Investigation on Data Identification Problem for Data-Dependent Superimposed Training

Abstract

In data-dependent superimposed training (DDST) scheme, the data-induced interference in channel estimation is eliminated at the sacrifice of data distortion. Unfortunately, data distortion causes data identification problem (DIP) for DDST scheme, which results in error floor phenomenon in bit error rate (BER). Although some literatures have analyzed the DIP, the DIP is still an open problem without solution. In this work, we firstly review the data identification problem from the view of sub-space. The analysis result inspires us to introduce a precoding matrix for resolving the DIP in DDST scheme. In order to prevent the advantages in DDST scheme being reduced by the precoding matrix, we introduce several constraints on the precoding matrix. Subsequently, we derive the requirement of the precoding matrix for solving the DIP in DDST system by using singular value decomposition. Furthermore, an appropriate precoding matrix is developed based on Zadoff-Chu sequence, which is shown to satisfy all conditions derived in this work. Finally, simulation results are conducted to verify that the precoding matrix removes the error floor in BER for DDST scheme.