Memristive Signal Classification for Noisy Low-Amplitude Inputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification technologies for electric signals, particularly neuro-signals and cardio-signals, face challenges with high noise levels and low amplitudes, leading to information loss when digitalized, and require signal conversion, which is inefficient and time-consuming.
Innovation Solution
A feature extraction device utilizing a memristive structure for analog data processing, allowing non-linear current output based on input and reference signals, enabling high accuracy and recognition rates without digital conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If electric signals are digitalized for classification, then processing can be performed using standard digital systems, but information loss occurs particularly for low amplitude signals with high noise levels
Solution Approach 1:
The patent replaces digital signal processing systems with an analog neural network system that processes electric signals in their native analog form. The neural network comprises analog computing elements (operational amplifiers, resistors, capacitors) that perform classification operations directly on the continuous electric signal without requiring digitization, thereby preserving all signal information including subtle variations in low amplitude signals.
2Adaptability or versatility
If signal conversion is performed for classification, then signals can be processed by digital systems, but the process is inefficient and time-consuming
Solution Approach 1:
The patent eliminates the signal conversion step by substituting digital processing with analog processing. The neural network is designed to accept electric signals directly and perform classification in the analog domain, removing the time-consuming analog-to-digital conversion process while maintaining versatility in processing different types of electric signals.
Solution Approach 2:
The neural network weights and biases are pre-trained offline to capture the classification logic. During actual signal processing, the pre-trained network directly classifies incoming electric signals without requiring any conversion or preliminary processing steps, achieving ultrafast classification performance.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The device achieves ultrafast and accurate classification of electric signals with high noise levels and low amplitudes by processing signals analogously, maintaining signal integrity and enhancing recognition rates.
Implementation Method 1
The recognition device may include a memristive structure to cause a non-linear current output as a function of an applied input signal and a reference input signal
Data Source
AI summary
A classification device for classifying an electric input signal may include: an input node to receive the electric input signal; a reference circuit configured to generate an electric reference signal; a memristive structure configured to cause an electric response signal as a function of both the electric input signal received at the input node and the electric reference signal generated by the reference circuit; an electric analysis circuit coupled to the memristive structure to determine one or more characteristics of the electric response signal caused by the memristive structure; and one or more processors configured to determine a class associated with the electric input signal based on the one or more characteristics of the electric response signal determined by the electric analysis circuit.