Method and apparatus for designing filter for signal processing

Abstract

The present disclosure provides an apparatus for designing a filter for signal processing, the apparatus for designing a filter comprising: a preprocessing unit for decomposing a test signal into a plurality of decomposed signals by using a preset signal decomposition algorithm; a filter configuration unit for configuring a neural network-based filter by training a neural network by having the test signal as an input and the plurality of decomposed signals as targets; and a filter adjustment unit for, when decomposing an actual signal through the neural network-based filter, performing optimization of the neural network-based filter by using reinforcement learning.

Description

Method and apparatus for designing a filter for signal processing

[0001] The present disclosure relates to a technique for designing an artificial intelligence-based filter used to decompose a signal.

[0002] In the field of signal processing, signals are decomposed using bandpass filters or predefined frequency bands. A bandpass filter is a type of frequency-selective filter that allows a frequency band between a defined lower frequency limit (f1) and an upper frequency limit (f2) to pass through. This method is effective when one wants to separate components of a specific frequency range from a signal. For example, it is used in applications such as separating only the human voice from a voice signal or extracting only data from a specific channel from a communication signal.

[0003] However, to effectively utilize a bandpass filter, the frequency band of the signal to be separated must be accurately known in advance. This can be problematic when the characteristics of the signal are not well known or when dealing with complex signals containing various frequency components. Furthermore, because the signal is decomposed based on a fixed frequency band, it may be difficult to respond to dynamic changes in the signal or the appearance of unforeseen frequency components. In other words, if the signal characteristics change, it becomes necessary to select a new frequency band and design a new filter. Additionally, for signals containing various frequency components that change over time, it is difficult to extract sufficient information using a single frequency band filter. This can be particularly challenging when analyzing signals arising from natural phenomena or complex signals originating from various sources.

[0004] For example, it is difficult to predefine clear frequency bands for signals generated in environments where various equipment coexists, such as vibration or sound signals within a factory. Consequently, existing signal decomposition methods may have limitations in such environments. Accordingly, there is a need for specific measures that can decompose signals more consistently and efficiently, even in environments where diverse equipment is mixed, such as factories.

[0005] The present embodiments aim to provide a method and apparatus for designing a filter for signal processing that can provide high flexibility and adaptability to signals of various domains.

[0006] An embodiment of the present disclosure, derived to solve the aforementioned problem, may provide a filter design device for designing a filter for signal processing, comprising: a preprocessing unit that decomposes a test signal into a plurality of decomposed signals using a preset signal decomposition algorithm; a filter configuration unit that takes the test signal as input and targets the plurality of decomposed signals to train a neural network to configure a neural network-based filter; and a filter adjustment unit that performs optimization of the neural network-based filter using reinforcement learning when decomposing an actual signal through the neural network-based filter.

[0007] In addition, one embodiment may provide a filter design method for designing a filter for signal processing, comprising the steps of: decomposing a test signal into a plurality of decomposed signals using a preset signal decomposition algorithm; training a neural network with the test signal as input and the plurality of decomposed signals as targets to construct a neural network-based filter; and, when decomposing an actual signal through the neural network-based filter, performing optimization of the neural network-based filter using reinforcement learning.

[0008] According to the present embodiment, a method and apparatus for designing a filter for signal processing capable of providing high flexibility and adaptability to signals of various domains can be provided.

[0009] In addition, by selecting the optimal combination of signal decomposition filters through reinforcement learning, the performance of a prediction model based on the decomposed signal can be optimized.

[0010] In addition, it is possible to improve the performance of filters used in signal processing, increase the accuracy of analysis, and secure the possibility of application in various environments.

[0011] FIG. 1 is a diagram illustrating the configuration of a filter design device for signal processing according to one embodiment.

[0012] FIG. 2 is a diagram illustrating the operation of a filter design device for signal processing according to one embodiment.

[0013] FIGS. 3 to 5 are drawings for illustrating specific examples of a filter design device for signal processing according to one embodiment.

[0014] FIG. 6 is a diagram illustrating the procedure of a filter design method for signal processing according to one embodiment.

[0015] FIG. 7 is a diagram illustrating a specific procedure for a filter design method for signal processing according to one embodiment.

