second database unit 10, and is realized by, for example, the database circuit 26 shown in FIG.
The first database unit 9 stores the digital learning radio signal output from the signal receiving unit 1.
The second database unit 10 stores the fluctuation of the learning Fourier transform signal calculated by the feature amount calculation unit 7 as a feature amount related to a known wireless device.

The wireless device identification unit 11 is realized by, for example, the wireless device identification circuit 27 shown in FIG.
The wireless device identification unit 11 compares the feature amount of the wireless device to be identified calculated by the feature amount calculation unit 7 with the feature amount of the known wireless device stored by the second database unit 10. Perform the process.
The wireless device identification unit 11 carries out a process of identifying the wireless device to be identified based on the comparison result of the feature amount of the wireless device to be identified and the feature amount of the known wireless device.

In FIG. 1, the signal receiving unit 1, the Fourier transform unit 2, the rising edge detecting unit 3, the parameter setting unit 4, the feature amount calculating unit 7, the database unit 8 and the wireless device identification unit 11 which are components of the wireless device identification device However, it assumes what is implement | achieved by the special purpose hardware as shown in FIG. That is, what is realized by the signal reception circuit 21, the Fourier transform circuit 22, the rising edge detection circuit 23, the parameter setting circuit 24, the feature amount calculation circuit 25, the database circuit 26, and the radio device identification circuit 27 is assumed.

Here, the database circuit 26 is, for example, nonvolatile or random access memory (RAM), read only memory (ROM), flash memory, erasable programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM) or the like. A volatile semiconductor memory, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disc) corresponds to this.
Further, the signal reception circuit 21, the Fourier transform circuit 22, the rise detection circuit 23, the parameter setting circuit 24, the feature quantity calculation circuit 25 and the wireless device identification circuit 27 are, for example, a single circuit, a composite circuit, a programmed processor, parallel A programmed processor, an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of these is applicable.

The components excluding the signal receiving unit 1 in the wireless device identification device are not limited to those realized by dedicated hardware, and the components excluding the signal receiving unit 1 in the wireless device identification device are software, firmware, or , And may be realized by a combination of software and firmware.
The software or firmware is stored as a program in the memory of the computer. A computer means hardware that executes a program, and for example, a central processing unit (CPU), a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, or a digital signal processor (DSP). Do.

FIG. 3 is a hardware configuration diagram of a computer in the case where components excluding the signal receiving unit 1 in the wireless device identification device are realized by software or firmware.
When components other than the signal receiving unit 1 in the wireless device identification apparatus are realized by software or firmware, the database unit 8 is configured on the memory 31 of the computer, and the Fourier transform unit 2, the rising edge detecting unit 3, parameter setting A program for causing the computer to execute the processing procedure of the unit 4, the feature value calculation unit 7 and the wireless device identification unit 11 is stored in the memory 31, and the processor 32 of the computer executes the program stored in the memory 31. do it.

Also, FIG. 2 shows an example in which each of the components of the wireless device identification device is realized by dedicated hardware, and FIG. 3 shows an example in which the wireless device identification device is realized by software or firmware. However, some of the components of the wireless device identification device may be realized by dedicated hardware, and the remaining components may be realized by software, firmware or the like.

Next, the operation will be described.
<Process at the time of learning>
First, the contents of processing when learning a radio signal transmitted from a known radio will be described.
FIG. 4 is a flowchart showing a learning process in the wireless device identification method according to the first embodiment of the present invention.
The signal receiving unit 1 receives a wireless signal transmitted from a known wireless device as a learning wireless signal, and converts the received learning wireless signal from an analog signal to a digital signal (step ST1 in FIG. 4).
It is assumed that a serial number, which is individual information identifying a known wireless device, is added to the learning wireless signal.
The signal receiving unit 1 outputs the digital learning wireless signal s _g (n) to the Fourier transform unit 2 and the first database unit 9. n in s _g (n) is a sample number of a learning radio signal. For example, FIG. 9 shows a learning radio signal having a sample number n of 1 to 4.
The first database unit 9 stores a learning radio signal s _g (n) to which a serial number is added.
FIG. 9 is an explanatory view showing a learning radio signal to which the serial number stored in the first database unit 9 is added.
In the example of FIG. 9, a plurality of learning radio signals in which reception conditions such as the SNR at the time of reception or the sampling frequency are different are stored even if the radios have the same serial number.

The Fourier transform unit 2 performs STFT on the digital learning wireless signal s _g (n) output from the signal receiving unit 1 as shown in the following equation (1), and outputs the learning wireless signal s _g (n). A learning Fourier transform signal S _g (m, k), which is a Fourier transform signal indicating the STFT result, is output to the rise detection unit 3 and the feature quantity calculation unit 7 (step ST2 in FIG. 4).

