Adaptive identification system, adaptive identification device, and adaptive identification method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing adaptive identification systems face reduced identification accuracy due to noise disturbances, and methods to improve accuracy, such as the approximate steepest descent algorithm, increase calculation complexity.
Innovation Solution
An adaptive identification system that generates an identification input signal with a frequency component that is an integer multiple of a fundamental frequency, satisfying a persistently exciting condition, and uses a moving average and diagonal matrix to update the adaptive filter coefficients, reducing calculation complexity while maintaining identification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the step size in LMS algorithm is reduced to avoid decrease in identification accuracy, then identification accuracy is maintained, but adaptation speed decreases and it takes time to obtain final identification result
Solution Approach 1:
The patent applies dynamics by making the step size adaptive rather than fixed. The step size is dynamically adjusted based on the eigenvalue distribution of the input signal's correlation matrix. When eigenvalues are concentrated, a larger step size is used to speed up adaptation; when eigenvalues are spread out, a smaller step size is used to maintain accuracy. This dynamic adjustment resolves the contradiction between identification accuracy and adaptation speed.
Solution Approach 2:
The patent changes the step size parameter based on the condition number of the input signal's correlation matrix. By calculating the eigenvalues and determining the concentration ratio, the system automatically adjusts the step size parameter to optimize both convergence speed and identification accuracy, resolving the trade-off between these two parameters.
2Loss of time
If an approximate steepest descent algorithm is used to reduce time to obtain final identification result, then adaptation speed increases, but calculation amount significantly increases compared with LMS algorithm
Solution Approach 1:
The patent changes the step size parameter based on the condition number of the input signal's correlation matrix. By calculating the eigenvalues and determining the concentration ratio, the system automatically adjusts the step size parameter to optimize both convergence speed and identification accuracy, resolving the trade-off between these two parameters.
3Measurement precision
If noise disturbance is present in the environment, then identification accuracy is reduced, but increasing algorithm complexity to compensate increases calculation amount
Solution Approach 1:
The patent uses feedback by continuously monitoring the eigenvalue distribution of the input signal's correlation matrix and adjusting the step size accordingly. This feedback mechanism allows the system to adapt to changing environmental conditions including noise, maintaining identification accuracy without requiring complex noise filtering algorithms.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
An adaptive identification system (110), for identifying a propagation system (11) characteristic by an adaptive filter (111), includes a signal generator that generates an identification input signal including a frequency component of an integer multiple of a fundamental frequency and having a periodicity satisfying a condition of being persistently exciting, a setting unit that sets moving average time to a fundamental period of the identification input signal, and an adaptive algorithm execution unit that uses a moving average value and a diagonal matrix to update a coefficient of the adaptive filter, the moving average value being obtained by calculating a moving average of a cross-correlation vector of a vector of the identification input signal and an observation signal with the moving average time, and the diagonal matrix being obtained by diagonalizing a matrix obtained by calculating a moving average of an autocorrelation matrix of the vector of the identification input signal with the moving average time.