KR102279563B1 - Radar signal prcessing apparatus and radar signal processing method - Google Patents

Abstract

The radar signal processing apparatus according to the embodiment includes a virtual reception signal acquisition unit configured to acquire a virtual reception signal by reflecting a virtual transmission signal transmitted to a synthesis target from the synthesis target, and a virtual reception signal acquisition unit configured to obtain a virtual reception signal transmitted to a real target corresponding to the synthesis target. A radar reception signal acquisition unit for acquiring a radar reception signal in which a radar transmission signal is reflected from the real target, a generator for generating a verification signal corresponding to the radar reception signal using the virtual reception signal and the radar reception signal, and the It may include a discriminator for determining the accuracy of the verification signal by comparing the radar received signal and the verification signal.

Description

Radar signal processing apparatus and radar signal processing method {RADAR SIGNAL PRCESSING APPARATUS AND RADAR SIGNAL PROCESSING METHOD}

The present invention relates to a radar signal processing apparatus and a radar signal processing method, and more particularly, to provide a method for accurately detecting an actual target by generating a large amount of processed radar signal even when only a small amount of real radar signal is detected, It relates to a radar signal processing apparatus and a radar signal processing method.

There is a limitation that a lot of verification data cannot be obtained for the real environment. To solve this problem, a model that simulates data in the real environment from the simulated transmission/reception signals created through software is needed. A representative model among the models used in this process is the generative model.

A generative model refers to a model that creates a distribution of new data having a similar distribution when the distribution of training data is given.

One of the characteristics of these generative models is that they require a large amount of data for learning. However, there are so many characteristics that real data can have (waveform shape according to the number of wings of a drone), so it is difficult to train a generative model by making all of them data under real conditions. That is, it means that a large amount of training data is insufficient.

Therefore, a generative model capable of transfer learning for very few samples (real environment data) should be trained under simulation data.

The present invention has been derived to solve the above problems, and its purpose is to generate a large amount of simulated data that is close to the real even if there is a small amount of real data so that a small amount of real data can accurately identify the detected real target. there is.

In particular, the present invention enables to perform the above process through a deep learning learning process based on generative adversarial networks (GAN), so that the target can be automatically and quickly detected with accurate target identification. There is a purpose.

A radar signal processing apparatus according to an embodiment includes a virtual reception signal acquisition unit configured to acquire a virtual reception signal by reflecting a virtual transmission signal transmitted to a composite target from the composite target, and a radar transmitted to a real target corresponding to the composite target. A radar reception signal acquisition unit for acquiring a radar reception signal in which a transmission signal is reflected from the real target, a generator for generating a verification signal corresponding to the radar reception signal using the virtual reception signal and the radar reception signal, and the radar and a discriminator that compares the received signal with the verification signal to determine the accuracy of the verification signal.

The generator may generate a final verification signal based on a result of determining the accuracy of the verification signal generated by the discriminator.

The generator and the discriminator may perform machine learning using at least one of a convolutional neural network (CNN) and a recurrent neural network (RNN).

The virtual reception signal acquisition unit may perform the virtual reception signal acquisition by referring to image information including the synthesis target and real environment information corresponding to the image information.

Further comprising a signal analysis unit for analyzing the virtual received signal and the radar received signal, and using the analysis result to perform the process of generating the verification signal by the generator and determining the accuracy of the verification signal by the discriminator there is.

The signal analyzer may convert at least one of the virtual received signal and the radar received signal into a form of a frequency corresponding to time to generate a converted signal for each, and perform the analysis process with the converted signal .

A radar signal processing method according to an embodiment includes, by a virtual reception signal acquisition unit, a virtual reception signal acquisition step of acquiring a virtual reception signal by reflecting a virtual transmission signal transmitted to a composite target from the composite target, a radar reception signal A radar reception signal acquisition step in which a radar transmission signal transmitted to a real target corresponding to the synthesized target is reflected from the real target by an acquisition unit, a radar reception signal acquisition step, by a generator, the virtual reception signal and the radar The method may include a generating step of generating a verification signal corresponding to the radar received signal using a received signal and a determining step of comparing the radar received signal with the verification signal to determine the accuracy of the verification signal by a discriminator. .

According to the present invention, even if there is a small amount of real data, a large amount of simulated data close to the real can be generated to accurately identify an actual target in which a small amount of real data is detected.

In particular, the present invention enables the above process to be performed through a deep learning learning process based on generative adversarial networks (GAN), so that the target can be automatically and quickly detected with accurate target identification.

According to the present invention, in a situation where pilot signals for the UAV cannot be obtained in the first place, such as from an enemy UAV in a military environment, the shape and characteristics of an enemy UAV can be recovered through the barely detected signal, so that it can be easily identified when detected again in the future. be able to provide evidence.

In the present invention, RADAR, one of the signal measurement technologies used in autonomous driving, which has relatively high versatility but has a weakness in precise shape estimation, generates various simulation signals, and solves various problem situations that may occur in autonomous driving situations. Prediction and analysis can be performed.