In equation (1), w (n−m) is a window function, and for example, a rectangular window or a Hamming window can be considered.
m is the number of samples for shifting the window function w (n-m), m ₀ is the first sample number of the application interval of the window function w (n-m), and M is the window function w (n-m) The number of samples, k, is a frequency bin number.
Note that each of the number of samples M of the window function w (n-m) and the number of samples m for shifting the window function w (n-m) is set in advance by the second parameter setting unit 6.

Here, FIG. 5 is an explanatory view showing the STFT of the learning radio signal s _g (n) and the identification radio signal s _d (n) by the Fourier transform unit 2.
The learning wireless signal s _g (n) and the identifying wireless signal s _d (n) are two-dimensional signals indicating the relationship between time and amplitude. The identification radio signal s _d (n) will be described later.
The learning Fourier transform signal S _g (m, k) indicating the STFT result and the discrimination Fourier transform signal S _d (m, k) are three-dimensional signals indicating the relationship between time, frequency and power. The identification Fourier transform signal S _d (m, k) will be described later.
Each of the learning radio signal s _g (n) and the identification radio signal s _d (n) is STFT by the Fourier transform unit 2 to obtain the learning radio signal s _g (n) and the identification radio signal s _d ( Since each of n) is coherently integrated, an effect of improving SNR can be obtained.

Rising edge detection unit 3 performs a process of detecting the rise time _{T 0} of the learning radio signal _s g (n) (step ST3 in FIG. 4).
FIG. 6 is an explanatory view showing detection processing of the rising time T ₀ by the rising detection unit 3.
It will be specifically described below detection processing of the rising time T ₀ by the rising edge detection unit 3.

First, the rising edge detection unit 3 generates a square value | S _g (m, k) of the learning Fourier transform signal S _g (m, k) output from the Fourier transform unit 2 as shown in the following equation (2). ) | ² is calculated as spectrogram SP _g (m, k).

Next, the rising edge detection unit 3 calculates an energy time waveform E _g (m) for learning, which is a time waveform of energy in the spectrogram SP _g (m, k), as shown in the following equation (3).
The energy time waveform E _g (m) for learning is calculated by summing the powers at the same time among the powers at a plurality of frequencies included in the spectrogram SP _g (m, k). The total value of the power at the time of

In equation (3), k ₁ is a frequency bin number indicating the lower limit frequency of the frequencies for summing power among a plurality of frequencies included in the spectrogram SP _g (m, k), and k _n is the spectrogram SP _g It is a frequency bin number which shows the upper limit frequency of the frequency which totals electric power among several frequencies contained in (m, k).
k ₌ k 1, ···, a _{k n.}

Next, the rising edge detection unit 3 compares the energy time waveform E _g (m) for learning with the threshold value E _th .
As the threshold value E _th , for example, a value obtained by subtracting the set value E ₀ from the maximum value of the energy time waveform E _g (m) for learning can be considered.
The rising edge detection unit 3 refers to the comparison result of the energy time waveform E _g (m) for learning and the threshold value E _th, and at first, the energy time waveform E _g for learning that indicates the energy of the noise level It is determined whether (m) has reached the threshold value E _th .
The rise detection unit 3 detects the time when the energy time waveform E _g (m) for learning reaches the threshold value E _th as the rise time T ₀ of the learning radio signal s _g (n).
In the first embodiment, when the rising detection unit 3 detects the rising time T ₀ of the learning wireless signal s _g (n), the power at a plurality of frequencies included in the spectrogram SP _g (m, k) The powers at the same time are summed up. Thus, even if it contains frequencies vary greatly power at the rise time, the influence of the variation of the power is reduced, thereby improving the detection accuracy of the rise time T _0.

The feature amount calculation unit 7 recognizes a time zone (t ₁ to t ₂ ) including the rise time T ₀ and the set frequency bin k _b from the parameters for feature amount calculation set by the first parameter setting unit 5. .
The setting process of the parameter for feature amount calculation by the first parameter setting unit 5 will be described later.
Next, as shown in FIG. 6, the feature quantity calculation unit 7 learns from the learning Fourier transform signal S _g (m, k) output from the Fourier transform unit 2, the learning detected by the rising edge detection unit 3. use to extract the radio signal _s g a time zone including the rising time _{T 0} of the _{_{(n) (t 1 ~ t}} 2) learning Fourier transform signal _s g of _{(n) t1 ~ t2.}

Next, as shown in FIG. 7, the feature quantity calculation unit 7 _{selects one} of the complex time signals included in the extracted training Fourier transform signal s _g (n) _{t 1 to t 2} for the set frequency bin k _b . The complex time signal _sg hat (n) _{t1 to t2} are cut out.
For convenience of the electronic application, in the text of the specification, since it is not possible to add the symbol “^” on the letter s, it is written as s _g hat (n) _{t 1 to t 2} .
FIG. 7 is an explanatory view showing calculation processing of the feature amount by the feature amount calculation unit 7.
The feature amount calculation unit 7 calculates the variation of the extracted learning Fourier transform signal s _g (n) _{t 1 to t 2} as the feature amount from the extracted complex time signal _sg hat (n) _{t 1 to t 2} (FIG. 4 Step ST4).

