Carla Purdy

Hardware implementation of genetic algorithm modules for intelligent systems

A genetic algorithm (GA) is an intelligent search strategy supported by operations inspired by biological evolution. Although a GA is able to find very good solutions for a variety of applications, it typically requires many computations and iterations to be effective, and the amount of time consumed by these computations and iterations is enormous. Thus, software implementations of GAs applied to increasingly complex problems and large search spaces can cause unacceptable delays. An alternative to this approach is the hardware implementation of GAs in order to achieve tremendous speedup over software counterparts by exploiting the inherent parallelism of the GA paradigm. This paper presents the design of libraries of hardware modules for a GA system - one library in the Hardware Description Language (HDL) Verilog HDL and one in the language VHDL. Each library is based on a widely used MATLAB library

Sensor data processing using genetic algorithms

An off-the-shelf package is commonly used to generate an approximating polynomial from partial sensor data. The floating-point coefficients are rounded to the target architectures size. Rounding errors can actually be due to this solution space translation. A genetic algorithm can be used to find the optimal coefficient set in the restricted target space.

Data-specific signal denoising using wavelets, with applications to ECG data

We discuss performance improvements for a hybrid intelligent signal pattern classifier when denoising techniques are introduced before actually applying the classifier. We show how the efficiency of signal denoising for this classifier depends on several factors: the thresholding techniques chosen, the kind of wavelet used in denoising, the level of wavelets applied, and the synchronization between the wavelet selected and the input data. Results from experiments on ECG data which employ different kinds of wavelets (Haar, Daubechies, Symmlet and Coiflet) show that, for this data, level 3 decomposition using Symmlet wavelets with soft thresholding using the minimax principle gives the best results. We also show that further improvements in performance can be obtained by using vector quantization of wavelet coefficients before thresholding.

Combining genetic algorithms and neural networks to build a signal pattern classifier

In this paper we show how genetic algorithms and neural networks are combined to build a high-performance signal pattern classifier (GNSPC). Signal patterns are intrinsic to many sensor-based systems. The goal of GNSPC is to differentiate among large numbers of signal pattern classes with low classification cost and high classification performance. Classification performance is measured by the correct classification of noisy signal patterns despite using pure signal patterns for building the classifier. GNSPC is basically a decision tree classifier with similarity classification rules. The rules are used to test the similarity of signal patterns. A combination of a genetic algorithm and a neural network is used to find the best rules for the decision tree. This combination provides powerful classification capabilities with great tuning flexibility for either performance or cost-efficiency. Learning techniques are employed to set the genetic algorithm global parameters and to obtain training data for the neural network.

Design automation of MEMS systems using behavioral modeling

We propose a behavioral approach to designing MEMS devices. This approach differs from much current research in that this approach would not require dimensional parameters for the device, but instead would require a high level, functional or behavioral description. This paper examines how such an approach would work using a case study of an optical processor manually designed using the MUMPs process.

Bio-inverter model and interface to digital hardware

Conventionally, VLSI circuits are built using CMOS transistors and the circuits are fabricated on silicon. But the looming crisis created by continually shrinking CMOS features sizes is pushing scientists to research alternate technologies for future generations of circuits. One active field of research is the development of bio-circuits. In this paper we discuss one such bio-circuit, the bio-inverter. This inverter is implemented using the switching mechanism observed in the phage lambda virus. This switching mechanism has been well documented in work by Ptashne and by Weiss. We have developed a model for such an inverter both in Matlab and also in VHDL-AMS. Our simulations show that the bio-inverter possesses good rise and fall time characteristics, switching threshold and gain, even though the gate switching time is approximately 10/sup 2/ seconds. We have also discussed the practical limitations of building bio-circuits and their applications as sensors for drug research.

A multistrategy signal pattern classifier

Signal patterns are intrinsic to many sensor-based systems. Here we present a multi-strategy signal pattern classifier MSPC that can differentiate among large numbers of signal pattern classes with low classification cost. MSPC combines decision tree concepts, fuzzy rules, genetic algorithms and neural networks to achieve its goal. This combination provides powerful classification capabilities with great tuning flexibility for either performance or cost-efficiency. Fuzzy rules measure similarities between signal patterns. A genetic algorithm finds the best fuzzy classification rule. Neural networks are used in the final classification step and in noise elimination. Machine learning concepts are applied to set the overall algorithm parameters.

Approximating functions for embedded and ASIC applications

Often embedded programmers and application specific integrated circuit (ASIC) designers are frustrated by the inability to realize near floating-point accuracy. in a fixed-point application. The problem is not limited to function approximation but also impacts FIR filter design. In this paper we examine the problem of approximating a known function on a closed interval and show that a genetic algorithm (GA) may be used to obtain results superior to those obtained by implementing floating-point algorithms, such as Taylor series and Chebyshev polynomials, in fixed-point directly.

Efficient FPGA implementation of a generic function approximator and its application to neural net computation

Typically, digital sigmoid implementations for neural nets have low accuracy or unwieldy memory requirements. The authors presented a highly accurate, memory-efficient sigmoid calculator, designed using a genetic algorithm. The VHDL design, implemented in an Altera Flex10K device, is easily reconfigurable for any sigmoid slope or for computing other required system-on-a-chip functions

Formal methods for safety critical system specification

For safety critical systems, hardware is often preferred over software because it is easier to achieve safety goals in hardware alone and because hardware is considered more reliable than software. But as systems become more complex, software solutions will also be important. Here we demonstrate, using a simple example, that formal methods are a useful tool for developing software specifications for safety critical systems, since they reduce ambiguity in the design and can be proven consistent. Using formal methods for specifications will enable the development of dependable, high-performance, reliable hardware/software safety critical systems. The method we describe is the first step in our work to establish a hardware/software development process for safety critical systems.