1A is a block diagram illustrating a radar signal processing apparatus 1 according to an embodiment of the present invention.
1B is a flowchart illustrating a radar signal processing method according to an embodiment.
2A is a diagram referenced to describe a ray tracing simulation method according to an embodiment.
2B is a flowchart illustrating a process of obtaining a virtual received signal by the virtual received signal obtaining unit 110 .
FIG. 2C is a diagram referenced to explain a pair of various types of waveforms that can be transmitted by a virtual transmitter and a waveform that is reflected by a composite target generated using a computer program and received by a virtual receiver.
3 is a diagram referenced to explain the operation of the radar reception signal acquisition unit 120 .
4 is a diagram referenced to explain the operation of the signal analyzer 130 according to the embodiment.
FIG. 5 is a diagram referenced to explain a process of generating a verification signal with high accuracy corresponding to a virtual received signal using the generator 140 and the discriminator 150 .
6 and 7 are diagrams referenced for comparing signal verification for an arbitrary target.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0023] Reference is made to the accompanying drawings, which show by way of illustration specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein with respect to one embodiment may be embodied in other embodiments without departing from the spirit and scope of the invention. In addition, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the present invention. Accordingly, the detailed description set forth below is not intended to be taken in a limiting sense, and the scope of the invention, if properly described, is limited only by the appended claims, along with all scope equivalents to those claimed. Like reference numerals in the drawings refer to the same or similar functions throughout the various aspects.

1A is a block diagram illustrating a radar signal processing apparatus 1 according to an embodiment of the present invention, and FIG. 1B is a flowchart illustrating a radar signal processing method of the radar signal processing apparatus 1 .

Hereinafter, a radar signal processing method according to an embodiment will be described with reference to FIGS. 1A and 1B together.

The radar signal processing device 1 may be a computing device including a central control processor, and may include a personal computer, a workstation, or the like.

As shown in FIG. 1A , the radar signal processing apparatus 1 may include a control unit 100 , a communication unit 200 , and a memory unit 300 .

The control unit 100 controls each component constituting the radar signal processing apparatus 1 as a whole, and includes a virtual reception signal acquisition unit 110 , a radar reception signal acquisition unit 120 , a signal analysis unit 130 , It may include a generator 140 and a discriminator 150 .

The communication unit 200 is for performing data communication between the radar signal processing apparatus 1 and another external device (not shown), and may perform wired/wireless communication.

The communication unit 200 may receive a virtual reception signal from an external device and transmit it to the virtual reception signal acquisition unit 110 .

The communication unit 200 may receive a radar reception signal in which a radar transmission signal transmitted from an external device or the radar signal processing device 1 is reflected by an actual target and transmit the received radar signal acquisition unit 120 .

The memory unit 300 may store a program for processing and control of the controller 100 , and may store data input to or output from the radar signal processing apparatus 1 .

The memory unit 300 stores the neural network, and the controller 100 may update the neural network stored in the memory unit 300 by machine learning.

The virtual reception signal acquisition unit 110 may acquire a virtual reception signal reflected from a synthetic target created using a computer program (s210). A virtual transmission signal may be transmitted to the synthesis target and reflected from the synthesis target, and the virtual transmission signal transmission and virtual reception signal acquisition may be performed by performing a simulation in a virtual environment.

Transmission of a virtual transmission signal may be performed by a virtual transmitter, and acquisition of a virtual reception signal may be performed by a virtual receiver. In addition, the virtual transmission signal and the virtual reception signal are defined as virtual signals generated for simulation, not actual physical propagation signals.

According to an embodiment, the virtual reception signal acquisition unit 110 may directly transmit a virtual transmission signal to the synthesis target through the virtual transmitter and acquire the virtual reception signal reflected from the synthesis target through the virtual receiver. . In addition, according to another embodiment, when the radar signal processing apparatus 1 and a separate external device (not shown) transmit a virtual transmission signal and receive a virtual reception signal, the virtual reception signal obtaining unit 110 is externally It may be obtained by receiving a virtual reception signal from the device. According to the latter, a virtual reception signal may be received and obtained from an external device through the communication unit 200 .

For reference, virtual signal transmission and reception through simulation may be performed through a ray tracing process.

2A is a diagram referenced to describe a ray tracing simulation method according to an embodiment.

Referring to FIG. 2A, some of the virtual transmission signals transmitted from one or more virtual transmitters S1, S2, and S3 are transmitted to a synthesis target and reflected from the synthesis target. signal) can be obtained by receiving the virtual receiver (R).

The virtual reception signal acquisition unit 110 may transmit a virtual transmission signal and acquire the virtual reception signal by referring to image information including a synthesis target and real environment information corresponding to the image information.

2B is a flowchart illustrating a process of obtaining a virtual received signal by the virtual received signal obtaining unit 110 .

Specifically, as shown in FIG. 2B , the controller 100 may acquire image information including a synthesis target (s111).

The controller 100 may acquire 2D image information and 3D map information. In an embodiment, the two-dimensional image information may include additional information corresponding to the image, and the additional information may include photographing information including location information at which the image was captured, direction information, and an angle of view. 3D map information corresponding to 2D image information may be determined based on the additional information. In addition, the 3D map information may include location information and 3D map information corresponding thereto. Such information may include information on the shape of buildings, structures, and plants on the surface or water surface, and may include information related to at least one of a virtual transmitter candidate location and a virtual receiver candidate location according to an embodiment.

