The device receives an input signal xt and an observation signal yt,
one component of which is the response of a system to the input signal. It determines
an error signal et at an instant t in accordance with the equation et=yt-XtTHt-1,
where Ht-1 is a vector made up of L coefficients of an identification
filter having a finite impulse response representative of the impulse response
of the system and XtT=(xt, xt-1, .
. . , xt-L+1). A predictive analysis of the input signal is performed
by a computation applied to frames of the input signal. The L coefficients of the
identification filter are adapted by adding to the vector Ht-1 a vector
proportional to
##EQU1##
where Ut is a vector made up of the L values of a prediction residue
of the input signal at the instant t and at the preceding L-1 instants, and
is a positive or zero coefficient.
