An Analog Hardware Classifier for Estimating Sea State or Wave Height from Inertial Sensor Data

The related research presents a novel methodology for implementing power-efficient analog classifiers, achieving a power consumption of merely 176nW. These classifiers efficiently process multiple input features while sustaining a high degree of precision and minimizing power consumption. It is based on Threshold Machine Learning model, and integrates sigmoid activation circuit along with current comparators. Validation of the implementation was performed employing a real-time vessel dataset, achieving a good accuracy of 82.95% in predicting true waves from inertial sensor data. Furthermore, a comparative analysis was conducted with other analog classifiers employing the identical dataset. All related models were trained using a software-based equivalent one. The executed design was implemented using a TSMC 90nm CMOS process. It is simulated with the Cadence IC Suite.