The controller 100 may acquire real environment information corresponding to the image information. (s112) The real environment information may include an object located on a communication path and characteristics of the object. More specifically, by analyzing the two-dimensional image information, it is possible to determine the characteristics of objects that can be located on the communication path based on the analysis. The properties of the objects may include at least one of the material of the surface of the object and the external shape of the object, and in the case of an object capable of transmitting radio waves, information related to the shape of the object and the degree of signal attenuation during transmission.

The controller 100 may obtain mapping information by mapping the real environment information to image information. (s113) For example, when mapping to 3D map information, based on additional information included in 2D image information Additional information obtained through the 2D image information may be mapped to an object corresponding to the 3D map information.

The virtual reception signal acquisition unit 110 of the control unit 100 may perform a ray tracing simulation in which the virtual reception signal is acquired by reflecting the virtual transmission signal from the synthesis target with reference to the mapping information. (s114) )

According to an embodiment, the ray tracing simulation considers a beam (virtual transmission signal) in a specific direction, and performs a ray tracing simulation corresponding thereto while sequentially changing beam information, or omnidirectional transmission that can be transmitted from a virtual transmitter. It is assumed that the beam is transmitted within the same time period, and a ray tracing simulation corresponding thereto may be performed. As a result of the ray tracing simulation, it is possible to predict and analyze the signal quality that can be received by the virtual receiver by reflecting the path through which the signal transmitted from the virtual transmitter takes to be received by the virtual receiver and the real environment information located on the path. can

However, as described above, when the radar signal processing device 1 and a separate external device (not shown) transmit a virtual transmission signal and receive a virtual reception signal, the virtual reception signal obtaining unit 110 is an external device. It may be obtained by receiving a virtual reception signal from the . In this case, s111 to s114 may be performed by an external device, and the virtual reception signal obtaining unit 110 may receive and acquire a virtual reception signal from the external device.

FIG. 2C is a diagram referenced to explain a pair of various types of waveforms that can be transmitted by a virtual transmitter and a waveform that is reflected by a composite target generated using a computer program and received by a virtual receiver. Referring to FIG.

라고 할 때, 가상의 수신기가 받아들이는 신호

는 다음과 같이 나타내어진다.According to an embodiment, a system including a virtual transmitter, a synthetic target reflecting a signal transmitted from the virtual transmitter, and a virtual receiver forms a Linear time-variant (LTV) system when the synthetic target is not fixed. Let's say the synthesis target consists of S different scatterers. The impulse response of each scatterer

, the signal received by the virtual receiver

is expressed as

(t: time, the time elapsed after the virtual transmission signal is transmitted to the composite target and reflected from the composite target

τ: Delay time, in general, the virtual received signal is the sum of the delay (τ) and Doppler shift (ν), and the virtual transmitted signal changed by multiplying the amplitude of the signal by b (attenuation factor) is shown The τ variable used here represents one of three characteristics for modeling a virtual received signal. [For reference, b (attenuation coefficient) is determined according to the size and permittivity of the scatterer, the distance between the virtual transmitter and the scatterer, and the distance between the scatterer and the virtual receiver, which will be described in detail later.]

: τ 만큼 지연된 송신 신호)

: transmit signal delayed by τ)

는 CW, FMCW, Pulse, QPSK 등의 여러 waveform 중에서 주어질 수 있다.Here, the virtual transmission signal

can be given among several waveforms such as CW, FMCW, Pulse, and QPSK.

에 Inverse Fourier transform (IFT) 를 취하면 다음과 같다.here

Taking the Inverse Fourier transform (IFT) on

: j 번째 산란점 (scatterer)의 응답 함수로 정의할 수 있다. 그리고, 응답 함수를 정의한 이유는 다음과 같다. 전술한 바와 같이, 일반적으로 가상의 수신 신호는 지연 (τ)및 도플러 천이 (ν) 및 신호의 크기가 변화된 (b) 가상의 송신 신호의 합으로 나타낼 수 있다. 주파수 이동과 감쇠계수의 의미를 부여하면 다음과 같다. 먼저 지연이 발생하는 원리는 다음과 같다. 송신기에서 방사된 신호가 점 scatter 에 반사된 후 수신기를 통해 물리적으로 이동을 하므로 신호가 이동한 거리를 빛의 속도로 나누어준 만큼 지연이 발생한다. 도플러 천이가 발생하는 이유는 산란점이 공간상에서 움직이므로 수신 신호의 파장이 변한다. 가상의 수신 신호의 주파수 또한 변하는데 그 이유는 자유 공간에서 빛의 속도는 일정하고 주파수는 빛의 속도를 파장으로 나눈 값이기 때문이다. 마지막으로 감쇠 계수의 의미는 다음과 같다. 감쇠 계수는 수신 신호의 크기를 결정하는 변수이다. 가상의 수신 신호의 변화를 주는 요소는 가상의 송신기, 가상의 수신기, 및 산란점 간의 거리, 산란점의 물리적 특성 (유전율, 투자율 등등) 에 영향을 받는다. (Note that,

