KR20240019538A - Sensor fusion device and and multi-sensor system having same - Google Patents

Abstract

The present invention includes a differentiator that receives first data from a first sensor and the output of a sensor fusion device and calculates an error between them; a PI controller that calculates an output value from the error from the differencer; an integrator that receives the output value of the PI controller and second data from the second sensor and integrates them; A sensor fusion device including an inverter that inverts the output of the integrator and a multi-sensor system including the same are presented.

Description

Sensor fusion device and multi-sensor system having the same {Sensor fusion device and and multi-sensor system having same}

The present invention relates to a sensor fusion device, and particularly to a low-power sensor fusion device that performs a sensor fusion process using an analog circuit and a multi-sensor system equipped therewith.

Mobile objects such as drones and autonomous vehicles are equipped with multiple sensors to determine the speed, direction, and location of the moving object in real time. However, all sensors have their own unique disadvantages. To compensate for these shortcomings, data from sensors with opposing strengths and weaknesses are fused, which is called sensor fusion. By using sensor fusion, accuracy can be improved by reducing noise included in data. In other words, sensor fusion can overcome the limitations of individual sensors by fusing data collected from multiple sensors to generate more reliable information with less uncertainty. Additionally, sensor fusion is widely used in various systems where sensors are used because it does not require the use of high-performance sensors.

Sensor fusion is being actively applied, for example, to unmanned aerial vehicles (UAVs), also called drones. Drones are equipped with various sensors such as GPS, gyroscope, accelerometer, magnetometer, and pressure sensor to determine the drone's exact location or direction in real time. Here, the angular velocity measured by the angular velocity inevitably includes noise in the low frequency band. Conversely, acceleration measured by an accelerometer includes noise in high frequency bands. Therefore, by fusing the data from the two sensors using an appropriate algorithm, a highly accurate signal with noise removed can be obtained. In other words, theoretically, the current posture of the drone can be predicted using only the measured angular velocity, and the posture can be predicted using only the measured acceleration. However, if the posture is predicted using data obtained from a single sensor, there is a lot of error from the actual posture. Sensor fusion is a method used to overcome these limitations. By fusing data obtained from two or more sensors through an appropriate algorithm, noise is removed and highly accurate data can be obtained.

The most widely used algorithms for sensor fusion are the Kalman filter and complementary filter. The performance of these algorithms has already been verified for decades, and as a result, they are widely used in most systems that utilize sensors, such as drones and autonomous vehicles. However, these existing sensor fusion algorithms continuously repeat calculations each time the sensor detects a signal, so the calculation process is complicated and often requires a complex mathematical model. For this purpose, existing sensor fusion algorithms are performed on a digital processor. In other words, the sensor fusion result is obtained by calculating the discrete time-based Kalman filter algorithm on a digital processor. Since the Kalman filter algorithm is performed on a digital processor, there is a problem with excessive computation and, consequently, large power consumption. However, drones, IoT devices, and mobile devices operate with limited batteries, so it is difficult to install high-performance processors. Therefore, it is burdensome to perform the algorithm for sensor fusion using a processor with limited performance.

Meanwhile, in order to perform the existing sensor fusion algorithm, a process of converting analog signals to digital signals is necessary. In other words, the data obtained from the sensor is an analog signal, but a process of converting it into a digital signal is necessary to perform the sensor fusion algorithm. In other words, the signal output from the sensor goes through a process of being converted into a digital signal before being input to the digital processor for the sensor fusion algorithm. This conversion process requires additional energy consumption and causes a certain amount of latency.

Korean Patent No. 10-1880940

The present invention provides a sensor fusion device capable of performing a sensor fusion algorithm by overcoming the shortcomings of the existing sensor fusion algorithm performed on a digital processor, and a multi-sensor system equipped therewith.

The present invention provides a sensor fusion device that can overcome the shortcomings of existing sensor fusion algorithms by performing a sensor fusion algorithm using an analog circuit, and a multi-sensor system equipped therewith.

The present invention provides a sensor fusion device based on an analog-digital hybrid circuit that can reduce power consumption by converting signals using a digital processor and performing a sensor fusion process using an analog circuit, and a multi-sensor system including the same.

A low-power sensor fusion device according to an embodiment of the present invention includes a differencer that receives first data from a first sensor and the output of the sensor fusion device and calculates an error between them; a PI controller that calculates an output value from the error from the differencer; an integrator that receives the output value of the PI controller and second data from the second sensor and integrates them; and an inverter that inverts the output of the integrator.

The first sensor includes an accelerometer and the first data includes Euler angles (E _mea (t)), and the second sensor includes a gyroscope and the second data includes an angular velocity (W _mea (t)). It includes, and the output of the sensor fusion device includes an Euler angle prediction value (E _est (t)).

)을 계산한다.The differencer is _an error value ₍

) is calculated.

The PI controller includes a differential amplifier having first and second input terminals and an output terminal, a memristor provided between the output terminal of the differencer and the first input terminal of the differential amplifier, and the differential amplifier. It includes a resistor and a capacitor connected in series between an output terminal and the first terminal.

의 출력값(-u(t))을 계산하며, G_PI는 멤리스터의 전도도값, R_PI는 저항의 저항값, C_PI는 캐패시터의 캐패시턴스이다.The PI controller receives the error value (e(t)) from the differencer and

Calculate the output value (-u(t)), where G _PI is the conductivity value of the memristor, R _PI is the resistance value of the resistor, and C _PI is the capacitance of the capacitor.

The output value (-u(t)) of the PI controller is proportional to the error value (e(t)) from the differencer.

The PI controller controls the gain value by controlling the conductivity of the memristor by matching the Kalman coefficient (K) to the conductivity of the memristor.

