: This thesis applies neural network feature selection techniques to multivariate time series data to improve prediction of a target time series. Two approaches to feature selection are used. First, a subset enumeration method is used to determine which financial indicators are most useful for aiding in prediction of the S&P 500 futures daily price. The candidate indicators evaluated include RSI, Stochastics and several moving averages. Results indicate that the Stochastics and RSI indicators result in better prediction results than the moving averages. The second approach to feature selection is calculation of individual saliency metrics. A new decision boundary-based individual saliency metric, and a classifier independent saliency metric are developed and tested. Ruck's saliency metric, the decision boundary based saliency metric, and the classifier independent saliency metric are compared for a data set consisting of the RSI and Stochastics indicators as well as delayed closing price values. The decision based metric and the Ruck metric results are similar, but the classifier independent metric agrees with neither of the other metrics. The nine most salient features, determined by the decision boundary based metric, are used to train a neural network and the results are presented and compared to other published results. (AN)

/pdf/nonlinear-time-series-analysis-1j3a35n4y3.pdf

Nonlinear Time Series Analysis.

Modeling and computer-aided design (CAD) techniques are essential for microwave design, especially with the drive towards first-pass design success. We have described neural networks for microwave modeling and design. Neural networks are suitable when modeling a required relationship for which analytical formulas are hard to derive, or for which the computational effort is too high. This relationship can be either of the IO relationship of the overall model (straight neural network model), the output-input relationship (inverse model), a relationship between existing model and desired data (neuro-SM), or relationship of a subpart of the overall model (knowledge based neural network). Neural networks are fast to evaluate, and the neural network formulas are easy to implement into microwave CAD. The simplicity of adding input neurons or hidden neurons makes neural network flexible in handling functions of different dimensions and of different degree of nonlinearity. We have also demonstrated that neural networks are helpful in developing parametric or scalable models for passive and active microwave devices.

Smart Modeling of Microwave Devices

In this paper, a novel two hidden layers artificial neural network (2HLANN) model is proposed to predict the dynamic nonlinear behavior of wideband RF power amplifiers (PAs). Starting with a generic low-pass equivalent circuit of the PA, several circuit transformations are applied in order to build an appropriate artificial neural network structure and improve the modeling accuracy. This approach culminates in the development of a real-valued and feed-forward 2HLANN-based model. The parameters (number of neurons, memory depth, etc.) of the proposed model and the back propagation learning algorithm (learning rate, momentum term, etc.) used for its training were carefully studied and thoughtfully chosen to ensure the generality of the constructed model. The validation of the proposed models in mimicking the behavior of a 250-W Doherty amplifier driven with a 20-MHz bandwidth signal is carried out in terms of its accuracy in predicting its output spectrum, dynamic AM/AM and AM/PM characteristics, and in minimizing the normalized mean square error. In addition, the linearization of the Doherty PA using the 2HLANN enabled attaining an output power of up to 46.5 dBm and an average efficiency of up to 40% coupled with an adjacent channel power ratio higher than 50 dBc.

Physically Inspired Neural Network Model for RF Power Amplifier Behavioral Modeling and Digital Predistortion

Due to dynamic changes of supply voltage, envelope-tracking (ET) power amplifiers (PAs) exhibit very distinct characteristics in different power regions. It is very difficult to compensate the distortion induced by these amplifiers by employing conventional digital predistortion techniques. In this paper, by introducing a new piecewise Volterra model based on a vector threshold decomposition technique, we first set several thresholds in the input power level according to the PA characteristics, and decompose the input complex envelope signal into several sub-signals by using these thresholds. We then process each sub-signal separately by employing the dynamic deviation reduction-based Volterra series, and finally recombine them together to produce the predistorted output. Experimental results show that by using this new decomposed piecewise digital predistorter model, the distinct characteristics of the ET system at different signal power levels can be accurately modeled, and thus, the distortion, including both static nonlinearities and memory effects, caused by the amplifier nonlinear behavior can be effectively compensated.

Digital Predistortion for Envelope-Tracking Power Amplifiers Using Decomposed Piecewise Volterra Series

© Cambridge University Press 2007 and Cambridge University Press, 2009.This 2007 book is a comprehensive exposition of FET modeling, and is a must-have resource for seasoned professionals and new graduates in the RF and microwave power amplifier design and modeling community. In it, you will find descriptions of characterization and measurement techniques, analysis methods, and the simulator implementation, model verification and validation procedures that are needed to produce a transistor model that can be used with confidence by the circuit designer. Written by semiconductor industry professionals with many years' device modeling experience in LDMOS and III-V technologies, this was the first book to address the modeling requirements specific to high-power RF transistors. A technology-independent approach is described, addressing thermal effects, scaling issues, nonlinear modeling, and in-package matching networks. These are illustrated using the current market-leading high-power RF technology, LDMOS, as well as with III-V power devices.

Modeling and Characterization of RF and Microwave Power FETs

In this paper, we present an envelope-domain behavioral model of a high-power RF amplifier. In this modeling approach, we use the signal envelope information, and the behavioral model is generated using an established nonlinear time-series approach to create a time-domain model that operates in the envelope or signal domain. We have generated a model of a 200-W Doherty amplifier from measured IQ data taken using a wideband code-division multiple-access excitation; the amplifier was driven from the linear regime into saturation. The time-series model was created using a time-delay embedding identified from auto-mutual information analysis, and an artificial neural network was used to fit the multivariate transfer function. The model has been validated using measured and simulated data, and it has been used in the development of a system-level design of a digital pre-distorter

Envelope-domain time series (ET) behavioral model of a Doherty RF power amplifier for system design

A modeling procedure was developed to generate electrical package models for metal-ceramic packages These models are capable of accounting for package effects associated with the package lead capacitance, the self and mutual inductances of the bond wires, and the coupling between the input and output of the package A combination of full-wave electromagnetic simulation and equivalent-circuit model extraction allows accurate model generation and efficient circuit simulation Measured S-parameters were used to verify the overall modeling methodology It has been demonstrated that the package effects play an important role in the accurate prediction of the packaged transistor performance

Equivalent-circuit modeling and verification of metal-ceramic packages for RF and microwave power transistors

The use of substrate integrated waveguides (SIW) for microwave and millimeter wave integrated components has increased dramatically over the last decade. They mimic the performance of conventional metallic waveguides and they are fabricated using printed circuit boards using the top and bottom metallisation with two rows of vias forming the side walls. This creates a low profile, compact, and light weight alternative to conventional metallic waveguides, and they allow a direct interconnection with printed circuit boards and active components. This paper reviews the fundamental theory, documents the research that has been performed over the past decade, and summarises progress up to the recent state-of-the-art including novel SIW structures for passive circuits and antennas as well as new applications for reconfigurable and printed circuits using SIW technology.

Peter H. Aaen

Papers

Smart Modeling of Microwave Devices

Modeling and Characterization of RF and Microwave Power FETs

Envelope-domain time series (ET) behavioral model of a Doherty RF power amplifier for system design

Equivalent-circuit modeling and verification of metal-ceramic packages for RF and microwave power transistors

Recent Advances in Theory and Applications of Substrate-Integrated Waveguides: A Review