: It can be defined as the response function of the j-th scatterer. And, the reason for defining the response function is as follows. As described above, in general, the virtual reception signal may be expressed as the sum of the delay (τ), the Doppler shift (ν), and (b) the virtual transmission signal in which the magnitude of the signal is changed. The meaning of frequency shift and attenuation coefficient is given as follows. First, the principle of delay is as follows. Since the signal emitted from the transmitter is reflected by the point scatter and then physically moves through the receiver, a delay occurs as much as the distance traveled by the signal divided by the speed of light. The reason for the Doppler shift is that the scattering point moves in space, so the wavelength of the received signal changes. The frequency of the virtual reception signal also changes because the speed of light in free space is constant and the frequency is the speed of light divided by the wavelength. Finally, the meaning of the damping coefficient is as follows. The attenuation coefficient is a variable that determines the magnitude of the received signal. Factors that change the virtual received signal are affected by the virtual transmitter, the virtual receiver, and the distance between the scattering points and the physical properties of the scattering points (permittivity, permeability, etc.).

번째 산란점이 가상의 송신 신호를

만큼 지연시키고

만큼 도플러 천이를 일으키며

만큼 신호 감쇠를 일으킨다고 가정해보자. 그러면 응답함수 식은

이다. 이 식의 의미를 서술하면 다음과 같다. Previously, it was explained that the scattering point causes both delay and Doppler shift attenuation of the virtual transmission signal. The response function is designed to mathematically define this phenomenon. This function takes a delay time and a Doppler shift as input, and outputs a single number as output. if

The second scattering point represents the virtual transmission signal.

delay as much

causing a Doppler shift

Assume that it causes signal attenuation by Then the response function is

am. The meaning of this expression is described as follows.

, Dirac-delta function) 의 곱으로 표현되어 있다. 본 발명에서 정의한 함수

는 디랙 텔타 함수이며 정의는 다음과 같다. 함수의 입력이 0이면 무한대이고 0이 아닐 경우 0을 출력한다. 또한 디랙 델타 함수를 실수 전체 영역에 대해 적분했을 때 값은 1이다. 이 응답 함수를 수신 신호 식

에 대입 할 경우 수신 신호

는 정확히 지연

, 도플러 천이

와 감쇠

가 적용된 송신 신호,

이다.)This formula consists of the damping coefficient and two Dirac delta functions (

, is expressed as a product of the Dirac-delta function). Functions defined in the present invention

is a Dirac delta function and its definition is as follows. If the input of the function is 0, it is infinity and if it is not 0, 0 is output. Also, when the Dirac delta function is integrated over the whole real area, the value is 1. Convert this response function to the received signal expression

Receive signal when substituting into

is exactly the delay

, Doppler shift

with attenuation

The transmitted signal to which

am.)

Substituting this into the above expression

를 가지고 (가상의 송신기 입장에서의 속도+가상의 수신기 입장에서의 속도)에 비례하는 단일한 Doppler shift

, 그리고 scatterer의 유전율과 크기에 비례하며, 송신기와 수신기에서 각각 scatterer로의 거리의 제곱에 반비례하는 감쇠 계수 (attenuation factor)

를 가진다. 이를 수식으로 나타내면 다음과 같다.becomes like Each target is made of S point scatterers, and these point scatterers have a single delay proportional to each incoming signal from the virtual transmitter (distance from the virtual transmitter + distance from the virtual receiver).

A single Doppler shift proportional to (velocity at the point of the virtual transmitter + speed at the point of the virtual receiver) with

, and an attenuation factor proportional to the permittivity and size of the scatterer and inversely proportional to the square of the distance from the transmitter and receiver to the scatterer, respectively.

have This can be expressed as a formula as follows.

: 산란 물체의 응답 함수 (1,2,..S 번째 산란점 (scatterer)의 응답 함수의 합)(

: The response function of the scattering object (sum of the response functions of the 1, 2,..S-th scatterers)

: j 번째 산란점 (scatterer)의 응답 함수

: Response function of the j-th scatterer

: 지연 시간

: delay time

: 도플러 천이

: Doppler shift

j 번째 산란점의 도플러 천이

Doppler shift of the j-th scattering point

: 디랙 델타 함수)

: Dirac delta function)

, 가상의 수신기의 위치를

, s번째 scatterer의 위치를

라고 했을 때

는Here, the location of the virtual transmitter

, the location of the virtual receiver

, the position of the s-th scatterer

when said

is

: 빛의 속도 (

: Speed of light

: 거리 함수

,

)

: distance function

,

)

input: two 3D coordinates, output: distance between the two coordinates)

는 거의 변하지 않는다고 가정한다.is represented by where the scatterer is

is assumed to change little.

, 가상의 수신기로부터 멀어지는 방향의 속도를

이라고 했을 때,

는In the same way, the speed in the direction away from the virtual transmitter of the s-th scatterer

, the velocity in the direction away from the virtual receiver

when said,

is

: j 번째 산란점의 도플러 천이

: Doppler shift of the j-th scattering point

는 이 신호의 carrier frequency이다.is represented by here

is the carrier frequency of this signal.

