Analyzing Reversible Lapped Transforms using Reng Probing

Lapped transforms provide the fundamental analysis framework for most of the modern lossy audio compression algorithms. Recently, there has been increasing interest in compression algorithms that can encode audio to true lossless quality and yet allow lossy reconstructions from selected subsets of the encoded bitstream. To achieve lossless compression within a transform-based coding system, the transformations must be reversible-i.e., the forward transform must map integer inputs to integer outputs. In this paper, we use multirate Reng probing to analyze and compare the performances of the reversible modified discrete cosine transform (RMDCT) and the integer modified discrete cosine transform (IntMDCT).