Texas Instruments article page 1 A Parallel DSP architecture running Genetic Algorithms applied to acoustic noise cancellation. Jose Sotelo Jr. email: sotelousp.br Silvio Bistafa James Cunha Werner email: jamwerusp.br Eduardo Ribeiro de Castro Ronaldo Coza Giorgi Depto. Eng. Mecanica USP Av. Prof. Mello Moraes, 2231 CEP 05508-900 So Paulo- SP Brazil tel. (55)(011)8185330 fax (55)(011)8131886 University of So Paulo ABSTRACT: This paper addresses the problem of actively control acoustic noise in ducts through the application of genetic algorithm. Conventional techniques, like adaptive or classical control, have convergence problems due to the acoustic feedback, propagation delay and secondary paths. The software was designed to work in a parallel DSP TMS320C44 architecture, managing real time interrupts and communication with a shared memory. An UNIX simulator was developed with a simplified model to test the software. The noise reduction obtained in simulations is around 40 dB and the experimental results is 20 dB down. 1. INTRODUCTION Sound pollution is an emerging problem everywhere, and our objective is the development of an algorithm that permit the noise cancellation through the generation of an appropriate signal (active noise control). The superposition principle establishes that two waves would be added, canceling their pressure effect into the space. This could reduce the sound physical feeling. We select an experimental setup with a rectangular duct, because it has a nice experimental condition (plane waves), nice boundary condition for acoustic, easy construction and a mathematical models solution using the variable separation methods (Fig.1). The real time control require an algorithm with fast convergence in order to adapt itself for noise time variations, due to temperature and pressure fluctuation, acoustic feedback and broadband features. The extensive use of DSPs require a development methodology to help the designer to look for the bottle necks, the dead locks and the job division between the processors. 2. THE GENETIC ALGORITHM. The evolution and improvement of life occurs through reproduction, when each individual contributes with its own genetic information building a new one with fitness to the environment and more surviving chances. With the objective of obtaining the sound components to be used for the cancellation of undesirable noise in the interior of ducts, a computer software which mimics the genetic process was developed. Each individual of the generation represents a signal component with a different frequency, phase, and amplitude. It creates a population randomly distributed and evaluates the fitness of each individual. The best individuals are selected, and crossover and mutation take place. Following few generations, the population converges to the mix of components that better reduce the acoustic noise. Two different approaches based on genetic algorithms (GA) were studied. In the Simple Genetic Algorithm (SGA) each individual of a generation represents a specific frequency, phase and amplitude used in cancellation of noise, and the fitness function is the average energy of the signal. The Successive Approach Genetic Algorithm (SAGA) is a modification of SGA, where a first level procedure searches for most probable candidate frequencies and a second level improves them between fixed limits. Texas Instruments article page 2 3.TRANSACTION DIAGRAM. There are different diagramming techniques to develop mono-processor commercial software: data flow diagram (DFD), functional decomposition, Warnier-Orr diagram, flowchart, Nassi-Schneider action diagram ER,etc. Their application into conventional processing and with relational databases were largely studied and automated. The main feature is the presentation of data flow and processing sequence. Yourdon began to develop a graphic methodology, pointed to the inadequacies of a purely data -flow oriented approach to real-time systems designs. Parallel algorithms were presented into structured English with a hard view of critical processing situation. Our proposal is the definition of some rules and primitive symbols to design real-time parallel software, where resources contest, CPU waiting and deadlocks are under control and visible to the programmer. They represent the data flow, state sequence of each processor and the interactio n between the processors into the parallel real-time processing. A computer program is an ordered sequence of statements and procedures to answer to external events. A processor changes its state continuously, and each state is associated with a transaction. The transaction diagram is a graphic presentation where each state is specified and its processing is detailed into the horizontal way, beginning into the state definition and finishing into the conditional flow control. Vertical flow occurs only with conditional flow control. SGA and SAGA transaction diagrams are drawn int