In cooperative communications, some nodes act as relays to assist other nodes to transmit their information. Relay technology extends radio range by processing wireless signals between the transmitter and the receiver. Relay networks can be classified into two categories: one-way relay networks (OWRNs) and two-way relay networks (TWRNs). We focus on channel-unaware TWRNs. Since in TWRNs both nodes transmit at the same time, the received signal at each relay contains messages originating from both nodes. The combination of the two messages is processed and broadcast from each relay. Consequently, each node receives not only the desired signal sent from the other node but also an SI component sent from the same node in the previous phase. The existence of SI has a deleterious effect on the system performance, and hence must be cancelled prior to detection. SI cancellation depends on CSI availability at the nodes. When perfect CSI is available its cancellation is straightforward. However, the main challenge in TWRNs is to cancel SI perfectly when no CSI available at either the nodes or the relays. Several approaches for mitigating the effect of SI in channel-unaware TWRNs have been proposed in the literature. These proposed schemes rely on CSI, SI or noise estimation which result in estimation error, imperfect SI cancellation and error propagation. In order to overcome the aforementioned drawbacks, we develop a novel relaying scheme for channel-unaware TWRNs with multiple-antenna half-duplex (HD) amplify-and-forward (AF) relays. This scheme enables the self-interference to be cancelled perfectly by joint space-time processing of the relay received signals. To overcome the restriction of having even number of active antennas at each relay, we developed another new signalling scheme that enables self-interference to be perfectly cancelled in the absence of CSI when HF AF relays have a single antenna each. This scheme relies on mode-switching technique.To analyze the effectiveness of the proposed schemes, we derive upper bounds on the pairwise error probability. Using these upper bounds, we compare two schemes and provided a nonintuitive fact that increasing the number of relays and using SP(2) code does not necessarily yield better performance than using less relays with the Alamouti code.