JP7294031B2 - Information processing device, information processing method, program, prediction device and prediction method - Google Patents

Description

The present invention relates to an information processing device, an information processing method, a program, a prediction device, and a prediction method.

In recent years, various techniques have been known for performing noise reduction on signals. For example, a technique of detecting an amplitude error from a signal (I signal and Q signal) containing both a phase error and an amplitude error and correcting the signal so as to reduce the amplitude error is disclosed ( For example, see Patent Document 1.). However, in such a technique, an amplitude error is detected from a signal containing a phase error. If correction is made to reduce the error, the amplitude will change along with the correction. That is, even if an amplitude error detected from a signal containing a phase error is used, amplitude correction cannot be performed with high accuracy.

On the other hand, there has been disclosed a technique for reducing noise with higher accuracy by detecting a phase error and an amplitude error and correcting a signal based on the detected phase error and amplitude error (see, for example, Patent Document 2).

JP 2004-266416 A JP 2009-118114 A

A signal observed by a given sensor (observed signal) may have a phase change, and a phase error may occur at a phase change point. For example, when capturing a physical change as a phase change of an observation signal, capturing the phase change more accurately leads to capturing the physical change with higher accuracy. However, neither Patent Literature 1 nor Patent Literature 2 mentions a technique for capturing a phase change.

Therefore, in general, in order to capture phase changes of a signal more accurately, the signal accuracy is improved by increasing the signal sampling period, or by increasing the amount of data and averaging many signals. need to be raised. However, increasing the sampling period or increasing the number of data requires high performance hardware (or enormous computational resources) or a long processing time.

Therefore, it is desirable to provide a technique that enables noise reduction on signals with higher accuracy.

In order to solve the above problem, according to one aspect of the present invention, output data output from the neural network based on inputting a plurality of analytic signals as input data to the neural network, and the output data and a learning unit that updates the parameters of the neural network based on the phase corresponding to the input data, the noise removal signal based on the input data , and the training data including the phase of the correct answer; and the neural network after the parameters are updated. and a model output unit that outputs the network as a trained model.
Further, the learning unit calculates an inter-phase loss based on the phase corresponding to the output data and the correct phase, and calculates the inter-signal loss and the inter-phase loss based on the output data and the noise removal signal. The parameters may be updated based on and.

According to another aspect of the present invention, output data output from the neural network based on inputting a plurality of analytic signals as input data to the neural network, and a phase corresponding to the output data. , updating the parameters of the neural network based on the noise-removed signal based on the input data and teacher data including the phase of the correct answer ; and using the neural network after the parameters have been updated as a trained model. and outputting.

According to another aspect of the present invention, a computer is caused to input a plurality of analytic signals as input data to a neural network, and output data output from the neural network, and a learning unit for updating the parameters of the neural network based on the phase to be performed, the noise-removed signal based on the input data, and teacher data including the phase of the correct answer ; and the neural network after the parameters have been updated. and a model output unit that outputs a trained model.

According to another aspect of the present invention, output data output from a neural network corresponding to input data, a phase corresponding to the output data, a noise removal signal based on the input data , and a correct phase are Acquire the neural network whose parameters have been updated based on the included teacher data as a trained model, and obtain the output data output from the trained model corresponding to the analytic signal as a noise removal signal. A prediction device is provided, comprising a processing unit and a data output unit for outputting the denoised signal.

According to another aspect of the present invention, output data output from a neural network corresponding to input data, a phase corresponding to the output data, a noise removal signal based on the input data , and a correct phase are acquiring the neural network after the parameters have been updated as a trained model based on the included teacher data, and obtaining the output data output from the trained model corresponding to the analytic signal as a noise removal signal. and outputting the denoised signal.

According to another aspect of the present invention, a computer is provided with output data output from a neural network corresponding to input data, a phase corresponding to the output data, a noise elimination signal based on the input data , and a correct answer. The neural network after the parameters have been updated is acquired as a trained model based on teacher data including the phase of and the output data output from the trained model corresponding to the analytic signal is a noise removal signal and a data output unit for outputting the noise-removed signal.

As described above, according to the present invention, there is provided a technique that enables noise reduction for signals to be performed with higher accuracy.

