Dynamics and Control 2, 113-129 (1992)
© 1992 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands.
Vector-Channel Lattice Filters for the Delta Operator:
Complete Derivation
FARYAR JABBARI
Department of Mechanical and Aerospace Engineering, University of California, Irvine, CA 92717
Received May 22, 1990; revised November 12, 1990, August 29, 1991
Editor: H. Stalford
Abstract. The numerical difficulties associated with high sampling rates have been the main motivationfor the
recent interest in application of delta-operator formulation in adaptive identification and control techniques, as
an alternativeto the traditional shift operator. In this article, the derivationof the vector-channellattice algorithm
for least-squares parameter estimation of an input/output model based on the delta operator is presented. Since
the vector-channelformulationis a generalizationof the standard mtdtichannelforms, results presented here apply
to the multichannel lattices as a special case. In addition to the basic derivation, algorithms for residual errors
and autoregressive coefficients, a simple example, and issues regarding implementation are discussed.
1. Introduction
In this article, the complete derivation of the vector-channel lattice filters for the delta-
operator-based least-squares estimation is presented. The main motivation is the combina-
tion of superior numerical characteristics of lattices with advantages of the delta-operator
approach. Furthermore, lattices allow order-recursive implementation, which provides the
opportunity for on-line order identification and/or change of order. The derivation here
treats the vector-channel case in which there could bep measurement channels, each with
m independent measurements. The vector-channel structure forces the sensors (or
measurements) in each channel to "see" the same dynamics (see [31 and [81 for details).
The vector-channel structure is a generalization of the standard multichannel forms. In-
deed, by setting m = 1 in the derivation, the standard least-squares lattice is obtained.
In [4], the derivation for a single scalar-channel delta lattice is presented. Though the
derivation in that reference captures the essence of the delta lattice, it is not suited for con-
trol and identification applications, since even a SISO system requires a two-channel struc-
ture, one for the input and one for the output. This article presents the complete derivation
for the vector-channel lattice filter, which reduces to the standard multichannel form as
a special case. Algorithms for the residual filters and the AR coefficients are included also.
The resulting algorithms can be used for MIMO system as a special case (i.e., m = 1).
A simple example and some details regarding the embedding technique and other implemen-
tation issues are provided in Section 2.4.
Since the main contribution of this article is the derivation itself, much of the discussion
on advantages of the lattices and delta-operator approach are left to references such as [3]
and [6]. In the following, however, the motivation behind this work is briefly outlined.