Fast blind adaptive equalisation for multiuser CDMA systems

  • Adel Daas

Student thesis: Doctoral Thesis

Abstract

In order to improve communication over a dispersive channel in a CDMA system, we have to re-establish the orthogonally of codes which are used when combining input signals from many users onto a single communication path, as otherwise the performance of such system is limited significantly by inter-symbol interference (ISI) and multiuser access interference (MAI). In order to achieve this, adaptive filters are employed. A variety of adaptive schemes to remove ISI and MAI have been reported in the literature, some of which rely on training sequences, such as the Least Mean Squares (LMS) and Recursive Least Squares (RLS) algorithms, or on blind adaptation, such as the Constant Modulus Algorithm (CMA) or the Decision Directed algorithm (DD), which has similar convergence properties as the LMS in the absence of decision errors, the CMA is relatively slow compared to the DD algorithm but more robust in converging to a suitable solution. This thesis is concerned with developing a new robust and low-complexity blind multiuser equalisation over frequency selective channels. A robust pilot-assisted equalisation strategy is developed for the partially loaded time division duplex (TDD) component of the universal mobile telecommunications system (UMTS). In addition to training-based equalisation performed using the midamble of a data packet, some of the unused spreading codes are exploited to upload pilots in order to perform anadditional semi-blind adaptation over the payload of a packet. The latter ensures continuious adaptation and better tracking performance. The affine projection concept along with the concurrent constant modulus algorithm (CMA) and decision-directed (DD) mode are implemented to update the equaliser weights. Computer simulations are used to assess the performance of the proposed adaptation strategy over various UMTS TDD time bursts.A new low complexity adaptive technique is derived for blind multiuser equalisation basedon fitting the probability density function (PDF) of the equalizer output to the desired PDF of the corresponding symbol alphabet, i.e. matched-PDF. The cost function of the proposed technique can be measured by a stochastic gradient descent approach. The performance of the proposed adaptation strategy is assessed by a number of simulations, and benchmarked against FIRMER-CMA under QPSK modulation. The matched-PDF algorithm is used for the equalisation of Space-Time Block Coding (STBC) and Time-Reversal Space Time Block Coding (TR-STBC) signals transmitted over dispersive MIMO channels. The performance is demonstrated in a number of simulations and benchmarked against other blind schemes such as: CMA, Newton's method, and the Conjugate Gradient method. A thorough evaluation is carried out taking into consideration the complexity of each implementation in terms of multiply-accumulate (MAC) operations required per iteration. Finally, some variations of matched-PDF algorithm are proposed to improve the equaliser performance, including concurrent matched-PDF and decision directed, matched-PDF with affine projection algorithm, as well as pilot assisted equalisation based on matched-PDF algorithm for partially loaded systems.
Date of Award1 Oct 2009
LanguageEnglish
Awarding Institution
  • University Of Strathclyde
SupervisorStephan Weiss (Supervisor) & Lina Stankovic (Supervisor)

Cite this