It is a figure which shows the structural example of the noise reduction system (information processing system) which concerns on embodiment of this invention. It is a figure which shows the functional structural example of the trained model production|generation apparatus which concerns on the same embodiment. It is a figure which shows the structural example of the neural network which concerns on the same embodiment. 6 is a flow chart showing an operation example of the trained model generation device according to the same embodiment. It is a figure which shows the functional structural example of the noise removal signal generation apparatus which concerns on the same embodiment. 4 is a flowchart showing an operation example of the noise removal signal generation device according to the same embodiment; It is a figure which shows the hardware constitutions of the information processing apparatus as an example of the trained model production|generation apparatus which concerns on embodiment of this invention.

Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In the present specification and drawings, constituent elements having substantially the same functional configuration are denoted by the same reference numerals, thereby omitting redundant description.

In addition, in this specification and drawings, a plurality of components having substantially the same functional configuration may be distinguished by attaching different numerals after the same reference numerals. However, when there is no particular need to distinguish between a plurality of constituent elements having substantially the same functional configuration, only the same reference numerals are used. Also, similar components in different embodiments may be distinguished by attaching different alphabets after the same reference numerals. However, when there is no particular need to distinguish between similar components of different embodiments, only the same reference numerals are used.

(1. Details of embodiment)
Next, details of embodiments of the present invention will be described.

(1-1. Description of system configuration)
FIG. 1 is a diagram showing a configuration example of a noise reduction system (information processing system) according to an embodiment of the present invention. As shown in FIG. 1, the noise reduction system 1 includes a trained model generation device 10 (information processing device) and a noise-removed signal generation device 20 (prediction device).

In the embodiment of the present invention, the trained model generation device 10 and the noise removal signal generation device 20 are configured to be able to communicate with each other. is transmitted to the noise removal signal generation device 20 and used for noise reduction by the noise removal signal generation device 20. FIG. However, the trained model generated by the trained model generation device 10 may be provided to the noise removal signal generation device 20 in any manner. For example, the trained model generated by the trained model generation device 10 is written to a predetermined recording medium by the trained model generation device 10, and the trained model is read from the recording medium by the noise removal signal generation device 20. good too.

Furthermore, in the embodiments of the present invention, it is mainly assumed that the trained model generation device 10 and the noise removal signal generation device 20 are configured by different computers. However, the trained model generation device 10 and the noise removal signal generation device 20 may be configured by the same computer.

(1-2. Configuration of Trained Model Generating Device)
Next, functions of the trained model generation device 10 according to the embodiment of the present invention will be described. FIG. 2 is a diagram showing a functional configuration example of the trained model generation device 10 according to the embodiment of the present invention. As shown in FIG. 2, the trained model generation device 10 according to the embodiment of the present invention includes a data acquisition unit 110, a preprocessing unit 120, a learning unit 130 and a model output unit 140.

The trained model generation device 10 includes an arithmetic device such as a CPU (Central Processing Unit), and a program stored in a memory (not shown) is expanded in a RAM (Random Access Memory) by the CPU and executed, thereby generating the program. Functionality can be realized. At this time, a computer-readable recording medium recording the program may also be provided. Alternatively, the trained model generation device 10 may be configured with dedicated hardware, or may be configured with a combination of multiple pieces of hardware.

The data acquisition unit 110 acquires learning data and teacher data as digital signals. For example, when learning data and teacher data are recorded in advance on a predetermined recording medium, the data acquisition unit 110 acquires the learning data and teacher data from the recording medium. The learning data includes input signals obtained by a given sensor. Here, the type of sensor is not limited. For example, the sensor may be a fiber optic sensor or an acoustic sensor. On the other hand, the teacher data includes a signal (noise-removed signal) after noise has been removed from the input signal. The data acquisition unit 110 acquires a plurality of combinations of such input signals and noise-removed signals.

The preprocessing unit 120 performs various preprocessing on a plurality of input signals. More specifically, the preprocessing unit 120 shapes each of the plurality of input signals into an analytic signal. The analytic signal is composed of a combination of the I signal and the Q signal. Therefore, preprocessing section 120 obtains a plurality of combinations (analysis signals) of I and Q signals. For example, when the input signal is only the I signal, the preprocessing unit 120 converts the input signal into the analytic signal by Hilbert transform or the like.