[0016] Hereinafter, some embodiments of the present disclosure will be described in detail with reference to the exemplary drawings. In assigning reference numerals to the components of each drawing, the same components may have the same reference numeral as much as possible, even if they are shown in different drawings. Furthermore, in describing the embodiments, if it is determined that a detailed description of related known components or functions may obscure the essence of the technical concept, such detailed description may be omitted. Where terms such as "comprising," "having," or "consisting of" are used in this specification, other parts may be added unless "only" is used. Where a component is expressed in the singular, it may include a plural unless otherwise specified.

[0017] Additionally, terms such as first, second, A, B, (a), (b), etc., may be used to describe the components of the present disclosure. These terms are used merely to distinguish the components from other components, and the nature, order, sequence, or number of the components are not limited by such terms.

[0018] In describing the positional relationship of components, where it is stated that two or more components are "connected," "combined," or "joined," it should be understood that while the two or more components may be directly "connected," "combined," or "joined," they may also be "connected," "combined," or "joined" with other components "intervened." Here, the other components may be included in one or more of the two or more components that are "connected," "combined," or "joined" with one another.

[0019] In describing the temporal flow relationship regarding components, methods of operation, or methods of production, for example, when the temporal or sequential relationship is described using "after," "following," "next," or "before," it may include cases where the relationship is not continuous unless "immediately" or "directly" is used.

[0020] Meanwhile, where numerical values ​​or corresponding information regarding a component (e.g., levels, etc.) are mentioned, even without separate explicit notation, the numerical values ​​or corresponding information may be interpreted as including a range of error that may occur due to various factors (e.g., process factors, internal or external shocks, noise, etc.).

[0021] Conventional signal decomposition methods using bandpass filters have limitations because it is difficult to predefine clear frequency bands for vibration or sound signals generated in environments where various equipment coexists, such as factories.

[0022] Empirical mode decomposition (EMD) is proposed as one method to overcome the limitations of such bandpass filters. Empirical mode decomposition is a method that decomposes a signal into multiple intrinsic mode functions (IMFs) based on the signal's local frequency characteristics. This can be usefully applied when the signal exhibits nonlinear and anomalous characteristics that vary over time. By utilizing empirical mode decomposition, various frequency components that naturally appear in a signal can be flexibly decomposed according to the signal's dynamic changes. Since this does not rely on a fixed frequency band, it allows for the identification and analysis of the signal's intrinsic characteristics without prior knowledge regarding the frequency band.

[0023] However, in the case of empirical mode decomposition, the results can vary significantly depending on the original signal. This means that even signals within the same domain can be decomposed into different frequency bands. Consequently, consistent analysis or filtering may become problematic. Furthermore, the empirical mode decomposition process may involve various heuristics and subjective decisions. For example, it may include decisions regarding whether a specific eigenmode function should be regarded as noise or as a significant signal component. This can reduce the reproducibility and consistency of the analysis.

[0024] The present disclosure proposes a method and apparatus for automatically designing signal filters based on reinforcement learning and empirical mode decomposition for signal processing. Accordingly, signals can be effectively decomposed without information regarding defined frequency bands. Furthermore, it overcomes the disadvantages of empirical mode decomposition, provides a consistent decomposition method for any signal, and enables data analysis by presenting an optimized filter combination through the application of reinforcement learning to the decomposed signal.

[0025] Hereinafter, a filter design apparatus for signal processing according to embodiments of the present disclosure will be described in detail with reference to the relevant drawings.

[0026] FIG. 1 is a diagram illustrating the configuration of a filter design device for signal processing according to one embodiment. FIG. 2 is a diagram illustrating the operation of a filter design device for signal processing according to one embodiment.

[0027] Referring to FIG. 1, a filter design device (100) for signal processing may include a preprocessing unit (110) that decomposes a test signal into a plurality of decomposed signals using a preset signal decomposition algorithm, a filter configuration unit (120) that takes a test signal as input and targets a plurality of decomposed signals to train a neural network to form a neural network-based filter, and a filter adjustment unit (130) that performs optimization of the neural network-based filter using reinforcement learning when decomposing an actual signal through the neural network-based filter.

[0028] The preprocessing unit (110) can receive a test signal. Depending on the example, the test signal may be collected from various environments. For example, vibration signals or sound signals generated from various equipment, workers, or work vehicles in factories or steel mills may be collected as test signals. The preprocessing unit (110) may receive a test signal from an external device or read a test signal stored in a separately provided storage device.

