Study On New Technology For DS-CDMA Blind Multiuser Detection

Code Division Multiple Access (CDMA) technology became one of thecommunication technologies with developing prospect, and has being applied tomilitary communications and civilian communications extensively. But CDMA is aninterference-limited system, where Multi-Access Interference (MAI) is existed as themain factors restricting CDMA system capacity and performance. Because it is difficultfor spread spectrum codes of users to simultaneously acquire excellent self-correlationcharacteristic and cross-correlation characteristic. Multiuser Detection (MUD)technique makes full use of information containing MAI to combine with signalprocessing technologies. It is resistant to MAI and Near-far Effect, lowers therequirement of power control, and increases the capacity of CDMA system. Blindadaptive implementation of MUD is an attractive research hotspot because it doesn’trequire training sequences and interference information of other users.In this paper, we study on the blind multiuser detection (BMUD) for DS-CDMAsystem. The traditional blind adaptive multiuser detection algorithms based on LMS andRLS are improved. A new blind multiuser detector based on regularization method isproposed and analyzed theoretically. To solve the ambiguity of conventionalIndependent Component Analysis (ICA), an improved algorithm for blind multiuserdetection based on regularized ICA is put forward.The author’s major contributions are summarized as follows.1. For blind multiuser detector based on Constraint Minimum Output Energy(CMOE), we derived another representation on the basis of canonical presentation one,which is obvious in physical significance and easy for adaptive implementation. Basedon it, we realized blind multiuser detection using variable step size normalizedleast-mean-square (NLMS) algorithm. NLMS algorithm is fast and robust, which isbetter than the fundamental LMS algorithm. Meanwhile, we also proposed a variableforgetting factor RLS (VF-RLS) algorithm to adaptively implement BMUD, which hasfaster tracking ability of time varying parameters and better steady-state performance. Itis suitable for dynamic environment application.2. Regularization theory was creatively applied to CDMA blind multiuser detection.We proposed several blind multiuser detection strategies based on regularization. Theadvantages of BMUD based on Tikhonov regularization, which is also called diagonalloading, are robust and have some tolerance to signature waveform mismatch. We alsogave generalization form of Tikhonov regularization BMUD. Otherwise, by modifying condition number of covariance matrix, such as multiplying covariance matrix elementsby different weights, attenuating elements far from main diagonal, which is calledcovariance matrix tapering (CMT), also belongs to regularization theory.3. Performance of BMUD based on Tikhonov regularization was analyzed intheory. For some special cases, such as equal correlation signals with equal interferencepower, equal correlation signals with ideal power control and orthogonal signals withideal power control, expressions are given with Signal Interference and Noise Ratio(SINR) as the performance measure. And we also analyzed the effect of each parameteron its SINR performance. Some important conclusions are obtained. Simulations arecarried out to compare bit error ratio and verify that the Tikhonov regularization BMUDhas some tolerance to signature waveform mismatch.4. The traditional Independent Component Analysis (ICA) method has inherentdrawbacks, i.e., permutation ambiguity and amplitude ambiguity. And it doesn’t makefull use of the given information, such as signature waveform of the desired user. Wemodified it, and introduced regularization functional in the cost function, whichcontained the prior information of desired user’s spread spectrum code. It can improvethe ambiguity problem and accelerate the convergence of the algorithm. We usedstochastic gradient descent method to optimize the cost function, and obtained thealgorithm of estimating detector weights. Regularized ICA (regularICA) method isflexible and robust, since we can choose different cost functions and regularizationfunctional under different optimization rules. It doesn’t need data whitening, and is ofsimple calculation.