Abstract âThis paper proposes a distributed TDMA slot allocation protocol that relies on zero amount of explicit information exchange among the participating ...
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
Zero-Exposure Distributed TDMA Using Time-coded Packet Transmissions Debasmit Banerjee, Mahmoud Taghizadeh, Subir Biswas Electrical and Computer Engineering Department, Michigan State University Abstract –This paper proposes a distributed TDMA slot allocation protocol that relies on zero amount of explicit information exchange among the participating nodes. This novel property allows the protocol to work in restricted environments such as in privacy-sensitive body area networks and various military networks in which nodes may need to cooperate in distributed TDMA but are not allowed to explicitly exchange any information in order to preserve their anonymity. This paper introduces an innovative approach of time-coded packet transmissions for implicitly exchanging slot timing information. It is shown that using such implicit information, together with a notion of pattern based shadow packet, the nodes are able to selfallocate collision-free TDMA slots in a Zero Exposure manner. The protocol is evaluated and its performance has been shown using extensive simulation models. Keywords - Privacy, Secure wireless network; distributed TDMA; Medium Access Control; anonymous networking
I.
INTRODUCTION
A. Zero Exposure TDMA Wireless ad-hoc and sensor networks are used in emerging applications such as body area sensing, intrusion and event monitoring, and for a slew of military applications in privacysensitive environments. A critical component of wireless networks is the medium access control (MAC) protocol that is responsible for administering the wireless channel access. Two broad categories of wireless MAC protocols are TDMA based [1][4][6] and CSMA contention based [2][7]. The main advantage of TDMA over CSMA is its contention-free nature and guaranteed fairness, which are especially attractive in highly dense networks with real-time constraints on message latency. With TDMA, nodes do not need to contend for slots, leading to collision-free access. Although TDMA slot allocation can be simply managed by a centralized access point, such centralized administration is not usually applicable for dynamic and ad hoc networks. As a result, majority of the TDMA protocols for sensor and ad hoc networks in the literature have focused on distributed slot allocation solutions which rely on in-band [1][9] or out-of-band [10][11][12] control mechanisms for slot-allocation. Such control mechanisms require network nodes to exchange their individually occupied slot information so that they can individually adjust their slot occupancy based on a distributed allocation algorithm. Traditional slot allocation mechanisms used in distributed TDMA protocols may not be applicable in highly privacysensitive environments in which exposing a node’s identity in any form is not acceptable. Wireless networks in such environments may require packets to be fully encrypted in all protocol layers (i.e. from physical layer to all the way to the application layer [3]) of the network stack. From a MAC perspective, this implies that no header information, including MAC source and destination addresses, can be exposed. We term such extreme constraint as Zero Exposure networking.
Such extreme privacy may be required for military networks and other applications including Body Area Networks (BAN) which often deal with sensitive patient information. For example, a BAN composed of wearable sensors mounted on a person may use an encryption mechanism such that all sensors on that BAN share a pre-distributed key which is different from that used by the sensors on a different person’s BAN. When those two different BANs share the same wireless channel and come within communication range, then all the sensors in those BANs need to settle on a set of collision-free TDMA slots. And that is without being able to exchange any explicit slot allocation information which may expose their individual identities. None of the existing solutions in the literature can perform distributed MAC allocation with such Zero Exposure constraint. This paper sets out to accomplish that. Other examples of Zero Exposure environment can be within military ad hoc deployments in which multiple wireless networks, running independent privacy models, geographically co-exist and share a single wireless channel. In this paper we propose a protocol for distributed TDMA slot allocation with zero information exposure in multi-hop wireless networks. The key concept is to use an innovative approach of time-coded packet transmissions for implicitly exchanging slot timing information without having to rely on any explicit information exchange among the network nodes. Using only such implicit information, the nodes are able to selfallocate collision-free TDMA slots in a Zero Exposure manner. B. Key Challenges Slot allocation in TDMA protocols for multi-hop wireless networks need to satisfy the constraint that at steady state, no two 1-hop or 2-hop neighbors can have partially or completely overlapping transmission slots. Overlap between slots of 1-hop neighbors cause direct collisions and overlap between slots of two hop neighbors can cause hidden collisions. In a Zero Exposure environment, exchanging slot occupancy information among 1-hop or 2-hop neighbors is not feasible since two nodes may belong to two different privacy domains which may prevent them from deciphering each other’s packets. Therefore, the primary challenge for Zero Exposure distributed slot allocation is the unavailability of any explicit means to disseminate the slot occupancy information within up to 2-hop wireless neighborhoods. The second challenge stems from the absence of network-wide time synchronization without which the task of cross-node slot alignment is non-trivial [5]. C. Contributions The primary contributions of the paper are as follows. First, we develop a time-coded packet transmission mechanism for implicitly exchanging slot occupancy information in a distributed manner. Second, a pattern based shadow packet mechanism is developed for disseminating slot occupancy information up to 2-hop wireless neighborhood. Third, using these two key concepts, a distributed slot allocation algorithm is developed for fast allocation convergence in the absence of network time synchronization. Finally, through extensive ns2
978-1-4244-9268-8/11/$26.00 ©2011 IEEE
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
simulations, the proposed Zero Exposure allocation mechanism is functionally validated and its performance is evaluated under varying topology and network dynamics. II.