More specifically, let the first analytic signals be Ix1 and Qx1, the second analytic signals be Ix2 and Qx2, . , each analytic signal is represented by the following equation (1).

Ix1(t)=cos(ωt+φ1+θi1)+zi1
Qx1(t)=sin(ωt+φ1+θq1)+zq1
Ix2(t)=cos(ωt+φ2+θi2)+zi2
Qx2(t)=sin(ωt+φ2+θq2)+zq2
・・・
Ixn(t)=cos(ωt+φn+θin)+zin
Qxn(t)=sin(ωt+φn+θqn)+zqn
... (1)

After that, if there is a phase shift between the analytic signals, the preprocessing unit 120 aligns the phases of the analytic signals. As a result, the learning unit 130 can average the plurality of analytic signals in the direction of the number of signals while preventing attenuation of the plurality of analytic signals. The first analytic signals after the phases are aligned are I1 and Q1, the second analytic signals after the phases are aligned are I2 and Q2, . . . n after the phases are aligned When In and Qn are the second analytic signals, the respective analytic signals after the phases are aligned are represented by Equation (2).

I1(t)=cos(ωt)+zi1
Q1(t)=sin(ωt)+zq1
I2(t)=cos(ωt)+zi2
Q2(t)=sin(ωt)+zq2
・・・
In(t)=cos(ωt)+zin
Qn(t)=sin(ωt)+zqn
... (2)

Note that equation (2) shows an example in which each phase is aligned with ωt. However, each phase may be adjusted to another value (reference phase). For example, each phase may be aligned to ωt+π/2 or ωt+φ1. In the following, I1(t), Q1(t), I2(t), Q2(t), . Also called At this time, the noise removal signals Iy and Qy included in the teacher data are expressed as shown in Equation (3) below.

Iy=cos(ωt)
Qy=sin(ωt)
... (3)

Next, a configuration example of a neural network according to an embodiment of the present invention will be described. FIG. 3 is a diagram showing a configuration example of a neural network according to an embodiment of the invention. As shown in FIG. 3, the embodiment of the present invention assumes that the neural network includes an encoder N1 followed by a decoder N2. The encoder N1 functions to generate a feature amount by compressing the data amount based on the input data, and the decoder N2 functions to restore the data amount based on the feature amount input from the encoder N1.

For example, the encoder N1 is constructed by repeating convolutional and pooling layers, and the decoder N2 is constructed by repeating convolutional and upsampling layers. However, the specific configuration of the encoder N1 is not limited, nor is the specific configuration of the decoder N2.

Here, examples of the neural network including the encoder N1 and the decoder N2 include AE (Auto Encoder), VAE (Variational Auto Encoder), U-net, and the like. However, the neural network including encoder N1 and decoder N2 is not limited to such examples.

Furthermore, the neural network according to embodiments of the present invention is not limited to including encoder N1 and decoder N2. For example, a neural network according to embodiments of the present invention may include convolution layers that perform convolution operations. According to such a convolution operation, a feature amount can be calculated from the input data (averaging of a plurality of analytic signals in the signal number direction can be performed).

The learning unit 130 inputs a plurality of analytic signals as input data I and Q to the neural network. More specifically, the learning unit 130 inputs the plurality of phase-aligned analytic signals as input data I and Q to the neural network. When input data I and Q are input to the neural network, output data I' and Q' corresponding to the input data I and Q are output from the neural network. The learning unit 130 updates the parameters (weight parameters) of the neural network based on the output data I' and Q' and the teacher data Iy and Qy (including the noise removal signal based on the input data). More specifically, the respective parameters of encoder N1 and decoder N2 are updated.

A parameter update method is not limited. For example, the learning unit 130 calculates the loss between the I signals of the output data and the teacher data (that is, the loss between the I signals between I' and Iy), and the loss between the Q signals of the output data and the teacher data. (ie, the Q inter-signal loss between I' and Iy). Learning section 130 then updates the parameters of the neural network based on the I-signal loss and the Q-signal loss using error backpropagation or the like.

