RNN Sensor Noise Reduction in Actuator Controllers Without Phase Delay
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing controllers using low-pass filters to reduce noise in sensor signals suffer from incomplete noise reduction in the low-frequency band and phase delay in the high-frequency band, degrading control performance.
Innovation Solution
A controller equipped with a noise reduction processing unit utilizing a recurrent neural network trained to learn the correspondence between noisy and noise-reduced sensor signals, which includes a signal processing unit for differential processing and a control processing unit for actuator control, effectively reducing noise across all frequency bands and preventing phase delay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a low-pass filter is used to reduce noise in sensor signals, then noise in the high-frequency band is reduced, but noise in the low-frequency band cannot be completely reduced and phase delay occurs in the high-frequency band
Solution Approach 1:
The patent transforms the noise reduction approach by changing from traditional filter parameters (cutoff frequency, order) to neural network parameters (weights, biases, activation functions). The recurrent neural network learns optimal parameter configurations during training to simultaneously achieve noise reduction across all frequency bands while preserving phase accuracy, resolving the contradiction between noise reduction and phase precision.
Solution Approach 2:
The patent replaces the mechanical/physical filtering system (low-pass filter with fixed characteristics) with an intelligent computational system (recurrent neural network with adaptive learning). This substitution enables the system to dynamically adjust its noise reduction behavior based on learned patterns, achieving superior performance in both noise reduction and phase preservation compared to traditional mechanical filtering approaches.
2Object-affected harmful factors
If a low-pass filter is used to reduce noise, then high-frequency noise is attenuated, but control performance degrades due to incomplete noise reduction and phase delay
Solution Approach 1:
The recurrent neural network incorporates feedback mechanisms through its recurrent connections, allowing it to utilize temporal information from previous states in the sensor signal. This feedback capability enables the network to learn complex noise patterns across different frequencies and time scales, achieving comprehensive noise reduction while maintaining the phase relationships necessary for accurate control performance.
Solution Approach 2:
The neural network is trained in advance using labeled training data that represents various noise conditions and corresponding clean signals. This preliminary training action allows the network to pre-learn optimal noise reduction strategies and phase preservation techniques, so that during actual operation, the controller can directly apply these learned patterns to maintain high control performance while reducing noise across all frequency bands.
Data Source
AI summary
A controller includes: a noise reduction processing unit that acquires a sensor signal, which is based on an output from a sensor that detects time-series data and includes a noise, and that reduces the noise included in the sensor signal on the basis of a recurrent neural network trained so as to learn a correspondence relationship between a first signal including the noise corresponding to the sensor signal and a second signal indicating the first signal from which the noise has been removed; and a control processing unit that controls an actuator on the basis of an output from the noise reduction processing unit.