ZERO EXPOSURE DISTRIBUTED TDMA
A. Time-Coded Packet Transmission Once a TDMA slot (of duration τ) is self-selected, a node periodically sends packets in that slot once in every TDMA frame (of duration T). A node’s slot timing is implicitly detected by all its 1-hop neighbors by simply noting the node’s packet transmission time. Using such implicit information, each node in the network can learn the slot timing of all its 1-hop neighbors without actually knowing the identification of the neighbor nodes. In the absence of time synchronization, each node maintains its own TDMA frame, starting at the node’s own slot time. Fig. 1(a) shows how the nodes in a three-node network maintain information about the slot occupancy of the other nodes with respect to their own individual TDMA frames. We introduce the concept of a phase (φ) which represents the location (i.e. timing) of a node’s slot from the perspective of a global observer that maintains a global frame. For example, to a global observer, the phase of node A is φA as shown in Fig. 1(b). Without inter-node time synchronization, the absolute phase does not have any meaning from an individual node’s (i.e. that of a local observer) standpoint. However, a node can detect its phase difference with another node. Phase difference between two nodes is the time difference between their respective slots. For example, the quantity δAB in Fig. 1(a) represents the phase difference between nodes B and A as perceived by node B. B. Shadow Packets for 2-hop Informtion Dissemination Achieving collision-free slot assignment in TDMA requires the slot timing information of a node to be disseminated up to its 2-hop neighbor nodes. In the absence of any explicit mechanism for such information transfer (i.e. Zero Exposure), we introduce a new mechanism of shadow packets which is used as follows. A node sends shadow packets during its 1-hop neighbors’ slots in order to propagate its 1-hop neighbors’ slot timing information to all other 1-hop neighbors. By noting the start timing of a shadow packet, a node can infer the slot timing of one of its 2-hop neighbor. The end result of this shadow packet transmission process is that all nodes are informed about the slot timing of their 2-hop neighbors. This shadow packet based 2-hop information dissemination, coupled with the timecoded 1-hop information dissemination in Sec. II:A, makes a node aware of the slot timing of all nodes up to its 2-hops - the basic requirement for any distributed slot selection algorithm. A shadow packet is distinguishable from the regular packets because of its slightly shorter length, and may be used for sending useful data. The primary difference is that a regular packet is sent by a node during its own TDMA slot, whereas a shadow packet is sent by the node during the slots of its 1-hop neighbors. The slot size is assumed to be the same as the regular packet size to accommodate both packet types. If not managed carefully, the shadow and regular packets can cause collisions in the following manner. Consider a node Ni that sends shadow packets during the slots of one of its 1-hop
Fig. 1: (a) Local slot occupancy view to individual nodes (b) Slot occupancy view to a global observer (c) Regular and shadow packet sending pattern
neighbors Nj. Now, those shadow packets sent by node Ni will collide with the regular packets sent by Nj (i.e. at the same time slots) at a receiver node Nk, which is a common neighbor of both Ni that Nj. We introduce the following pattern based shadow scheduling to avoid such collisions. C. Pattern Based Shadow Scheduling Based on a globally pre-determined frame-level pattern, a node skips sending regular packets in its own TDMA slot. For example, when the pattern is set to ‘1110’, a node sends packets in its slot in three consecutive TDMA frames and then it skips the fourth frame; the cycle continues for a super-frame of four TDMA frames. The fourth frame in the super-frame is skipped so that all of that node’s 1-hop neighbors can