Real-Time Temporal Signals Separation and Identification in Noisy Environments

This proposed project combines ideas from neuro-biological modeling of the cochlea, engineering designs of optimal filters, the theoretical formulation using the calculus of variations, and optimal filtering and control. Several competing formulations will be developed. A potential breakthrough entails a formulation that develops (i) a neural module that separates mixed signals or signal and noise from measurements in dynamic and cluttered environments, with very minimum assumptions, (ii) subsequent modules that initially preprocess the signals via a bank of filters in line with the cochlea's models, followed by an identification stage of the static "features." For part (ii), an alternative approach will pursue a direct formulation based on optimal filtering and control to develop identification networks for temporal signals. The Phase I effort will validate and evaluate the proposed explicit algorithms within a simulated mixing environment containing a combination of dynamically changing signals at varying levels of interference. The developed algorithms will be coded in a tailored C/C++ development code with graphics capabilities for demonstration. A real-time computing environment for separating mixed signals/signals and noise is expected. This algorithm is novel and departs from known methods in industry. The benefits include the ability to eliminate noise from speech signals or sounds of different vehicles/aircraft/missiles.