Furthermore, in the embodiment of the present invention, it is assumed that the teacher data includes not only the noise-removed signals Iy, Qy but also the correct phase values Py (Iy, Qy) corresponding to the noise-removed signals Iy, Qy. At this time, when the noise elimination signals are represented by Iy and Qy in the above equation (3), the correct phase values Py (Iy, Qy) corresponding to the noise elimination signals Iy and Qy are ωt. be.

Based on the phase P' (I', Q') corresponding to the output data I', Q' and the phase correct value Py (Iy, Qy) corresponding to the noise removal signals Iy, Qy, the learning unit 130 Then, the loss (loss between phases) is further calculated. Learning section 130 updates the parameters based on the I-signal loss, Q-signal loss, and phase-phase loss. In this way, the phase-to-phase loss is also reflected in the parameter update, so that noise reduction that can cope with the phase change is performed.

Note that each of these losses may be calculated by a loss function. The loss function may be MAE (Mean Absolute Error), MSE (Mean Square Error), or other functions. Also, the loss between I signals, the loss between Q signals, and the loss between phases may be calculated with different loss functions. For example, the loss between I signals and the loss between Q signals may be calculated by MAE, and the loss between phases may be calculated by MSE.

Each of the I-signal loss, the Q-signal loss and the phase-to-phase loss may be equally reflected in the parameter update of the neural network, or may be reflected in the parameter update of the neural network based on individually set weights. good. That is, learning section 130 may update the parameters based on the weighted sum of the I-signal loss, Q-signal loss, and phase-phase loss. where α is the weight of the I-signal loss, β is the weight of the Q-signal loss, γ is the weight of the phase-to-phase loss, J is the loss function, and MAE is used as the loss function. , the loss function J is expressed by the weighted sum as shown in the following equation (4).

J=α|Iy−I′|+β|Qy−Q′|+γ|Py−P′| (4)

The learning unit 130 evaluates the accuracy of the neural network (model) after the parameters have been updated. Then, when the accuracy of the model is not sufficient, the learning unit 130 re-inputs the input data to the neural network, and updates the parameters again based on the output data and the teacher data. On the other hand, when the accuracy of the model is sufficient, learning section 130 outputs the model to model output section 140 as a learned model. As an example, whether or not the accuracy of the model is sufficient may be determined based on whether or not the loss function is smaller than a threshold.

The model output unit 140 outputs the neural network after the parameters have been updated as a trained model. As described above, in the embodiment of the present invention, it is assumed that the model output unit 140 transmits the generated learned model to the noise removal signal generation device 20 . However, the model output unit 140 may write the generated learned model to a predetermined recording medium. In such a case, the trained model is read from the recording medium by the noise removal signal generator 20 .

(1-3. Operation of Trained Model Generating Device)
Next, the operation of the trained model generation device 10 according to the embodiment of the present invention will be explained. FIG. 4 is a flow chart showing an operation example of the trained model generation device 10 according to the embodiment of the present invention. Since the configuration of the trained model generation device 10 has been described in detail as above, the description of the operation of the trained model generation device 10 here is simpler than the description of the configuration of the trained model generation device 10. change it.

In the trained model generation device 10, the data acquisition unit 110 acquires learning data and teacher data as digital signals (S11). The learning data includes input signals obtained by a given sensor. On the other hand, the teacher data includes a signal (noise-removed signal) after noise has been removed from the input signal. The data acquisition unit 110 acquires a plurality of combinations of such input signals and noise-removed signals.

The preprocessing unit 120 performs various types of preprocessing on a plurality of input signals (S12). More specifically, the preprocessing unit 120 shapes each of the plurality of input signals into an analytic signal. The analytic signal is composed of a combination of the I signal and the Q signal. Therefore, preprocessing section 120 obtains a plurality of combinations (analysis signals) of I and Q signals. Then, if there is a phase shift between the analytic signals, the preprocessing unit 120 aligns the phases of the analytic signals. As a result, the learning unit 130 can average the plurality of analytic signals in the direction of the number of signals while preventing attenuation of the plurality of analytic signals.

The learning unit 130 generates a model based on the phase-aligned multiple analytic signals (S13). More specifically, the learning unit 130 outputs the output data I′, Q ' get. The learning unit 130 updates the parameters (weight parameters) of the neural network based on the output data I' and Q' and the teacher data Iy and Qy (including the noise removal signal based on the input data).