send shadow packets corresponding to the node in question. In other words, the 1-hop neighbors of a node send shadow packets in a pattern (i.e. 0001 in this case) that is 1’s complement of the globally pre-set pattern. This is how the collisions between shadow and regular packets are avoided. For example, in Fig. 1(c), the global pattern is set to ‘101’. Observe that node B sends shadow packets corresponding to its 1-hop neighbor A with a 1’s complement pattern ‘010’ to avoid collisions with node A’s regular packets. Although the pattern is globally pre-determined, the nodes maintain their individual pattern sequences asynchronously. As a result, a node first needs to learn about the pattern phase of each of its 1-hop neighbors by listening to their transmissions over multiple super-frames. Once the pattern phase for a 1-hop neighbor is detected, the node then starts sending the corresponding shadow packets following the 1’s complement of the detected pattern phase. A longer pattern length (i.e. one ‘0’ for a large number of ‘1’s) causes a smaller bandwidth loss due to the shadow packets before a node actually stops sending them. On the other hand, a longer pattern means longer duration needed for the node to
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
Fig. 2: Types of collisions. (a) (i) and (ii) Shadow-shadow, (b) Shadow-regular, (c) Regular-regular
learn the pattern. Therefore, the pattern should be chosen carefully in order to strike a balance between the above two factors. The shortest possible pattern ‘10’ is not feasible since in this case from an observer node’s standpoint both shadow and regular packets appear in the same sequence. This prevents the observer node from detecting (and resolving) a regularregular packet collision which will be discussed in Sec. II:F. It turns out that the shortest functionally feasible pattern is ‘101’ which is used for protocol implementation and performance evaluation in this paper. With ‘101’ the bandwidth loss is 33% between when a node starts sending shadow packet due to a network perturbation, and when it stops sending them. The shadow packet transmission process, as described above, starts in the events of network topology changes including node additions. Using the shadow packets, once a node is able to gather the slot timing information about its 2hop neighborhood, the slot self-allocation process, described in Sec. II:D, takes place. If there are no more events of network topology changes within its neighborhood, the node stops its shadow packet transmissions after a predefined time period, and it starts sending continuous regular packets without skipping frames for the shadow packets. Note that the shadow packets can also be used for sending useful data since the content of a packet is not actually used for the protocol to work. In other words, when Nk, a 1-hop neighbor of node Nz, sends shadow packets in Nz’s slot, the former can send its own data in those packets. As a result, although there is a temporary throughput loss for Nz, for the node Nk there is a temporary gain in throughput during the shadow sensing period. D. Slot Self Allocation Upon joining the network, a node chooses an arbitrary time as its frame starting point and gathers the slot occupancy information about its up to 2-hops neighborhood by simply listening to the regular and shadow packet transmissions within its 1-hop neighborhood. The minimum duration after which such occupancy information can be deterministically identified is one super-frame. For e.g., when the pattern is set to ‘101’, the minimum duration would be three TDMA frames, since a node receives a shadow packet only once in three frames. After all the occupied slots (within its 2-hop neighborhood) are identified, a node self-allocates an available slot and defines it to be the new starting point of its own frame. At this point, the node starts sending regular packets in that slot, and shadow packets in its 1-hop neighbors’ slots as described in Sec. II:B.
