Study On BSS Methods And Applications In DS-CDMA Multiuser Detection

In recent years, the study of blind source separation methods has become a noticeable problem in the signal processing fields, and related technologies have also been applied to various fields, especially using it to process DS-CDMA multiuser detection. The communication of DS-CDMA has its merits, such as low probability of intercept, strong anti-interference ability and achieving code division multiple access, so it has been widely applied in the anti-interference communication and civilian mobile communication. The most important task of it is the blind detection and separation of DS-CDMA signal in the noise and interference environment. Traditional detection technology carries out spread spectrum code matching respectively to each user's signal absolutely according to the classic direct sequence spread spectrdm theory, so it's less able to interfere with anti-MAI.In this thesis, applied research has been done based on the depth discussion of the blind source separation method. Two improved algorithms based on the theory of multiuser detection of DS-CDMA have been proposed, and meaningful results have been gained combined the algorithm with blind multiuser detection based on the theory of statistics signal processing. The main related contents and conclusions can be summaried as followings:(1) Related theory and methods of ICA and PCA have been discussed aimed at BSS problem. Natural gradient ICA learning algorithm and it's properties have been discussed; the theory and method of the whitening processing of the signal using PCA method have been studied; the signal, which is mixed by two ideal sequence signal without noise interference, has been separated by natural gradient ICA learning algorithm, and mixed sound of the actual car traveling conditions has been separated. The experimental results show that the time-domain signal didn't change before and after separation, and the signal amplitude increased by the separation of the natural gradient ICA algorithm, and it was found that the volume of the separated voices became significantly larger by contrasting through audition. Natural gradient ICA algorithm can effectively eliminate the background noise at the same time of preservating of most of the speech signal energy, and the separated spectrum was more similar to the pure language of the speech signal spectrum. The natural gradient ICA algorithm could achieve better effect when it was applied in the actual BSS fields.(2) Many BSS algorithms have worse effect when they proceed blind source separation to the high-dimentional mixed signal. This thesis has discussed the linear mutilayer ICA algorithm, and analyzed its stability, and also compared the property of MaxKurt algorithm with it. The basic idea of the linear mutilayer ICA algorithm is that local ICA stage and mapping stage were performed in every layer of the algorithm. For local ICA stage, every neighboring mapping stage was separated by MaxKurt algorithm which was in the ICA algorithm. For mapping stage, one dimensional mapping was formed by overall mapping analysis and higher relevance signal. By implementing this two processes cyclarly, linear mutilayer ICA algorithm could almost extract all the dependent components. The experimental results show that, linear mutilayer ICA algorithm has higer efficiency than other standard ICA algorithms.(3) In order to better predict and improve multiuser detection problem of DS-CDMA, the mathematical model of multiuser DS-CDMA signal and the form of observational data were given, and the model of observational data was given by combining with the spread of multipath and the influence of fading signal to the systems. Based on the DS-CDMA downlink of natural gradient, BMUD can carry out blind estimation directly to all user's symbol sequences by feedback BMUD technology from the received CDMA signals. This technology can be convolution, or even blind source separation signals of non-linear environment. For feedback BMUD their structures can be constrainted by natural gradient algorithm and can apply to the linear convolution of mixed environment by minimizing the mutual information. Forward and backward feedback structures are both suitable for the natural gradient BMUD. Combining of specific update rules of forward and backward feedback structures, algorithm of based TDS feedback DS-CDMA multiuser detection has been proposed, and DS-CDMA model of existing multichannel access has been improved. The basic idea of the improved algorithm is: in the processing DS-CDMA signal of the downlink channel, continuous data firstly sampled in the rate of chip, making each symbol becomes C interval samples. The width of dealt window was two symbols, and these interval discrete data samples R[m] form a 2C-dimensional vector Rm. Since the sampling process and the symbol is asynchronous Rm contains 3 symbols which emerged one after another, and each user is identified according to the unique symbol, and this symbol together with other allocated users' symbols are quasi-orthogonal, and original symbol is estimated by the information of both the original symbol and the nearest two symbols before and after the original symbol during the DS-CDMA transporting. So we can see that DS-CDMA signal system model expressed linear mixed of delay and the source of the convolution with a certain period of time delay. Therefore, by using the maximum entropy principle and the natural gradient algorithm of convolution mixture, we can take feedback system model structure for processing to this semi-blind source separation problem which mixed matrix and symbol sequences are both unknown. In the existing algorithm, learning parameterμwith iterative process-related is the key point of influencing algorithm's convergence, so in the proposed based TDS feedback DS-CDMA multiuser detection, learning parameterμis set to value, and this decreased the complexity of the calculation, and this also could make algorithm faster convergence. The simulation experiment results show that the proposed improved algorithm can match better with the DS-CDMA model, and it is effective to the DS-CDMA multiuser detection problem. It also can detect user symbol exactly with lower undetected rate and better practicability.(4) Under the channel environment of MAI and high SNR, combined the mind of improving algorithm based TDS feedback DS-CDMA multiuser detection, DS-CDMA blind source separation system was constructed based the method of FastICA-TDS. The basic idea is: each user occupies the same frequency band simultaneously in the CDMA system. Symbol is unique for each user, so user is identified by their symbols. When the public frequency band delivers signals, they should put the user's unique symbols into the data symbols. Different users's signals are mixed together during the transmission. The unique symbol is detected by the signal receiver, each user's delivering signals are identified from the mixed signals. This process is similar to the process that ICA separates the mixed signals. Firstly, we change the traditional DS-CDMA signal model into the basic model of ICA. Secondly, mixed signals are separated by the FastICA algorithm. The final vector derived from algorithm is equal to one row of the orthogonal mixed matrix which is composed of user's chips from the DS-CDMA system, which means the separate a non-Gaussian signal. This signal is one of the user's symbols. If the two adjacent received values is equal or very little difference, the iteration process will be end. In addition, each symbol should be subtracted from the mixed-signal after the symbol being extracted, so it is implementation until all the symbols are separated. Because FastICA algorithm is non-gaussian and can eliminate high-order correlation of the input data, so the proposed DS-CDMA blind source separation system which is based on the FastICA-TDS method has robustness to gassian noise. Also because the signal which is delivered in the transmission channel is genenal from a binary number of a limited alphabet, moreover the definition of the symbol expantion is known, so the blind source separation to DS-CDMA is with priori information, is not blind. Different detector's BER properties are compared by experiment, and the results show that the system has lower undetected rate, and is obviously better than MMSE and MF in the system power controlling. DS-CDMA blind source separation system which is based on FastICA-TDS is certain novelty and ingenuity, and has better referential significance to the researchers of this field.