Intelligent Methods in Signal Processing and Communications

This paper presents a discussion of the applications of adaptive array processing methods in mobile communication systems. A brief history of the development of adaptive arrays is presented together with a summary of current methodologies used for implementation. A “best estimate” of what adaptive applications axe likely in communication systems in the future is provided together with a description of the significant technical challenges. The paper is intended as an overview and does not contain detailed mathematical development.

In this article we describe a novel system for the detection of a target signal in the presence of additive interference. The system operates under the premise that the statistics of the target signal and the interference are uknown and that both may be nonstationary. The design of the system is guided throughout by the information preservation rule.

The system consists of two independent channels fed from a common module that computes the Wigner-Ville distribution of the received signal.One channel, termed the interference channel, consists of a principal components analyzer followed by a multilayer perceptron classifier; these two components are tarined on different realizations of the received signal known to contain interference only. The other channel, termed the target channel, has a similar composition, except that this time the training data consist of different realizations of the received signal known to consist of a target signal plus interference.The two channels are linearly combined into a single output node where the final decision is made, whether a target is present or not.

Eperimental results, using real life ground-truthed data collected with an instrument-quality radar, are presented demonstrating the superior performance of the new detection system(receiver) over a conventional Doppler constant false-alarm receiver.

Recently, with technological advance on microelectronic and vision system, true verification of individual identities has now become possible. This technology is based on a field called biometrics. Biometric systems are automated methods of verifying or recognizing the identity of a living person on the basis of some physiological characteristic, like a fingerprint or face pattern, or some aspect of behavior, like handwriting or keystroke patterns. The objectives of this chapter are to investigate various biometric identification methods and to develop useful techniques for implementing good biometric identification systems. In particular, we focus on two types of biometric methods, namely, face recognition and palm print recognition. Among all the biometric identification methods, face recognition has attracted much attention in recent years because it has potential to be most non-intrusive and user-friendly. In this chapter we propose an integrated face recognition system based on probabilistic decision-based neural networks (PDBNN)[33]. The face recognition system consists of three modules: First, a face detector finds the location of a human face in an image. Then an eye localizer determines the positions of both eyes in order to genef.ate meaningful feature vectors. The facial region proposed contains eyebrows, eyes, and nose, but excluding mouth. (Eye-glasses will be allowed.) Lastly, the third module is a face recognizer. The PDBNN can be effectively applied to all the three modules. It adopts a hierarchical network structure with nonlinear basis functions and a competitive credit-assignment scheme. This chapter demonstrates a successful application of PDBNN to face recognition on the public ORL face database. Regarding the performance, experiments on three different databases all demonstrated high recognition accuracies as well as low false rejection and false acceptance rates. As to the processing speed, the whole recognition process (including PDBNN processing for eye localization, feature extraction, and classification) consumes approximately one second on a Sparcl0, without using a hardware accelerator or co-processor. A new biometric identification scheme using human palm print information is proposed. This scheme extracts discriminant features from a grayscale palm image by edge filtering and Hough transform, and recognizes the input pattern by a novel structural matching algorithm.

face recognition system demonstrated the power of the PDBNN pattern classifier, and the palm print recognition system indicates the importance of a good feature extractor. Due to their non-intrusiveness, both systems are very suitable for gateway access control and computer security management.

Our main result is a theorem which gives necessary and sufficient conditions under which discrete-space multidimensional myopic input-output maps with vector-valued inputs drawn from a certain large set can be uniformly approximated arbitrarily well using a structure consisting of a linear preprocessing stage followed by a memoryless nonlinear network. Noncausal as well as causal maps are considered. Approximations for noncausal maps for which inputs and outputs are functions of more than one variable are of current interest in connection with, for example, image processing.

We present a systematic method for incorporating prior knowledge (hints) into the learning-from-examples paradigm. The hints are represented in a canonical form that is compatible with descent techniques for learning. We focus in particular on the monotonicity hint, which states that the function to be learned is monotonic in some or all of the input variables. The application of monotonicity hints is demonstrated on two real-world problems-a credit card application task, and a problem in medical diagnosis. We report experimental results which show that using monotonicity hints leads to a statistically significant improvement in performance on both problems. Monotonicity is also analyzed from a theoretical perspective. We consider the class M of monotonically increasing binary output functions. Necessary and sufficient conditions for monotonic separability of a dichotomy are proven. The capacity of M is shown to depend heavily on the input distribution.

