Algoritmo adaptativo para melhoria de desempenho do arranjo de antenas inteligentes 5G

Abstract

The LMS (Least Mean Squares) algorithm recursively calculates the weights of an array of smart antennas. In its conventional form, the LMS initializes its weight vector with zero values, performs its execution with the number of configured iterations and, after these iterations, generates a coverage array factor function for user and interference angles. After that, the algorithm uses the weight values from the last iteration of the previous run, as initials weights for the first iteration of the current run, in order to generate the new angles due to the displacement of the user. Theoretically, the greater the number of iterations, the longer the algorithm will need to converge to zero or to the smallest possible minimum mean square error (MMSE) between the output of the algorithm and the reference signal. However, this premise does not actually occur in practice after one or a set of many iterations. According to communication quality standards provided by the network, it is known what is the acceptable MMSE for communication between user and network to occur, even during the displacement of the user. This master’s dissertation work proposes the F-LMS algorithm (Fast - LMS), a modification of the LMS algorithm that ends its iterations at the moment the desired MMSE value is reached. When stopping the iterations at the time of the MMSE, the error value used to adjust the weights will be the smallest possible within the requests for network quality, up to the moment iteration, causing the output and coverage of the F-LMS have superior precision in relation to other algorithms. Due to the displacement of the user, the F-LMS adapts its coverage beam according to the new angular positions and, through the criterion of relationship between the magnitude of the coverage signal at the desired user angle and magnitude of the coverage signal at the interference angle, decide whether to reduce or increase the number of operating antennas in the array in order to save time processing and decrease the number of mathematical operations of the algorithm. The results were obtained through simulations in the MATLAB computational tool. The results obtained from the use of the F-LMS proposal were compared with the results obtained from the use of LMS, L-LMS (Leaky - LMS) and VSS-LMS (Variable Step Size - LMS) algorithms. The FLMS proved to be faster, more accurate and with a smaller number of mathematical operations than the LMS, L-LMS and VSS-LMS algorithms.