by
D. Leonard Anair
Project Advisor: C. R. Johnson Jr.
Degree Date: January 1999
Performance of various linear multiuser detectors for Direct Sequence Code Division Multiple Access(DS-CDMA) is addressed in this Masters of Engineering project. A system model for DS-CDMA, including channel ISI(Inter-Symbol Interference), user asynchronism, non-orthogonal spreading codes, and additive white Gaussian noise(AWGN) is developed. Derivations of the Matched Filter Detector, Zero-forcing Detector, and the Minimum Mean Squared Error(MMSE) Detector utilizing the developed system model are presented. Adaptive detectors, utilizing the Least Mean Squares(LMS) Algorithm and Constant Modulus Algorithm(CMA) are also investigated. These detectors are implemented in a comprehensive software package for the purpose of performance comparison and analysis. The software includes a Graphical User Interface(GUI) to control received data generation and simulation of the implemented detectors. The generation of realistic receiver data includes parameters for varying SNR, spreading codes, power distribution, and multipath for both in-cell and out-of-cell users. Documentation of the software and the descriptions of the various detectors will be available on-line.
Executive Summary: The goals of this project as stated in the abstract, I am pleased to say, were achieved over the course of two semesters. The final software package contains implementations of three fixed detectors and two adaptive implementations, controlled by a graphical user interface. To develop a software package of this nature, I had to become confident with many details of communications systems which are easily overlooked in a classroom presentation of the topic. For this reason I have included in the report detailed descriptions and derivations of the detectors that I implemented.
Through extensive research, I developed a system model along with
other members of the BERG 
that was essential to implementing the different
detectors in software. The system model, which includes channel ISI and
arbitrary asynchronism, allows for relatively
straight forward derivations for the Zero-Forcing and Minimum Mean
Squared Error Detectors. The derivations of the various detectors is
addressed in the report. Along with the fixed detectors, LMS and CMA
were implemented as the adaptive algorithms. The LMS implementation proved
very useful due to the fact that the algorithm does not depend on the
individual user's spreading codes. The implementation of CMA is based
on a minimum entropy initialization technique and receiver
pre-whitening. Both adaptive implementations proved to be quite
successful in detection of users in a DS-CDMA system.
Another important part of the software was creating realistic channel models. A mobile communication system, which is the prime motivation for CDMA, has highly time-varying channels, as well as adjacent channel interference, adjacent cell interference, and in the multiuser case, multiple access interference. In order to make the task less daunting, time varying channels were left for the next version of software. A good model was still produced which include pulse shaping, varying power distribution, different spreading codes, asynchronism, and Additive White Gaussian Noise. Also included are the effects of multipath and out-of-cell interference. With this model, reliable simulation results were produced.
The Graphical User Interface proved to be a challenging design problem due to the number of parameters associated with multiuser detection. A modular design was developed for easy implementation and upgradeability. The final design consists of two sections, one controlling the system parameters and the other controlling the detectors and simulations. The software is all available on the web at http://backhoe.ee.cornell.edu/BERG/ along with help documentation.
Acknowledgements: I would like to say ``thanks'' to the few people that made this project possible: Rick Brown for helping me workout a lot of the gory details, even if it was on the golf course; Phil Schniter for giving me extensive Matlab advice on creating ROBO and SPANC-ME; Raúl Casas for letting me squat at his desk for 4 months; and Rick Johnson for inspiring me to take on this project and setting the occasional deadline to keep me motivated.