LDPC codes have become the most popular error control code in various fields like telecommunication, magnetic recording etc. due to their high error correcting capability. The Approximate Lower Triangular (ALT) encoding is the most commonly used encoding technique of LDPC codes. This technique though elegant suffers from the shortcoming that when a particular sub matrix of the H matrix obtained after ALT encoding is singular, the technique fails. In this paper a new algorithm is proposed to handle this shortcoming. In this work a new algorithm is proposed to bring any rectangular sparse LDPC matrix into a rectangular part and a square upper triangular part. This algorithm is implemented to encode a general H matrix. This algorithm is further appended to the ALT format of LDPC encoding and another efficient encoding technique is proposed for LDPC code. Both the proposed techniques have a pre-processing step followed by the actual encoding step. Scatter plots of the H matrices (after preprocessing) are shown. BER performance of the new Modified ALT technique is compared with the existing Systematic Approximate Lower Triangular method. The proposed algorithm gives better BER performance.