We have identified the primary tasks of cooperative networks, namely message decoding, cooperative link establishment, and cooperative routing. These tasks must be completed by peer protocols at the link layer, MAC sublayer, and network layer. In any wireless network, signaling channels are defined to facilitate these protocols. For example, a packet header that identifies the source, destination and sequence number of each packet represents an embedded control channel at the link layer. Control packets such as RTS and CTS in 802.11 systems are a type of control channel signaling at the MAC sublayer. Similarly, the ad hoc routing protocols DSR and AODV employ RREQ and RREP packets for network layer signaling. Alternatively, these control channels may be embedded in the PHY layer, e.g., special code sequences in 2G and 3G CDMA cellular systems or specific bits in frame headers.
In all wireless systems, the design of the signaling channels is an important engineering problem. However, from the point of view of system design, the key consideration is that the information exchanged for link establishment must be small relative to the data that is subsequently transmitted. Assuming such reasonable control mechanisms exist, the question is: How are network resources allocated?
With respect to the design of network protocols, some preliminary considerations are as follows:
• Protocols are needed to allow a source and multiple relays to decide when and how to establish a cooperative link.
• Although the payload of a packet transmission may be received unreliably, a specified set of nodes in a link must recognize and act upon this transmission. This requires a method for a node to signal a packet transmission. Conceptually, this could be done using a control message akin to RTS that is transmitted at sufficiently low rate to enable decoding by all nodes. However, a packet header must still contain an acquisition sequence, i.e., a sequence of known bits that allows receiving nodes to acquire the symbol timing of the arriving packet. This header might as well be extended to include a low-rate data field that associates the packet with a particular message. Perhaps the header also identifies the nodes that are expected to decode the message in question.
• To handle repeated unsuccessful decoding, a mechanism is needed to allow a node to give up trying to decode and instead send a NAK. These NAKs may also be used to reconfigure a cooperative link.
• Multi-terminal link optimization will include distributed scheduling of relay transmissions, coding and power control suitable for cooperation, as well as adaptation of the forwarding strategy, e.g., AF, CF, or DF.
• Routing protocols must account for cooperative links. Suitably abstracted routing metrics must be developed.
In the following, we consider node and network architectures that begin to address the above requirements. In order to limit the discussion, we remark that although full duplex relaying and beamform-ing (coherent transmission) from multiple nodes benefit cooperation [33, 58], realizing these modes of operation in mobile wireless settings is at the very least challenging and arguably impossible. We thus focus on the implementation of half-duplex diversity and network coding mechanisms. In the following, we examine how cooperative communication can be implemented in the physical and link layers. We start by reviewing a conventional PHY layer architecture in Section 6.4. We then examine a cooperative PHY layer architecture in Section 6.5.