E. Allocation Convergence Once a node self-allocates a slot, it might defer its selected slot to resolve temporary collisions due to conflicting slots with any of its 2-hop neighbors. Allocation convergence for a node (for given event - e.g. a node addition) is defined as the time instant when the node finalizes its slot location. After allocation convergence, a node constantly monitors for any change in slotoccupancy within its 1-hop neighbors and if there are no changes in occupancy for a predefined time period, the node stops sending the shadow packets for its 1-hop neighbors. This happens in a distributed manner since decisions can be made by each node autonomously. After a node stops sending shadow packets for its 1-hop neighbors, it also starts sending regular packets in its own slot in every frame. F. Collision Detection and Resolution A unique feature of the proposed protocol is that since it relies only on the packet transmisison times (i.e. not relying on packet content), it works even when the received packets are corrupted due to channel errors or packet collisions. The only requirement is that when a node receives a corrupted packet in a slot (due to collision or channel error), it should be able to mark the corresponding slot as occupied before discarding the corrupted packet. Collisions are in fact used as a source of valuable information as it can indicate overlapping slot ownership within a 2-hop neighborhood. Specific collision types are identified and resolved as described below. 1) Collision Types A collision can be among shadow packets, shadow and regular packets, or regular packets. All three types are shown in Fig. 2. From Fig. 2(a), it can be seen that the collision between two shadow packets can be due to: (i) the propagation of a single node’s slot occupancy information, or (ii) propagation of slot occupancy information of two different nodes more than two hops away. In either case, the collision between two shadow packets does not violate the requirement for slot occupancy that nodes within two hops should not share the same slot. Similarly, collision between regular and shadow packets (Fig. 2(b)) does not result in the violation of the requirement, either. Although this can result in reducing the throughput of the node whose regular packet is colliding with the shadow packet from another node, once the node stops sending shadow packets this situation does not exist anymore. The third scenario (Fig. 2(c)), happens when there is collision between regular packets. This situation needs to be avoided as it causes direct violation of the 2-hop slot occupancy requirement because collision between two regular packets means that two nodes (within 2-hops) have overlapping slots. 2) Detection of regular-regular collision Detection of a regular-regular collision is a two-step procedure as follows. First a node needs to detect that there is a collision among its neighbors, and then it needs to distinguish this type of collision from the two other types. Collision detection is done by checking the duration of the received signal. Since the size of the regular packets is the largest among different packet sizes, any received signal that lasts longer than
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
Fig. 3: (a) Regular packet collision pattern (b) Random jitter and effective slot-size
regular packet duration can be inferred as an overlapping transmission or collision at the receiver. Identifying the type of collision is done by monitoring the pattern of collisions over multiple frames. We explain this using the following example. Let C indicate a collision or corrupt packet reception, N indicate a successful packet reception (i.e. no collision), and E indicate a silent or empty slot, i.e. no packet reception. Then for 101 as the pattern, a regular-regular collision follows the sequence of ..CNNCNNC.. or ..CECCEC.. over multiple frames as shown in Fig. 3(a). These are the only two possible patterns for regular-regular collision. Although the collided slot size can range from one-slot size (i.e. fully overlapped transmissions) to two-slot sizes (i.e. slightly overlapped transmissions), the node detecting the collision considers only the starting time of the first slot to mark the slot as collided. 3) Collision Resolution To resolve 1-hop collisions, every node senses the channel before sending a regular packet and if the channel is found to be busy, the node defers its slot and transmits the regular packet only after the channel becomes free. 2-hop collisions are resolved by interrupt packets (same size as regular packets) which are sent by a node which detects a regular-regular collision between its 1-hop neighbors. We explain the 2-hop collision resolution mechanism using the example in Fig. 4. In the example, nodes 1 and 3, which are 2-hops away from each other, share overlapping slots. Node 2 detects a regular-regular collision from the collision pattern and marks the starting time of node 1’s slot as the collided slot start time, as pointed out in Fig. 4(b). Node 2 sends an interrupt packet immediately after the start of node 1’s slot. The interrupt packet keeps the channel busy for node 3 which causes it to defer its slot until the end of the slot occupied by node 1 and hence resolves the collision between nodes 1 and 3. Collisions among more than two nodes are resolved by multiple interrupt packets.
Fig. 4: Collision detection and resolution: (a) Neighbors of 2, i.e. 1 and 3 are using overlapping slots for regular packet transmission. (b) When 2 detects regular-regular collision from the pattern, it sends an interrupt packet immediately after the start of overlapping slots. (c) This causes 3 to delay its slot and settle on a non-overlapping slot within its 2-hop neighborhood
Note that the interrupt packets work successfully only if the conflicting slots are not exactly overlapped. In the proposed protocol a random jitter has been introduced, due to which the probability of exact overlap of two allocated slots (belonging to different nodes) is extremely low. When a node selects a slot in the beginning, or it selects a slot after its previous slot gets delayed, it adds a very small random jitter to the newly selected slot time. To avoid exact slot overlapping over time due to clock drift or due to merging two disconnected networks, jitter is also used after a predefined number of frames. The range of the random jitter has been chosen to be between 0.006τ and 0.06τ, where τ is the slot size. The minimum value of the jitter needs to be large enough to account for channel propagation delay (for the interrupt packet to reach the other nodes). TABLE I : Baseline System Parameters in Simulation Propagation model Channel bandwidth Transmission range Network size Regular & Interrupt packet size Shadow packet size Slot duration Shadow Pattern
III.
