a second-order backward adaptive lattice structure predictor is used in aac main profile, and it requires a large amount of computations and memory, among some AAC decoder (ex:FAAD,FFMPEG),the predictor states are stored as float, and its range acrosses large, who knows how to implement the prediction by fixed-point? Your any advice will be much appreciated.
AAC Main Profile is dead and forgotten, so you might not have much luck finding someone who has worked with it.
Any reason you're interested in it?