The gain value of the PI controller increases as the conductivity of the memristor increases.

The memristor includes a gate, a source, and a drain, and a channel layer is formed between the gate, the source, and the drain, and the channel layer is a SnS ₂ layer formed in a polycrystalline state with a grain boundary between the crystal layers. Includes.

The integrator receives and integrates the output value of the PI controller and the angular velocity (W _mea (t)) from the angular velocity.

The inverter receives and inverts the output of the integrator to output the predicted Euler angle value (E _est (t)), and the predicted Euler angle value (E _est (t)) is fed back to the differencer.

A multi-sensor system according to another embodiment of the present invention includes two or more sensors; A sensor fusion device that inputs and fuses data from the two or more sensors; and a digital processor, at least a portion of which converts the output of the sensor fusion device.

The two or more sensors output analog data, the sensor fusion device fuses the analog data from the two or more sensors to output an analog signal, and the digital processor converts the analog signals of the sensor fusion device into digital signals.

The sensor fusion device includes a differentiator that receives first data from a first sensor and an output of the sensor fusion device and calculates an error between them; a PI controller that calculates an output value from the error from the differencer; an integrator that receives the output value of the PI controller and second data from the second sensor and integrates them; and an inverter that inverts the output of the integrator.

The first sensor includes an accelerometer and the first data includes Euler angles (E _mea (t)), and the second sensor includes a gyroscope and the second data includes an angular velocity (W _mea (t)). It includes, and the output of the sensor fusion device includes an Euler angle prediction value (E _est (t)).

)을 계산한다.The differencer is _an error value ₍

) is calculated.

The PI controller includes a differential amplifier having first and second input terminals and an output terminal, a memristor provided between the output terminal of the differencer and the first input terminal of the differential amplifier, and the differential amplifier. It includes a resistor and a capacitor connected in series between an output terminal and the first terminal.

의 출력값(-u(t))을 계산하며, G_PI는 멤리스터의 전도도값, R_PI는 저항의 저항값, C_PI는 캐패시터의 캐패시턴스이다.The PI controller receives the error value (e(t)) from the differencer and

Calculate the output value (-u(t)), where G _PI is the conductivity value of the memristor, R _PI is the resistance value of the resistor, and C _PI is the capacitance of the capacitor.

The PI controller controls the gain value by controlling the conductivity of the memristor by matching the Kalman coefficient (K) to the conductivity of the memristor.

The integrator receives and integrates the output value of the PI controller and the angular velocity (W _mea (t)) from the angular velocity.

The inverter receives and inverts the output of the integrator to output the predicted Euler angle value (E _est (t)), and the predicted Euler angle value (E _est (t)) is fed back to the differencer.

Sensor fusion devices according to embodiments of the present invention are composed of analog circuits. That is, the present invention performs a sensor fusion process using a Kalman filter or complementary filter algorithm using an analog circuit. At this time, in the present invention, the analog signal from the sensor is directly input to the analog sensor fusion device to perform the sensor fusion process.

Additionally, the present invention uses a digital processor to perform only a limited role of converting signals to be input to an analog circuit. In other words, the digital processor only serves to deliver signals to the analog sensor fusion device or to convert the analog output signal from the analog sensor fusion device into a digital signal so that it can be used for later control. In other words, in the present invention, a digital processor performs a signal transmission or signal conversion function, and an analog circuit performs a sensor fusion algorithm.

Therefore, by performing the sensor fusion process using an analog circuit, the present invention can solve the computational problem of a digital processor compared to performing a sensor fusion algorithm using a conventional digital processor, thereby enabling less power consumption. .

In addition, the present invention performs a sensor fusion process by directly inputting the analog signal from the sensor, thereby eliminating the need for a process of converting the analog signal to a digital signal, thus eliminating the problem of additional energy consumption and latency compared to the prior art. No.

1 is a schematic diagram of a multi-sensor system to which a sensor fusion device according to an embodiment of the present invention is applied.
Figure 2 is a configuration diagram of a sensor fusion device according to an embodiment of the present invention.
3 to 6 are diagrams for explaining a memristor applied to a sensor fusion device according to an embodiment of the present invention.
Figure 7 shows data experimentally confirming the change in output value of the complementary filter circuit according to the conductivity value of the memristor.
Figure 8 is a graph comparing the results of performing the Kalman filter algorithm using the sensor fusion device according to the present invention and the results of performing the Kalman filter algorithm using a conventional digital processor.
Figure 9 is a graph comparing the results of performing the complementary filter algorithm using a low-power sensor fusion device according to the present invention and the results of performing the complementary filter algorithm using a conventional digital processor.
Figure 10 is a photograph of a drone equipped with a sensor fusion device according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. However, the present invention is not limited to the embodiments disclosed below and will be implemented in various different forms. These embodiments only serve to ensure that the disclosure of the present invention is complete and to convey the scope of the invention to those skilled in the art. It is provided for complete information.

1 is a schematic diagram of a system to which a sensor fusion device according to an embodiment of the present invention is applied. That is, Figure 1 is a schematic diagram of a system in which a sensor fusion device that includes at least two sensors and fuses data from the sensors is applied. These systems may include, for example, drones, IoT devices, mobile devices, etc. that are equipped with at least two sensors and perform sensor fusion.