Two-ray ground 2Mbps 240m 3-100 nodes 1024 Bytes 512 Bytes 4.096ms 101
EVALUATION
We have implemented the proposed Zero-Exposure TDMA allocation algorithm within the ns2 MAC simulation module. The baseline simulation parameters are shown in Table 1. The protocol has been evaluated for different network topologies including linear, loop, fully-connected, grid and random mesh with network size ranging from 3 to 100 nodes. Network dynamics have been tested by joining two converged subnetworks. The protocol has also been tested for two different node deployment policies – one where all network nodes are introduced at the same time, and the other in which nodes have been introduced incrementally once the remaining network has converged. However, due to space constraints, only results for the former deployment policy have been included in the paper. A. Functionality Tests The functionality of the protocol has been tested by running simulations for static as well as dynamic networks with linear and fully-connected topologies. Fig. 5 shows the functionality graphs for fully-connected and linear networks. The y-axis represents the phase (normalized by frame size) of the nodes with respect to a global observer (see Sec. II:A), and the x-axis represents the time in terms of the number of frames. For all functionality experiments, the frame size is set to 5 slots which represent the maximum number of up to 2-hop neighbors for all the topologies that were experimented with. An inter-node phase difference of less than 0.2, which is the normalized phase equivalent of the slot size τ (4.096 ms), indicates that two nodes have overlapping slots. Specific events of the protocol functionality have been pointed out in the figures using the alphabetical markings in Fig. 5. 1) Static Network Functionality tests for static networks have been performed by introducing all nodes in the network at the same time. Fig. 5(a) depicts the phase of the nodes in a fully-connected
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
Fully Connected - Static 0.9 0.8
E2
Phase
0.7 0.6 0.4
Node1 Node2
0.5
E1
seen in the figure. It can be also observed from the figure that all network nodes eventually stop sending shadow packets in a completely distributed fashion at different times.
Node3
B. Performance Characterization The convergence time is defined as the interval from when 0 a node is added to the network to when the node has a fixed 0 10 20 30 40 50 60 Frame Count slot assigned to it. F-ratio is defined as the ratio of the chosen (a) frame size (i.e. number of slots) and the minimum frame size Linear Joining needed for a network to reach convergence. Minimum number 1 E1 E2 E3 of required slots is determined by the maximum number of 0.9 Node1 0.8 nodes within up to 2-hop neighborhood. For example, F-ratio Node2 0.7 = 1 is the smallest possible frame size and demonstrates the Node3 0.6 0.5 situation in which the frame size is exactly equal to the Node4 0.4 Node5 number of required slots. Note that due to slot sharing among 0.3 Node6 0.2 nodes farther than 2-hops, the minimum frame size is usually 0.1 Node7 much lesser than the actual number of nodes in the network. 0 55 65 75 85 95 105 115 125 Note that for all the functionality results in Sec. III:A, the FFrame Count ratio was chosen to be 1. In other words, the protocol (b) functionality was evaluated in the most stringent situation Fig. 5: Functionality tests for: (a) Static fully-connected topology possible. (b) Dynamic network - two linear subnets joined through node 4 1) Protocol Convergence network with 5 nodes. Initially nodes 1 and 3 have overlapping Since the allocation process is distributed, the convergence slots until frame count = 2. Node 1 detects the channel to be time can be different for different nodes. The graphs in Fig. 6 occupied (by node 3) while sending its regular packet and show the average and worst case convergence time with 95% defers its slot to 0.2074, which causes its slot to be overlapped confidence range based on the variance of 300 simulation runs with that of node 4, as shown by the marking E1 in the figure. of linear, grid and fully-connected topologies where all nodes This results in node 4 deferring its slot and selecting a non- are simultaneously introduced in the network. The x-axis overlapping slot at frame count = 4. After these changes in slot represents the network size and the y-axis represents the time to occupancy, all the nodes have converged with fixed allocated converge in terms of the number of frames. The average case slots and they continue to transmit regular and shadow packets convergence shows the network-wide average of the according to the chosen pattern ‘101’ as marked in the figure convergence time and the worst case convergence time is for node 5 as E2. After there has been no event of network plotted for the node which takes the longest to converge. topology change for a predefined time period, the nodes stop From the graphs, it can be observed that convergence times following the pattern which can be seen from the solid lines for F-ratio = 1 is always higher than that for F-ratio = 1.5. Also after 30 frames. for linear and grid topologies with higher network sizes, the worst case convergence is considerably higher than the average 2) Dynamic Network Fig. 5(b) demonstrates the functionality of the protocol case convergence. This was caused because with an increase in when two isolated converged sub-networks with linear topology network size, even though all other nodes in the network have are joined due to the insertion of a new node. After the two self-allocated slots, there are always one or two nodes which subnets (each with