The asymptotic Cramér-Rao bound for 2-D superimposed exponential signals (Q1612685)

2-D complex-valued superimposed exponential signals are considered, having the representation \[ y(m,n)=x(m,n,\theta)+\varepsilon(m,n),\quad m=1,\dots,M;\;n=1,\dots,N, \] where \[ x(m,n,\theta)= \sum_{k=1}^{P}\alpha_{k}\text{exp}[i(\omega_{k}m+\eta_{k}n+\phi_{k})] \] and \(\theta\) denotes the signal parameter vector of interest \[ \theta=[\alpha_{1},\phi_{1},\omega_{1},\eta_{1},\dots, \alpha_{P},\phi_{P}, \omega_{P},\eta_{P}]^{T}. \] Since the exact Cramér-Rao bound (CRB) matrix requires burdensome computation and its expression is not readily interpretable, the paper focusses on obtaining of the asymptotic CRB matrix for the signal parameter vector \(\theta\). It is shown that the asymptotic CRB matrix has a simple form and it is easily interpretable as opposed to the exact CRB. It is emphasized that the Maximum Likelihood (ML) method attains the performance corresponding to the asymptotic CRB as \(M\) and \(N\) increase. It is shown that the nonlinear least squares estimate asymptotically achieves the same performance as the ML estimate.