Functions of banded Hermitian block Toeplitz matrices in signal processing (Q875035)

From MaRDI portal





scientific article; zbMATH DE number 5141651
Language Label Description Also known as
default for all languages
No label defined
    English
    Functions of banded Hermitian block Toeplitz matrices in signal processing
    scientific article; zbMATH DE number 5141651

      Statements

      Functions of banded Hermitian block Toeplitz matrices in signal processing (English)
      0 references
      0 references
      0 references
      10 April 2007
      0 references
      Let \({\mathcal F}\) be the set of all infinite block matrices with only finitely many nonzero blocks and let \({\mathcal H}\) be the set of all infinite block matrices \(A=(a_{jk})_{j,k=1}^\infty\) with the property that there exist constants \(0<M<\infty\) and \(0<\sigma<1\) such that \(\| A_{jk}\| \leq M\sigma^{j+k}\) for all \(j,k\geq 1\). For a natural number \(n\), the operators \(P_n\) and \(W_n\) are defined on \(\ell_N^2\), the space of columns of length \(N\), by \(P_n:\{x_1,x_2,\dots\}\mapsto\{x_1,\dots,x_n,0,0, \dots\}\) and \(W_n:\{x_1,x_2,\dots\}\mapsto\{x_n,\dots,x_1,0,0,\dots\}\), respectively. Consider the Toeplitz-like matrices \(A_n=T_n(a)+P_nUP_n+ W_nVW_n\), where \(T_n(a)\) is an \(nN\times nN\) block Toeplitz matrix generated by a matrix Laurent polynomial \(a\) and \(U,V\in{\mathcal F}\). The main result says that if \(a,U,V\) are selfadjoint, \(\alpha,\beta\in{\mathbb R}\) are such that \(\alpha<\beta\), the spectrum of \(a\) and the spectra of \(A_n\) are contained in \([\alpha,\beta]\), and \(f\) is an analytic function in an open neighborhood of \([\alpha,\beta]\), then \(f(A_n)=T_n(f(a))+P_nKP_n+W_nLW_n+C_n\), where \(K,L\in{\mathcal H}\) and \(\| C_n\| =O(\tau^n)\) as \(n\to\infty\) for some \(\tau\in(0,1)\). This result is applied to get a sufficiently precise information on the entries \([f(A_n)]_{ij}\) of the matrix \(f(A_n)\). For some concrete functions \(f\) and Hermitian Toeplitz matrices, the obtained results are of significant importance in signal processing and information theory. Some such applications are considered.
      0 references
      block Toeplitz matrix
      0 references
      MMSE linear equalizer
      0 references
      MMSE linear estimator
      0 references
      MIMO channel
      0 references
      multivariate stationary process
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references