three nodes connected in a linear fashion) take more time to self-allocate a slot. This in turn increases the have converged and stopped following the pattern, node 4 worst-case convergence time. The effect is more severe when enters the network at frame count = 57 and joins these two F-ratio = 1, as the frame size with this F-ratio is at its maximum capacity (i.e. the absolute minimum frame size for the given subnets to form a network of seven nodes connected linearly. Different events shortly before and after the joining are network). Also, with this conservative F-ratio, the jitter from one node sometimes causes its neighbor to defer its slot. This described below using various markers in the figure: E1: From the collision pattern, node 4 detects that two of its effect, however, is not observed network-wide as it vanishes neighbors (nodes 3 and 5) have overlapping slots and sends an with the distance from the affected node due to the jitter from other nodes. Another observation from Fig. 6(a) and (b) is that interrupt packet to resolve the collision. E2: The interrupt packet sent by node 4 causes node 3 to delay the average case convergence for linear networks is almost its slot and to select a non-overlapping slot at frame count = 72. constant, unlike grid networks, which show a slight increase in E3: The slot initially selected by node 4 has a slight overlap the average convergence time. The reason behind this is the with node 6’s slot. When node 6 senses the channel before topological difference between the two networks. A grid sending its regular packet, it finds the channel to be busy due to topology is more complex in nature with the presence of loops the shadow packet sent by node 5 for its 1-hop neighbor node 4. which results in circular dependencies among the nodes which This causes node 6 to adjust its slot and to select a slot that is results in a minor delay to find a self-allocated slot. Fig. 6(c) shows the convergence graphs for a fullynon-overlapping with node 4’s slot. The change in node 6’s slot also necessitates node 7 to make adjustments to its own slot as connected network for F-ratio=1 and 1.5. It can be observed 0.3 0.2
Phase
0.1
Node4 Node5
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE Globecom 2011 proceedings.
Convergence worst (F-ratio = 1) Convergence avg.(F-ratio = 1) Convergence worst (F-ratio = 1.5) Convergence avg. (F-ratio = 1.5)
Convergence worst (F-ratio = 1) Convergence avg.(F-ratio = 1) Convergence worst (F-ratio = 1.5) Convergence avg. (F-ratio = 1.5)
200
50
Frame Count
Frame Count
Frame Count
100 150
Grid
100 50
5
25
45
Network size
65
300 200 100 0
0
0
Convergence worst (F-ratio = 1) Convergence avg.(F-ratio = 1) Convergence worst (F-ratio = 1.5) Convergence avg. (F-ratio = 1.5)
400
9
29
49
69
89
5
Network size
25
45
Network size
Fig. 6. Convergence characteristics for (a) Linear (b) Grid and (c) Fully-connected topology with F-ratio = 1 and 1.5
that for F-ratio = 1, there is a rapid increase in the convergence time until network size = 20. Then a sudden drop of convergence time is seen after which the time increases with a less steep slope. The reason behind this is as follows. The maximum node-degree of a fully connected network is equivalent to the network size and with an increase in network size; there is a sharp increase in convergence time. But when the network size becomes greater than 20 nodes, the accumulative jitter added by each node results in the effective F-ratio to be larger than the actual F-ratio (i.e. 1). The increase in the effective F-ratio results in the decrease in convergence time as can be seen by the steep drop in the graph. As the network size is further increased, the convergence time also increases, but the accumulative jitter plays an additional part which results in a less steep increase than seen when the network consisted of 5 to 20 nodes. This phenomenon is not observed for F-ratio = 1.5 as the frame-size is already large enough for fast slot-allocation convergence. IV. RELATED WORK Several TDMA protocols have been proposed in literature. Classical centralized TDMA mechanisms face the roadblock of non-scalability and inefficiency for large ad hoc and infrastructure-less sensor networks. To address the scalability issue, Funneling-MAC [13] uses TDMA in high-traffic parts of a network, and CSMA in the low-traffic parts. However, this protocol relies on a central sink to coordinate the TDMA slot allocation. Any centralized protocol suffers from single point of failure weakness. TreeMAC [1] and TDMA-W [10] address this issue by using a non-centralized approach. However they are dependent on time-synchronization which is a non-trivial task for distributed ad-hoc networks. Z-MAC [8] works in the presence of loose time-synchronization, but it still needs global time-synchronization at startup. The third category of TDMA mechanisms are fully distributed and work in the absence of time synchronization. ISOMAC [9] and DRAND [11] are examples of such distributed TDMA protocols which work independent of network-wide time synchronization and are ideal for completely distributed ad hoc networks. These approaches rely on control packets to forward 2-hop neighbors allocated slot information in the network. Therefore they are not suitable for highly privacy-sensitive networks especially when two or more wireless networks with different privacy domains are collocated and share the same wireless channel. To the best of our knowledge, the proposed mechanism is the first distributed TDMA protocol which works in the absence of zero in-packet control information dependency (in-band [1][9] or out-of-band [10][11][12]) and network timesynchronization.
