Independent component analysis for blind signal separation in MIMO-OFDM systems

In the communications industry, the endless struggle engineers are faced with is how to increase the data rate for reliable transmission with the constraint of a fixed bandwidth. Claude Shannon, often referred to as the father of information theory, derived a theoretical limit for the capacity of communication systems with additive white Gaussian noise. With this concept defining the best possible performance for a given communications system, designers strived to create a system capable of results closer and closer to this upper bound. The limit however, is based upon a system consisting of one transmit and one receive antenna. As it turns out, if multiple antennas are employed, the capacity increases proportionately to the number of antennas used. The system setup known as Multiple-Input Multiple-Output offers significant improvement in bandwidth utilization which warrants its use in new wireless standards such as 802.11n. In addition to the spatial diversity provided by MIMO, improved bandwidth efficiency can be achieved via frequency diversity through a multiplexing scheme. For the current 802.11g and new 802.11n standards Orthogonal Frequency Division Multiplexing is used.; The wireless MIMO channel is initially unknown to both the transmitter and receiver. As the signals traverse the link they are mixed and attenuated. What is obtained from each receiving antenna is a weighted combination of each transmitted signal. The primary task then is to separate the received signals into the original data streams. Many methods exist that accomplish this task, but most incorporate some sort of training sequence which is used for channel estimation. The training data use bandwidth that could be used to transmit data bits if the channel estimation can be performed without this extra data and the demand exists for algorithms that perform data recovery in a blind manner where the use of training data is not necessary.; This thesis contributes such a blind scheme for signal separation. Independent component analysis is used to separate the data streams in MIMO-OFDM frequency selective channels given the valid assumption that the original symbol streams on separate antennas are mutually independent and non-Gaussian. The use of OFDM as a signaling scheme leads to the analysis of the system in the frequency domain which presents the problem as a set of linear systems of equations for different frequency points which are solved using ICA.; Fractional sampling in the frequency domain is presented as a method for increasing the number of data points with which the results of the ICA algorithm are improved. The improvement is achieved by combining the oversampled data with the Nyquist rate data to increase the systems resiliency to channel noise.