[0029] The preprocessing unit (110) can take a test signal as a raw signal and decompose it into a plurality of decomposed signals using a predetermined signal decomposition algorithm. That is, the preprocessing unit (110) can determine the plurality of decomposed signals as outputs. Specifically, the preprocessing unit (110) can decompose the test signal, which is the raw signal, into a plurality of decomposed signals (210), as shown in FIG. 2.

[0030] According to one example, the signal decomposition algorithm may include an empirical mode decomposition (EMD) algorithm. Additionally, the signal decomposition algorithm may include multivariate empirical mode decomposition (MEMD) which performs mode decomposition on a multivariate signal, noise-assisted multivariate empirical mode decomposition (NA-MEMD) which performs mode decomposition by adding noise during signal processing, partially noise-assisted multivariate empirical mode decomposition (PNA-MEMD) which performs mode decomposition by adding noise only to specific parts, and signal-assisted multivariate empirical mode decomposition (SA-MEMD) which performs mode decomposition using additional signals. However, this is merely an example, and other known signal decomposition algorithms may be applied if the technical concept according to the present disclosure can be applied.

[0031] When an empirical mode decomposition algorithm is applied as a signal decomposition algorithm, multiple decomposed signals corresponding to the output may include intrinsic mode functions (IMFs). An intrinsic mode function consists of a single frequency component and may be a separation of various frequency components of the original signal.

[0032] In addition, if the test signal is a multivariate signal, that is, if the test signal is composed of multiple channels according to one example, the multiple decomposed signals may each be composed of the same number of channels as the test signal. For example, if the original signal is composed of 12 channels, the multiple eigenmode functions decomposed by the multivariate empirical mode decomposition algorithm may each be composed of 12 channels.

[0033] The filter component (120) can take a test signal as input and target multiple decomposition signals to enable a neural network to learn and simulate a signal decomposition process according to a signal decomposition algorithm. That is, if the signal decomposition algorithm is an empirical mode decomposition algorithm, the filter component (120) can enable a neural network to learn and simulate an empirical mode decomposition process targeting multiple decomposition signals with a test signal as input. Specifically, as illustrated in FIG. 2, the filter component (120) can simulate a filter that performs empirical mode decomposition using a CNN learning algorithm (220). However, the CNN learning algorithm is not limited to a specific algorithm as long as the technical concept of the present disclosure can be applied as an example.

[0034] Accordingly, the filter configuration unit (120) can configure a neural network-based filter by training a neural network. The neural network-based filter can be configured to perform the operation of decomposing an input signal into multiple decomposed signals by applying a signal decomposition algorithm to the input signal. That is, the neural network-based filter can implement signal decomposition logic by mimicking empirical mode decomposition. In this case, the filter configuration unit (120) can configure the neural network-based filter so that it is possible to reconstruct a test signal using multiple decomposed signals.

[0035] Subsequently, an actual signal may be input. In one example, the actual signal may be collected and input from a location such as a factory. When the filter adjustment unit (130) decomposes the actual signal through a neural network-based filter, it may perform optimization of the neural network-based filter using a reinforcement learning algorithm. In one example, the reinforcement learning algorithm used to perform optimization may be configured based on a known Deep Q-Network (DQN) algorithm. However, this is merely an example, and as long as the technical concept of the present disclosure can be applied, the algorithm used for reinforcement learning is not limited to a specific algorithm.

[0036] The filter adjustment unit (130) can determine a reward based on the result of adjusting the parameters of the neural network-based filter when the actual signal is decomposed through the neural network-based filter, and can perform optimization of the neural network-based filter using the reward. Specifically, as shown in FIG. 2, the filter adjustment unit (130) can act as an agent of reinforcement learning to perform model-based prediction and select an optimal combination of filters by reflecting the reward based on the result.

[0037] For example, when the DQN algorithm is applied, state s can be composed of the current state of the signal or the parameters of the neural network-based filter. Additionally, action a can be composed of a change to the parameters of the neural network-based filter. The reward function can be composed of criteria for evaluating the performance of the neural network-based filter. For example, the accuracy of signal decomposition, the signal-to-noise ratio (SNR), or the quality of the decomposed modes can be used as the reward.