Hereinafter, the process of calculating the feature amount by the feature amount calculating unit 7 will be specifically described.
From the extracted complex time signal _sg hat (n) _{t1 to t2} , the feature quantity calculator 7 calculates the instantaneous amplitude a of the complex time signal _sg hat (n) _{t1 to t2} as shown in the following equation (4). Calculate _g (n).

In equation (4), I _g (n) is the real part of the complex time signal s _g hat (n) _{t 1 to t 2} , and Q _g (n) is the imaginary part of the complex time signal s _g hat (n) _{t 1 to t 2} It is a department.

The feature quantity calculation unit 7, the complex time signal cut _{s g} hat _{(n) t1 ~ t2,} as shown in the following equation (5), the instantaneous complex time signal _{s g} hat _{(n) t1 ~ t2} The phase φ _g (n) is calculated.

The feature quantity calculation unit 7, the complex time signal _{s g} hat _{(n) t1 ~ t2} cut, as shown in the following equation (6), the instantaneous complex time signal _{s g} hat _{(n) t1 ~ t2} The frequency f _g (n) is calculated.

Next, the feature quantity calculation unit 7 calculates the dispersion of the instantaneous amplitude a _g (n) at the complex time signal s _g hat (n) _{t 1 to t 2} as the first feature quantity as shown in the following equation (7) Calculate the value _pg1 .

The feature quantity calculation unit 7 calculates the skewness p _g2 of the instantaneous amplitude a _g (n) in the complex time signal s _g hat (n) _{t 1 to t} 2 as the second feature value as shown in the following equation (9) Calculate

The feature quantity calculation unit 7 calculates the kurtosis p _g3 of the instantaneous amplitude a _g (n) at the complex time signal s _g hat (n) _{t 1 to t 2} as the third feature quantity as shown in the following equation (11) Calculate

Next, the feature quantity calculation unit 7 calculates the dispersion of the instantaneous phase φ _g (n) in the complex time signal _sg hat (n) _{t1 to t2} as the fourth feature quantity as shown in the following equation (12) Calculate the value _pg4 .

As the fifth feature value, the feature value calculation unit 7 calculates the skewness p _g5 of the instantaneous phase φ _g (n) in the complex time signal s _g hat (n) _{t 1 to t 2} as the fifth feature value. Calculate

The feature quantity calculation unit 7 calculates the kurtosis p _g6 of the instantaneous phase φ _g (n) at the complex time signal s _g hat (n) _{t 1 to t 2} as the sixth feature quantity as shown in the following equation (16) Calculate

Next, the feature quantity calculation unit 7 calculates the dispersion of the instantaneous frequency f _g (n) at the complex time signal s _g hat (n) _{t 1 to t 2} as the seventh feature quantity, as shown in the following equation (17) Calculate the value _pg7 .

The feature quantity calculation unit 7 calculates, as the eighth feature quantity, the skewness p _g8 of the instantaneous frequency f _g (n) at the complex time signal s _g hat (n) _{t 1 to t 2} as represented by the following equation (19) Calculate

As shown in the following equation (21), the feature quantity calculation unit 7 uses the ninth feature quantity as the kurtosis p _g9 of the instantaneous frequency f _g (n) at the complex time signal s _g hat (n) _{t 1 to t 2} Calculate

The feature amount calculation unit 7 sets a weighting parameter indicating the weight w _i of the first to ninth feature amounts p _gi (i = 1, 2,..., 9) set by the second parameter setting unit 6. get.
Among the first to ninth feature quantities p _gi , there are feature quantities effective for identification of a wireless device and feature quantities not effective for identification. Therefore, a large weight is set to the feature amount effective for identification, and a small weight is set to the feature amount not effective for identification.
For example, w _i (i = 1, 2,..., 9) is a value of 0 or more and 1 or less, and w ₁ + w ₂ +... + W ₉ = 1.
Setting the weighting parameter by the second parameter setting unit 6 is not particularly limited, for example, in the first dispersion value of the ninth feature amount p _gi of the feature quantity p _gi ninth normalized from the first A method may be considered in which a value obtained by normalizing the first to ninth feature quantities p _gi is used as a weight.

The feature amount calculation unit 7 multiplies the first to ninth feature amounts p _gi and the weights w _i as shown in the following equation (22) to obtain the first to ninth feature amounts p _gi. Perform weighting.
p ' _gi = p _gi × w _i (22)
i = 1, 2, ..., 9
Feature amount calculation unit 7, a vector from the first the weighted arranged feature quantities p _'gi ninth, calculated as a feature vector p _g of the known radio.
Feature amount calculation unit 7, a feature quantity vector p _g of the calculated known radio, together with the serial number is added to the learning radio signal is stored in the second database section 10 (step of FIG. 4 ST5).
The feature quantity calculation unit 7 outputs information indicating that the feature vector p _g of the known radio stored in a second database section 10 to the first parameter setting unit 5.

