1.- Givens rotation

2.- QR

Thanks for the reply, I tried implementing DCD algorithm proposed in this -  http://www.gamedev.net/page/resources/_/technical/math-and-physics/matrix-inversion-using-lu-decomposition-r3637 

but not getting the correct results.

can you suggest some reading for fixed point implementation of Givens rotation and QR decomposition.

Givens rotation uses division and square root operation, in both operations precision loss would be there in fixed point. will it still be good method for beamforming application ?

138

Generally, the most common and well-established matrix

inversion methods in the literature are  (Least Squares, QR Decomposition, Singular Value Decomposition (SVD)). However, each one them might be applicable for some situations while it is not for other cases  (i.e. exact matrix inversion can't be exactly obtained ).  However, the best method I have found in the literature so far has recently been published under this name "A New Approach for Inversion of Large Random Matrices in Massive MIMO Systems" . This method can give exact solution for matrix inversion problem without any conditions.  Kindly  see and download the attached paper

if the valúes of the correlación matriz are too small, you try to increase the amount of bits that you are using for represent the fraccionary part and to use full precission in your operations. Also, you can observe the SQNR after each math operation in order to identify where are you increase the fraccionary bits.

Dear Arpit,

Due to some numerical problems (such as near zero elements), matrices could appear nearly singular in MATLAB preventing their exact inversion. To get around this problem,

pseudoinverse (which is a type of generalized inverse) could be helpful. MATLAB does this through the "pinv" command which calculates the Moore-Penrose pseudo inverse of a singular or even a non-square matrix A. It is based on singular value decomposition (SVD).

Its use is very simple:

B=pinv(A);

as opposed to exact inversion

B=inv(A).

Have you considered using a stochastic gradient adaptive algorithm such as CLMS in order to avoid the matrix inversion?