[0038] The filter adjustment unit (130) can initialize Q values ​​for all state-actions. Subsequently, the filter adjustment unit (130) can select an action that changes at least one parameter among the current parameters of the neural network-based filter. The filter adjustment unit (130) can use a Convolutional Neural Network (CNN) to take a state as input, approximate the Q function values ​​for various actions in the current state, and select the optimal action. According to one example, the filter adjustment unit (130) can select action a using an exploration or exploitation strategy in the current state s. In this case, the filter adjustment unit (130) can select the optimal action based on an epsilon-greed (ε-greedy) policy. In the exploration phase, the filter adjustment unit (130) can randomly select an action according to the probability ε. That is, the filter adjustment unit (130) can randomly change at least one parameter among the current parameters of the neural network-based filter. The filter adjustment unit (130) can select the optimal action based on the Q-value learned so far according to the 1-ε probability during the utilization stage. That is, the filter adjustment unit (130) can select the action with the highest Q value. According to one example, the probability ε value can be set to gradually lower as learning progresses. That is, the probability ε value can be set high at the beginning to select various combinations, and then the probability ε value can be set low to strengthen the utilization stage.

[0039] When the filter adjustment unit (130) performs action a, it transitions to a new state s′ and receives a reward r. That is, when a change in the parameters of the neural network-based filter is performed, the neural network-based filter is configured with the changed parameters, and decomposition of the actual signal can be performed through the newly configured neural network-based filter. When a reward is obtained according to a reward function for the decomposition signals generated according to the decomposition, it can be reflected in the existing Q value. In this case, the known Bellman equation or Q-learning equation can be used for updating the Q value. For example, if the reward function is configured as SNR, the SNR of the decomposed signal is determined, and the higher the determined SNR, the higher the reward can be obtained.

[0040] The filter adjustment unit (130) can perform optimization of the Q value by repeatedly performing the update process of the Q value for various states and actions. Optimization of the Q value involves learning the Q value to select the optimal action in a given environment, and when the Q value is optimized, the action having the highest Q value in each state can be selected. That is, the action that maximizes the total reward expected in that state can be selected. For example, the filter adjustment unit (130) can maximize the SNR of the decomposed signals by performing parameter adjustment of a neural network-based filter corresponding to the optimal action in the current state of the signal.

[0041] According to this, a method and apparatus for designing a signal processing filter capable of providing high flexibility and adaptability for signals in various domains can be provided. Furthermore, by selecting the optimal combination of signal decomposition filters through reinforcement learning, the performance of a prediction model based on the decomposed signal can be optimized. In addition, it is possible to improve the performance of filters used in signal processing, increase the accuracy of analysis, and ensure the possibility of application in various environments.

[0042] FIGS. 3 to 5 are drawings for illustrating specific examples of a filter design device for signal processing according to one embodiment.

[0043] Referring to FIG. 3, an operation for training a Deep-EMD model is illustrated according to one example. Deep-EMD can utilize the capabilities of a deep neural network for time series data analysis. To this end, an input signal (a) consisting of 12 channels can be decomposed into 8 eigenmode functions (b), each consisting of 12 channels, using the MEMD algorithm (310).

[0044] The Deep-EMD model (320) can take an input signal (a) consisting of 12 channels as input and target decomposed eigenmode functions (b) to simulate and learn the signal decomposition process according to the MEMD algorithm. That is, by creating a model that reconstructs the original signal through eigenmode functions generated through empirical mode decomposition, the consistency of signal decomposition can be improved. Once the neural network is trained, it can perform decomposition in a consistent manner for various signals within the same domain. Accordingly, the subjectivity and consistency problems of the empirical mode decomposition method can be significantly mitigated.

[0045] Referring to FIG. 4, the trained Deep-EMD model (320) can decompose a real input signal (c) consisting of 12 channels into eight eigenmode functions (d), each consisting of 12 channels. In this case, according to one example, the Deep-EMD model (320) can be configured with a structure as illustrated in FIG. 5. A separate Deep-EMD model can be configured for each eigenmode function. The first Deep-EMD model is configured to use the original input signal x(t), and subsequent models can be configured to use the eigenmode function output from the Deep-EMD model, denoted as DIMFn(t), and the residual signal, denoted as rn(t). This can represent the difference between rn(t) and DIMFn(t). By integrating all DIMFs and the residual signal, the original input signal can be reconstructed. The learning approach using a neural network can minimize subjective decisions during the decomposition process. Because the model learns directly from the data, it can automatically determine which eigenmode functions are important, which can improve the objectivity and reproducibility of the analysis. In addition, once the neural network model is trained, the computational cost required to perform signal decomposition is relatively low, which may be more suitable for processing large datasets and real-time signal processing requirements.

