Wearable Pulse Sensor Signal Quality Estimation Using Classifier Cascade

The patent changes the parameter being measured from filtered signal characteristics to transition matrix properties. Instead of filtering the signal to remove motion artifacts, the method uses a sample-to-sample transition matrix that captures the probabilistic relationship between consecutive signal samples. This matrix approach transforms the signal quality assessment into a parameter-based evaluation that is inherently robust to motion artifacts without requiring complex filtering operations.

The patent segments the confidence assessment into multiple hierarchical levels corresponding to different data window sizes. Instead of requiring a single large data window for any confidence assessment, the system evaluates confidence at progressively larger scales (e.g., individual pulses, sequences of pulses, extended periods). This segmentation allows the system to provide timely feedback at smaller time scales while still achieving high measurement precision when larger windows are ultimately analyzed.

The patent implements periodic feedback provision based on confidence level thresholds rather than continuous feedback. The system monitors signal quality continuously but provides user feedback only when confidence levels cross predefined thresholds or when placement corrections are needed. This periodic action pattern reduces energy consumption associated with continuous display updates and processing while still effectively guiding users to proper sensor placement through timely feedback.