As shown in Figure 1, the system to which the sensor fusion device according to the present invention is applied is a sensor fusion device that fuses at least two sensors 10 and 20 and data output from at least two sensors 10 and 20. It may include (30) and a digital processor (40) that performs a signal conversion function. Here, the sensors 10 and 20 may include at least two of sensors including GPS, gyroscope, accelerometer, magnetometer, and pressure sensor. At this time, in an embodiment of the present invention, the sensors 10 and 20 may be an angular velocity sensor and an accelerometer, respectively. In addition, the sensor fusion device 30 is composed of an analog circuit and fuses the output signals from the sensors 10 and 20, that is, the sensing data of each sensor 10 and 20. That is, the sensor fusion device 30 is composed of an analog circuit and directly inputs sensing data from the sensors 10 and 20 and fuses these data. In addition, the digital processor 40 performs a function of converting a signal input to the sensor fusion device 30 or a signal output from the sensor fusion device 30. That is, the digital processor 40 controls the overall driving and operation of a multi-sensor system including a plurality of sensors, such as a drone, and performs a signal conversion function as a part of the present invention.

As described above, in the present invention, the sensor fusion device 30 is composed of an analog circuit to perform a sensor fusion process, and the digital processor 40 only performs a signal conversion function. That is, the sensor fusion device 30 of the present invention can implement the sensor fusion algorithm, that is, the Kalman filter or complementary filter, performed in a conventional digital processor, as an analog circuit. Therefore, the present invention can reduce power consumption compared to the conventional method of performing a sensor fusion algorithm using the digital processor 40, thereby enabling a low-power sensor fusion device.

Figure 2 is a configuration diagram of a sensor fusion device according to an embodiment of the present invention. This fusion sensor device is an analog circuit that implements the sensor fusion algorithm, that is, the Kalman filter or complementary filter, performed in a conventional digital processor. A sensor fusion device inputs data from at least two sensors, and the present invention fuses data from an angular velocity sensor and an accelerometer, using a drone as an example. Here, W _mea is the angular velocity measured from the angular velocity, and E _mea is the attitude of the drone predicted using the acceleration measured from the accelerometer, that is, the Euler angle. The drone's attitude is expressed through three Euler angles called roll, pitch, and yaw, and contains a lot of noise compared to the Euler angles of an actual drone. In addition, E _est is a signal from which noise has been removed from the drone's attitude (Euler angle prediction value) predicted through the sensor fusion process, that is, E _mea . As shown in FIG. 2, the low-power sensor fusion device according to an embodiment of the present invention is composed of an analog circuit that performs a sensor fusion process by directly inputting output signals from at least two sensors. That is, the low-power sensor fusion device directly inputs the Euler angle (E _mea (t)) from the accelerometer and the angular velocity (W _mea (t)) from the gyroscope and outputs the predicted Euler angle value (E _est (t)). The present invention can prevent additional energy consumption and latency compared to converting the analog signal output from a conventional sensor into a digital signal by directly inputting the analog signal output from the sensor.

Referring to FIG. 2, the sensor fusion device according to an embodiment of the present invention includes an acceleration value from an accelerometer, that is, the Euler angle (E _mea (t)), and an Euler angle predicted value (E _est (t)), which is an output value of the sensor fusion device. ) is fed back and input to calculate the difference between them, that is, the error value (e(t)), and the output value (- A PI controller 200 that calculates u(t)), inputs the output of the PI controller 200 (-u(t)) and the angular velocity (W _mea (t)) from the angular velocity, integrates them, and outputs them. (-E _est (t)) Includes an integrator 300 and an inverter 400 that inverts the output (-E _est (t)) of the integrator 300 and outputs the predicted Euler angle value (E _est (t)). can do. The sensor fusion device according to an embodiment of the present invention calculates a real-time Euler angle change value (i.e., Euler angle predicted value (E _est (t)) according to the drone's attitude change from the Euler angle calculated value (i.e., Euler angle (E _mea (t)). ))) can be predicted and used for attitude control of the drone. The sensor fusion device according to an embodiment of the present invention will be described in more detail for each configuration as follows.

1. Calm down

)를 출력한다. 다시 말하면, 차분기(100)는 필터를 통해 노이즈가 제거된 오일러 각도 예측값(E_est(t))와 현재의 오일러각(Emea(t))의 차이를 계산한다.The differencer 100 receives feedback and inputs the acceleration from the accelerometer, that is, the Euler angle (E _mea (t)), and the predicted Euler angle value (E _est (t)), which is the output value of the sensor fusion device. That is, the differencer 100 inputs the measured Euler angle (E _mea (t)) and the estimated Euler angle (E _est (t)). At this time, the differencer 100 inputs the Euler angle (E _mea (t)) to the inverting input terminal (-) and inputs the fed-back Euler angle prediction value (E _est (t)) to the non-inverting input terminal (+). It may include a differential amplifier 110. Additionally, the differencer 100 includes a plurality of resistors R11, R12, R13, and R14. That is, the first resistor (R11) connected to the inverting input terminal (-), the second resistor (R12) connected to the non-inverting input terminal (+), and the third resistor (R13) connected between the non-inverting input terminal and the ground terminal. ) and a fourth resistor (R14) connected between the output terminal and the inverting input terminal (-), that is, a feedback resistor. At this time, the first to fourth resistors R11, R12, R13, and R14 of the differential 100 may have the same resistance value, for example, 10 MΩ. This differential 100 amplifies the difference between the two signals input to the inverting input terminal (-) and the non-inverting input terminal (+) of the differential amplifier 110 and outputs the amplified difference. In _other words, the differencer 100 calculates _the error (

) is output. In other words, the differencer 100 calculates the difference between the predicted Euler angle value (E _est (t)) from which noise has been removed through a filter and the current Euler angle (Emea (t)).

