In sphere decoding techniques, it was seen that the generalized sphere decoder algorithms have been applied to decode the MIMO systems. The transmitted vector is determined by decoding a sequence of determined sub problems.In this paper, the proposed sphere decoder has promised considerable performance as compared to the generalized sphere decoder. The proposed sphere decoding algorithm follows an adaptive radius selection approach for reducing the computational complexity as compared to the generalized sphere decoding algorithm and with the ideal ML decoder which applies on the QPSK signaling. Also it has proven that the proposed sphere decoder has given the performance near optimal like the ML decoder. The performance comparison is done under the Flat Rayleigh fading environment. We substantiate our new proposed reduced complexity near optimal sphere decoding algorithm with simulations.