Performance of data-dependent superimposed training without cyclic prefix

A recent improvement in superimposed training for channel estimation, where the training sequence is actually added to the information data, is called data-dependent superimposed training (DDST). Here we show (theoretically and via simulations) that the performance of DDST (both for channel estimation and equalisation) may suffer minimal degradation when implemented without the use of a cyclic prefix (which carries only redundant information).