Interferer Detection and Cancellation in TDMA-Systems

Fredrik Nordström


Centre for Mathematical Sciences
Mathematical Statistics
Lund Institute of Technology,
Lund University,
2001

ISSN 1404-028X
Abstract:
In this thesis we investigate algorithms that reduce the influence of interference in receivers in current Time Division Multiple Access systems for mobile communications. We study the important case when the receiver has just one antenna since this is the most common case in mobile receivers of today. Further we have assumed that there is only one strong interferer.
First we determine if an interferer is present or not. We solve the problem by tests based on assumptions on normality and properties of the covariance function of the signal. With the assumption that the interferer transmits during several bursts in a row we can detect and reject the hypothesis of interference or no interference with high probability by using several of the tests in combination.
When an interferer is detected we detect its training sequence and its relative position to the desired signal. We have invented an algorithm that can detect the sequence and the position with high probability for interesting signal to interference ratio. The only assumptions are that the interferer must be active during several bursts in a row and that the drift in time of the interferer relative to the desired signal is slow.
Under the idealized case, i.e. synchronized interferer and signal and known interferer training sequence, we show what can be gained in bit error rates with two algorithms that are extensions of the Viterbi decoder, compared with the Viterbi decoder. These algorithms take the interferer into account and with that information the bit error rate is reduced dramatically for low signal to interference ratio.
We also show what is lost in bit error rates when the two training sequences only are partly overlapping compared with the synchronized case. Since the bit error rate strongly depends on the channel estimates we improve the channel estimates with a more powerful algorithm.
The most difficult channel estimation problem is when the two training sequences do not overlap. First we show that the naive iterative estimation method does not work. Then we give an algorithm that estimate both channel and symbols at the same time which leads to satisfying channel estimates. That algorithm can reduce the bit error rate dramatically for some cases of phase-shift modulated signal compared with the one that are obtained with the Viterbi decoder.