Here, the setting process of the parameter for feature amount calculation by the first parameter setting unit 5 will be described.
FIG. 8 is an explanatory view showing a relationship between parameters for feature amount calculation and feature amount vectors of a plurality of known wireless devices.
In the example of FIG. 8, 〇, the feature quantity vector p _g according to known radio types _(A), ◇ is feature vector p _g according to known radio type _(B), △ the type It is the feature-value vector pg which _concerns on the known radio | wireless machine of (C).
In the example of FIG. 8, since the number of wireless devices belonging to the type (A) is 10, 10 feature quantity vectors 〇 are obtained.
Further, since the number of wireless devices belonging to type (B) is 10, 10 feature quantity vectors are obtained, and the number of wireless devices belonging to type (C) is 10 10 feature quantity vectors Δ are obtained.

In FIG. 8A, since the parameter for calculating the feature amount is not appropriate, the feature amount vector 係る for the wireless device of type (A), the feature amount vector ◇ for the wireless device of type (B), and the type (C) An example is shown in which the feature quantity vector 係る associated with the wireless device of the present invention partially overlaps.
When the feature vector 〇 for a known wireless device, the feature vector ◇, and the feature vector △ overlap, these feature vectors 〇, ,, と and the feature vector for the wireless device to be identified However, among the feature amount vectors ,, ,, Δ, it is not possible to specify with high accuracy the feature amount vector that is most similar to the feature amount vector of the wireless device to be identified.
Therefore, in order to specify with high accuracy the feature quantity vector that is most similar to the feature quantity vector of the wireless device to be identified among the feature quantity vectors ,, ,, △, the feature quantity of the known wireless apparatus It is desirable that the vector と, the feature vector ◇, and the feature vector ベクトル do not overlap.
In FIG. 8B, since the parameters for calculating the feature amount are appropriate, the feature amount vector 係る for the wireless device of type (A), the feature amount vector ◇ for the wireless device of type (B), and the type (C) An example is shown in which the feature amount vector 係る associated with the wireless device in (1) is not overlapped.

The first parameter setting unit 5 sets a parameter indicating a time range Δt which is a length of a time zone (t ₁ to t ₂ ) and a parameter indicating a set frequency bin k _b as an initial setting of a parameter for feature amount calculation. And set any value.
The time zone (t ₁ to t ₂ ) may be set based on the energy time waveform E _g (m) for learning.
For example, at a time before the rise time T _{0 of} the learning wireless signal s _g (n), a time at which the energy is 30 dB lower than the maximum value of the learning energy time waveform E _g (m) is t ₁ Do. Also, at a time after the rise time T _{0 of} the learning radio signal s _g (n), the time at which the energy is 30 dB lower than the maximum value of the learning energy time waveform E _g (m) is t ₂ Do.
As for the t _2, a method of setting on the basis of the variation in the frequency direction of the learning Fourier transform signal s _{g (n)} is considered.
For example, it is assumed that the time at which the frequency fluctuation during the transient response of the learning Fourier transform signal s _g (n) falls within a specific frequency range is t ₂ .

The set frequency bin k _b, the energy time waveform E _g for learning _(m), the average power is considered a method that employs a frequency bin having the maximum value.
Also, the set frequency bin k _b, the energy time waveform E _g for learning _(m), to detect the more frequency bins average power threshold, an average of the frequency of the one or more frequency bins detected It is conceivable to adopt a frequency bin of the average frequency.

The first parameter setting unit 5 receives the information indicating that stored from the feature amount calculation unit 7 feature vector p _g of the known radio in the second database section 10, from the second database 10 , it acquires the feature quantity vector p _g plurality of known radio.

The first parameter setting unit 5 determines whether there is overlap between the feature vector p _g according to several known radio, for example, as shown in FIG. 8B, a plurality of known radio without overlap between the feature vector p _g according to, as an effective parameter for feature calculation is set to above, and ends the parameter setting processing.
The first parameter setting unit 5 sets, for example, as shown in Figure 8A, if there is overlap between the feature vector p _g according to several known radio, as a parameter for the feature amount calculation, first At least one of the time range Δt being set or the set frequency bin k _b is updated.
Feature amount calculation unit 7, using the parameters for the updated feature quantity calculation by the first parameter setting unit 5, again, it is calculated several known feature quantity vectors p _g radios respectively.
The first parameter setting unit 5 determines whether there is overlap between the feature vector p _g according to several known radio calculated again by the feature amount calculation unit 7.
Until the overlap is eliminated between the feature vector p _g according to several known radio, and updating the parameters for feature calculation by the first parameter setting unit 5, feature vector p by the feature amount calculating section 7 The calculation process of _g is repeated.
Note that the update processing of the parameter, repeating the process of calculating the feature quantity vectors p _g, if not eliminated overlap according to several known radio, the overlap is the smallest parameter as appropriate parameters , End the parameter setting process.

