Blind identification of a multiple input single output multiuser channel

CECILIA GALARZA

Rosario

Conferencia; Workshop on Signal Processing, WSP06; 2006

Universidad Nacional de Rosario - IEEE Signal Processing Society

A multiuser channel is composed by mutually interfering digital information streams. The superposition of transmitted signals may be due to nonideal characteristics of the transmission medium, as in high speed digital data transmission through phone cable, or may be part of the modulation scheme, as in wireless transmission like CDMA. The present work deals with multiple input single output (MISO) channels. This is the case when a single receiver has to decode the mixture generated by multiple users sharing the same transmission medium. Generally, these systems are underdetermined because they have fewer degrees of freedom than users. Also, time and frequency resources are not sufficiently separated among users to conveniently split the information coming from each one at the receiver end. This lack of diversity makes decoding a complex task. Nonetheless, receiver design techniques require estimating the MISO channel dynamics. Note that for a better use of the available bandwidth the identification should be performed on a blind or semi-blind fashion, adding complexity to the whole problem. In the past, several alternatives have been proposed to solve the channel identification problem. In particular, C. Aldana and J. Cioffi presented a solution for a multicarrier modulation (MCM) system when all the users are synchronized. In this case, each user has a known finite alphabet to transmit the information through M carriers. All users transmit though the same carriers and the dynamics of the MISO channel may be decomposed in M ideal MISO channels in parallel. For this case, the authors propose an iterative identification scheme using the Expectation Maximization (EM) algorithm. The present work follows these lines of thinking. First, an alternative algorithm using the LMS algorithm is proposed. It is shown that although the initial solution using the EM algorithm has a better performance than the option with LMS, both performances are similar for high signal to noise rations. Hence, the identification strategy using LMS becomes a feasible low computational cost strategy for certain range of SNR. On a second part, we revise the hypothesis of the original problem. Specifically, we analyze two instances of non-synchronization among users. On the first case, the users have different unknown delays; on the second case, the symbol periods of the users are different. For these situations, appropriate models are obtained. It is shown that when there are different delays for each user, a new LTI dynamics may be added to the system to account for this situation. Finally, a new algorithm to identify the original MISO channel and the delays is proposed.