Robust Coherent Sensor Combining With Decision Directed Reversal Correction

Abstract

Measurement of a single source by multiple sensors can produce signals that are synchronous but reversed: one signal rises while the other falls. Combining the signals can produce a better signal quality than a single sensor, but detection and correction of the reversals is crucial to avoid destructive interference. Conventional time delay estimation (TDE) methods are good detection tools but may be hampered by conditions such as drift, low signal to noise ratios, and burst interference. Furthermore, these methods are designed for stationary periodic signals, and are susceptible to nonstationary behavior present in the physical world. A method of detection based on the signal slope and a fused reference is presented. Simulations for signals with varying levels of interference were analyzed. A tradeoff between immunity to drift and immunity to noise was controllable by a window length parameter. The proposed method was robust to the simulated interference, withstanding more than 10-dB higher noise and drift than conventional TDE methods. A real-world experiment to extract breathing from a bed-based pressure sensor array was also run. Mean correlations of extracted respiration to gold standard respiration curves were 0.863, achieving a 32% performance improvement over not detecting and correcting the reversals, and a 7% improvement over the conventional detection methods.