Kernel Adaptive Filtering for DUT Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional models for computer modeling of electronic devices struggle to accurately incorporate infinite order nonlinearity and memory effects with manageable computational complexity.
Innovation Solution
The use of kernel adaptive filtering, which enables modeling of electronic devices with nonlinearity of infinite order and incorporates memory effects through kernel algorithms like kernel recursive least squares (KRLS) and kernel least mean squares (KLMS), allowing for the generation of simulated responses to stimulus inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional models (Volterra series, memory polynomial, X-parameter) are used to model DUT, then the model can capture nonlinearity and memory effects, but the computational complexity becomes unmanageable when incorporating infinite order nonlinearity
Solution Approach 1:
The patent transforms the modeling approach by changing the parameter representation from conventional polynomial coefficients to kernel function parameters. This allows the model to capture infinite order nonlinearity through the kernel function's inherent properties while maintaining computational efficiency through the reduced parameter space required by the kernel-based representation.
Solution Approach 2:
The patent substitutes the traditional mechanical/mathematical polynomial-based modeling system with a kernel-based system that uses similarity functions. This substitution enables the model to handle infinite order nonlinearity more efficiently by leveraging the kernel trick, which maps inputs to a higher-dimensional feature space where nonlinear relationships become linearly separable and computationally tractable.
2Reliability
If conventional models incorporate memory effects, then the model can accurately represent DUT behavior, but the model structure becomes more complex and harder to manage
Solution Approach 1:
The patent merges the handling of nonlinearity and memory effects into a unified kernel-based framework. Instead of treating these as separate model components that increase complexity, the kernel function simultaneously captures both effects through its ability to represent temporal dependencies and nonlinear transformations in a single integrated structure.
Solution Approach 2:
The kernel function serves multiple functions simultaneously: it models nonlinear transformations, captures memory effects through temporal kernels, and provides a flexible framework that can adapt to different DUT characteristics. This multi-functionality reduces the need for separate model components and simplifies the overall model structure.
Data Source
AI summary
A method provides modeling a DUT and generating a simulated response. The method includes receiving a first portion of a stimulus signal generated by a signal generator, a second portion of the stimulus signal being input to the DUT; receiving a response signal output by the DUT in response to a second portion of the stimulus signal; digitizing the received first portion and the received response signal; correcting the digitized signals; measuring training input series data of the digitized first portion of the stimulus signal and training output series data of the digitized response signal; and utilizing kernel adaptive filtering for extracting a device model from the training input and output series data, and for generating simulated responses of the DUT to subsequent stimulus inputs, respectively. The kernel adaptive filtering may include a kernel least mean squares algorithm, a kernel Affine projection algorithm or a recursive least squares algorithm.