Innovation-based detector using second-order approximation model in ocean acoustic signal processing

The model-based approach is applied in the shallow ocean acoustic signal detection problem. Based on a state-space representation of the normal mode propagation model and a vertical linear array measurement system, the extended Kalman filter (EKF) is used to accomplish the shallow ocean environment identification process, in which one of the outputs is the innovation sequence. When the model does not match the environment, the innovation sequence becomes non-zero mean and/or non-white. A second-order approximation state-space model is proposed in the model-based processing scheme, resulting in smaller amount of computation and better model accuracy. Several statistics for testing the properties of the innovation sequence are outlined and analyzed, composing an innovation-based detector which will declare a model mismatch if an anomaly (possibly a target) emerges. Simulations under a typical shallow ocean environment are performed, giving the receiver operating characteristic (ROC) curves with regard to different SNRs and parameters in the test statistic weighted sum squared residual (WSSR), showing the overall detection performances of these test statistics of the innovation sequence.