When the parameter for calculating the feature amount becomes an appropriate value by updating the parameter for calculating the feature amount by the first parameter setting unit 5, the parameter is stored in the second database unit 10 as shown in FIG. 8B. that overlap between the plurality of known feature vector p _g of the radio is eliminated, or, overlap smallest.
After the parameters for feature calculation is updated, when a feature vector p _g according to several known radio calculated again by the feature amount calculation unit 7 is stored in the second database section 10, improper feature vector p _g according to several known radio that is calculated in accordance with the parameters for the feature quantity calculation is deleted from the second database 10.
FIG. 10 is an explanatory view showing an example of the feature amounts related to a plurality of known wireless devices stored in the second database unit 10.
In FIG. 10, the known types of radios show examples of type (A) and type (B), and the radios belonging to type (A) are individual of serial number (1), serial number An example of the individual of (2) is shown.
Further, FIG. 10 shows an example of parameters for feature amount calculation and weighting parameters.

In the first embodiment, an example is shown in which the parameter for calculating the feature amount is updated by the first parameter setting unit 5 until the parameter for calculating the feature amount becomes an appropriate value. The update timing of the parameter for is not limited to this.
For example, when the learning wireless signal newly transmitted from the known wireless device is received by the signal receiving unit 1, the first feature amount is calculated by the feature amount calculating unit 7. The parameter setting unit 5 may update the parameter for feature amount calculation.

<Process at the time of identification>
Next, processing contents in identifying a wireless device will be described using a wireless signal transmitted from the wireless device to be identified.
FIG. 11 is a flowchart showing an identification process in the wireless device identification method according to the first embodiment of the present invention.
The signal receiving unit 1 receives a radio signal transmitted from a radio to be identified as an identification radio signal, and converts the received identification radio signal from an analog signal to a digital signal (step ST11 in FIG. 11).
The signal receiving unit 1 outputs the digital radio signal for identification s _d (n) to the Fourier transform unit 2. n in s _d (n) is a sample number of the identification radio signal.

The Fourier transform unit 2 STFTs the digital identification radio signal s _d (n) output from the signal reception unit 1 as shown in the following equation (23) (step ST12 in FIG. 11).
The Fourier transform unit 2 outputs the discrimination Fourier transform signal S _d (m, k), which is a Fourier transform signal indicating the STFT result of the identification radio signal s _d (n), to the detection unit 3 and the feature amount calculation unit 7 (Step ST12 in FIG. 11). FIG. 5 also shows the STFT of the identification radio signal s _d (n).

Rising edge detection unit 3 performs a process of detecting the rise time _{T 0} of the identification radio signal _s d (n) (step ST13 in FIG. 11).
It will be specifically described below detection processing of the rising time T ₀ by the rising edge detection unit 3.

First, the rising edge detection unit 3 generates a square value | S _d (m, k) of the identification Fourier transform signal S _d (m, k) output from the Fourier transform unit 2 as shown in the following equation (24). ) | ² is calculated as spectrogram SP _d (m, k).

Next, the rising edge detection unit 3 calculates an energy time waveform E _d (m) for identification, which is a time waveform of energy in the spectrogram SP _d (m, k), as shown in the following equation (25).
The energy-time waveform E _d (m) for identification is calculated by summing the powers at the same time among the powers at a plurality of frequencies included in the spectrogram SP _d (m, k). The total value of the power at the time of

In equation (25), k ₁ is a frequency bin number indicating the lower limit frequency of frequencies for summing power among a plurality of frequencies included in spectrogram SP _d (m, k), and k _n is a spectrogram SP _d It is a frequency bin number which shows the upper limit frequency of the frequency which totals electric power among several frequencies contained in (m, k).
k ₌ k 1, ···, a _{k n.}

Next, the rising edge detection unit 3 compares the energy time waveform E _d (m) for identification with the threshold value E _th .
The rising edge detection unit 3 refers to the comparison result of the energy-time waveform E _d (m) for identification and the threshold value E _th to determine the energy-time waveform E _d for identification that initially indicated the energy of the noise level. It is determined whether (m) has reached the threshold value E _th .
The rising edge detection unit 3 detects the time when the energy time waveform E _d (m) for identification reaches the threshold value E _th as the rising time T ₀ of the identification wireless signal s _d (n).
In the first embodiment, when the rising detection unit 3 detects the rising time T ₀ of the identification radio signal s _d (n), the power at a plurality of frequencies included in the spectrogram SP _d (m, k) The powers at the same time are summed up. Thus, even if it contains frequencies vary greatly power at the rise time, the influence of the variation of the power is reduced, thereby improving the detection accuracy of the rise time T _0.

