On the performance of combined detection estimation in convex uncertainty spaces

The performance of combined detector estimators (CDE's) in the adaptive estimation problem is discussed. They provide better performance than robust single estimator designs. Suboptimal realizations are also shown to be useful in the linear quadratic (LQ) discrete-time case. Finally, a posteriori characterizations of suboptimal CDE risk are shown to be useful in determining whether to improve the near optimality by adding more estimators.