Link Allocation

Link allocation algorithms and protocols try to solve the link scheduling problem.

They allocate unique time-slots in the TDMA cycle to a node for each directed link to its neighbors. Thus, the link method allows two neighboring nodes to transmit si-multaneously whenever the destination nodes are not neighbors of both transmitting nodes. A better spatial reuse is expected from link allocation schemes. In addition, every node can send a single packet to each of its neighbors during every cycle.

In a sense, the link allocation scheme attempts to emulate a wireline network so that all higher layer protocols may be used without modification. However, specific advantages of node allocation techniques like short TDMA cycles and fast delivery of multi-cast traffic are lost.

In this section, we use the same classification as in the previous one and distin-guish four categories of schemes: centralized, distributed with deterministic reser-vation, random access, and time-spread schemes.

Centralized Algorithms

One of the first approaches to find a sub-optimal algorithm for link scheduling is the protocol proposed by Nelson and Kleinrock in [154] and calledspatial TDMA.

Node locations are assumed to be fixed and known. Authors translate the scheduling problem in the maximal clique problem. A clique is a set of links allowing all its elements to transmit simultaneously successfully. A single slot can thus be assigned to a clique. A maximal clique is one in which no additional links can be added without creating a conflict. The schedule is a set of maximal cliques which contains all the links of the network topology.

As for node allocation, a neural approach can be used to solve the link scheduling problem [49]. Finally, the aforementioned unified algorithm in [166] can be applied to link scheduling.

Deterministic Reservation

Cidon and Sidi propose in [73] an extension of their protocol for link allocation.

Again, the channel is split into control and transmission channels. The control channel consists of a request segment and a confirmation segment (see figure 1.24).

CONFIRMATION SEGMENT

1 2 1 2 1 2 1 2

tc(1) tc(2) tc(i) tc(N)

2 i N

1 tr(2)

tr(1) tr(i) tr(N)

1 2 i N

REQUEST SEGMENT

Figure 1.24: Control channel of the Cidon & Sidi algorithm.

Each segment is divided in N mini-slots where N is the number of nodes. Mini-slots in the request segment are used to specify the intended receiver. A confirmation mini-slot is divided in two parts. The first part is used to transmit a deletion signal.

If this signal is sent by some of its neighbors, a node cannot transmit on this time-slot. The second part is used by a node to transmit a confirmation signal and to specify to its neighbors that it will send data on the time-slot.

[161] proposes also a protocol where the channel is divided into a data segment and a control segment and called Dynamic distributed Time-Slot Assignment Pro-tocol (DTSAP). But the originality of this paper is that it addresses the problem of the topological dynamics. Here, three underlying problems arise: the network synthesis problem (establishing correct assignment tables at the genesis of the net-work and inclusion of new nodes), the new neighbors problem (two nodes become neighbors), and the connectivity loss (two neighbors lose their connectivity). Here, nodes exchange connectivity tables.

[48] uses the aforementioned NCR method to construct a link allocation protocol (LAMA) and a pairwise link allocation protocol (PAMA). But, in these cases, spread spectrum is assumed. In LAMA codes are attributed to receivers, whereas in PAMA, codes are attributed to sender-receiver pairs. The main drawback of this approach is that the one-hop and two-hop away neighbor identities are assumed to be known at each node.

Random Access Reservation

Tang and Garcia-Luna-Aceves suggest [187] that all previous algorithms and proto-cols are designed either for broadcasting or unicasting, but not for both. In addition, TSMA protocols suffer from two limitations: The sender is unable to know which neighbor can correctly receive the packet it sends in a particular slot, and these protocols are not scalable because the frame length must be larger than the number of nodes in a two-hop neighborhood.

Thus, they propose in [187] the Collision Avoidance Time Allocation protocol (CATA). CATA allows nodes to contend for and reserve time slots by means of a distributed reservation and handshake mechanisms. CATA ensures that no colli-sions occur in successfully reserved time-slots, and reservations support unicasting, multicasting, and broadcasting.

In CATA, a frame is divided in L slots. In each slot, four control mini-slots (CMS) are followed by a data segment (see figure 1.25).

CMS1 CMS2 CMS3 CMS4

Slot 1 Slot 2 Slot 3 ... Slot L Frame

RTS NTS Data

Data RTS CTS NTS

SR CL

Figure 1.25: CATA frame structure.

Every node receiving data in a given slot transmits a slot reservation (SR) packet in CMS1, this is a busy tone to senders attempting to establish transmissions. Every node that sends data transmits a RTS during CMS2 to jam any possible RTS ad-dressed to its neighbors. Both sender and receiver send a not-to-send packet (NTS) during CMS4 in order to jam any broadcast or multicast reservation. The reser-vation of a unicast transfer is made as follows. The sender sends an RTS during CMS2 if the channel is clear during CMS1. It detects a successful reservation with the reception of a CTS during CMS3. The sender of a broadcast or multicast RTS detects the failure of its request when it either receives an NTS or noise during CMS4. If the channel is clear during CMS4, the reservation is considered successful.

CTS/BI

Frame Slot 1 ... Slot i ... Slot O ms 0 ... ms m

Slot

RTS/BI

Figure 1.26: DPRMA frame structure.

The idea, taken from MACA, to use RTS/CTS handshaking in TDMA based protocols is also exploited in [121] for mixed voice and data traffic. The frame struc-ture of the Distributed Packet Reservation Multiple Access (DPRMA) protocol is shown in figure 1.26.

As in MACA, nodes contend on the mini-slot 0 using the RTS/CTS handshake.

In case of failure, the contention continue on themremaining mini-slots. Otherwise, data transmission takes place. Voice terminals start this contention process on mini-slot 0 with probability 1, while data terminals start with a probabilitypsmaller than 1. Moreover, winning voice terminals reserve the slot for several frames. On the contrary, data terminals can use only one slot.

A part of the RTS and CTS time intervals are used for carrier sensing. A terminal having won the slot transmits, during this part, a busy tone signal to prevent any contention on this slot.

DPRMA introduces the notion of quality of service by giving a higher priority to voice calls. It also solves the hidden terminal problem but leaves aside the exposed terminal issue. Another protocol for speech communication is proposed in [151].

Finally, to complete the picture, Space-time division multiple access (SDMA) [59] is discussed. Here, the nodes are assumed to know their location and time-slot are associated to locations in the network area. Inside a so called space time-slot, reservation is made in a random fashion.

Time Spread Protocols

Initially designed for node allocation in TDMA networks, TSMA can be used for the design of link allocation protocols [70]. [71] defines the concept of threaded TSMA. The proposed protocol is link oriented: Every neighbor of a sender receives

the transmitted packet, but it is destined to a single user. In the original TSMA protocol, called GRAND, the frame length is dependent on the maximal degree in the network. In paper [71], the authors try to overcome this limitation. The basic idea is the interleaving of several different TSMA protocols on a time-sharing basis to obtain a threaded TSMA (T-TSMA).

. Chou & Li

Algorithms Random AccessReservation Time Spread Centralized

. Spatial TDMA . CATA

Reservation Deterministic Link Allocation

. TSMA/CDMA . Cidon & Sidi

. T−TSMA . DTSAP

. LAMA & PAMA . SDMA

. DPRMA . UxDMA

. Neural Algos

Figure 1.27: A classification of the link allocation algorithms and protocols.