[0046] According to one example, the internal block hierarchy of each Deep-EMD model may be configured as shown in Table 1 below. However, this is merely an example, and if the technical concept of the present disclosure can be applied, it may be configured with a different hierarchy.

[0047] LayerStructureFilters / Neurons1Fully-connected + LeakyReLU (0.3)5122Bi-LSTM + LeakyReLU (0.3)2563Dropout (0.3)-4Conv + LeakyReLU (0.3)645Conv + LeakyReLU (0.3)326Conv + LeakyReLU (0.3)167Conv + PReLU88Fully-connected1

[0048] Each Deep-EMD model can share the same structure across all layers, as shown in Table 1. Fully-connected layers are fully connected, where every input node can be connected to every output node. Bi-LSTM layers are integrated to capture the temporal characteristics of the signal. One-dimensional convolutional layers can be utilized to identify the shape features of time-series signals. Leaky ReLU can be used between each layer, excluding the final layer, to prevent the exclusion of negative regions during the training process. Dropout can randomly disable some neurons. In the final layer of the model, a single-neuron fully connected layer can be used to generate a single-channel output signal. Parameterized ReLU (PReLU) can be used before the final layer to fine-tune the amplitude of the output signal. Subsequently, the corresponding input signals and decomposed signals can be used to optimize Q-values ​​through reinforcement learning. Reinforcement learning can automatically select the optimal filter combination during the signal decomposition and processing. This allows for maximizing the efficiency of signal decomposition and fine-tuning the system for specific signal processing objectives (e.g., optimizing classification performance).

[0049] Hereinafter, a method for designing a filter for signal processing capable of performing some or all of the embodiments described with reference to FIGS. 1 to 5 will be described with reference to the drawings. The foregoing description may be omitted to avoid redundant descriptions, and in such cases, the omitted content may be applied substantially identically to the following description, provided that it does not contradict the technical concept of the invention.

[0050] FIG. 6 is a diagram illustrating the procedure of a filter design method for signal processing according to one embodiment. FIG. 7 is a diagram illustrating the specific procedure of a filter design method for signal processing according to one embodiment.

[0051] Referring to FIG. 6, the filter design device can decompose a test signal into a plurality of decomposed signals using a preset signal decomposition algorithm (S610).

[0052] The filter design device can receive a test signal. Depending on the example, the test signal may be collected from various environments. For instance, vibration signals or sound signals generated from various equipment, workers, or work vehicles in factories or steel mills may be collected as test signals. The filter design device can receive the test signal from an external device or read the test signal stored in a separately provided storage device.

[0053] A filter design device can take a test signal as a raw signal and decompose it into multiple decomposed signals using a predetermined signal decomposition algorithm. That is, the filter design device can determine the multiple decomposed signals as outputs. Specifically, as illustrated in FIG. 7, the filter design device can decompose the test signal, which is the raw signal, into multiple decomposed signals (IMF) (S710).

[0054] According to one example, the signal decomposition algorithm may include an empirical mode decomposition (EMD) algorithm. Additionally, the signal decomposition algorithm may include multivariate empirical mode decomposition (MEMD), noise-assisted multivariate empirical mode decomposition (NA-MEMD), partial noise-assisted multivariate empirical mode decomposition (PNA-MEMD), and signal-assisted multivariate empirical mode decomposition (SA-MEMD), etc. However, this is merely an example, and other known signal decomposition algorithms may be applied if the technical concept according to the present disclosure can be applied.

[0055] When an empirical mode decomposition algorithm is applied as a signal decomposition algorithm, multiple decomposed signals corresponding to the output may include intrinsic mode functions (IMFs). An intrinsic mode function consists of a single frequency component and may be a separation of various frequency components of the original signal.

[0056] In addition, if the test signal is a multivariate signal, that is, if, for example, the test signal is composed of multiple channels, the multiple decomposition signals can each be composed of the same number of channels as the test signal.

[0057] Referring again to FIG. 6, the filter design device can configure a neural network-based filter by training a neural network with a test signal as input and a plurality of decomposition signals as targets (S620).