를 적용했을 때, 가상의 수신기의 신호는 최종적으로Over

When , the signal of the virtual receiver is finally

:

만큼 지연된 가상의 송신 신호(

:

virtual transmit signal delayed by

: 감쇠 계수

: 가상의 송신 신호 i:허수 단위 )t: time

: damping coefficient

: virtual transmission signal i: imaginary unit )

becomes like Finally, by applying this equation, the received signal of the virtual receiver for each virtual transmitter is synthesized.

마다 값을 가지는 수열로 저장되는데 수열로 저장되는 형태의 수신 신호는 다음과 같다.Since the received signal synthesized in this way must be stored in the device, a constant sampling rate

Each value is stored as a sequence, and the received signal stored as a sequence is as follows.

: 가상의 송신 신호(

: virtual transmission signal

: 신호 샘플링 주기

: signal sampling period

: 이 수식은 컴퓨터에서 처리하는 신호는 모두 디지털이기 때문에 시간에 따라 연속인 수신 신호를 일정 시간 간격으로 샘플링하는 과정이 필요하기 때문에 도입되었다. 여기서

은 가상의 수신 신호를

시간 간격으로 샘플링 했을 때 샘플링 시점을 나타낸 것이다.)

: This formula was introduced because all of the signals processed by the computer are digital, so it is necessary to sample the received signal, which is continuous according to time, at regular time intervals. here

is a virtual received signal.

It indicates the sampling point when sampling at time intervals.)

Finally, the virtual reception signal acquisition unit 110 in which such an algorithm is designed generates a large amount of virtual reception signals primarily necessary for radar verification through sufficiently many types of transmission waveform designs and scatterer designs.

3 is a diagram referenced to explain the operation of the radar reception signal acquisition unit 120 .

The radar reception signal acquisition unit 120 may acquire a radar reception signal in which a radar transmission signal transmitted to an actual target corresponding to the synthesized target is reflected from the actual target (s220).

Unlike the virtual reception signal acquisition unit 110 , the radar reception signal acquisition unit 120 may acquire a radar reception signal that is an actual physical radio signal instead of acquiring a virtual reception signal.

According to an embodiment, a radar transmission signal is generated through the radar signal processing device 1 and a separate external transmitter (eg, the signal transmission device and the RF signal generation device of FIG. 3 ), and a radar transmission signal to an actual target through a transmission antenna is transmitted, the radar reception signal reflected from the actual target may be received by the RF signal reception device through the reception antenna and stored in the signal reception/storage device. That is, the RF signal reception apparatus and/or the signal reception/storage apparatus of FIG. 3A may correspond to the radar reception signal acquisition unit 120 of the present invention.

However, according to another embodiment, the radar reception signal acquisition unit 120 generates a radar transmission signal and transmits the radar transmission signal to an actual target to implement all of a series of processes of receiving the radar signal reflected from the actual target. may be

According to an embodiment, a series of processes of transmitting a radar transmission signal and receiving a radar reception signal may be performed using software defined radio (SDR), a communication technology that processes a physically configured modulation/demodulation circuit in software.

3 is composed of a signal transmitting apparatus, an RF signal generating apparatus, a transmitting antenna for transmitting the same, a receiving antenna, an RF signal receiving apparatus, and an apparatus for storing the received signal. Except that the received signal is an actual signal, the equation is the same as that introduced in FIG. 2C. The RF signal generating apparatus may generate an RF signal corresponding to the virtual transmission signal received from the signal transmitting apparatus.

4 is a diagram referenced to explain the operation of the signal analyzer 130 according to the embodiment.

The signal analyzer 130 analyzes the virtual received signal and the radar received signal, and uses the analysis result to generate a verification signal by the generator 140 and determine the accuracy of the verification signal by the discriminator 150 so that the process is performed. (s230) For example, the method may include a process of demodulating a received signal to a baseband and performing autocorrelation on the demodulated signal using an auto-correction function.

According to an embodiment, the signal analysis by the signal analyzer 130 may analyze the waveform itself of the generated signal, or according to another embodiment, may be performed by converting the generated signal into a spectrogram form. In the latter case, when the object has periodic motion, the received signal may also have periodicity. When the received signal has periodicity, the spectrogram has a sparse form when the received signal is converted into a spectrogram. (The converted received signal is expressed by a small number of large portions (or significant portions) and a large number of small portions (or insignificant portions). Thus, the signal spectrogram can highlight important parts of the received signal, which can help generators and discriminators learn.

In the latter case, specifically, the signal analysis unit 130 may convert at least one of a virtual received signal and a radar received signal into a frequency form corresponding to time to generate a converted signal for each.

Specifically, the spectrogram is a visual or vector/matrix form of a change in the frequency component of a signal with time. This is suitable for expressing the change form when the aspect of the target and environment changes according to time.

That is, a spectrogram divides a signal into several times and records all frequency components at each time, and consists of a time axis and a frequency axis. The most basic example of a spectrogram is the Short-time Fourier transform (STFT). STFT is a Fourier transform of a given sampling signal over several time intervals.