2. PI controller

The PI controller (Proportional-Integral controller) 200 inputs the output of the differencer 100 and calculates the output value (-u(t)) required for control. That is, the PI controller 200 calculates an output value by inputting the error e(t) of the Euler angle E _mea (t) and the predicted Euler angle value E _est (t) from the differencer 100. This PI controller 200 may include a memristor 210 and a differential amplifier 220. The memristor 210 is provided between the output terminal of the differential amplifier 100 and the inverting input terminal (-) of the differential amplifier 220. In addition, the differential amplifier 220 receives the output of the memristor 210 to the inverting input terminal (-), and the ground potential is input to the non-inverting input terminal (+). Additionally, the output of the differential amplifier 220 is input to the integrator 300, and the output terminal of the differential amplifier 220 is connected to the inverting input terminal (-) through a resistor (R21) and a capacitor (C21).

라는 출력(-u(t))을 생성한다. 여기서, G_PI는 멤리스터(210)의 전도도값, R_PI는 저항(R21)의 저항값, C_PI는 캐패시터(C21)의 캐패시턴스이다. 한편, PI 제어기(200)는 마이너스 값, 즉 -u(t)를 출력하는데, 그 이유는 센서 퓨전 알고리즘의 3번째 과정, 즉 적분기(300)에서 W_mea(t) 신호와 u(t)를 빼주는 과정이 필요하기 때문에 -u(t)를 PI 제어기(200)의 출력으로 생성한다.This PI controller 200 receives the error value (e(t)) from the differencer 100 and

It generates an output called (-u(t)). Here, G _PI is the conductivity value of the memristor 210, R _PI is the resistance value of the resistor (R21), and C _PI is the capacitance of the capacitor (C21). Meanwhile, the PI controller 200 outputs a negative value, that is, -u(t), because the third process of the sensor fusion algorithm, that is, the integrator 300 combines the W _mea (t) signal and u(t). Because a subtraction process is necessary, -u(t) is generated as the output of the PI controller 200.

Additionally, the PI controller 200 can control the gain value by controlling the conductivity of the memristor 210 according to changes in circumstances. For example, the initial gain value can be set by applying the memristor 210 having a predetermined initial conductivity, and the gain value can be controlled by controlling the conductivity of the memristor 210 according to changes in situations. That is, if the PI controller 200 determines that it is necessary to adjust the gain value of the PI controller 200 that calculates the estimated Euler angle due to the influence of environmental variables such as wind, it can adjust the conductivity of the memristor 210. At this time, the present invention corresponds the Kalman coefficient (K) to the conductivity of the variable resistance element, that is, the memristor 210. For example, if K=1, the conductivity of the memristor 210 is expressed in an analog circuit as 1S[simen]. At this time, the memristor 210 is not a device that operates in a digital manner with only a value of 0 or 1 like existing semiconductor devices, but is a device that can freely change the value between 0 and 1 using next-generation semiconductor materials. Therefore, any K value can be expressed as the conductivity value of the memristor 210.

3 to 6 are diagrams for explaining the memristor 210 applied to the PI controller 200 of the present invention. That is, FIG. 3 is a cross-sectional schematic diagram of the memristor, FIG. 4 is a top view of the memristor, FIG. 5 is a cross-sectional photograph of the memristor, and FIG. 6 is a graph for explaining the change in conductivity of the memristor. As shown in FIG. 3, the memristor can be manufactured as a transistor structure having a gate (G), source (S), and drain (D). At this time, as shown in FIG. 4, the source (S) and the drain (D) may be formed in a direction that intersects the gate (G) at a predetermined distance apart. Meanwhile, as shown in FIGS. 3 and 5, the gate (G) may be made of a doped silicon substrate, and the source (S) and drain (D) may be made of a stacked structure of Ti and Au. However, the gate (G) may be made of a conductive material formed on a silicon substrate, and the source (S) and drain (D) may be made of at least one conductive material as well as a stacked structure. For example, the gate (G), source (S), and drain (D) include at least one of various conductive materials such as Al, Au, Cu, Ir, Ru, Pt, Ti, TiN, Ta, TaN, Cr, etc. can do. Additionally, a gate insulating film made of Al ₂ O ₃ or the like may be formed on the gate G, and a channel layer made of SnS ₂ or the like may be formed on the gate insulating film. Here, the gate insulating film may be formed of various insulating films such as an oxide film, a nitride film, and an oxynitride film in addition to Al ₂ O ₃ . In addition, the channel layer may be formed of transition metal dichalcogenide in addition to SnS ₂ , MoS ₂ , MoSe ₂ , MoTe ₂ , WS ₂ , WSe ₂ , WTe ₂ , TaS ₂ , TaSe ₂ , TiS ₂ , TiSe ₂ , ZrS ₂ , ZrSe ₂ , HfS ₂ , HfSe ₂ , SnSe ₂ , GeS ₂ , GeSe ₂ , GaS ₂ , GaSe ₂ , GaSe, GaTe, InSe, In ₂ Se ₃ , Bi ₂ S ₃ , Bi ₂ It may include at least one selected from the group consisting of Se ₃ and Bi ₂ Te ₃ . When a voltage is applied to the gate (G) of this memristor 210, a channel through which current can easily flow is formed in the SnS ₂ layer, resulting in current flowing between the source (S) and drain (D). do. At this time, the SnS ₂ layer is formed by overlapping several layers of SnS ₂ single crystal atomic layers. That is, the SnS ₂ layer is formed in a poly-crystalline state. In this polycrystalline layer, a grain boundary exists between the crystal layers, and a trap in which electrons can be trapped is formed in this area. Therefore, by applying a specific voltage, electrons can be intentionally filled or extracted from the grain boundaries that exist between the SnS ₂ layers. In that case, the amount of current flowing through the channel may increase or decrease. This is the mechanism that controls the conductivity value of the SnS ₂ memristor. The graph in FIG. 6 is actually measured data, and when an electrical pulse is applied to the SnS ₂ memristor, the conductivity value is between 0.1 and 1.1 μS. It shows that it can be controlled.

