Signal Processing Based On SVM And Antenna Selection In Multiple Elements Of Antenna Wireless Systems

The rapid development of the various services in the wireless communication such as the multimedia service requires more resources for communicating. The multiple elements of antenna (MEA) is an effective technique for using the space resource, which totally expands the traditional resources including frequency, time and code. Therefore, the using of MEA can improve the signal transporting and increase the system capacity. In CDMA systems with the smart antenna, the adaptive antenna array can maximize the antenna gain on the direction of signal with the realtime beam-forming of the transmited/received signal. So it can save the power and enhance the robustness of the system. As the extension of smart antenna, Multiple input multiple output (MIMO) system uses multiple element antenna in both the transmitter and the receiver, which makes the received signal more strong. The more important is that MIMO can simultaneously transmit different data independently in the rich scatter environment. Therefore, MIMO can be used to suppress the interference and increase the capacity of the CDMA systems effectively.However, the channel in the real communications is variational because of the different environment and the floating users, which shows that the wireless signal processing must be realtime. In smart antennas, the multiple access interference (MAI) can't be totally suppressed and it is more forceful interference than the environment noise although it provides the different beams for the different users. So the multiple user detection (MUD) as an intelligent method becomes one of the key techniques in CDMA system. Now, many of MUD methods have been suggested such as the optimal detector, decorrelating and MMSE MUD algorithms. But they are too complex to satisfy the realtime application, or the performance of them is limited which can't use the communication resources well. To resolve this problem, some more resultful means are raised such as the blind adaptive methods and the intelligent detection based on SVM introduced later.Support vector machine (SVM) is one of the most effective machine learning methods, which are based on principles of structural risk minimization and statistical learning theory. Recently, SVM has become a new advanced technique for the signal processing of the wireless communication system. However, the conventional SVM needs to resolve the NP-complex quadratic programming problem, the time resuming of which makes it difficult to apply the realtime processing, especially with the large training data set. As the result, some improved algorithms such as SMO, LS-SVM and so on have been suggested. But they have the limited application condition. So we raised the fast online support vector classification (FOSVC) algorithm which only uses the training data violating K.K.T conditions, which leads to faster speed. On the other hand, FOSVC can periodically retrain the various sample set that is updated by the new sample, which accords to the real signal processing. Simulation results showed that the FOSVC outperforms the other SVMs in term of the number of SVs and training time while keeping the comparable classification errors.the object of multiuser detection is to estimate the transmitted signals accurately as the received signals are known. This is a typical binary classification problem about the signal value (+1) and (-1). To solve this problem effectively with the given model, a multiuser detection method based on FOSVC algorithm is proposed in the dissertation. It gets the detector by retraining with the successively added new batch of data and don't need complex computing such as resolving the inverse of covariance matrix, which simplified the training procedure. Simulation results illustrated that the performance of the FOSVC detector was much better than that of MMSE detector because it can suppress MAI and noise well.As we have known, MIMO has robustness of signal and the higher capacity with using multiple element of antenna(MEA) in both the receiver and the transmitter. However, MEA needs more the cost of the RF chains hardware and increases the complexity of the signal processing. Antenna selection is a low-cost low-complexity alternative to capture many of the advantages of MIMO systems, which selects partial antennas from all antennas for the following signal processing. MIMO can be used to increase space diversity and channel capacity. Accordingly, antenna selection has different criterion and methods for the two targets. In real systems, the channels of MIMO is correlated which leads to the rank of channel matrix is reduced. Antenna selection only uses some effective antennas through the channel decomposition.The optimal algorithms of antenna selection is to find the best subset of the channel through exhaust searching, which has the exponential complexity degree and lead to be unfit for the realtime signal processing. To reduce the complexity of the the antenna selection algorithms with the high channel capacity, the antenna selection algorithms based on the eigenvectors are proposed in this dissertation with using the decomposition of the channel. The eigenvectors-based gradually elimination(EVBGE) selection algorithm iteratively erases the worst element of antennas until the rest is required. The simulation result illustrated the algorithm outperformed the optimal algorithm in wasting time, and whose outage capacity almost equaled to that of the optimal.In addition, the simple algorithm of EVBGE is also raised to make the complexity more lower, which only needs once SVD computing. The simple algorithm selects the elements of antenna according to the nonzero eigenvalues and those eigenvectors of channel matrix. It is equal to the norm-based selection(NBS) algorithm in eigen-space. The experiment result indicated that the simple algorithm had the best executing performance in three selection algorithms, and that its outage capacity was very closed to that of the optimal, which provides an alternative method for applying the antenna selection in the real systems.