V. SUMMARY AND ONGOING WORK We have presented a novel Zero Exposure distributed TDMA protocol in which slot occupancy information is disseminated by time-coded packet transmissions, and that is without using any explicit information in the packet. Packet arrival time is used as a source of information to exchange slot occupancy of 2-hop neighbors. The protocol does not rely on time-synchronization and it works in wireless ad-hoc networks where nodes belong to different privacy domains where the packets are fully encrypted. The functionality and evaluation of the protocol has been demonstrated using NS2 simulations. Ongoing work on this topic includes the evaluation of the protocol in the presence of packet loss and clock-drift. Other extensions include developing an energy efficient approach and an analytical model to estimate its performance. VI.
REFERENCES
[1] W. Song, R. Huang, B. Shirazi, and R. LaHusen, “TreeMAC: Localized TDMA MAC protocol for real-time high-data-rate sensor networks,” IEEE PerCom 2009 [2] IEEE Standard 802.11, Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification, IEEE, 1997. [3] Nucrypt LLC, Alpha EtaTM Noise Based Physical Layer Encryption, http://www.nucrypt.net/documents/nucrypt_sell_sheet.pdf [4] Y. Zhang, S. Zheng, and S. Xiong, “A Scheduling Algorithm for TDMA-Based MAC Protocol in Wireless Sensor Networks,” IEEE ETCS '09 [5] J. Elson, L. Girod, and D. Estrin, “Fine-grained network time synchronization using reference broadcasts,” ACM SIGOPS Operating Systems Review, vol. 36, 2002, pp. 147–163. [6] J. Degesys, I. Rose, A. Patel, and R. Nagpal, “DESYNC: SelfOrganizing Desynchronization and TDMA on Wireless Sensor Networks,” ACM/IEEE IPSN 2007 [7] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MAC protocol for wireless sensor networks,” IEEE INFOCOM 2002. [8] I. Rhee, A. Warrier, M. Aia, J. Min, and M. Sichitiu, “Z-MAC: A Hybrid MAC for Wireless Sensor Networks,” IEEE Transactions on Networking, vol. 16, 2008, pp. 511-524. [9] F. Yu, T. Wu, and S. Biswas, “Toward In-Band Self-Organization in Energy-Efficient MAC Protocols for Sensor Networks,” IEEE Trans. on Mobile Computing, vol. 7, 2008, pp. 156-170. [10] Z. Chen and A. Khokhar, “Self organization and energy efficient TDMA MAC protocol by wake up for wireless sensor networks,” IEEE SECON 2004. [11] I. Rhee, A. Warrier, J. Min, and L. Xu, “DRAND: Distributed Randomized TDMA Scheduling for Wireless Ad Hoc Networks,” IEEE Trans. on Mobile Computing, vol. 8, 2009, pp. 1384-1396. [12] J. Chen and S. Jiang, “Improvement of Slots Utilization with a Stealing-TDMA Protocol for Ad Hoc Network,” IEEE VTC 2006 [13] G. Ahn, S.G. Hong, E. Miluzzo, A.T. Campbell, and F. Cuomo, “Funneling-MAC: a localized, sink-oriented MAC for boosting fidelity in sensor networks,” ACM Sensys 2006
