Bayesian Sequential Joint Detection and Estimation under Multiple Hypotheses

TL;DR

This paper develops a Bayesian sequential method for joint hypothesis testing and parameter estimation, minimizing sample size while controlling error levels, with solutions characterized by a Bellman equation and validated through numerical examples.

Contribution

It introduces a novel Bayesian sequential framework for joint detection and estimation under multiple hypotheses, with a new characterization of the optimal solution via a Bellman equation and practical methods for finding optimal coefficients.

Findings

Optimal schemes are derived and validated through Monte Carlo simulations.

The solution involves a non-linear Bellman equation parametrized by cost coefficients.

Two approaches (linear programming and gradient ascent) effectively find optimal coefficients.

Abstract

We consider the problem of jointly testing multiple hypotheses and estimating a random parameter of the underlying distribution. This problem is investigated in a sequential setup under mild assumptions on the underlying random process. The optimal method minimizes the expected number of samples while ensuring that the average detection/estimation errors do not exceed a certain level. After converting the constrained problem to an unconstrained one, we characterize the general solution by a non-linear Bellman equation, which is parametrized by a set of cost coefficients. A strong connection between the derivatives of the cost function with respect to the coefficients and the detection/estimation errors of the sequential procedure is derived. Based on this fundamental property, we further show that for suitably chosen cost coefficients the solutions of the constrained and the unconstrained problem coincide. We present two approaches to finding the optimal coefficients. For the first approach, the final optimization problem is converted into a linear program, whereas the second approach solves it with a projected gradient ascent. To illustrate the theoretical results, we consider two problems for which the optimal schemes are designed numerically. Using Monte Carlo simulations, it is validated that the numerical results agree with the theory.