At the physical level we will be using the orderwire to exchange position and link quality information and to setup the wireless connections. The process of setting up the wireless connections involves setting up links between edge nodes and between edge and remote nodes.
The network will have one master switch (EN), which will run the topology configuration algorithm  and distribute the resulting topology information to all the connected ENs over point-to-point packet radio links. The point-to-point link layer is AX.25 .
The master EN could initially be the first active EN, and any EN would have the capability of playing the role of the master.
The first EN to become active would initially broadcast its callsign(where callsign = radio address) and start-up-time in a MYCALL packet (Table 2), and listen for responses from any other ENs. Since it is the first active EN, there would be no responses in a given time period, say T. At the end of T time, the EN could rebroadcast its MYCALL packet and wait another T seconds. At the end of 2T seconds, if there are still no responses from other ENs, the EN assumes that it is the first EN active and takes on the role of the master. If the first two or more ENs start up within T seconds of each other, at the end of the interval T, the EN could compare the start-up times in all the received MYCALL packets and the EN with the oldest start-up time would become the master.
Each successive EN that becomes active would initially broadcast its callsign in a MYCALL packet. The master on receipt of a MYCALL packet would extract the callsign of the source of the packet, establish a point-to-point link to the new EN and send it a NEWSWITCH packet (Table 3). The new EN on receipt of the NEWSWITCH packet over a point-to-point link, would obtain its position from its GPS receiver and send its position to the master as a SWITCHPOS packet (Table 4) over the point-to-point link. On receipt of a SWITCHPOS packet, the master would record the position of the new EN in its "switch position" table (table of EN positions), and run the topology configuration algorithm , to determine the best possible interconnection of all the ENs. The master would then distribute the resulting information to all the ENs in the form of a TOPOLOGY packet (Table 5) over the point-to-point links. The EN can then use this information to setup the high-speed links as specified by the topology algorithm. The master would also distribute a copy of its ``switch position'' table to all the ENs (over the point-to-point links), which they can use in configuring RNs as discussed below. This sequence of operations is illustrated in Figure 3 and Figure 4. Also, the EN can then use the callsign information in the ``switch position'' table to setup any additional point-to-point packet radio links (corresponding to the high-speed links) required to exchange any link quality information. Thus this scheme would result in point-to-point packet radio links from the master to every EN (a point-to-point star network with the master as the center of the star) and also between those ENs that have a corresponding high-speed link, as shown in Figure 1.
In the event of failure of the master node (which can be detected by listening for the AX-25 messages generated on node failure), the remaining ENs exchange MYCALL packets, elect a new master node, and the network of ENs is reconfigured using the topology configuration algorithm . The efficiency of this method of handling failure of the master node versus maintaining a hot backup for the master node is to be studied.
Figure 3: State Diagram for Master.
Figure 4: State Diagram for EN not serving as Master.
Each RN that becomes active would obtain its position from its GPS receiver and broadcast its position as a USER_POS packet (Table 6). This packet would be received by all the ``nearby'' ENs. Each candidate EN would then compute the distance between the RN and all the candidate ENs (which is possible since each EN has the positions of all the other ENs from the ``switch position'' table). An initial guess at the best EN to handle the RN would be the closest EN. This EN would then feed the new RN's position information along with the positions of all its other connected RNs to a beamsteering algorithm that returns the steering angles for each of the beams on the EN so that all the RNs could be configured. If a time slot and/or beam is available to fit in the new RN (this information will be returned by the beamforming algorithm), the EN would steer its beams so that all its connected RNs and the new RN are configured, record the new RN's position in its ``user position'' table (table of positions of connected users), establish a point-to-point link to the new RN and send it a HANDOFF packet (Table 7) with link setup information indicating that the RN is connected to it. If the new RN cannot be accommodated, the EN would send it a HANDOFF packet with the callsign of the next closest EN, to which the RN could send another USER_POS packet over a point-to-point link. This EN could then use the beamform algorithm to determine if it could handle the RN, and so on. Figure 5 shows the states of operation and transitions between the states for a RN.
This scheme thus uses feedback from the beamforming algorithm together with the distance information to configure the RN. It should be noted that the underlying AX.25 protocol  ensures error free transmissions over point-to-point links. Also the point-to-point link can be established from either end and the handshake mechanism for setting up such a link is handled by AX.25. If the RN does not receive a HANDOFF packet within a given time it can use a retry mechanism to ensure successful broadcast of its USER_POS packet.
Figure 5: State Diagram for RN.
The point-to-point orderwire links would be retained as long as a RN is connected to a particular EN and a corresponding high-speed link exists between them (to enable exchange of link quality information). The link can be torn down when the mobile RN migrates to another EN in case of a handoff.