로 된다. 여기서, K_P는 비례 게인(gain), K_I는 적분 게인이다. 따라서,

,

가 되기 때문에 G_PI를 조절하면 PI 제어기(200)의 성능을 조절할 수 있다. 즉, 멤리스터(210)의 전도도를 변동시킴으로써 PI 제어기(200)의 게인값을 변동시킬 수 있다. 도 7은 멤리스터의 전도도값에 따른 상보필터 회로의 출력값 변화를 실험적으로 확인한 데이터이다. G_PI값이 0.5μS에서 2μS로 커지게 되면, K_P=1, K_I=1에서 K_P=4, K_I=4로 커지게 된다. 즉, G_PI값이 0.5μS에서 K_P=1, K_I=1이지만, G_PI값이 2μS로 커지게 되면 K_P=4, K_I=4로 커지게 된다. 이렇게 비례 게인과 적분 게인이 커지면 PI 제어기(200)의 출력값이 커지게 된다. 이 경우, 상대적으로 가속도계의 측정값(E_mea(t))이 각속도계의 측정값(W_mea(t))보다 더 많이 최종 상보필터의 출력(E_est(t))에 반영된다. 즉, 센서 퓨전 장치의 최종 출력값이 주로 가속도계의 측정값을 따라 결정된다. 가속도계는 측정값에 고주파 노이즈가 포함되는 단점이 있다. 따라서, 상보필터가 가속도계의 측정값을 주로 반영할 경우, 초록색 그래프처럼 출력값에 노이즈가 포함된다. 반대로, G_PI값이 0.5μS에서 0.1μS으로 작아지게 되면, K_P=1, K_I=1에서 K_P=0.2, K_I=0.2로 작아지게 된다. G_PI값이 0.5μS에서 K_P=1, K_I=1이지만, G_PI값이 0.1μS로 작아지게 되면 K_P=0.2, K_I=0.2로 작아지게 된다. 이 경우 상대적으로 가속도계의 측정값(E_mea(t))보다 각속도계의 측정값(W_mea(t))이 더 많이 최종 상보필터(E_est(t))의 출력값에 반영된다. 각속도계의 측정값(W_mea(t))은 짧은 시간 범위 안에서는 유효하지만 긴 시간동안 측정하면 드리프트(drift) 현상이 발생하면서 측정값이 어느 한쪽으로 서서히 치우치는 단점이 있다. 따라서, 빨간색 그래프처럼 측정값의 최대/최소값이 들쑥날쑥한 것을 볼 수 있다. 결국, 상보필터에 적절한 PI 제어기(200)의 게인을 정해주어야 이상적인 출력값을 얻을 수 있는데, 기존에는 게인값을 몇번의 시행착오(trial and error)를 거쳐 임의로 정해주고, 고정된 값을 계속 사용하였다. 그러나, 본 발명은 PI 제어기(200)의 게인값을 가변할 수 있는 멤리스터(210)의 전도도값에 대응시켰기 때문에 주변 상황이 변하여 상보필터의 특성을 변화시킬 필요가 발생한 경우 PI 제어기(200)의 게인값을 적절히 변화시킬 수 있다. 그리고, 칼만필터와 마찬가지로 저전력으로 상보필터 알고리즘을 계산할 수 있다.As described above, the PI controller 200 can adjust the performance of the PI controller 200 by adjusting the conductivity of the memristor 210, that is, G _PI . The output of a typical PI controller is

It becomes. Here, K _P is the proportional gain and K _I is the integral gain. thus,

,

Since , the performance of the PI controller 200 can be adjusted by adjusting G _PI . That is, the gain value of the PI controller 200 can be changed by changing the conductivity of the memristor 210. Figure 7 is data experimentally confirming the change in output value of the complementary filter circuit according to the conductivity value of the memristor. When the G _PI value increases from 0.5μS to 2μS, it increases from K _P =1, K _I =1 to K _P =4, K _I =4. In other words, when the G _PI value is 0.5μS, K _P = 1 and K _I = 1, but when the G _PI value increases to 2 μS, it increases to K _P = 4 and K _I = 4. As the proportional gain and integral gain increase, the output value of the PI controller 200 increases. In this case, relatively more of the measured value of the accelerometer (E _mea (t)) is reflected in the output (E _est (t)) of the final complementary filter than the measured value of the angular velocity (W _mea (t)). In other words, the final output value of the sensor fusion device is mainly determined by the measured value of the accelerometer. Accelerometers have the disadvantage that their measurements include high-frequency noise. Therefore, when the complementary filter mainly reflects the measured value of the accelerometer, noise is included in the output value, as shown in the green graph. Conversely, when the G _PI value decreases from 0.5μS to 0.1μS, it decreases from K _P =1, K _I =1 to K _P =0.2, K _I =0.2. When the G _PI value is 0.5μS, K _P = 1 and K _I = 1, but when the G _PI value decreases to 0.1 μS, it decreases to K _P = 0.2 and K _I = 0.2. In this case, the angular velocity measurement value (W _mea (t)) is relatively more reflected in the output value of the final complementary filter (E _est (t)) than the accelerometer measurement value (E _mea (t)). The measured value of the angular velocity (W _mea (t)) is valid within a short time range, but when measured over a long period of time, a drift phenomenon occurs and the measured value gradually becomes biased to one side. Therefore, you can see that the maximum/minimum values of the measured values are uneven, as shown in the red graph. In the end, an ideal output value can be obtained only by determining the gain of the PI controller 200 appropriate for the complementary filter. In the past, the gain value was arbitrarily determined through several trials and errors, and the fixed value was continued to be used. . However, since the present invention corresponds the gain value of the PI controller 200 to the conductivity value of the variable memristor 210, if the surrounding conditions change and it is necessary to change the characteristics of the complementary filter, the PI controller 200 The gain value can be changed appropriately. And, like the Kalman filter, the complementary filter algorithm can be calculated with low power.

