Data Augmentation Framework for Smart Sensor System Development Using the Sensor-in-the-Loop Prototyping Platform

Sensor subsystems are becoming more complex as they increasingly take on various tasks, from signal processing to pattern recognition. This off-loading of continuous processes can decrease power consumption of the entire system, as the application processor of e.g. a smartwatch can spend more time in low-power modes. The design and validation of these sensor subsystems are thus becoming more difficult and time-consuming. This development process relies heavily on the availability of suitable sensor signals for testing. We propose a modular component-based framework to generate a multitude of tests using either prerecorded sensor signals or artificial signals to allow for a faster and more thorough development process and prototyping.Using the recently proposed Sensor-in-the-Loop architecture, the work at hand demonstrates not only the ability to test and develop the software offline in a simulation, but to also use the augmented sensor signal data directly on the hardware prototype at run-time. The framework can be used in various testing- and development setups to simulate sensor characteristics, processing steps and errors. We show, based on three comprehensive examples, that the proposed framework is able to simulate the common errors found in inertial MEMS sensors, generate signal traces to develop, train, and evaluate gesture recognition algorithms, and simulate pre-processing steps in order to evaluate their feasibility before they are implemented in the sensor firmware.