Approximation of dual Gabor frames, window decay, and wireless communications (Q5946563)

Given \(a,b>0\) and a function \(g\in L^2( R)\), the associated Gabor system consists of the functions \[ g_{n,m}(x)= e^{2\pi i mbx}g(x-na), \quad m,n \in Z. \] If there exist constants \(A,B>0\) such that \[ A\|f\|^2 \leq \sum_{m,n \in Z} |\langle f, g_{n,m}\rangle|^2 \leq B\|f\|^2 \] for all \(f\in L^2( R)\), then \(\{g_{n,m}\}_{m,n\in Z}\) is called a Gabor frame. In this case the frame operator \(Sf= \sum_{m,n \in Z} \langle f, g_{n,m}\rangle g_{n,m}\) is well-defined and invertible on \(L^2( R)\). With \(\gamma := S^{-1}g\), every \(f\in L^2( R)\) can be represented as \[ f=\sum_{m,n \in Z} \langle f, \gamma_{n,m}\rangle g_{n,m}. \] In order for this to be useful, one needs to find \(\gamma\), i.e., to invert \(S\), which is usually difficult. In the present paper a method for approximation of \(\gamma\) using finite-dimensional techniques is developed. In case \(g\) as well as its Fourier transform has exponential decay, the obtained rate of convergence towards \(\gamma =S^{-1}g\) is exponential, but with an unknown exponent. It is demonstrated that the results are useful in wireless communication.