Ai Hui Tan

Design of Ternary Signals for MIMO Identification in the Presence of Noise and Nonlinear Distortion

A new approach to designing sets of ternary periodic signals with different periods for multi-input multi-output system identification is described. The signals are pseudo-random signals with uniform nonzero harmonics, generated from Galois field GF(q), where q is a prime or a power of a prime. The signals are designed to be uncorrelated, so that effects of different inputs can be easily decoupled. However, correlated harmonics can be included if necessary, for applications in the identification of ill-conditioned processes. A design table is given for q les 31. An example is presented for the design of five uncorrelated signals with a common period N = 168 . Three of these signals are applied to identify the transfer function matrix as well as the singular values of a simulated distillation column. Results obtained are compared with those achieved using two alternative methods.

Design of computer-optimized pseudorandom maximum length signals for linear identification in the presence of nonlinear distortions

The design of pseudorandom maximum length (PRML) signals for linear identification in the presence of nonlinear distortions is considered. For this application, it is advantageous for the signal to have harmonic multiples of two and three suppressed, in order to minimize the effect of nonlinearity, thus resulting in a better estimate of the underlying linear dynamics. Such signals may be designed through exhaustive search for the sequence-to-signal conversions. However, for signals generated from Galois fields GF(q) with q large, this method is computationally inefficient. An alternative technique is proposed where a primitive version of the signal, the period of which is considerably shorter than that of the required PRML signal, is first generated as a computer-optimized signal. The primitive signal is then used to define the conversions for the generation of the required PRML signal, which is a member of a new class of hybrid signal.

Modeling of direction-dependent Processes using Wiener models and neural networks with nonlinear output error structure

The modeling of direction-dependent dynamic processes using Wiener models and recurrent neural network models with nonlinear output error structure is considered. The results obtained are compared for several simulated first-order and second-order processes and using three different types of input signals: a pseudorandom binary signal, an inverse-repeat pseudorandom binary signal and a multisine (sum of harmonics) signal. Experimental results on a real system, namely an electronic nose system, are also presented to illustrate the applicability of the techniques discussed.

Identification of an Electric Resistance Furnace

In this paper, the identification of an electric resistance furnace is described. Physical modeling is applied to obtain a mathematical description that captures the static and dynamic behavior of the furnace. Experimental data are collected and used for estimating the parameters of the furnace, which is found to be a dominantly first-order system. First-order linear, bilinear, and direction-dependent models are employed to describe the characteristics of the furnace. A comparison between the quality of these models is made. It was found that the bilinear model provides the best fitting among the three models considered for this particular application.

Identification of wiener-hammerstein models with cubic nonlinearity using lifred

Abstract The identification of Wiener-Hammerstein models using linear interpolation in the frequency domain (LIFRED) is extended from models with quadratic nonlinearity to models with cubic nonlinearity. The modifications to the algorithm are discussed and a simulation example is presented. A further technique is proposed which enables the estimation of the gain response of the first linear subsystem from estimation lines in the output which are distorted due to contributions of several combinations of the input harmonics. This new technique is less susceptible than the first approach to the effects of noise and a possible reason for this is discussed.

Perturbation signal design

Abstract This tutorial paper focuses on a number of designs for perturbation (input) signals for system identification; all of the signals can be designed using software readily available on the World Wide Web. Pseudorandom signals have fixed spectra, and both binary and multilevel signals based on maximum-length sequences are discussed. Other classes of pseudorandom binary signals that greatly increase the number of available signal periods are described. Computer-optimized signals have spectra that can be specified by the user, and the paper deals with three types – multisine (sum of harmonics) signals, and binary and multilevel multiharmonic signals. Perturbation signal quality measures are also considered in the paper.

A guide to the design and selection of perturbation signals

There are now many types of perturbation signal that can be used for system identification. These include signals with fixed power spectra, computer-optimized signals for which the user specifies the required harmonics, and hybrids of the two. With so many types now available, it is often difficult for the user to know how to select a signal that will be most appropriate for a particular application. In this paper, the authors provide a general guideline to this selection in a number of different experimental situations, giving the reasons for the particular selection.

Comparison of Three Proportional-Integral-Derivative Based Controllers on a Bilinear Electric Resistance Furnace

In this paper, three proportional-integral-derivative controller designs are explored and implemented to control the operation of a single zone electric resistance furnace. The furnace is modeled using a bilinear model with uncertainty in the parameters. One of the controllers is tuned using Ziegler-Nichols tuning rules and the other two are designed using internal model control based on linear and bilinear process models, respectively. Controller performance is evaluated in the presence of actuator saturation, as well as model uncertainty. Comparison is made based on integral absolute error, integral square error, and percentage maximum overshoot of the process output.

Direction-dependent system modeling approaches exemplified through an electronic nose system

The modeling of processes exhibiting direction-dependent behavior is considered. Depending on the application, different models may be suitable. This brief is concerned with the use of Wiener models and piecewise-linear (PWL) models. These approaches are applied to data from an electronic nose system, for which knowledge of the physical principles is combined with system identification methods. Both models are found to provide close approximations to the behavior of the system itself.

Identification of systems with direction-dependent dynamics

In this paper, the identification of systems with direction-dependent dynamics by means of bilinear models and Wiener models is considered. It is shown that when such a system is perturbed by a pseudo-random binary signal based on a maximum-length sequence, distinctive patterns are observed in the cross-correlation function between the system input and the system output. These patterns are not present when other kinds of pseudo-random binary signals are used. The patterns obtained for bilinear models and Wiener models are similar, and both depend on the characteristic polynomial of the maximum-length sequence used. For the case in which the dynamics involved are first-order, analytical results are obtained which allow the patterns to be compared in detail. The results expected when the pseudo-random signals used are inverse-repeat are also described. It is concluded that both kinds of model are suitable for use in this application, provided that the model parameters are appropriately chosen.