Stochastic Optimization of Space-Time Constellations

For wireless data communication systems employing multiple antennas, space-time codes play crucial roles for fast transmission of data with accuracy and bandwidth efficiency. Motivated by the large size of constellations of space-time codes and the resultant computational complexity, we develop a stochastic approach for the optimization of space-time constellations. We use union bounds of block error rate as performance measures of the space-time codes. To overcome the computational complexity, we propose to transform the performance measure into the mean of a bounded random variable and establish a statistical method for the estimation of such mean and its gradients with respect to parameters. A stochastic gradient descent method is developed for optimizing space-time codes. Such stochastic techniques are applied to obtain high performance space-time codes of large constellation sizes.