[0058] The filter design device can take a test signal as input and target multiple decomposition signals, and enable a neural network to learn and simulate a signal decomposition process according to a signal decomposition algorithm. That is, if the signal decomposition algorithm is an empirical mode decomposition algorithm, the filter design device can enable a neural network to take a test signal as input and target multiple decomposition signals to learn and simulate an empirical mode decomposition process. Specifically, as illustrated in FIG. 7, the filter design device can generate a simulate filter that performs empirical mode decomposition (S720).

[0059] Accordingly, the filter design device can configure a neural network-based filter by training a neural network. The neural network-based filter can be configured to perform the operation of decomposing an input signal into multiple decomposed signals by applying a signal decomposition algorithm to the input signal. That is, the neural network-based filter can implement signal decomposition logic by mimicking empirical mode decomposition. In this case, the filter design device can configure the neural network-based filter to enable the reconstruction of a test signal using multiple decomposed signals.

[0060] Referring again to FIG. 6, when the filter design device decomposes an actual signal through a neural network-based filter, it can perform optimization of the neural network-based filter using reinforcement learning (S630).

[0061] Subsequently, an actual signal may be input. In one example, the actual signal may be collected and input from a location such as a factory. When the filter design device decomposes the actual signal through a neural network-based filter, it may perform optimization of the neural network-based filter using a reinforcement learning algorithm. In one example, the reinforcement learning algorithm may include a Deep Q-Network (DQN) algorithm. However, this is merely an example, and the algorithm is not limited to a specific algorithm as long as the technical concept of the present disclosure can be applied.

[0062] When a real signal is decomposed through a neural network-based filter, the filter design device determines a reward based on the result of adjusting the parameters of the neural network-based filter and can perform optimization of the neural network-based filter using the reward. Specifically, as illustrated in FIG. 7, the filter design device can decompose a signal through a neural network-based simulation filter (S730). That is, the filter design device can decompose the original signal and signals of the same domain using a learned neural network model. Additionally, the filter design device can perform optimization by performing model-based prediction on the decomposed signals (S740). That is, the filter design device can select the optimal filter combination for a new signal through a reinforcement learning model and optimize the performance of the prediction model based on this. Additionally, the filter design device can select the optimal combination of filters by reflecting the reward based on the result (S750). That is, the filter design device can try various filter combinations and find the optimal combination. In this case, the combination can decompose the original signal most effectively and optimize the performance of the prediction model.

[0063] For example, when the DQN algorithm is applied, state s can be composed of the current state of the signal or the parameters of the neural network-based filter. Additionally, action a can be composed of a change to the parameters of the neural network-based filter. The reward function can be composed of criteria for evaluating the performance of the neural network-based filter. For example, the accuracy of signal decomposition, the signal-to-noise ratio (SNR), or the quality of the decomposed modes can be used as the reward.

[0064] The filter design unit can initialize Q-values ​​for all state-action pairs. Subsequently, the filter design unit can select an action that modifies at least one parameter among the current parameters of the neural network-based filter. The filter design unit can use a Convolutional Neural Network (CNN) to take the state as input, approximate the Q-function values ​​for various actions from the current state, and select the optimal action. For example, the filter design unit can select action a from the current state s using an exploration or exploitation strategy. In this case, the filter design unit can select the optimal action based on an epsilon-greed (ε-greedy) policy. In the exploration phase, the filter design unit can randomly select an action according to probability ε. That is, the filter design unit can randomly modify at least one parameter among the current parameters of the neural network-based filter. In the exploitation phase, the filter design unit can select the optimal action based on the Q-values ​​learned so far according to probability 1-ε. That is, the filter design unit can select the action with the highest Q-value. For example, the probability ε value can be set to gradually lower as learning progresses. That is, the probability ε value can be set high in the beginning to select various combinations, and then the probability ε value can be set low to strengthen the utilization stage.

[0065] When the filter design device performs action a, it transitions to a new state s′ and receives a reward r. That is, when the parameters of a neural network-based filter are modified, the neural network-based filter is configured with the modified parameters, and decomposition of the actual signal can be performed through the newly configured neural network-based filter. When a reward is obtained according to a reward function for the decomposed signals generated by the decomposition, it can be reflected in the existing Q value. In this case, the known Bellman equation or Q-learning equation can be used to update the Q value. For example, if the reward function is configured as SNR, the SNR of the decomposed signal is determined, and a higher reward can be obtained as the determined SNR increases.

