234 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 47, NO. 1, JANUARY 1999 by the corresponding versions of the JADE algorithm, i.e., (21) and (23). The SINR results obtained by the proposed technique are presented by solid lines, whereas those obtained by the JADE algorithm are presented by dashed lines. The upper line in each pair represents the full-fledged version, whereas the lower line represents the computationally simpler version. In the first experiment, the sources were FM modulated with SNR of 10, 15, 20, and 5 dB, respectively. Fig. 1 presents the results for the source at Notice that the performance gain of the full-fledged version over the corresponding version of the JADE algorithm is 3 dB at and reduces to 0.5 dB at The second experiment was identical to the third except that the sources were QAM16 modulated. The results corresponding to the source at are presented in Fig. 2. Notice that the performance difference between the full-fledged versions varies from 5 dB at to 1 dB at In addition, notice that in contrast to the FM modulated signals in the third experiment, the QAM16 signals require more samples for the same performance level. ACKNOWLEDGMENT The authors wish to thank the reviewers for pointing out the identity of the derived criterion with the Comon criterion and for their suggestions and comments that improved the quality of the paper. REFERENCES [1] A. Bunse-Gerstner, R. Byers, and V. Merhmann, “Numerical methods for simultaneous diagonalization,” SIAM J. Matrix Anal. Appl., vol. 14, pp. 927–949, 1993. [2] J.-F. Cardoso, “On the performance of source separation algorithms,” in Proc. EUSIPCO, Edinburgh, U.K., 1994, pp. 776–779. [3] J.-F. Cardoso and A. Souloumiac, “Jacobi angles for simultaneous diagonalization,” SIAM J. Matrix Anal. Appl., vol. 17, pp. 161–164, 1996. [4] , “Blind beamforming for non-Gaussian signals,” Proc. Inst. Elect. Eng. F, vol. 140, pp. 362–370, 1993. [5] P. Comon, “Analyze en composantes independantes et identification aveugle,” Traitement du Signal, vol. 7, 1990. [6] , “Independent component analysis, A new concept?,” Signal Process., vol. 36, pp. 287–314, 1994. [7] A. Dinc and Y. Bar-Ness, “Bootstrap: A fast blind adaptive signal separator,” in Proc. ICASSP, 1992, vol. 2, pp. 325–328. [8] M. Ga´ eta and J. L. Lacoume, “Source separation without a priori knowledge: The maximum likelihood solution,” in Proc. EUSIPCO, 1990, pp. 621–624. [9] E. Gonen and G. Mendel, “Optimum cumulant-based blind beamforming for coherent signals and interferences,” Proc. ICASSP, 1995, vol. 3, pp. 1908–1911. [10] O. Shalvi and E. Weinstein, “New criteria for blind deconvolution of nonminimum phase channels,” IEEE Trans. Inform. Theory, vol. 36, pp. 312–321, 1990. [11] S. Shamsunder and G. Giannakis, “Modeling non-Gaussian array data using cumulants: DOA estimation of more sources with less sensors,” Signal Process., vol. 30, pp. 279–297, 1993. [12] A. Souloumiac and J.-F. Cardoso, “Comparaison de m´ ethodes de s´ eparation de sources,” in Proc. GRETSI, Juan-Les-Pins, France, 1991, pp. 661–664. [13] , “Performances en s´ eparation de sources,” Proc. GRETSI, Juan- Les-Pins, France, 1993, pp. 321–324. [14] L. Tong, R. W. Liu, V. C. Soon, and Y. H Fang, “Indeterminacy and identifiability of blind estimation,” IEEE Trans. Circuits Syst., vol. 38, pp. 499–509, 1991. [15] L. Tong, Y. Inouye, and R. W. Liu, “Waveform-preserving blind estima- tion of multiple independent sources,” IEEE Trans. Signal Processing, vol. 41, pp. 2461–2470, 1993. [16] M. Wax and J. Sheinvald, “A least squares approach to joint diagonal- ization,” IEEE Signal Processing Lett., vol. 4, pp. 52–53, 1997. [17] I. Ziskind and M. Wax, “Maximum likelihood localization of multiple Sources by Alternating Projection,” IEEE Trans. Acoust., Speech, Signal Processing, vol. 36, pp. 1533–1560, 1988. Non-Gaussian Characterization of DS/CDMA Noise in Few-User Systems with Complex Signature Sequences Andrea Teschioni, Claudio Sacchi, and Carlo S. Regazzoni Abstract— In this correspondence, an higher order statistics (HOS) based analysis of the non-Gaussian characteristics of the DS/CDMA global noise and a non-Gaussian evaluation of the expected BER, in few- user systems with complex signature sequences, are performed. Different error-rate performances provided by binary Gold and four- phase even-odd-equivalent-Gold (EOE-Gold) spreading sequences are also considered. I. INTRODUCTION The problem of the precision of the Gaussian approximation for the bit-error-rate (BER) in direct sequence code division multiple access (DS/CDMA) systems has already been studied by Yao [1], using the exponential statistical moment. Laforgia et al. [2], and Lehnert and Pursley [3] addressed the problem to evaluate the bit error probability to a high degree of accuracy. In [2] and [3], the focus is on providing upper and lower bounds for the BER and not on assessing a precise analytical expression for the BER. On the other side, in our work, the problem of giving a precise analytical measure of the BER, when the Gaussian assumption cannot be made a priori (i.e., in the case of few-user systems), is considered. The assumption of Gaussian noise background in few-user systems is unjustified and leads to inaccurate estimates of BER’s. This problem is overcome in the current paper by using a non-Gaussian parametric approximation to actual noise (i.e., a nonnormal symmetric distribution). The reported analysis is higher order statistics (HOS) based, as the value of a higher order statistic (the normalized kurtosis) is tested to characterize the deviation from Gaussianity of the global DS/CDMA noise in few-user systems with a BPSK modulation and complex signature sequences (Section II). The statistical analysis of the global DS/CDMA noise is restricted to the case of two users, i.e., a reference user and a single interferer. This is the worst case of violation of the basic hypothesis of the central limit theorem concerning the Gaussianity of the multiuser interference [9] (i.e., only one additional noisy component). In this sense, the results here presented about the non-Gaussian characterization of the global DS/CDMA noise can be considered one of the most significant examples of few users systems. A non-Gaussian pdf model, i.e., the generalized Gaussian pdf, whose sharpness is linked directly to the value of the normalized kurtosis [8], is used as parametric distribution to provide an analytical approximation for Manuscript received March 26, 1997; revised July 7, 1998. This work was supported by Italian National Research Council (CNR) within the framework of the Progetto Finalizzato Trasporti 2 (PFT2) project. The associate editor coordinating the review of this paper and approving it for publication was Dr. Jonathon A. Chambers. The authors are with the Signal Processing and Telecommunication Group, Department of Biophysical and Electronic Engineering (DIBE), University of Genoa, Genoa, Italy (e-mail: andretes@dibe.unige.it). Publisher Item Identifier S 1053-587X(99)00156-7. 1053–587X/99$10.00  1999 IEEE