Abstract: In an amplify-and-forward (AF) cooperative relay network, the combined training scheme is proposed to acquire channel state information (CSI) of different links. The combined training mode consists time-division multiplexing (TM) training and superimposed training (ST), where the TM training is employed at source node to estimate the cascaded channel (S-R-D) while the relay training sequence is added on top of the information-bearing received signal to estimate the secondhop channel (R-D). In order to improve the estimation performance of R-D link, a novel cooperative interference suppression (CIS) scheme is proposed to remove cooperative data interference using detected symbols of S-D links. The new scheme employs combined training mode to make the individual channel and cascaded channel estimated independently, decreasing the complexity of training sequence design. Simulation results are presented to validate the performance of the proposed schemes.