[0066] The filter design device can perform optimization of the Q value by repeatedly performing the Q value update process for various states and actions. Optimization of the Q value involves learning the Q value to select the optimal action in a given environment, and when the Q value is optimized, the action having the highest Q value in each state can be selected. That is, the action that maximizes the total reward expected in that state can be selected. For example, the filter adjustment unit (130) can maximize the SNR of the decomposed signals by performing parameter adjustment of the neural network-based filter corresponding to the optimal action in the current state of the signal.

[0067] According to this, a method and apparatus for designing a signal processing filter capable of providing high flexibility and adaptability for signals in various domains can be provided. Furthermore, by selecting the optimal combination of signal decomposition filters through reinforcement learning, the performance of a prediction model based on the decomposed signal can be optimized. In addition, it is possible to improve the performance of filters used in signal processing, increase the accuracy of analysis, and ensure the possibility of application in various environments.

[0068] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0069] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention.

[0070] The methods and / or various embodiments described above may be realized in digital electronic circuits, computer hardware, firmware, software, and / or combinations thereof. Various embodiments of the present disclosure may be executed by a data processing device, for example, one or more programmable processors and / or one or more computing devices, or may be implemented as a computer program stored on a computer-readable recording medium and / or a computer program stored on a computer-readable recording medium. The computer program described above may be written in any form of programming language, including a compiled language or an interpreted language, and may be distributed in any form, such as a standalone program, a module, a subroutine, etc. The computer program may be distributed through a single computing device, a plurality of computing devices connected through the same network, and / or a plurality of computing devices distributed to be connected through a plurality of different networks.

[0071] The above-described methods and / or various embodiments may be performed by one or more processors configured to execute one or more computer programs that process, store, and / or manage any functions, functions, etc. by operating based on input data or generating output data. For example, the methods and / or various embodiments of the present disclosure may be performed by special-purpose logic circuits such as a Field Programmable Gate Array (FPGA) or an Application Specific Integrated Circuit (ASIC), and an apparatus and / or system for performing the methods and / or embodiments of the present disclosure may be implemented as a special-purpose logic circuit such as an FPGA or an ASIC.

[0072] One or more processors executing a computer program may include one or more processors of a general-purpose or special-purpose microprocessor and / or any type of digital computing device. The processor may receive instructions and / or data from each of read-only memory and random access memory, or receive instructions and / or data from read-only memory and random access memory. In the present disclosure, components of a computing device performing the methods and / or embodiments may include one or more processors for executing instructions and one or more memory devices for storing instructions and / or data.

[0073] According to one embodiment, a computing device may exchange data with one or more mass storage devices for storing data. For example, the computing device may receive and / or receive data from a magnetic disc or an optical disc, and may transfer data to a magnetic disc or an optical disc. A computer-readable storage medium suitable for storing instructions and / or data associated with a computer program may include, but is not limited to, any form of non-volatile memory including semiconductor memory devices such as EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable PROM), and flash memory devices. For example, a computer-readable storage medium may include magnetic discs such as internal hard disks or removable disks, optical magnetic discs, CD-ROMs, and DVD-ROMs.

[0074] To provide interaction with a user, the computing device may include, but is not limited to, a display device for providing or displaying information to the user (e.g., CRT (Cathode Ray Tube), LCD (Liquid Crystal Display), etc.) and a pointing device (e.g., keyboard, mouse, trackball, etc.) on which the user can provide input and / or commands, etc. on the computing device. That is, the computing device may further include any other type of device for providing interaction with the user. For example, the computing device may provide any form of sensory feedback to the user for interaction with the user, including visual feedback, auditory feedback and / or tactile feedback. In this regard, the user may provide input to the computing device through various gestures such as visual, vocal, and motion.

[0075] In the present disclosure, various embodiments may be implemented in a computing system comprising backend components (e.g., data servers), middleware components (e.g., application servers), and / or frontend components. In this case, the components may be interconnected by any form or medium of digital data communication, such as a communication network. For example, the communication network may include a Local Area Network (LAN), a Wide Area Network (WAN), etc.

