On optimal signal-filter pairs (Q2487092)

In the frame of the general theory of signals, defined on \(\{0,N- 1\}\), as it was developed by Trakhman, one consider the complex-valued, \(N\)-periodic function space function which is denoted by \(\mathbb{C}_N\) and whose elements are called signals. By introducing the multiplication by a complex scalar, the addition of two signals and the multiplication of two signals, \(\mathbb{C}_N\) becomes a commutative algebra algebra with unity (the signal with all samples equal to one, denoted by 1). The inverse of \(x\), if it exists, is defined by \(x^{-1}(k)= [x(k)]^{-1}\). A filter is a mapping \({\mathcal R}: \mathbb{C}_N\to\mathbb{C}_N\), defined by \(y= h* x\), where \(h\) is the impulse response (IR) of the filter, \(*\) is the cyclic convolution. By \(R_{xy}\), the cyclic correlation between \(x\) and \(y\) is denoted. A filter \({\mathcal R}\) together its IR \(h\) is matched to a given signal \(x\) if one has \({\mathcal R}[x]= h* x = R_{xx}\). It is shown that the matched filter to any signal exists. By \(X\), \(X={\mathcal F}_N[x]\), the discrete Fourier transform (DFT) on \(\mathbb{C}_N\) is denoted, whereas the signal the signal \(X\) is called the spectrum of \(x\). It is shown that the IR of the matched filter is unique iff its spectrum is invertible. As usually, by \(E\) the energy (over a period) is denoted. By \(\delta_N\), the \(N\) periodic impulse is denoted. In the class \(\mathbb{C}_N\) a special signal is introduced: it is the signal for which the following equality holds \([R_{xx}(0)]^{-1} R_{xx}=\delta_N\). Such a signal is called a delta correlated signal. Proposition 2 gives the canonical form of a delta correlated signal. Then, for a given signal, the side lobe filter (SLF) defined trough \(x* h= E(x)\delta_N\) is introduced. Proposition 3 shows the fact that for a given signal with a prescribed energy, the solution for a minimization problem, called the optimal signal-filter pair, is obtained for a delta correlated signal and the matched filter is the optimal SLF. The second part of the paper extends the results of the first part when the cyclic shifts and replaced by \(n\)-adic shifts and the DFT are done by the Vilenkin-Christensen transform. The necessary bases for this extension and detailed proofs of propositions are given.