Representation of operators in the time-frequency domain and generalized Gabor multipliers (Q967572): Difference between revisions

A Gabor multiplier is obtained by taking a Gabor window \(g \in {\mathcal S}(\mathbb R)\) with dual window \(h\), a \textit{symbol function} \(\mathbf{m}: \mathbb Z^2 \to\mathbb C\), and letting \[ S f = \sum_{n,k} \mathbf{m}(n,k) \langle f, M_{n \nu_0} T_{kb_0} g \rangle M_{n \nu_0} T_{kb_0} h. \] Here, \(T_x,M_\xi\) denote shifts in time and frequency, respectively, thus \(b_0, \nu_0\) are sampling densities in the time and frequency variables. Duality between \(g\) and \(h\) means that taking \(\mathbf{m} \equiv 1\) yields the identity operator for \(S\). Not all operators on \(\text{L}^2(\mathbb R)\) are well approximated by Gabor multipliers. This paper focuses on the optimal approximation of Hilbert-Schmidt operators by Gabor multipliers. The authors derive an explicit formula for the optimal approximation in terms of the windowed Fourier transform \(V_gh\) of the synthesis window \(h\) with respect to the analysis window \(g\), which is valid if the lattice dual to \(b_0\mathbb Z\times \nu_0 \mathbb Z\) is sufficiently dense. Estimates of the approximation error are also provided. These can be read as quantitative versions of the known statement that an operator is well-approximated by Gabor multipliers only if its time-frequency spreading function is under-spread. The authors then extend their scheme by considering sums of Gabor multipliers, and derive Hilbert-Schmidt-optimal approximations for this setting as well.