Furthermore, in the embodiment of the present invention, it is assumed that the teacher data includes not only the noise-removed signals Iy and Qy but also the phase correct values Py corresponding to the noise-removed signals Iy and Qy. Then, the learning unit 130 further calculates a loss (loss between phases) based on the phase P′ corresponding to the output data I′ and Q′ and the phase correct value Py corresponding to the noise removal signals Iy and Qy. . Learning section 130 updates the parameters based on the I-signal loss, Q-signal loss, and phase-phase loss. In this way, the phase-to-phase loss is also reflected in the parameter update, so that noise reduction that can cope with the phase change is performed.

The learning unit 130 evaluates the accuracy of the neural network (model) after the parameters have been updated. Then, if the accuracy of the model is not sufficient ("No" in S14), learning unit 130 returns to S13, re-inputs the input data to the neural network, and determines parameters based on the output data and teacher data. Update again. On the other hand, when the accuracy of the model is sufficient (“Yes” in S14), the learning unit 130 outputs the model as a learned model to the model output unit 140, and the model output unit 140 outputs the model to the model output unit 140. The subsequent neural network is output as a trained model (S15).

(1-4. Configuration of noise removal signal generation device)
Next, functions of the noise removal signal generation device 20 according to the embodiment of the present invention will be described. FIG. 5 is a diagram showing a functional configuration example of the noise removal signal generation device 20 according to the embodiment of the present invention. As shown in FIG. 5, the noise-removed signal generator 20 according to the embodiment of the present invention comprises a data generation section 210, a preprocessing section 220, a noise processing section 230 and a data output section 240. FIG.

The noise elimination signal generation device 20 includes an arithmetic device such as a CPU (Central Processing Unit), and a program stored in a memory (not shown) is developed in a RAM (Random Access Memory) by the CPU and executed, thereby performing Functionality can be realized. At this time, a computer-readable recording medium recording the program may also be provided. Alternatively, the noise removal signal generation device 20 may be configured with dedicated hardware, or may be configured with a combination of multiple pieces of hardware.

The data generation unit 210 generates data by capturing the state of the observation target as a physical change. For example, the data generator 210 is configured by a predetermined sensor. Here, the type of sensor is not limited. For example, the sensor may be a fiber optic sensor or an acoustic sensor.

The preprocessing unit 220 performs various preprocessing on the data generated by the data generation unit 210 . More specifically, the preprocessor 220 shapes the data generated by the data generator 210 into an analytic signal. Thereby, the preprocessing unit 220 obtains a combination (analysis signal) of the I signal and the Q signal. For example, when the input signal is only the I signal, the preprocessing unit 220 converts the data generated by the data generation unit 210 into an analytic signal by Hilbert transform or the like.

More specifically, let the first analytic signals be Ix1 and Qx1, the second analytic signals be Ix2 and Qx2, . , each analytic signal is represented by the following equation (5).

Ix1(t)=cos(ωt+φ1+θi1)+zi1
Qx1(t)=sin(ωt+φ1+θq1)+zq1
Ix2(t)=cos(ωt+φ2+θi2)+zi2
Qx2(t)=sin(ωt+φ2+θq2)+zq2
・・・
Ixn(t)=cos(ωt+φn+θin)+zin
Qxn(t)=sin(ωt+φn+θqn)+zqn
... (5)

After that, the preprocessing unit 220 aligns the phases of the plurality of analytic signals when there is a phase shift between the plurality of analytic signals. As a result, the learning unit 130 can average the plurality of analytic signals in the direction of the number of signals while preventing attenuation of the plurality of analytic signals. The first analytic signals after the phases are aligned are I1 and Q1, the second analytic signals after the phases are aligned are I2 and Q2, . . . n after the phases are aligned When In and Qn are the second analytic signals, the analytic signals whose phases have been adjusted are expressed as in Equation (6).