에 대해서, window 크기 K, 중첩 샘플 L이라고 하고 window 함수를

으로 놓았을 때 STFT 식은 다음과 같다.sampled signal

For , let the window size K, the overlapping sample L, and let the window function

When set to , the STFT expression is as follows.

The above formula is a formula for converting the sampled received signal into a spectrogram.

: 어떤 이산 시간 n, 이산 주파수 k에 대한 스펙트로그램. 각 이산 시간 n (실제 시간은

) 에서의 푸리에 변환 값을 저장한다.

: Spectrogram for any discrete time n, discrete frequency k. Each discrete time n (actual time is

) and store the Fourier transform value in .

: 이산 시간 n에서의 샘플링된 이후의 수신 신호. 여기서도 실제 시간은

이다.

: Received signal after sampling at discrete time n. Here, too, the actual time

am.

: window 함수. 시간

주변에서의 일정 영역에서만 양수값을 가지고 그 밖의 영역에서는 0 값을 가진다. 아래의 Hann window 함수 그래프가 그 예시다. 왼쪽의 시간에 따른 amplitude 그래프에 따르면 일정 시간 영역에서만 양수값을 가지고, 이는 주어진 샘플링된 신호 x 중에서 일부분만이 spectrogram

를 형성하는 데에 쓰이도록 한다.

: window function. time

It has a positive value only in a certain area around it and has a value of 0 in other areas. The Hann window function graph below is an example. According to the amplitude graph over time on the left, it has a positive value only in a certain time domain, which means that only a part of the given sampled signal x

to be used to form

) 이다. 시간에 대한 물리적인 의미를 담아 스펙트로그램 식을 쓸려면

으로 쓰는게 바람직하나 보통 STFT 에대해 수식을 쓸 때 편의상

을 생략 한다 >n: discrete time <actual time is nx (sampling period or

) to be. To write a spectrogram expression with the physical meaning of time

It is preferable to write it as , but for convenience when writing formulas for STFT

omit >

: 윈도우 길이. 위의 window 함수의 정의에 따라 window 함수의 값이 0이 아닌 영역의 크기를 의미한다.

: window length. According to the definition of the window function above, it means the size of the area where the value of the window function is not 0.

L: Samples to be superimposed on the time axis

exp: an exponential function whose base is the natural logarithm

) )이다. 주파수에 대한 물리적인 의미를 담아 스펙트로그램 식을 쓸려면

으로 쓰는게 바람직하나 보통 STFT 에 대해 수식을 쓸 때 편의상 1/(

) 을 생략 한다 >k: discrete frequency <actual frequency is kx (sampling frequency or 1/(

) )am. To write a spectrogram expression with the physical meaning of frequency

It is preferable to write it as 1/(

) is omitted.

: nL~nL+(K-1) 까지의 숫자가 될 수 있으며, 이러한 이유는 window function이 해당 영역에서만 0이 아닌 값을 가지기 때문이다.

: It can be a number from nL to nL+(K-1), because the window function has a non-zero value only in the corresponding area.

For example, when designing a quadratic chirp (a signal in the form of a sine wave whose frequency increases with time according to a quadratic function) that starts at 100 Hz and exceeds 200 Hz, it is used with a sampling rate of 1 kHz, window size of 128, and overlapping samples. If the length (the number of overlapping samples for each adjacent window) is set to 120 and STFT is taken, a spectrogram as shown in FIG. 4 appears.

However, this is an embodiment, and according to another embodiment, a signal shape may be converted and utilized by further using a Candence-Velocity Diagram (CVD) that measures how periodically the frequency change extracted from the STFT changes.

FIG. 5 is a diagram referenced to explain a process of generating a verification signal with high accuracy corresponding to a virtual received signal using the generator 140 and the discriminator 150 .

Since the received radar signal reflected from the actual target is generally small, it is difficult to accurately determine its identity. Therefore, by generating a large amount of processed signal corresponding to a small amount of radar reception signal and determining whether the large amount of processing signal is an actual true signal such as an actual radar reception signal, the accuracy of the generated large amount of processed radar signal is improved. is to make it Here, a large amount of processed radar signals are verification signals to be described later.

The verification signal generation and discrimination through the generator 140 and the discriminator 150 may be performed through a deep learning learning process based on a generative adversarial network (GAN).

A generative adversarial network (GAN) consists of a generator 140 that generates a virtual data sample and a discriminator 150 that determines whether the input data sample is real data, and the generator 140 and It refers to a machine learning model built through adversarial training between the discriminators 150 .

According to an embodiment, the generator 140 may generate a verification signal corresponding to the radar reception signal by using the virtual reception signal and the radar reception signal (s240).

The generator 140 may be a generation model manufactured by repeatedly performing machine learning that receives a virtual reception signal and a radar reception signal as an input node and outputs a verification signal corresponding to the radar reception signal as an output node. That is, the generator 140 may refer to the radar reception signal together in the process of generating the verification signal from the virtual reception signal. Specifically, the verification signal may be generated by combining the virtual reception signal and the radar reception signal.

