In the context of wideband communication systems, we can find communications channels modeled as either MIMO (Multiple Input Multiple Output) systems where several antennas are used in the transmitter (inputs) and several ones in the receiver (outputs), or one channel systems that can be modeled as the previous systems (multi-carrier systems or inter-carrier interference multichannel systems, multi-user systems with one or several mobile terminal antennas and optical communications over multimode fiber). These systems expect to reach bandwidth relative transmission capacity greater than SISO (Single Input Single Output) channel model. Nowadays, there exists an important researching activity, from the system implementation point of view, devoted to the development of a high complexity coding, equalization and \emph{detection} algorithms that help to get the promising capacities. The nowadays solutions to the \emph{detection} problem can be classified in three categories: suboptimal, ML (\emph{Maximum Likelihood}) or cuasi-ML, and iterative solutions. The last one uses explicit error control techniques with \emph{soft} information exchange between the detector and the decoder. In the ML or cuasi-ML solutions, a tree searching technique is used that can be optimized in order to get a polynomial complexity cost within certain signal-to-noise ratio interval. We can find in the suboptimal solutions, the zero forcing technique, minimum mean squared error and successive interference cancellation (SIC) techniques and its ordered version (OSIC). The suboptimal solutions do not reach the ML or cuasi-ML performance but they can provide the solution in a polynomial time. We have implemented in this thesis a method based on the current literature to solve the OSIC problem. Besides, we have developed and implemented an original method to solve this problem with better performance. This is one of the most interesting contributions of our work. The implementations have been parallelized, evaluated and compared. Both OSIC methods are based on methods that solve the recursive least squares problem (RLS) and are inspired in the Kalman filter. These methods have been studied, parallelized, evaluated and compared. Besides, an additional method based on the QR factorization updating to solve the RLS problem has been developed. The design of all the parallel algorithms has been oriented to a pipelined organization due to the intrinsic sequential character of the algorithms, the good efficiency got with this technique and the potential extrapolation to VLSI implementations.