Title :
Theorems and fallacies in the theory of long-range-dependent Processes
Author_Institution :
Dept. of Electr. & Comput. Eng., Univ. of Wisconsin, Madison, WI
fDate :
3/1/2005 12:00:00 AM
Abstract :
It is frequently claimed in the literature that long-range dependence has equivalent formulations in the time domain and the frequency domain. Although many researchers understand that this is only "operationally true," i.e., it holds in cases of interest, many state this equivalence as a mathematical theorem. In particular, it is claimed as a theorem in the literature that if a covariance function decays like one over a fractional power of n, then the corresponding power spectral density tends to infinity at the origin. It is shown here that the power spectral density need not exist. Conversely, if the power spectral density exists and tends to infinity at the origin, it is shown here that the covariance may not have the claimed decay. To conclude, a new theorem is proved that gives sufficient conditions on the power spectral density to guarantee that a process is asymptotically second-order self-similar (ASOSS). This result is used to provide a counterexample to the claim in the literature that asymptotic second-order self-similarity implies long-range dependence
Keywords :
asymptotic stability; covariance analysis; fractals; random processes; time-frequency analysis; ASOSS; asymptotically second-order self-similar; covariance function decay; long-range dependence process; mathematical theorem; power fraction; power spectral density; slowly varying function; time-frequency domain; Capacitive sensors; Computer networks; Conferences; Fourier series; H infinity control; Information theory; Mobile computing; Routing; Spread spectrum communication; Wireless sensor networks; Asymptotic second-order self-similarity; exact second-order self-similarity; long-range dependence; slowly varying functions;
Journal_Title :
Information Theory, IEEE Transactions on
DOI :
10.1109/TIT.2004.842768