According to an embodiment, the discriminator 150 may determine the accuracy of the verification signal by comparing the radar reception signal and the verification signal (s250).

The discriminator 150 may be a learning model manufactured by repeatedly performing machine learning that receives a radar reception signal and a verification signal as an input node and outputs an accuracy determination result of the verification signal as an output node. That is, the discriminator 150 may output the accuracy determination result by repeatedly performing machine learning to determine whether the verification signal is a real signal such as a radar reception signal or a fake signal. For example, if it is true, '0' may be output, and if it is false, it may be output as '1'.

Thereafter, when a false case is derived based on the accuracy determination result for the verification signal of the discriminator 150, the generator 140 converts all the verification signals to the real signal. The generation of the verification signal can be repeated in a direction that can be mistaken for And, according to this, by repeating the hostile process between the generator 140 and the discriminator 150, as a result, the probability distribution that the finally generated verification signal is a real signal increases.

In the process described above, each variable of each neural network constituting the generator 140 and the discriminator 150 may be updated.

The generator 140 and the discriminator 150 according to the embodiment may be implemented to perform machine learning using at least one of a convolutional neural network (CNN) and a recurrent neural network (RNN). . However, this is an embodiment and the present invention may be applied in the same/similar manner even when other types of neural networks are used.

Hereinafter, an algorithm for training the generator 140 and the discriminator 150 will be described.

First, the virtual reception signal is set to x, the verification signal is set to G(x), and the radar reception signal is set to y.

And, the loss function of the generator (140, G) is as follows.

(1)

(One)

(2)

(2)

here,

생성기

generator

Input: virtual incoming signal

Output: Verification signal

분류기

classifier

또는 레이더 수신 신호

Input: Verification signal

or radar reception signal

Output: Discriminant result (single number)

가상의 수신신호의 확률 분포

Probability distribution of virtual received signal

레이더 수신신호의 확률 분포

Probability distribution of radar reception signal

가상의 수신신호의 확률에 대한 평균값

Average value of probability of virtual received signal

가상의 수신신호와 레이더 수신신호의 결합확률분포

Combination Probability Distribution of Virtual Received Signal and Radar Received Signal

: 가상의 수신신호와 레이더 수신신호의 결합확률분포에 대한 평균값

: Average value of combined probability distribution of virtual received signal and radar received signal

1st constraint: Radar reception signal and verification signal must be the same

Second constraint: The virtual reception signal and the verification signal must be the same

제 1 제약 조건을 만족시키기 위한 라그랑지승수

Lagrangian multiplier to satisfy the first constraint

제 2 제약 조건을 만족시키기 위한 라그랑지승수

Lagrangian multipliers to satisfy the second constraint

시그모이드 함수

sigmoid function

L1 norm

L1 norm

L2 norm

L2 norm

생성기의 목적 함수

generator objective function

The discriminator determines that the verification signal is 'true', and measures whether the first constraint and the second constraint are satisfied.

To satisfy all three conditions, the objective function of the generator must be minimized.

In addition, the loss function of the discriminator 150, D is as follows.

판별기에 레이더 수신신호를 입력으로 넣었을 때 판단 결과 (단일 숫자)(

Judgment result when the radar reception signal is input to the discriminator (single number)

판별기의 목적함수

The objective function of the discriminator

Measures whether the discriminator judges the verification signal as 'false' and the radar reception signal as 'true'

To satisfy both conditions, the objective function of the generator must be minimized)

This is based on LSGAN, a generative model specialized for the recovery of local information of image data, and adds a regularizer term for style transfer. L1-term(1) was introduced to train the similarity between the sparse radar reception signal y and the verification signal G(x), and L2-term(2) is the virtual reception signal x and the verification signal G(x). It was introduced to have the same properties as enemies. For reference, in the present invention, as described above, three goals to be achieved by the generator were set, and the meaning and function expression were described in detail. Here, the description of the LSGAN is an existing technology, and the generator has an objective function designed only so that the discriminator determines that the verification signal is 'true'. However, the above method does not completely achieve style transfer from the virtual received signal to the verification signal, which is the goal of the generator. Therefore, the present invention designed two constraints that can directly measure whether the generator is performing style conversion. In addition, by using the Lagrangian multiplier, the objective function used in the existing LSGAN and the function measuring whether the constraint is satisfied were combined. As a result, the generator learned through the objective function of the generator proposed in the present invention has room to convert the virtual received signal into a verification signal better than the existing LSGAN method.

According to an embodiment, the controller 100 including the generator 140 and the discriminator 150 may be manufactured in the form of at least one hardware chip and mounted on the radar signal processing apparatus 1 . For example, the controller 100 may be manufactured in the form of a dedicated hardware chip for artificial intelligence (AI), or a conventional general-purpose processor (eg, CPU or application processor) or a graphics-only processor (eg, GPU) ) may be manufactured as a part and mounted on the radar signal processing device 1 .

6 shows, for example, a result of comparing the forms of a virtual reception signal, a radar reception signal, and a verification signal for an arbitrary drone. As shown in FIG. 6 , it can be seen that the verification signal G(x) generated from the virtual reception signal x is almost similar in shape to the radar reception signal y because it is generated with reference to the radar reception signal y.

