Reviewing the environment in which HD-TDMA must operate, three distinct member- ship scenarios exist. The most complex case occurs when a station enters an empty cell, either on its own or at approximately the same time as other stations. The other two cases, occur when at least one active station exists already in the cell. A station with no current slots will need to request a slot as will a station with a pre-existing slot
allocation seeking further transmission resources.
Each case is distinct in how the request to allocate or de-allocate the slot is made, however a significant commonality exists in the process followed to allocate the slot once the request has been made.
3.6.5.1 Bootstrapping
The bootstrapping case occurs when a station powers on in a cell which is has no currently active station, the empty cell case. Motivated by the approach used in TBMAC [Cun- ningham and Cahill, 2002] which employs an epidemic approach to achieving consensus, a modified approach is employed. The TBMAC implementation relied on aligning boot- strapping process to cycle starts, this is an arbitrary choice as the cycle start has no significance except to form a boundary between the CP and CFP TBMAC periods as well as equipping stations with the knowledge that the bootstrap process would always complete at the cycle start. All stations which have joined the cell in the preceding cycle then enter the bootstrap phase and contend equally. Stations which power up and detect ongoing bootstrapping defer until after the bootstrap phase is complete and use the non-empty cell mechanism to seek a slot in due course.
At worst-case, a wait of two full cycle times exists before beginning the bootstrap process, when in fact a station need only listen for one full cycle time to determine if the empty cell scenario applies. HD-TDMA instead aligns the bootstrapping function at the slot time level, thus a station needs to wait no more than one cycle and one slot time before entering the bootstrapping phase reducing the worst-case delay by almost a full slot time.
If a station in the bootstrapping phase, despite not transmitting in any of the slots it has chosen yet, receives a message from another station also bootstrapping it aborts its bootstrapping process and waits for the other station(s) in bootstrap to achieve consensus before requesting a slot through the non-empty cell mechanism.
to maximise the probability that the epidemic process within the bootstrapping protocol will result in a consistent membership view on all stations. A detailed discussion and evaluation of the epidemic bootstrapping has previously been presented by Cunningham et al [Cunningham, 2003].
Each beacon-frame contains a bitmap which details the stations local view of the slot allocation, four possible states are supported, OWNER, OTHER, COLLISION and AVAILABLE. In order to achieve a consistent view of the slots in use, each station must merge its local view of the slot allocation state with those views of other stations as they are received.
3.6.5.2 Empty Cell Scenario
During the bootstrapping phase, if a station detects the absence of an expected beacon- frame during a bootstrapping phase the most likely cause is that two or more stations have tried to transmit in the same slot, having chosen the same slots when the bootstrap phase commenced, which is possible during the bootstrap phase. If such a case occurs, the slot status is set to collision thus moving the membership view towards consensus faster. If a station fails during the bootstrap its absence will be detected through the lack of transmissions, the collision process will promptly remove the failed station.
In a worst-case, two or more stations may select the same slot in which to transmit in during the bootstrapping phase, which stations receive which transmissions is a complex function of the proximity of the stations to each transmitter and any local propagation conditions, e.g. obstructions in the way. In the majority of cases, no station receives any valid transmission as a result of each transmission interfering with each other.
However there is a possibility that a subset of the stations would receive the trans- mission of station A, a subset of stations the transmission of station B and so on with the remaining stations receiving nothing meaningful at all. It is therefore necessary to provide an exchange of views of slot allocation. Since, no information of use is retained by slots marked as Available, Collision and Owner as the mac address of the owner is
already in the header of each packet transmitted. The additional information portion of the beacon-frame contains the list of mac addresses and the slots allocated as viewed by each station.
The key concern during the bootstrap phase is to ensure it possible for all stations to achieve consensus, logically the fewer stations which take part in the process maximises the probability of each station achieving a consistent view.
The architecture of HD-TDMA provides no dedicated space for slot allocation func- tions unlike TBMAC, this introduces a small time penalty in the best-case slot allocation times. The clear benefit is that the average and worst-case times from the request to the first transmission in the allocated slot in HD-TDMA are better than in TBMAC. Also, when all slots have been allocated there are no idle resources which represents a key performance benefit.
If the received bitmap state for slot x is owner and the local bitmap for slot x is available or other where the mac address of the owner is unknown the local station updates its data structures to incorporate the information received. If a conflict occurs between the received data and the current local view the slot is marked as collision. 3.6.5.3 Non Empty Cell Scenario
Two discrete cases are be supported, a station may have no slot currently allocated to it and therefore cannot use a beacon-frame to convey its request or the station has at least one slot allocated to it and therefore can use the beacon-frame to convey its request.
HD-TDMA does not reserve a portion of the TDMA cycle for slot management functions, instead it relies on unallocated slots. This may appear counter intuitive however, in the non-empty cell case, a station with no slots can only seek a slot if there is in fact at least one unallocated slot available, thus it is logical to issue the request for a slot within a slot which is unallocated.
For a station which has at least one slot allocated to it it may utilise the slot manage- ment portion of the beacon-frame to broadcast its slot management request a number
of times. Thus with all slots allocated there is no portion of the cycle idle satisfying the need to maximise performance.
Slot allocations under non-empty cell conditions occur in band with the requesting station utilising any slot which it determines to be unallocated to broadcast its slot request. In order to ensure slot allocations and data transmissions are segregated as far is as is practical, HD-TDMA includes a provision which allows a station to indicate in the beacon transmission that the remainder of the slot is idle and that it has no data to transmit.
A station, upon receiving the request, packages it into an atomic broadcast and sends it in the additional information portion of its beacon frames until its delivery time arrives.