3. Integrator

The integrator 300 inputs the output of the PI controller 200 and the measured value (W _mea (t)) of the angular velocity meter. That is, the integrator 300 integrates the output of the PI controller 200 (-u(t)) and the measured value of the angular velocity (W _mea (t)). At this time, the output (-u(t)) of the PI controller 200 is the error value (e(t)), that is, the Euler angle (E _mea (t)) from the differencer 100 and the Euler angle predicted value (E _est (t)) is proportional to the error (e(t)), and this error value is ultimately proportional to the measured value of the accelerometer (E _mea (t)). Therefore, the integrator 300 adds the measured value of the accelerometer (E _mea (t)) and the measured value of the angular velocity (W _mea (t)), which measurement value will have a greater influence on the final complementary filter output. It is decided whether Meanwhile, the integrator 300 inputs the output of the PI controller 200 and the measured value (Wmea(t)) of the angular velocity into the inverting input terminal (-), and the ground potential is applied to the non-inverting input terminal (+). It may include an amplifier 310. At this time. The output of the PI controller 200 and the measured value (W _mea (t)) of the angular velocity meter may be input to the inverting input terminal (-) through the first and second resistors (R31 and R32) connected in parallel, respectively. That is, the output of the PI controller 200 is input through the first resistor (R31), and the measured value (W _mea (t)) of the angular velocity meter is input through the second resistor (R32) to the inverting input terminal of the differential amplifier 310. It can be entered as (-). At this time, the first and second resistors R31 and R32 have the same resistance value. Additionally, the output of the differential amplifier 310 is input to an inverter (unity gain inverter) 400. The output of the differential amplifier 310 is fed back to the inverting input terminal (-) through the third resistor R33 and capacitor C31 connected in parallel. Here, the resistance value of the third resistor R33 is smaller than the resistance values of the first and second resistors R31 and R32.

4. Inverter

The inverter 400 inputs the output of the integrator 300, inverts it, and outputs it. That is, the inverter 400 serves to eliminate the negative sign because the output of the integrator 300 is -E _est (t). In other words, the inverter 400 serves to match the sign so that the final output of the analog complementary filter circuit is E _est (t). Accordingly, the output of the inverter 400 becomes the predicted Euler angle value (E _est (t)). The inverter 400 inputs the output of the integrator 300 to the inverting input terminal (-), and the non-inverting input terminal (+) inputs the ground potential. At this time, the inverter 400 inputs the output of the integrator 300 through the first resistor (R41). Additionally, the output of the inverter 400 is fed back to the inverting input terminal (-) through the second resistor (R42). Accordingly, the resistance values of the first and second resistors R41 and R42 are set to be the same, and the input value is inverted and output without changing the amplitude. That is, the inverter 400 inverts the output of the integrator 300 without changing the amplitude and outputs it. The output of the inverter 400, that is, the predicted Euler angle value (E _est (t)), is fed back to the differencer 100.

As described above, the sensor fusion device according to an embodiment of the present invention is composed of an analog circuit. That is, the present invention performs a sensor fusion process using a Kalman filter or complementary filter algorithm using an analog circuit. At this time, in the present invention, the analog signal from the sensor is directly input to the analog sensor fusion device to perform the sensor fusion process. In addition, the present invention uses a digital processor to perform only a limited role of converting signals to be input to an analog circuit. In other words, the digital processor serves only to transmit signals to the analog sensor fusion device or to convert the analog output signal from the analog sensor fusion device into a digital signal so that it can be used for later control. In other words, in the present invention, a digital processor performs a signal transmission or signal conversion function, and an analog circuit performs a sensor fusion process. Accordingly, a sensor fusion device based on an analog-digital hybrid circuit can be implemented.

Therefore, by performing the sensor fusion process using an analog circuit, the present invention can solve the computational problem of a digital processor compared to performing a sensor fusion algorithm using a conventional digital processor, thereby enabling less power consumption. . In addition, the present invention performs a sensor fusion process by directly inputting the analog signal from the sensor, so there is no need for a process of converting the analog signal to a digital signal, so there is no problem of additional energy consumption and latency compared to the prior art. No.

In addition, the present invention uses a next-generation resistance change element, that is, a memristor 210, to enable the performance of sensor fusion to be variably adjusted according to the surrounding environment. For example, when there is a lot of wind and the noise increases compared to a predetermined standard noise, the performance of sensor fusion can be adjusted by adjusting the conductivity value of the memristor.

In comparison, the sensor fusion method using a conventional digital processor adjusts variables that determine performance (e.g., Kalman variable K in the Kalman filter) to optimize sensor fusion performance. For example, in the Kalman filter, the K value is updated through algorithmic calculation every time new data is input from the sensor. In other words, the process of recalculating the K value occurs more than dozens of times per second, and the power consumption required for this process is significant. In the case of complementary filters, performance is optimized with two variable values (proportional gain K _P and integral gain K _I ), and these two values are selected by repeating several trial and error processes.

However, in the sensor fusion device of the present invention, the K value is stored in the conductivity value of the memristor, and changing the conductivity value of one element consumes relatively very little power. This is also one of the reasons why the analog-digital hybrid method consumes less power than performing sensor fusion through existing software. In other words, the variable that determines the performance of existing sensor fusion is stored in the conductivity value of the variable resistance element, that is, the memristor, and because changing the conductivity of the memristor requires relatively very little power consumption, as a result, the power required for sensor fusion Consumption can be reduced.

