文档介绍：© Digital Stock 1996
dentifying an unknovvn systenlhas been a central is• In this review article we focus on a particular type of
sue in various application areas such as control, system-the linear Finite Impulse Response (FIR) sys•
channel equalization, echo cancellation mu• tem. In other words, the class of models in which the
\vorks and teleconferencing, geophysi• search for the optimum filter is conducted assumes that
cal signal processing, and many others. Identification is each output value is determined by a bina•
the procedure of specifying the tion of a fixed, finite number,
unknown model in tenns ofthe M, of past values of the input
available experimental evidence) signal.
that is, a set ofmeasurements of The petformance of an algo•
a) the input-output desired re• rithm can be measured by a
sponse signals, and b) an appro• number offactors such as
priately chosen error cost function that is optimized with A The accuracy of the obtained solution with respect to
respect to the unknovvn model parameters. Adaptive iden• the theoretically expected set up
tification refers to a particular procedure where we learn iJ!, Its convergence speed
more aboutthe model as each ne\v pair ofmeaSl1rClllents is A Its tracking ability withrespect totime-varying statistics
received, and we update ourknowledge to incorporate the A plexity
newly received information. A Its robustness to rOlUld offerror accumulation
JULY 1999 IEEE SIGNAL PROCESSING MAGAZINE 13
l053-5888/99/©1999IEEE
vector and matrix quantities are denoted as lowercase
bold roman and uppercase bold roman respectively.
The Wiener Filter
Adaptive filtering and system-identification algorithms
deal with the estimation ofa set ofparameters ofa model
ofan W1known plant, which, once estimated, gives suffi•
cient information about the system dynamics. Linear sys•
tem parameterization is an important class of system
modeling with a wide area ofapplications. The most pop•
A.