I1(t)=cos(ωt)+zi1
Q1(t)=sin(ωt)+zq1
I2(t)=cos(ωt)+zi2
Q2(t)=sin(ωt)+zq2
・・・
In(t)=cos(ωt)+zin
Qn(t)=sin(ωt)+zqn
... (6)

Note that Equation (6) shows an example in which each phase is aligned with ωt. However, each phase may be adjusted to another value (reference phase). For example, each phase may be aligned to ωt+π/2 or ωt+φ1.

The noise processor 230 inputs the analytic signal obtained by the preprocessor 220 to the trained model. More specifically, the noise processing unit 230 inputs the analytic signals whose phases have been adjusted to the learned model. When the analytic signal is input to the trained model, output data corresponding to the analytic signal is output from the trained model. The noise processor 230 generates output data from the trained model as a noise removal signal.

The data output section 240 outputs the noise elimination signal generated by the noise processing section 230 . Here, the output destination of the noise elimination signal is not limited. In the embodiments of the present invention, it is mainly assumed that the data output unit 240 outputs the noise elimination signal to a predetermined recording medium, thereby writing the noise elimination signal to the recording medium. However, the data output unit 240 may output to an output destination according to the type of sensor. As an example, if the sensor type is an acoustic sensor, the noise elimination signal may be output to a speaker.

(1-5. Operation of noise removal signal generator)
Next, the operation of the noise elimination signal generation device 20 according to the embodiment of the present invention will be described. FIG. 6 is a flow chart showing an operation example of the noise removal signal generation device 20 according to the embodiment of the present invention. Since the configuration of the noise-removed signal generation device 20 has been described in detail as above, the description of the operation of the noise-removed signal generation device 20 here is simpler than the description of the configuration of the noise-removed signal generation device 20. change it.

In the noise removal signal generation device 20, the data generator 210 generates data by capturing the state of the observation target as a physical change (S21). The preprocessing unit 220 performs various preprocessing on the data generated by the data generation unit 210 (S22). More specifically, the preprocessor 220 shapes the data generated by the data generator 210 into an analytic signal. Thereby, the preprocessing unit 220 obtains a combination (analysis signal) of the I signal and the Q signal. Furthermore, if there is a phase shift between the I signal and the Q signal, the preprocessing unit 220 aligns the phases of the I signal and the Q signal.

The noise processor 230 inputs the analytic signal obtained by the preprocessor 220 to the trained model. More specifically, the noise processing unit 230 inputs the analytic signals whose phases have been adjusted to the learned model. When the analytic signal is input to the trained model, output data corresponding to the analytic signal is output from the trained model. The noise processor 230 generates output data from the trained model as a noise removal signal (S23). The data output unit 240 outputs the noise elimination signal (S24).

(2. Hardware configuration example)
Next, a hardware configuration example of the trained model generation device 10 according to the embodiment of the present invention will be described. However, the hardware configuration example of the noise removal signal generation device 20 according to the embodiment of the present invention can be similarly realized.

An example hardware configuration of the information processing device 900 will be described below as an example hardware configuration of the trained model generation device 10 according to the embodiment of the present invention. Note that the hardware configuration example of the information processing device 900 described below is merely an example of the hardware configuration of the trained model generation device 10 . Therefore, as for the hardware configuration of the trained model generation device 10, unnecessary configurations may be deleted from the hardware configuration of the information processing device 900 described below, or new configurations may be added.

FIG. 7 is a diagram showing the hardware configuration of an information processing device 900 as an example of the trained model generation device 10 according to the embodiment of the present invention. The information processing device 900 includes a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, a RAM (Random Access Memory) 903, a host bus 904, a bridge 905, an external bus 906, and an interface 907. , an input device 908 , an output device 909 , a storage device 910 and a communication device 911 .

The CPU 901 functions as an arithmetic processing device and a control device, and controls general operations within the information processing device 900 according to various programs. Alternatively, the CPU 901 may be a microprocessor. The ROM 902 stores programs, calculation parameters, and the like used by the CPU 901 . The RAM 903 temporarily stores programs used in the execution of the CPU 901, parameters that change as appropriate during the execution, and the like. These are interconnected by a host bus 904 comprising a CPU bus or the like.

The host bus 904 is connected via a bridge 905 to an external bus 906 such as a PCI (Peripheral Component Interconnect/Interface) bus. Note that the host bus 904, the bridge 905 and the external bus 906 do not necessarily have to be configured separately, and these functions may be implemented in one bus.

