What Are Multidimensional Systems?

Multidimensional systems are dynamic systems whose behavior depends on two or more independent variables, in contrast to classical one-dimensional systems governed by a single variable such as time. A 1-D control system evolves along a single axis; an n-D system evolves along a spatial grid, a spatiotemporal volume, or another multi-index domain. The field draws on linear algebra, polynomial ring theory, and functional analysis to extend the well-developed tools of classical systems theory into these higher-dimensional settings.

The study of multidimensional systems emerged as a distinct discipline in the 1970s, when researchers recognized that image processing, seismic data analysis, and iterative learning processes all demanded a framework that classical single-variable system theory could not provide. Work by R. P. Roesser, and separately by E. Fornasini and G. Marchesini, produced the foundational state-space models that remain central to the field today.

State-Space Models and Realization Theory

The Roesser model and the Fornasini-Marchesini (FM) second model are the two principal local state-space descriptions for n-D systems. Both represent system evolution as a function of multi-index shifts rather than a single time index. Realization theory asks when a given n-D transfer matrix can be expressed through one of these models, and what the minimal realization order is. Unlike the 1-D case, minimal realization for n-D systems is not always straightforward, because the required order is sensitive to the structure of the transfer matrix and the polynomial coefficients involved. Research on Fornasini-Marchesini state-space realization has shown that elementary-operation methods on n-D polynomial matrices can systematically derive FM model realizations from matrix fractional descriptions.

Stability Analysis

Stability in multidimensional systems cannot be reduced to the single-variable criteria that suffice in 1-D control. A 2-D system is bounded-input bounded-output (BIBO) stable only if its transfer function has no poles within or on the unit polydisc in the multi-complex variable domain, a condition that requires testing polynomials in several variables. For nonlinear 2-D systems described by Roesser or FM models, vector Lyapunov function methods have been used to establish exponential stability conditions, as detailed in work published in SIAM Journal on Control and Optimization. Extensions to systems with saturation, delays, and overflow arithmetic form an active research area.

Multidimensional Signal Processing

The systems framework connects directly to multidimensional signal processing, where signals are functions of two or more independent variables such as spatial coordinates, time, or frequency. Filter design in the n-D setting generalizes 1-D FIR and IIR methods to separable and non-separable filter banks, sampling on non-rectangular lattices, and transforms such as the 2-D DFT and the Radon transform. Spatial-temporal signal processing, array processing, and tomographic reconstruction are areas that rely on both the signal-theoretic and systems-theoretic sides of the discipline. The journal Multidimensional Systems and Signal Processing, published by Springer, is the primary peer-reviewed venue for the field.

Applications

Multidimensional systems have applications in a wide range of disciplines, including:

• Image and video processing, including filtering, restoration, and compression
• Medical imaging, particularly computed tomography and MRI reconstruction
• Seismic data processing for subsurface structure estimation
• Radar and sonar array processing
• Iterative learning control in industrial robotics
• Wireless channel modeling with spatial and temporal dimensions