This work is based on a study of blind source separation techniques in order to estimate coefficients in transmission systems using Alamouti codification with two transmit antennas and one receive antenna. Most of present standards include pilot symbols to estimate the channel in reception. Since these symbols do not deliver user's data, their use decrease transferring quantity and also the system capacity. On the other hand, algorithms of blind separation are less precise when estimating channel coefficients than those supervised, but achieving a higher transferring rate.
In this work we will deal with Alamouti codification system as a typical problem of blind sources separation where the signals transmitted and the channel coefficients must be estimated according to lineal and instantaneous mixtures (observations). Orthogonal structure required by Alamouti codification allows us to solve this problem by decomposing eigenvalues and eigenvectors of matrices calculated from different statistics of the observations. These algorithms could be classified as those using second order statistics and those using higher order statistics.
Algorithms based on second order statistics work with correlation matrix of observations. They are computationally less expensive, but require a lineal precoder in order to balance the power of the signals transmitted. One of our contributions is being able to determine in an empirical way how the power decompensation should be done in order to reduce the proabibility of error in the system.
On the other hand, algorithms dealing with high level statistics are based on diagonalize one or several high level cumulant matrices deriving into a major computational cost in the receiver. As an advantage we must point out that they do not require to include a lineal precoder to do the power decompensation. In this work we will prove that the output of these techniques depends on the level of eigenvalue of the diagonalized matrix spreading. This idea will be used by us in order to achieve the optimal cumulant matrix and also to propose a new algorithm that increases the output in relation to those already proposed by other authors.
Another important contribution of this present study is to propose a detailed comparison between channel estimation techniques in simulated scenarios, considering channels with Rayleigh and Rice distribution, and in real scenarios in ISM of 2.4 GHz band, by using a MIMO testbed developed in Universidade da Coruña.