The feature amount calculation unit 7 recognizes a time zone (t ₁ to t ₂ ) including the rise time T ₀ and the set frequency bin k _b from the parameters for feature amount calculation set by the first parameter setting unit 5. .
Parameters for the feature amount calculated by the first parameter setting unit 5, as the overlap between the feature vector p _g according to several known radio is eliminated, or set so that the overlap is minimized Parameters that have been
Next, as shown in FIG. 6, the feature quantity calculation unit 7 identifies the detection detected by the rising edge detection unit 3 from the identification Fourier transform signal S _d (m, k) output from the Fourier transform unit 2. use to extract the radio signal _s d time zone including the rising time _{T 0} of the _{_{(n) (t 1 ~ t}} 2) identifying the Fourier transform signal _s d of _{(n) t1 ~ t2.}

Next, as shown in FIG. 7, the feature quantity calculation unit 7 calculates the set frequency bin k _b among the complex time signals included in the extracted Fourier transform signal s _d (n) _{t 1 to t 2} . The complex time signal _sd hat (n) _{t1 to t2} are cut out.
For convenience of the electronic application, in the text of the specification, since it is not possible to attach the symbol “^” on the letter s, it is written as s _d hat (n) _{t 1 to t 2} .
The feature amount calculation unit 7 calculates the variation of the extracted Fourier transform signal s _d (n) _{t 1 to t 2} as the feature amount from the extracted complex time signal s _d hat (n) _{t 1 to t 2} (FIG. 11 Step ST14).

Hereinafter, the process of calculating the feature amount by the feature amount calculating unit 7 will be specifically described.
From the extracted complex time signal s _d hat (n) _{t 1 to t 2} , the feature amount calculation unit 7 calculates the instantaneous amplitude a of the complex time signal s _d hat (n) _{t 1 to t 2} as shown in the following equation (26) Calculate _d (n).

In equation (26), I _d (n) is the real part of complex time signal s _d hat (n) _{t 1 to t 2} , and Q _d (n) is the imaginary part of complex time signal s _d hat (n) _{t 1 to t 2} It is a department.

The feature quantity calculation unit 7, the complex time signal cut _{s d} hat _{(n) t1 ~ t2,} as shown in the following equation (27), the instantaneous complex time signal _{s d} hat _{(n) t1 ~ t2} The phase φ _d (n) is calculated.

The feature quantity calculation unit 7, the complex time signal _{s d} hat _{(n) t1 ~ t2} cut, as shown in the following equation (28), the instantaneous complex time signal _{s d} hat _{(n) t1 ~ t2} The frequency f _d (n) is calculated.

Next, the feature quantity calculation unit 7 calculates the dispersion of the instantaneous amplitude a _d (n) in the complex time signal s _d hat (n) _{t 1 to t 2} as the first feature quantity as shown in the following equation (29) Calculate the value p _d1 .

The feature quantity calculation unit 7 calculates the skewness p _d2 of the instantaneous amplitude a _d (n) in the complex time signal s _d hat (n) _{t 1 to t} 2 as the second feature quantity as shown in the following equation (31) Calculate

The feature quantity calculation unit 7 calculates the kurtosis p _d3 of the instantaneous amplitude a _d (n) at the complex time signal s _d hat (n) _{t 1 to t 2} as the third feature quantity as shown in the following equation (33) Calculate

Next, the feature quantity calculation unit 7 calculates the dispersion of the instantaneous phase φ _d (n) in the complex time signal s _d hat (n) _{t 1 to t 2} as the fourth feature quantity as shown in the following equation (34) Calculate the value p _d4 .

The feature quantity calculation unit 7 calculates the skewness p _d5 of the instantaneous phase φ _d (n) in the complex time signal s _d hat (n) _{t 1 to t 2} as the fifth feature quantity, as shown in the following equation (36) Calculate

The feature quantity calculation unit 7 calculates the kurtosis p _d6 of the instantaneous phase φ _d (n) at the complex time signal s _d hat (n) _{t 1 to t 2} as the sixth feature quantity as shown in the following equation (38) Calculate

Next, the feature quantity calculation unit 7 calculates the dispersion of the instantaneous frequency f _d (n) at the complex time signal s _d hat (n) _{t 1 to t 2} as the seventh feature quantity as shown in the following equation (39) Calculate the value p _d7 .

The feature quantity calculation unit 7 calculates, as the eighth feature quantity, the skewness p _d8 of the instantaneous frequency f _d (n) at the complex time signal s _d hat (n) _{t 1 to t 2} as represented by the following equation (41) Calculate

The feature quantity calculation unit 7 calculates the kurtosis p _d9 of the instantaneous frequency f _d (n) at the complex time signal s _d hat (n) _{t 1 to t 2} as the ninth feature quantity as shown in the following equation (43) Calculate

The feature amount calculation unit 7 sets a weighting parameter indicating the weight w _i of the first to ninth feature amounts p _di (i = 1, 2,..., 9) set by the second parameter setting unit 6. get.
The feature quantity calculation unit 7 multiplies the first to ninth feature quantity p _di and the weight w _i as shown in the following equation (44) to obtain the first to ninth feature quantity p _di. Perform weighting.
p ' _di = p _di × w _i (44)
i = 1, 2, ..., 9
The feature amount calculation unit 7 calculates a vector in which the weighted first to ninth feature amounts p ′ _di are arranged as the feature amount vector p _d of the wireless device to be identified.
The feature amount calculation unit 7 outputs the calculated feature amount vector p _d of the wireless device to be identified to the wireless device identification unit 11.

