In our globalised and digitalised world, fast and error-free communication is very important. To ensure this, we use error-correcting codes which should always be more efficient and have a higher error-correcting capacity. In this thesis, we investigate two types of LDPC codes which are obtained from Latin squares. The first is based on orthogonal Latin squares and the second is formed using Steiner 2-designs generated by a Latin square. For each code there is a bipartite graph in which cycles can occur. Such cycles can negatively influence the decoding efficiency. To counteract this, we will see how to locate and analyse these cycles and how to remove them.