A full generalized likelihood ratio test for source detection
Pei-Jung Chung, Kon Max Wong · 2012
This work presents a novel full generalized likelihood ratio test (GLRT) for signal detection in a sensor array environment. The multiple hypothesis test approach is well known to have excellent detection performance among several popular methods. Existing multiple test procedures consider the relation between two adjacent models. When the number of signals or the assumed number of signals is large, it tends to overestimate the number of signals. The proposed full GLRT procedure overcomes this disadvantage by employing complete information between candidate models and leads to gain in test power. A further advantage is that a confidence interval for the true number of signals can be constructed based on the outcome of the GLRT procedure. Numerical results show that the full GLRT procedure improves detection performance significantly in comparison with existing multiple test based approaches in challenging scenarios.