Estimating maxima of generalized cross ambiguity functions, and uncertainty principles (Q1940125): Difference between revisions

Motivated by the examples of radar cross ambiguity and wideband ambiguity functions, a generalized cross ambiguity function is defined as follows. Let \(S\subset\mathbb{R}^s\) for some \(s\geq 1\), \(X\) a Hilbert space, and let \(\{U(t), t\in S\}\), be a family of unitary operators depending weakly continuously on \(t\) and such that \(U(t_0)=I\) for some specific \(t_0\in S\). The generalized cross ambiguity function mapping \(A:X\times X\to \mathcal{C}(S)\) is defined as \(A(f,g)(t)=\langle f,\, U(t)^\ast g\rangle\). As is the case of radar ambiguity, certain parameter estimation problems can be framed in terms of maximization of \(A(f,g)\) for specific \(f,g\). Given \(F\in X\) and \(g=U(t_d)F\) for some fixed \(t_d\in S\), the unitarity of \(U\) gives \(A(f,G)(t_d)=\langle f,\, F\rangle\) and the quantity \(\Re A(f,G)(t)\) has a maximum value \(\|F\|^2\) at \(t_d\). Given \(t=(t_1,\dotsc, t_s)\in S\) and \(1\leq j\leq s\), one defines possibly unbounded skew-adjoint operators \(T_j: \mathcal{D}(T_j(t))\to X\) by \(\frac{\partial}{\partial t_j}A(f,g)(t)=-\langle f, T_j(t) g\rangle\) or, with suitable constraints, \(\frac{\partial}{\partial t_j} U(t) f=U(t) T_j(t) f\) and with \(F,G\) as above, \[ \Re A\Bigl(\frac{\partial}{\partial t_j}(T_k (t_d) F),G\Bigr)(t_d)=0=\Re A\Bigl(\frac{\partial}{\partial t_j}(T_k (t_d) F),F\Bigr)=0\, . \] It is proved that optimization properties are preserved under suitably defined differentiable mappings. It is also proved that, if there exist \(F_N\) and \(G_N\) converging to \(F\) and \(G\), respectively, then the \(t_d^N\) that maximize \(A(F_N,G_N)(t_d^N)\) converge to \(t_d\) that optimizes \(A(F,G)\). It is illustrated how this approach provides a practical means of approximating \(t_d\) in the case of radar and wide-band ambiguity functions for band-limited signals. The following inequality is established. Define a matrix \(B=B(F)\) by \(B_{jk}=\Re \langle T_j(t_d) F, T_k(t_d)F\rangle\) where \(F\in\mathcal{D} (T_j(t_d))\), \(j=1,\dotsc, s\). Also let \(\lambda_j=\langle T_j,F\rangle/\|F\|^2\) and \(C\) denote the matrix \(C_{jk}=\Re \langle T_j(t_d)F-\lambda_jF,\, T_k(t_d)F-\lambda_k F\rangle\). For a positive semidefinite matrix \(M\) and \(c>0\), define \(E_M(c)=\{x\in\mathbb{R}^s: x^TMx<c\}\). It is shown that \(E_B(c)\subset E_C(c)\). As a consequence of these facts the following uncertainty inequalities are established for symmetric operators \(S_1, S_2, S_3\) on \(X\) and \(F\in \mathcal{D}(S_jS_k)\cap\mathcal{D}(S_kS_j)\) for \(j,k=1,\dotsc, 3\). Let \(\xi=(\langle [S_2,S_3] F,F\rangle,\, \langle [S_3,S_1] F,F\rangle,\, \langle [S_1,S_2] F,F\rangle)^T\). Then, for \(B,C\) as above, one has \(\det B\geq \frac{1}{4}\overline{\xi}^T B\xi\) and \(\det C\geq \frac{1}{4}\overline{\xi}^T C\xi\).