7 shows the probability of detecting the correct object when the target needs to be detected through the radar in a situation where three different targets exist (in addition, the data generated by GAN is added to the probability of detecting the correct object) When (GANAug) and when not (STFT) are compared.

As a result of comparison, in the graph of FIG. 7 , it was found that the detection accuracy was increased as a result of comparing the performance after generating data in the mavic drone with a small amount of data. This means that the GAN successfully generated data for the mavic drone and converted it into STFT.

The embodiments described above may be implemented in the form of program instructions that can be executed through various computer components and recorded in a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination.

The program instructions recorded on the computer-readable recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software.

Examples of the computer-readable recording medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as a CD-ROM and DVD, and a magneto-optical medium such as a floppy disk. optical media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules for carrying out the processing according to the present invention, and vice versa.

Aspects herein may take the form of a computer program product embodied on one or more computer readable media embodied in hardware entirely, software entirely (including firmware, resident software, microcode, etc.) or computer readable program code. .

Features, structures, effects, etc. described in the above embodiments are included in one embodiment of the present invention, and are not necessarily limited to one embodiment. Furthermore, features, structures, effects, etc. illustrated in each embodiment can be combined or modified for other embodiments by those of ordinary skill in the art to which the embodiments belong. Accordingly, the contents related to such combinations and modifications should be interpreted as being included in the scope of the present invention.

In addition, although the embodiment has been described above, it is merely an example and does not limit the present invention, and those of ordinary skill in the art to which the present invention pertains are exemplified above in a range that does not depart from the essential characteristics of the present embodiment. It can be seen that various modifications and applications that have not been made are possible. For example, each component specifically shown in the embodiment can be implemented by modification. And differences related to such modifications and applications should be construed as being included in the scope of the present invention defined in the appended claims.

Claims (13)

a virtual reception signal acquisition unit configured to obtain a virtual reception signal by reflecting the virtual transmission signal transmitted to the synthesis target from the synthesis target;
a radar reception signal acquisition unit configured to acquire a radar reception signal in which a radar transmission signal transmitted to an actual target corresponding to the synthesized target is reflected from the actual target;
a generator for generating a verification signal corresponding to the radar reception signal by using the virtual reception signal and the radar reception signal; and
a discriminator that compares the radar reception signal with the verification signal to determine the accuracy of the verification signal;
The virtual reception signal acquisition unit,
acquiring the virtual reception signal with reference to image information including the synthesis target and real environment information corresponding to the image information,
Radar signal processing unit.
The method of claim 1,
The generator generates a final verification signal based on the accuracy determination result of the verification signal generated by the discriminator,
Radar signal processing unit.
The method of claim 1,
The generator and the discriminator,
Performing machine learning using at least one of a convolutional neural network (CNN) and a recurrent neural network (RNN),
Radar signal processing unit.
delete The method of claim 1,
Further comprising; a signal analysis unit for analyzing the virtual received signal and the radar received signal;
performing a process of generating the verification signal by the generator and determining the accuracy of the verification signal by the discriminator using the analysis result,
Radar signal processing unit.
6. The method of claim 5,
The signal analysis unit,
converting at least one of the virtual reception signal and the radar reception signal into a form of a frequency corresponding to time to generate a converted signal for each;
performing the analysis process with the converted signal,
Radar signal processing unit.
a virtual reception signal acquisition step of obtaining, by a virtual reception signal acquisition unit, a virtual reception signal by reflecting the virtual transmission signal transmitted to the synthesis target from the synthesis target;
a radar reception signal acquisition step of acquiring, by a radar reception signal acquisition unit, a radar reception signal in which a radar transmission signal transmitted to an actual target corresponding to the synthesized target is reflected from the actual target;
a generating step of generating, by a generator, a verification signal corresponding to the radar received signal using the virtual received signal and the radar received signal; and
A discriminating step of determining, by a discriminator, the accuracy of the verification signal by comparing the radar reception signal with the verification signal;
The virtual reception signal acquisition step includes:
acquiring the virtual reception signal with reference to image information including the synthesis target and real environment information corresponding to the image information,
Radar signal processing method.
8. The method of claim 7,
Generating a final verification signal based on the accuracy determination result of the verification signal generated by the discriminator; further comprising,
Radar signal processing method.
8. The method of claim 7,
The generating step and the determining step are,
Performing machine learning using at least one of a convolutional neural network (CNN) and a recurrent neural network (RNN),
Radar signal processing method.
delete 8. The method of claim 7,
An analysis step of analyzing the virtual reception signal and the radar reception signal by a signal analysis unit; further comprising,
performing a process of generating the verification signal by the generator and determining the accuracy of the verification signal by the discriminator using the analysis result,
Radar signal processing method.
12. The method of claim 11,
The analysis step is
Converting at least one of the virtual received signal and the radar received signal into a form of a frequency corresponding to time to generate a converted signal for each, and performing the analysis process with the converted signal,
Radar signal processing method.
A computer-readable recording medium for recording a computer program for executing the method according to any one of claims 7 to 9, 11 and 12.

Images