[0076] A computing device based on the embodiments thereof may be implemented using hardware and / or software configured to interact with a user, including a user device, a user interface (UI) device, a user terminal, or a client device. For example, the computing device may include a portable computing device such as a laptop computer. Additionally or alternatively, the computing device may include, but is not limited to, Personal Digital Assistants (PDAs), tablet PCs, game consoles, wearable devices, Internet of Things (IoT) devices, Virtual Reality (VR) devices, Augmented Reality (AR) devices, etc. The computing device may further include other types of devices configured to interact with a user. Furthermore, the computing device may include a portable communication device suitable for wireless communication through a network such as a mobile communication network (e.g., a mobile phone, a smartphone, a wireless cellular phone, etc.). A computing device may be configured to communicate wirelessly with a network server using wireless communication technologies and / or protocols such as radio frequency (RF), microwave frequency (MWF) and / or infrared frequency (IRF).

[0077] The foregoing description is merely an illustrative explanation of the technical concept of the present disclosure, and those skilled in the art to which the present disclosure pertains may make various modifications and variations within the scope of the essential characteristics of the technical concept. Furthermore, since these embodiments are intended to explain, not limit, the scope of the technical concept is not limited by these embodiments. The scope of protection of the present disclosure shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present disclosure.

[0078]

[0079] CROSS-REFERENCE TO RELATED APPLICATION

[0080] This patent application claims priority pursuant to Section 119(a) of the U.S. Patent Act (35 USC § 119(a)) to Korean Patent Application No. 10-2024-0151298 filed on October 30, 2024, all of which are incorporated by reference into this patent application. Furthermore, this patent application claims priority in countries other than the United States for the same reasons as above, all of which are incorporated by reference into this patent application.

Claims

1. In an apparatus for designing a filter for signal processing, A preprocessing unit that decomposes a test signal into multiple decomposition signals using a preset signal decomposition algorithm; A filter component that configures a neural network-based filter by training a neural network using the above test signal as input and targeting the above plurality of decomposition signals; and A filter design device comprising a filter adjustment unit that performs optimization of the neural network-based filter using reinforcement learning when decomposing an actual signal through the neural network-based filter.

2. In Paragraph 1, The above signal decomposition algorithm includes an empirical mode decomposition (EMD) algorithm, and The above plurality of decomposition signals are filter design devices including intrinsic mode functions (IMFs).

3. In Paragraph 1, The filter design device comprising the above filter component, which configures the neural network-based filter to enable the reconstruction of the test signal using the plurality of decomposition signals.

4. In Paragraph 1, The above test signal is composed of a plurality of channels, and The above plurality of decomposition signals are each configured with the same number of channels as the test signal, forming a filter design device.

5. In Paragraph 1, The filter adjustment unit is a filter design device that determines a compensation based on the result of adjusting the parameters of the neural network-based filter when the actual signal is decomposed through the neural network-based filter, and performs optimization of the neural network-based filter using the compensation.

6. In Paragraph 5, The above compensation is a filter design device determined based on at least one of the accuracy of signal decomposition, the signal-to-noise ratio (SNR), or the quality of the decomposed mode.

7. In a method for designing a filter for signal processing, A step of decomposing a test signal into multiple decomposed signals using a preset signal decomposition algorithm; A step of configuring a neural network-based filter by training a neural network using the above test signal as input and targeting the above plurality of decomposition signals; and A filter design method comprising the step of optimizing the neural network-based filter using reinforcement learning when decomposing an actual signal through the neural network-based filter.

8. In Paragraph 7, The above signal decomposition algorithm includes an empirical mode decomposition (EMD) algorithm, and The above plurality of decomposition signals are a filter design method including intrinsic mode functions (IMFs).

9. In Paragraph 7, A filter design method in which the step of configuring the above filter is to configure the neural network-based filter so as to enable the reconstruction of the test signal using the plurality of decomposition signals.

10. In Paragraph 7, The above test signal is composed of a plurality of channels, and A filter design method in which the above plurality of decomposition signals are each composed of the same number of channels as the above test signal.

11. In Paragraph 7, A filter design method comprising the step of performing optimization of the above filter, wherein when the actual signal is decomposed through the neural network-based filter, a reward is determined based on the result of adjusting the parameters of the neural network-based filter, and optimization of the neural network-based filter is performed using the reward.

12. In Paragraph 11, The above compensation is a filter design method determined based on at least one of the accuracy of signal decomposition, the signal-to-noise ratio (SNR), or the quality of the decomposed mode.

Images