A genetic algorithm for optimized reconstruction of quantized signals

This paper describes a genetic algorithm that evolves optimized sets of coefficients for one-dimensional signal reconstruction under lossy conditions due to quantization. Beginning with a population of mutated copies of the set of coefficients describing a standard wavelet-based inverse transform, the genetic algorithm systemically evolves a new set of coefficients that significantly reduces mean squared error (relative to the performance of the selected wavelet) for various classes of one-dimensional signals. The evolved transforms also outperform wavelets when subsequently tested against random signals from the same class