State Change Detection for Low-Power Sensor Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks consume extensive resources for classification, making it challenging to utilize them on real-time, energy-sensitive applications on low-power systems such as CPUs, DSPs, and MCUs, especially when dealing with streaming data that has substantial temporal similarity and requires repeated computations for successive data points with minimal changes.
Innovation Solution
A state change detection component is integrated into low-power devices to detect changes in input data streams, allowing the system to remain in a dormant state until a significant change is detected, triggering the classification process only when necessary, thereby reducing unnecessary computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for classification on low-power systems, then classification accuracy is improved, but energy consumption increases
Solution Approach 1:
The system uses periodic wake-up intervals where the neural network classifier is activated at predetermined times to perform classification on the buffer, rather than continuously processing every incoming data point. This periodic operation significantly reduces energy consumption while maintaining classification accuracy by processing representative samples of the data stream.
Solution Approach 2:
Data points are pre-buffered in a buffer before classification processing. This preliminary buffering allows the system to accumulate data and then process it in batches when the neural network is activated, reducing the frequency of expensive classification operations and lowering overall energy consumption.
2Speed
If continuous classification is performed on streaming data, then real-time processing capability is improved, but resource consumption increases
Solution Approach 1:
The system implements periodic classification at predetermined wake-up intervals rather than continuous classification. The neural network is activated at specific intervals to classify buffered data, maintaining real-time processing capability through periodic updates while dramatically reducing resource consumption compared to continuous operation.
Solution Approach 2:
The data stream is segmented into buffered portions that are processed together during wake-up intervals. Instead of processing every individual data point in real-time, the system segments the continuous stream into manageable batches, classifying representative samples while maintaining the ability to detect changes between wake-up periods.
3Productivity
If classification is performed on every data point, then processing completeness is improved, but computational redundancy increases
Solution Approach 1:
The system performs partial classification by processing only a subset of buffered data points rather than every single data point. During wake-up intervals, the neural network classifies representative samples from the buffer, which is sufficient to detect state changes and maintain processing completeness while avoiding the computational redundancy of classifying every individual point.
Solution Approach 2:
Data points are preliminarily buffered and filtered before classification. The system accumulates data in a buffer and applies preliminary processing to identify candidate points for classification, reducing the total number of classification operations needed while maintaining processing completeness through strategic selection of classification candidates.
Data Source
AI summary
A method for energy-efficient classification receiving, via a first circuit, an input data stream from one or more sensors. The first circuit detects, while a second circuit is in a dormant state, if a state change has occurred between a first input of the input data stream and a second input of the input data stream. The second input is a next succeeding input of the input data stream. The first circuit triggers the second circuit to perform a classification of the input data stream in response to detecting the state change.