Abstract-In this paper, the superimposed training strategy is introduced into the OFDM modulated amplify-and-forward (AF) two-way relay network (TWRN) to simplify the channel estimation at the destination, and the closed-form Bayesian Cramér-Rao lower bound (CRLB) is derived for the estimation of block-fading frequency-selective channels, which is used to guide the optimal training design. Through the superposition of an additional training vector at the relay under certain power allocation scheme, the separated channel information can be obtained directly at the destination. The Bayesian CRLB is derived for the random channel parameters, and orthogonal training vectors from the two source nodes are required to keep the Bayesian CRLB practical, due to the self-interference in the TWRN. A set of training vectors obtained from the minimization of the Bayesian CRLB are applied in a specific suboptimal channel estimation algorithm, and the mean-square error (MSE) performance is provided to verify the Bayesian CRLB results.
Keywords-Two-way relay, channel estimation, BayesianCramér-Rao lower bound(CRLB), training design, mean-square error.
I. INTRODUCTIONWith the combination of cooperative communication, twoway relay network (TWRN) emerged a few years ago, and has attracted a great deal of interest recently [1], [2], due to its improved spectral efficiency over one-way relay network (OWRN). In a TWRN, a major difficulty lies in how to effectively recover the data transmitted over an unknown fading channel from the other source terminal. Channel estimation in TWRN has been studied in [3]-[8]. Specifically, in [4] and [5], the cascaded source-relay-source channels were estimated using block-based training under the assumption of timeinvariant frequency-selective fading channels. Different from [4] and [5], where the relay only amplifies and forwards the received signal, [6] allowed the relay to first estimate the channel parameters and then allocated the powers for these parameters. The channel estimation problem was extended to the TWRN with multiple antennas at all the three nodes in [7]. A blind channel estimation algorithm based on the second order statistics of the received signal was proposed in [8] for AF TWRN. Inspired by the superimposed training in point-topoint communications, [9] and [10] designed a superimposed training strategy in AF OWRN.In this work, we introduce the superimposed training strategy into OFDM modulated AF TWRN to simplify the channel estimation at the destination, and derive the closedform Bayesian Cramér-Rao lower bound (CRLB) for the esti-