The wireless device identification unit 11 is a feature amount vector p _d for the wireless device to be identified calculated by the feature amount calculation unit 7 and a feature for a plurality of known wireless devices stored in the second database unit 10 comparing the amount vector p _g respectively.
The radio identification unit 11 includes a feature vector p _d according to the identification target radio, based on a result of comparison between the feature vector p _g according to several known radio, to be identified radios It identifies (step ST15 of FIG. 11).

Hereinafter, identification processing of a wireless device by the wireless device identification unit 11 will be specifically described.
FIG. 12 is an explanatory view showing identification processing of a wireless device by the wireless device identification unit 11.
In FIG. 12, only the first to third feature amounts are illustrated among the first to ninth feature amounts calculated by the feature amount calculating unit 7 for simplification of the description.

The wireless device identification unit 11 uses the feature quantity vector p _d for the wireless device to be identified and the plurality of known wireless devices stored by the second database unit 10 as shown in the following equation (45): The Mahalanobis distance d _{ij, h} between the feature quantity vector p _g is calculated.
Here, for convenience of explanation, the known radio is located H pieces, it is assumed that the feature vector p _g of the H-number of radio is respectively stored in the second database section 10.
The H wireless devices include, for example, a wireless device belonging to type (A), a wireless device belonging to type (B), or a wireless device belonging to type (C).

In equation (45), x _{i, h} is the position of the feature vector p _g of the known wireless device h (h = 1,..., H) in the n-dimensional feature space, and x _j is n This is the position of the feature quantity vector p _d of the wireless device to be identified in the feature quantity space of a dimension.
(X _{i, h-} x _j ) is the Euclidean distance between two points in the n-dimensional feature amount space, S is the feature amount space of the learning wireless signal s _g (n) transmitted from the known wireless device h Is the variance-covariance matrix of the distribution at.
The Mahalanobis distance d _{ij, h} is a distance in a space in which the influence of the distribution of the distribution of the learning wireless signal s _g (n) in the feature space is normalized.

Radio identification unit 11 outputs the calculated H-number of the Mahalanobis distance _{d ij,} compared to each other _h, H-number of the Mahalanobis distance _{d ij,} in _h, the minimum Mahalanobis distance _{d ij,} identifies the _h.
Among the H wireless devices, the wireless device identification unit 11 determines that the known wireless device related to the minimum Mahalanobis distance d _{ij, h} is most similar to the wireless device to be identified.
The wireless device identification unit 11 outputs the serial number of the known wireless device related to the minimum Mahalanobis distance d _{ij, h} as the identification result of the wireless device to be identified. This makes it possible to identify not only the type of wireless device but also the wireless device.
In the example of FIG. 12, the position of the feature vector p _g of the three known radio is shown. Among the three known radio, since the Mahalanobis distance between the feature vector p position of _g is known radio is ● is the minimum, radio identification target, the position of the feature vector p _g Identified as a ● radio.

Here, although the example in which the wireless device identification unit 11 identifies the wireless device using the Mahalanobis distance d _{ij, h} is shown, the present invention is not limited thereto. For example, linear identification by the least squares method, support vector The radio may be identified using a machine (SVM: Support Vector Machine) or a neural network.

As apparent from the above, according to the first embodiment, the time waveform of energy in the spectrogram of the Fourier transform signal output from the Fourier transform unit 2 is calculated, and the rise time of the wireless signal is detected from the time waveform of energy. The Fourier transform signal of the time zone including the rise time detected by the rise detection unit 3 is extracted from the rise detection unit 3 and the Fourier transform signal output from the Fourier transform unit 2 to extract the extracted Fourier transform signal The wireless device identification unit 11 is calculated by the feature amount calculation unit 7 by providing the feature amount calculation unit 7 that calculates the variation of the extracted Fourier transform signal as the feature amount from the complex time signal of the set frequency component included. Since the wireless device is configured to identify the wireless device based on the feature amount, the effect of being able to identify the individual of the wireless device Unlikely to.

In the present invention, within the scope of the invention, modifications of optional components of the embodiment or omission of optional components of the embodiment is possible.

The present invention is suitable for a wireless device identification device for identifying a wireless device and a wireless device identification method.

Reference Signs List 1 signal reception unit, 2 Fourier transform unit, 3 rising detection unit, 4 parameter setting unit, 5 first parameter setting unit, 6 second parameter setting unit, 7 feature amount calculation unit, 8 database unit, 9 first Database unit, 10 Second database unit, 11 radio identification unit, 21 signal reception circuit, 22 Fourier transform circuit, 23 rise detection circuit, 24 parameter setting circuit, 25 feature amount calculation circuit, 26 database circuit, 27 radio identification Circuit, 31 memories, 32 processors.