The input device 908 includes input means for the user to input information, such as a mouse, keyboard, touch panel, button, microphone, switch, and lever, and an input control circuit that generates an input signal based on the user's input and outputs it to the CPU 901 . etc. A user who operates the information processing apparatus 900 can input various data to the information processing apparatus 900 and instruct processing operations by operating the input device 908 .

The output device 909 includes, for example, a CRT (Cathode Ray Tube) display device, a liquid crystal display (LCD) device, an OLED (Organic Light Emitting Diode) device, a display device such as a lamp, and an audio output device such as a speaker.

The storage device 910 is a device for data storage. The storage device 910 may include a storage medium, a recording device that records data on the storage medium, a reading device that reads data from the storage medium, a deletion device that deletes data recorded on the storage medium, and the like. The storage device 910 is configured by, for example, an HDD (Hard Disk Drive). The storage device 910 drives a hard disk and stores programs executed by the CPU 901 and various data.

The communication device 911 is, for example, a communication interface configured with a communication device or the like for connecting to a network. Also, the communication device 911 may support either wireless communication or wired communication.

The hardware configuration example of the trained model generation device 10 according to the embodiment of the present invention has been described above.

(3. Summary)
As described above, according to the embodiment of the present invention, the output data output from the neural network based on inputting a plurality of analytic signals as input data to the neural network, and the noise based on the input data A learning unit 130 that updates the parameters of the neural network based on teacher data including a removal signal, and a model output unit 140 that outputs the neural network after the parameters are updated as a trained model. A finished model generator 10 is provided. According to such a configuration, noise reduction for signals can be performed with higher accuracy.

In addition, the output data and the teacher data each include an I signal and a Q signal. The parameters may be updated based on the signal-to-signal loss. At this time, the teacher data includes the correct value of the phase corresponding to the I signal and the Q signal of the teacher data, and the learning unit 130, based on the phase corresponding to the output data and the interphase loss based on the phase of the teacher data, Parameters may be updated.

According to such a configuration, since the parameters of the neural network are updated based on the I-signal loss, the Q-signal loss, and the phase-phase loss, the amplitude error and phase error of the observed signal can be corrected with higher accuracy. It will be done. In particular, even if the observed signal has a phase change, the phase change can be dealt with, and the observed signal can be corrected with higher accuracy.

Although the preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the present invention is not limited to such examples. It is obvious that a person having ordinary knowledge in the technical field to which the present invention belongs can conceive of various modifications or modifications within the scope of the technical idea described in the claims. It is understood that these also naturally belong to the technical scope of the present invention.

