Consider amplify-and-forward two-way relay networks with multiple relays and frequency selective fading channels, a low-complexity precoding scheme and semi-blind channel estimation method are proposed in this dissertation. Precoding is performed using a rotation-based matrix and decoding is low-complexity and easily scalable to the number of relays and/or the channel length since the decoding process requires only a circulant shifting of the received blocks. The composite channel impulse response of each source-relay-destination link is estimated based on the second-order statistics of the received signals. The ambiguity in the proposed channel estimates caused by the channel information of the direct link is eliminated using a small number of training blocks. The accuracy of the proposed semi-blind channel estimates is examined by deriving the mean square error of the channel estimates for two asymptotic cases, namely an infinite number of data blocks and a high signal-to-noise ratio (SNR) regime. The two cases thus provide a useful insight into the effects on the estimation performance of imperfect direct link information and statistical errors of the ensemble correlations, respectively. The asymptotic analyses are confirmed by the simulation results, which show that the normalized mean square error of the channel estimates varies inversely with the SNR and the number of data blocks. Keywords ─ ─ ─ ─ Two-way relay networks, semi-blind channel estimation, frequency-selective fading channels, low-complexity, precoding design, asymptotic analysis.iv