The robust strict positive real problem arises in identification and adaptive control, where strict positive realness is a sufficient condition for ensuring convergence of several recursive algorithms. The strict positive real (SPR) property of uncertain systems is analyzed in depth here, and some insightful theorems are provided for characterizing the SPRness of different types of uncertain sets. The design of appropriate compensators which enforce the SPRness of a given set is also addressed, through both algebraic and approximate procedures. Those compensators are widely used in the adaptive recursive filters field: they guarantee global convergence in many cases. The robust strengthened strict positive real problem is not excluded from our analysis, given its importance in some recursive algorithms: the problem arises as to how to design a compensator with a given constraint in its norm or its coefficients. Illustrative results are presented at the end of the paper.

In this chapter we introduce the use of Boundary Methods (BM) for distribution analysis. We view these methods as tools which can be used to extract useful information from sample distributions. We believe that Boundary Methods can be used for a number of applications, but here we restrict our attention to three applications. First, we discuss the use of boundary methods for determining the suitability of a particular feature set for pattern classification, i.e. we use the Boundary Methods to perform feature-set evaluation (FSE). We present results which establish the correspondence of Boundary Methods and the probability of error (Pe) for normal distributions. Second, we discuss the utility of Boundary Methods as a technique for sample-pruning (SP), and show how we can select samples, e.g., for progressive training of neural-networks. Finally, we state a theorem which relates Fisher’s Linear Discriminant (FLD) and Boundary Methods.

In this paper we study the theoretical limits of finite constructive convex approximations of a given function in a Hilbert space using elements taken from a reduced subset. We also investigate the trade-off between the global error and the partial error during the iterations of the solution. These results are then specialized to constructive function approximation using sigmoidal neural networks. The emphasis then shifts to the implementation issues associated with the problem of achieving given approximation errors when using a finite number of nodes and a finite data set for training.

The classification of a data collection using tree structures has been studied by statisticians and psychologists for many years, and it has shown to be an effective way of dividing a complex classification problem in a sequence of simpler decision tasks. The modularity of both learning and classification showed by these structures has attracted the attention of neural network researchers, looking for alternatives to the learning and computational problems of backpropagation networks. This paper overviews the current research on tree classification based on neural networks. Structure and learning algorithms are described; the implications of probabilistic interpretations of the network behaviour are discussed and some new learning rules that can speed up learning and reduce the final misclassification probability are proposed.

This work reviews the state of the art in techniques for communicating with chaos and highlights some of the engineering challenges in this field.

Subband coding is widely used to compress speech, audio and video signals. In some applications, for example audio compression, a non-uniform splitting of the spectrum of a digital signal can be preferable to a uniform one. Some techniques that allow to implement a non-uniform width subband decomposition are here discussed, design methods based on the cosine-modulation of more than one prototype are described and some examples of filter bank design are given.

Due to recent advances in display technology, three dimensional (3-D) imaging systems are becoming increasingly more common in applications such as computer vision, virtual reality, terrain mapping, navigation, and image understanding. To achieve 3-D perception, these systems use a stereo pair, which is a pair of images of the same scene acquired from different perspectives. Since there is an inherent redundancy between the images of a stereo pair, data compression algorithms can be employed to transmit and store these images efficiently.

In this chapter, we consider the problem of stereo image coding. We begin with a description of the stereo coding problem, and survey the current approaches to stereo coding. Then, we describe a new coding algorithm that is based on disparity compensation and subspace projection. This algorithm, called the Subspace Projection Technique (SPT), is an incomplete local transform with a data-dependent space-varying transform matrix. The advantage of the SPT approach over other techniques is that it is able to adapt to changes in the cross-correlation characteristics of stereo pairs locally.

In this chapter a methodology for the design of VLSI circuits for image and video coding applications is presented. In each section a different phase of the design procedure is discussed, along with a description of the involved software environments. An example of an area efficient single-chip implementation of a JPEG coder is presented to illustrate the methodology.