Mohit Garg

Dual Degree Student

Title:  Multi User Detection in Wireless Communication

Supervisor(s): Prof. U. B. Desai

Abstract:

In a multi-user environment, the performance of DS-CDMA communication systems is limited by the interference caused by other users (Multiple Access Interference). Multi-user signal processing schemes provide a way to overcome the effects of MAI and improve performance. Demodulators based on the Minimum Mean Squared Error (MMSE) criterion are optimal for ‘ideal’ AWGN channels. However, in the case of multiple access wireless multipath channels, wherein MAI and ISI are inherent, MMSE based algorithms do not offer the optimal framework for demodulation. Minimum Probability of Symbol Error (MPOSE) based detectors have been shown to perform significantly better than MMSE based approaches in these scenarios under a variety of modulation and multiplexing schemes.

In this work, we first propose a modification to two already published receiver based multi-user MPOSE algorithms whereby we not only achieve significant gains in computation but also improve the BER  performance – Multi-user Detection.

Next, in order to reduce the complexity of the receiver (which is usually mobile, and hence resource constrained, in the downlink), we develop two algorithms using a linear FIR pre-coding filter for jointly minimising the Probability of Error (MPOE) for all users – Multi-user Transmission. The first algorithm assumes full channel knowledge at the transmitter. The second filter design, on the other hand, is based only on the statistics of the channel (statistical channel model) and hence the knowledge of the actual channel coefficients is not required at the transmitter. This compromises on performance but saves on critical bandwidth on the reverse channel which is usually required for forward channel feedback to the transmitter.  Since the underlying theme is reduced receiver complexity, we consider a simple conventional single user detector at the receiver. In order to fully utilise the knowledge available at the transmitter, the filter weights are computed conditioned on the transmitted bit vector sequence. This also makes the computation of the optimal coefficients linear in the number of users as opposed to the exponential complexity otherwise.

Simulation results show that the proposed system exhibits fast convergence and performs well with just three taps in the transmit pre-coding filter. The performance of both the schemes is also significantly better than similar adaptive MMSE (steepest descent) based pre-filters. It is also shown that similar pre-filters cannot be optimised using RLS/LMS since a stochastic gradient cannot be computed at the transmitter.

Contact Address:   SPANN Lab,

                             Department of Electrical Engineering,

                             IIT Bombay,

                             Powai, Mumbai- 460 076

Email id: mohitgarg[at]ee.iitb.ac.in

Homepage : http://www.vectorstar.net/~mohitgarg