Next, by comparing the analog-digital hybrid-based sensor fusion device according to the present invention with the sensor fusion method using a conventional digital processor, it is shown using drawings and tables that the performance according to the present invention is similar to the existing one and power consumption can be reduced. Let me explain.

Figure 8 is a graph comparing the results of performing the Kalman filter algorithm using a low-power sensor fusion device according to the present invention and the results of performing the Kalman filter algorithm using a conventional digital processor. In Figure 8, the blue dotted line is the result of performing the Kalman filter algorithm using a conventional digital processor, and the red solid line is the result of performing the Kalman filter algorithm using the low-power sensor fusion device according to the present invention, and the two results are exactly the same. It can be seen that

In addition, Figure 9 is a graph comparing the results of performing the complementary filter algorithm using a low-power sensor fusion device according to the present invention and the results of performing the complementary filter algorithm using a conventional digital processor. In Figure 9, the black dotted line is the result of performing the complementary filter algorithm using a conventional digital processor, and the solid red line is the result of performing the complementary filter algorithm using the low-power sensor fusion device according to the present invention, and the two results are almost identical. You can see the similarity.

As can be seen in FIGS. 8 and 9, when comparing the results of sensor fusion performed with a low-power sensor fusion device according to the present invention and the results of sensor fusion performed with a digital processor in a conventional manner, there is almost no difference in performance. You can see that there is no difference.

Figure 10 is a photograph of a sensor fusion device according to an embodiment of the present invention mounted on a drone. In addition, Table 1 is a table showing the characteristics of the Kalman filter algorithm according to the existing method, and Table 2 is a table showing the characteristics of the Kalman filter algorithm according to the present invention. That is, Table 1 shows the results of measuring power consumption when the code for the Kalman filter algorithm is calculated using a drone's microcontroller, that is, a digital processor, as in the existing method, and Table 2 shows the results of measuring the power consumption using the digital processor and the digital processor according to the present invention. This is the result of measuring power consumption when calculating the Kalman filter algorithm using an analog-digital hybrid device consisting of an analog low-power sensor fusion device. That is, Table 2 shows the results of measuring power consumption when the digital processor performs signal conversion and the analog low-power sensor fusion device performs the sensor fusion process according to the Kalman filter algorithm. As shown in Figure 10, as a result of testing after mounting the low-power sensor fusion device according to the present invention on an actual drone, sensor fusion with the same performance was achieved with only 1/4 the power consumption compared to sensor fusion performed with a conventional digital processor. This is possible. This is explained using Table 1 and Table 2 as follows.

digital component V ⁺ +5V I _d(max) 12.3㎃ I _d(min) 7.5㎃

197mJ

digital component analog component V ⁺ 5V +5V V ^- - -5V I _d(max) 5.5㎃ - I _d(min) 2.2㎃ - I _a(max) - +41㎂ I _a(min) - +4㎂ Io _(max) - +37.5㎂ I _o(min) - -37.5㎂

53.7mJ
-

-
0.79mJ

54.5mJ

는 종래 방식으로 5초동안 칼만필터 알고리즘을 디지털 프로세서에서 계산했을 때 소모된 총 전력 소모량(전압×전류×시간)으로 197mJ이고,

는 본 발명에 따라 아날로그-디지털 하이브리드 컴퓨팅을 이용하여 칼만필터 알고리즘을 계산했을 때 디지털 프로세서에서 소모된 전력 소모량으로 53.7mJ이다. 그리고,

는 본 발명에 따라 아날로그-디지털 하이브리드 컴퓨팅을 이용해 칼만필터 알고리즘을 계산했을 때 아날로그 컴포넌트에서 소모된 전력 소모량으로 0.79mJ이고,

는 본 발명에 따라 아날로그-디지털 하이브리드 컴퓨팅을 이용해 칼만필터 알고리즘을 계산했을 때 소모된 전체 전력 소모량으로 54.5mJ이다.Power consumption (P) can be calculated from voltage × current × time. Here, the supply voltage is constant, but the current continues to change over time, so in this test, the power consumption for 5 seconds was calculated and compared. That is, the voltage × current value for 5 seconds was integrated over time to compare the power consumption according to the conventional method (P _soft-K ) and the power consumption according to the present invention (P _hybrid-K ). Here, V ⁺ is positive power (+5V voltage) supplied to the digital processor or low-power sensor fusion device, and V ^- is negative power (-5V voltage) supplied to the low-power sensor fusion device. Additionally, I _d(max) and I _d(min) are the maximum and minimum values of current supplied to the digital processor. However, because the I _d value is a value that continuously changes in real time, the I _d(max) and I _d(min) values do not have much meaning, and in the end, the power consumption value obtained through integration has a lot of meaning. At this time, the conventional I _d(max) and I _d(min) are 12.3 mA and 7.5 mA, respectively, and the I _d(max) and I _d(min) of the present invention are 5.5 mA and 2.2 mA, respectively. And, I _a(max) and I _a(min) are the maximum and minimum values of the current supplied to the low-power sensor fusion device, which are 41㎂ and 4㎂, respectively, and I _o(max) and I _o(min) are the low-power sensor fusion device. The maximum and minimum values of current output from the fusion device are +37.5㎂ and -37.5㎂, respectively. Meanwhile,

is the total power consumption (voltage

is the power consumption consumed by the digital processor when calculating the Kalman filter algorithm using analog-digital hybrid computing according to the present invention, which is 53.7mJ. and,

is the power consumption consumed by the analog component when calculating the Kalman filter algorithm using analog-digital hybrid computing according to the present invention, which is 0.79mJ;

is the total power consumption consumed when calculating the Kalman filter algorithm using analog-digital hybrid computing according to the present invention, which is 54.5mJ.

