Data Identifiability for Data-Dependent Superimposed Training

In channel estimation based on Data-Dependent Superimposed Training (DDST) certain frequency components are removed from the data symbols, prior to transmission. Since this means information is removed at the transmitter, the receiver may not find it possible to correctly recover the data. In this paper conditions for data identifiability are given when using a QAM constellation, and an analytical expression for the likelihood of correct detection is given for the noise-free case. A new detection method is then proposed, that can allow the use of larger constellations, and its performance is compared to the existing method.