Claims

A signal receiving unit for receiving a radio signal transmitted from a radio to be identified;
A Fourier transform unit that performs a short time Fourier transform on the wireless signal received by the signal reception unit, and outputs a Fourier transform signal indicating a short time Fourier transform result of the wireless signal;
A rise detection unit that calculates a time waveform of energy in a spectrogram of the Fourier transform signal output from the Fourier transform unit, and detects a rise time of the wireless signal from the time waveform of the energy;
From the Fourier transform signal output from the Fourier transform unit, a Fourier transform signal of a time zone including the rise time detected by the rise detection unit is extracted, and the set frequency component included in the extracted Fourier transform signal A feature amount calculation unit that calculates, as a feature amount, a fluctuation of the extracted Fourier transform signal from the complex time signal of
And a wireless device identification unit for identifying the wireless device based on the feature amount calculated by the feature amount calculation unit.
The signal receiving unit receives a wireless signal transmitted from a known wireless device,
The Fourier transform unit performs a short-time Fourier transform on the wireless signal of the known wireless device received by the signal receiving unit, and outputs a Fourier transform signal indicating a short-time Fourier transform result of the wireless signal of the known wireless device.
The rise detection unit calculates a time waveform of energy in a spectrogram of a Fourier transform signal of a known wireless device output from the Fourier transform unit, and a rise time of a wireless signal of the known wireless device from the time waveform of the energy To detect
The feature quantity calculation unit extracts a Fourier transform signal of a time zone including a rise time detected by the rise detection unit from among Fourier transform signals of a known wireless device output from the Fourier transform unit. The known radio calculated by calculating the fluctuation of the Fourier transform of the known radio extracted from the complex time signal of the set frequency component included in the Fourier transform signal of the extracted known radio as a feature amount Store the feature amount related to the machine in the database unit,
The wireless device identification unit compares the feature amount of the wireless device to be identified calculated by the feature amount calculation unit with the feature amount of a known wireless device stored by the database unit, The radio set identification device according to claim 1, wherein the radio set to be identified is identified based on the comparison result of (4).
The first parameter setting unit is configured to set a parameter indicating the length of the time zone and a parameter indicating the set frequency component as parameters for calculating the feature amount.
The feature amount calculation unit calculates each of the feature amount of the wireless device to be identified and the feature amount of the known wireless device using the parameter for calculating the feature amount set by the first parameter setting unit. The wireless device identification device according to claim 2, characterized in that:
The first parameter setting unit does not overlap the feature amounts calculated by the feature amount calculating unit when the signal receiving unit receives radio signals transmitted from different known wireless devices. 4. The wireless device identification apparatus according to claim 3, wherein the parameter for calculating the feature amount is updated such that an overlap of the feature amounts calculated by the feature amount calculating unit is minimized. .
The rise detection unit calculates the total value of the power at each time as the time waveform of the energy by summing the powers at the same time among the powers at a plurality of frequencies included in the spectrogram. The wireless device identification apparatus according to claim 1, wherein the time when the total value of the power reaches a threshold is detected as the rise time of the wireless signal.
The feature quantity calculation unit
As the first to third feature quantities, the dispersion value, skewness and kurtosis of the amplitude in the complex time signal are calculated, and as the fourth to sixth feature quantities, the dispersion value of the phase in the complex time signal, The radio according to claim 1, wherein the skewness and kurtosis are calculated respectively, and the dispersion value of the frequency in the complex time signal, the skewness and the kurtosis are calculated as the seventh to ninth feature quantities, respectively. Machine identification device.
A second parameter setting unit configured to set a weighting parameter indicating the weight of the first to ninth feature amounts;
The wireless device identification apparatus according to claim 6, wherein the feature amount calculation unit performs weighting of the first to ninth feature amounts in accordance with the weighting parameter set by the second parameter setting unit.
A signal reception unit receives a radio signal transmitted from the radio to be identified;
A Fourier transform unit performs a short time Fourier transform on the wireless signal received by the signal reception unit, and outputs a Fourier transform signal indicating a result of a short time Fourier transform of the wireless signal;
The rise detection unit calculates a time waveform of energy in a spectrogram of the Fourier transform signal output from the Fourier transform unit, and detects a rise time of the wireless signal from the time waveform of the energy;
The feature amount calculation unit extracts a Fourier transform signal of a time zone including the rise time detected by the rise detection unit from the Fourier transform signal output from the Fourier transform unit, and includes the extracted Fourier transform signal The fluctuation of the extracted Fourier transform signal is calculated as the feature amount from the complex time signal of the set frequency component being
A wireless device identification method, wherein a wireless device identification unit identifies the wireless device based on the feature amount calculated by the feature amount calculation unit.