For example, embodiments of the present invention can have the following configuration.
(1)
Based on output data output from the neural network based on inputting a plurality of analytic signals to the neural network as input data, and teacher data including a noise removal signal based on the input data, the neural network a learning unit that updates network parameters;
a model output unit that outputs the neural network after the parameters have been updated as a trained model;
An information processing device.
(2)
The information processing device includes a preprocessing unit that aligns the phases of the plurality of analytic signals,
The learning unit inputs the plurality of analytic signals after the phases have been adjusted to the neural network as the input data.
The information processing device according to (1) above.
(3)
the neural network includes an encoder and a decoder;
The information processing apparatus according to (1) or (2).
(4)
each of the output data and the teacher data includes an I signal and a Q signal;
The learning unit updates the parameter based on an I-signal loss between I-signals of the output data and the teacher data and a Q-signal loss between Q-signals of the output data and the teacher data.
The information processing apparatus according to any one of (1) to (3).
(5)
the teacher data includes correct phase values corresponding to the I signal and the Q signal of the teacher data;
The learning unit updates the parameter based on the phase corresponding to the output data and the loss between phases based on the phase of the teacher data.
The information processing device according to (4) above.
(6)
The learning unit updates the parameter based on a weighted sum of the I-signal loss, the Q-signal loss, and the phase-to-phase loss.
The information processing device according to (5) above.
(7)
Based on output data output from the neural network based on inputting a plurality of analytic signals to the neural network as input data, and teacher data including a noise removal signal based on the input data, the neural network updating network parameters;
outputting the neural network after the parameters have been updated as a trained model;
A method of processing information, comprising:
(8)
the computer,
Based on output data output from the neural network based on inputting a plurality of analytic signals to the neural network as input data, and teacher data including a noise removal signal based on the input data, the neural network a learning unit that updates network parameters;
a model output unit that outputs the neural network after the parameters have been updated as a trained model;
A program for functioning as an information processing device comprising
(9)
Acquire the neural network after parameters are updated as a trained model based on the output data output from the neural network corresponding to the input data and the teacher data including the noise removal signal based on the input data. a noise processing unit that obtains, as a noise-removed signal, output data output from the trained model corresponding to the analytic signal;
a data output unit that outputs the noise removal signal;
A prediction device, comprising:
(10)
The prediction device includes a preprocessing unit that aligns the phases of the analytic signals,
The noise processing unit inputs the phase-matched analytic signal to the trained model.
The prediction device according to (9) above.
(11)
Acquire the neural network after parameters are updated as a trained model based on the output data output from the neural network corresponding to the input data and the teacher data including the noise removal signal based on the input data. and obtaining output data output from the trained model corresponding to the analytic signal as a noise removal signal;
outputting the noise removal signal;
A prediction method, comprising:
(12)
the computer,
Acquiring the neural network after parameters are updated as a trained model based on the output data output from the neural network corresponding to the input data and the teacher data including the noise removal signal based on the input data. a noise processing unit that obtains, as a noise-removed signal, output data output from the trained model corresponding to the analytic signal;
a data output unit that outputs the noise removal signal;
A program for functioning as a prediction device with

1 noise reduction system 10 learned model model generation device 110 data acquisition unit 120 preprocessing unit 130 learning unit 140 model output unit 20 noise removal signal generation device 210 data generation unit 220 preprocessing unit 230 noise processing unit 240 data output unit N1 encoder N2 decoder

Claims (7)

A plurality of analytic signals as input data, output data output from the neural network based on input to the neural network, a phase corresponding to the output data, a noise removal signal based on the input data , and a correct answer a learning unit that updates parameters of the neural network based on teacher data including phase ;
a model output unit that outputs the neural network after the parameters have been updated as a trained model;
An information processing device. The learning unit calculates an interphase loss based on the phase corresponding to the output data and the correct phase, and calculates an interphase loss based on the output data and the noise removal signal and the interphase loss. updating said parameters based on
The information processing device according to claim 1 . A plurality of analytic signals as input data, output data output from the neural network based on input to the neural network, a phase corresponding to the output data, a noise removal signal based on the input data , and a correct answer updating parameters of the neural network based on training data including phase ;
outputting the neural network after the parameters have been updated as a trained model;
A method of processing information, comprising: the computer,
A plurality of analytic signals as input data, output data output from the neural network based on input to the neural network, a phase corresponding to the output data, a noise removal signal based on the input data , and a correct answer a learning unit that updates parameters of the neural network based on teacher data including phase ;
a model output unit that outputs the neural network after the parameters have been updated as a trained model;
A program for functioning as an information processing device comprising Parameters are updated based on output data output from a neural network corresponding to input data, a phase corresponding to the output data, and teacher data including a noise removal signal based on the input data and a correct phase . a noise processing unit that obtains the neural network after being processed as a trained model, and obtains output data output from the trained model in response to an analytic signal as a noise removal signal;
a data output unit that outputs the noise removal signal;
A prediction device, comprising: Parameters are updated based on output data output from a neural network corresponding to input data, a phase corresponding to the output data, and teacher data including a noise removal signal based on the input data and a correct phase . obtaining the neural network after being processed as a trained model, and obtaining output data output from the trained model in response to the analytic signal as a noise removal signal;
outputting the noise removal signal;
A prediction method, comprising: the computer,
Parameters are updated based on output data output from a neural network corresponding to input data, a phase corresponding to the output data, and teacher data including a noise removal signal based on the input data and a correct phase . a noise processing unit that obtains the neural network after being processed as a trained model, and obtains output data output from the trained model in response to an analytic signal as a noise removal signal;
a data output unit that outputs the noise removal signal;
A program for functioning as a prediction device with

Images