)이 197mJ이고, 표 2에 나타낸 바와 같이 본 발명에 따라 아날로그-디지털 하이브리드 컴퓨팅을 이용해 칼만필터 알고리즘을 계산했을 때 소모된 전체 전력 소모량(

)이 54.5mJ로서, 본 발명은 종래에 비해 약 1/4 정도의 전력을 소비함을 알 수 있다.As shown in Table 1, the total power consumption (

) is 197mJ, and as shown in Table 2, the total power consumption consumed when calculating the Kalman filter algorithm using analog-digital hybrid computing according to the present invention (

) is 54.5mJ, so it can be seen that the present invention consumes about 1/4 of the power compared to the prior art.

The technical idea of the present invention as described above has been described in detail according to the above-mentioned embodiments, but it should be noted that the above-described embodiments are for explanation and not limitation. Additionally, those skilled in the art will understand that various embodiments are possible within the scope of the technical idea of the present invention.

100: Differentiator 200: PI controller
300: integrator 400; inverter

Claims (21)

A differencer that receives the first data from the first sensor and the output of the sensor fusion device and calculates an error between them;
a PI controller that calculates an output value from the error from the differencer;
an integrator that receives the output value of the PI controller and second data from the second sensor and integrates them; and
A sensor fusion device including an inverter that inverts the output of the integrator.
The method of claim 1, wherein the first sensor includes an accelerometer and the first data includes an Euler angle (E _mea (t)), and the second sensor includes a goniometer and the second data includes an angular velocity (W). _mea (t)), and the output of the sensor fusion device includes an Euler angle prediction value (E _est (t)).
The method of claim 2, wherein the differencer is _an error value ₍

) a sensor fusion device that calculates
The method of claim 2, wherein the PI controller,
A differential amplifier having first and second input terminals and output terminals,
a memristor provided between the output terminal of the differential amplifier and the first input terminal of the differential amplifier;
A sensor fusion device including a resistor and a capacitor connected in series between the output terminal of the differential amplifier and the first terminal.
The method of claim 4, wherein the PI controller receives an error value (e(t)) from the differencer and

Calculate the output value (-u(t)) of the sensor fusion device, where G _PI is the conductivity value of the memristor, R _PI is the resistance value of the resistor, and C _PI is the capacitance of the capacitor.
The sensor fusion device of claim 5, wherein the output value (-u(t)) of the PI controller is proportional to the error value (e(t)) from the differencer.
The sensor fusion device of claim 6, wherein the PI controller controls the gain value by controlling the conductivity of the memristor by matching the Kalman coefficient (K) to the conductivity of the memristor.
The sensor fusion device of claim 7, wherein the gain value of the PI controller increases as the conductivity of the memristor increases.
The method of claim 8, wherein the memristor includes a gate, a source, and a drain, and a channel layer is formed between the gate, the source, and the drain, and the channel layer is in a polycrystalline state with a grain boundary between the crystal layers. Sensor fusion device containing _two layers of SnS formed.
The sensor fusion device of claim 2, wherein the integrator receives and integrates the output value of the PI controller and the angular velocity (W _mea (t)) from the angular velocity.
The method of claim 2, wherein the inverter receives and inverts the output of the integrator to output the Euler angle prediction value (E _est (t)), and the Euler angle prediction value (E _est (t)) is fed back to the differencer. Sensor fusion device.
Two or more sensors;
A sensor fusion device that inputs and fuses data from the two or more sensors; and
A multi-sensor system, at least a portion of which includes a digital processor that converts the output of the sensor fusion device.
The method of claim 12, wherein the two or more sensors output analog data, the sensor fusion device fuses the analog data from the two or more sensors to output an analog signal, and the digital processor converts the analog signals of the sensor fusion device into digital signals. A multi-sensor system that converts into signals.
The method of claim 13, wherein the sensor fusion device,
A differencer that receives first data from the first sensor and the output of the sensor fusion device and calculates an error between them;
a PI controller that calculates an output value from the error from the differencer;
an integrator that receives the output value of the PI controller and second data from the second sensor and integrates them; and
A multi-sensor system including an inverter that inverts the output of the integrator.
The method of claim 14, wherein the first sensor includes an accelerometer and the first data includes an Euler angle (E _mea (t)), and the second sensor includes a gyroscope and the second data includes an angular velocity (W). _mea (t)), and the output of the sensor fusion device includes an Euler angle prediction value (E _est (t)).
The method of claim 15, wherein the differencer calculates an _error value ₍

) A multi-sensor system that calculates
The method of claim 15, wherein the PI controller:
A differential amplifier having first and second input terminals and output terminals,
a memristor provided between the output terminal of the differential amplifier and the first input terminal of the differential amplifier;
A multi-sensor system including a resistor and a capacitor connected in series between the output terminal of the differential amplifier and the first terminal. The method of claim 16, wherein the PI controller receives an error value (e(t)) from the differencer and

Calculate the output value (-u(t)) of a multi-sensor system where G _PI is the conductivity value of the memristor, R _PI is the resistance value of the resistor, and C _PI is the capacitance of the capacitor.
The multi-sensor system of claim 18, wherein the PI controller controls the gain value by controlling the conductivity of the memristor by matching the Kalman coefficient (K) to the conductivity of the memristor.
The multi-sensor system of claim 15, wherein the integrator receives and integrates the output value of the PI controller and the angular velocity (W _mea (t)) from the angular velocity.
The method of claim 15, wherein the inverter receives and inverts the output of the integrator to output the Euler angle prediction value (E _est (t)), and the Euler angle prediction value (E _est (t)) is fed back to the differencer. Multi-sensor system.

Images