DynDSE: Automated Multi-Objective Design Space Exploration for Context-Adaptive Wearable IoT Edge Devices

Area of Science:

• Internet of Things (IoT)
• Edge Computing
• Wearable Technology
• Pattern Recognition

Background:

• Wearable IoT devices require efficient resource management for continuous operation.
• Context-adaptive pattern recognition is crucial for analyzing streaming sensor data.
• Optimizing designs for conflicting metrics like performance and energy consumption is challenging.

Purpose of the Study:

• To introduce DynDSE, a simulation-based procedure for Design Space Exploration of wearable IoT edge devices.
• To formally characterize the design space of such systems.
• To find optimal configurations balancing retrieval performance and resource consumption.

Main Methods:

• Formal characterization of the design space including system functionalities, components, and parameters.
• Iterative simulation-based search evaluating configurations against requirements using actual sensor data.
• Exploration of trade-offs between conflicting metrics: retrieval performance, execution time, energy consumption, memory demand, and communication latency.

Main Results:

• DynDSE successfully identified optimal configurations for electromyographic-monitoring eyeglasses.
• Achieved an F1 score above 80% for retrieval performance.
• Reduced energy consumption by 70% compared to non-optimized configurations.

Conclusions:

• The DynDSE approach effectively balances retrieval performance and resource consumption in wearable IoT systems.
• This methodology can be applied to diverse sensor-based applications requiring optimized wearable IoT designs.
• The study demonstrates significant energy savings and high accuracy in a real-world case study.