Sei sulla pagina 1di 5

A Cross-Layer Distributed TDMA Scheduling for

Data Gathering with Minimum Latency in


Wireless Sensor Networks
Punyasha Chatterjee #1, Nabanita Das*2
#
Information Technology Department, Govt. College of Engg. & Textile Technology, Serampore
Kolkata, India
1
tachical@gmail.com
*
Advanced Computing and Microelectronics Unit, Indian Statistical Institute
Kolkata, India
2
ndas@isical.ac.in

Abstract— In a wireless sensor network, given a routing tree least one time slot in each frame, in which, they can transmit
for data gathering from individual nodes to the sink node, this without any collision. Furthermore, a node depending on the
paper presents a simple, distributed algorithm for assigning time schedules of its neighbouring nodes may remain in the sleep
slot to each node for conflict-free communication, such that the mode when it is neither to transmit, nor to receive, i.e. can
maximum latency in data gathering at the sink is minimized. It
switch off their transceiver conserving appreciable amount of
requires just a one-time computation during the initialisation of
the network provided the nodes remain static. Simulation studies energy.
have been done to evaluate the performance in terms of latency, A proper TDMA scheduling is contention-less and
and the average duty cycle. collision-free as well. Since each node gets an equal sized slot,
fairness is ensured. It is useful, when nodes are transmitting
Keywords-TDMA (time-division multiple access) scheduling; stream of data, the network has high contention probability or
data gathering; distributed time slot assignment; minimum latency; requires real-life constraints such as message delay. Specially,
duty cycle; interference for sensor networks where energy is a scarce resource, TDMA
helps to conserve energy by avoiding collision and hence
I. INTRODUCTION
retransmission, and also by allowing nodes to remain in sleep
A typical multi-hop Wireless Sensor Network (WSN) mode whenever possible.
consists of a large number of static, autonomous, unattended But traditional TDMA has the following difficulties:
sensor nodes, densely deployed over a field to gather 1) Time synchronization and negotiation in slot
information about the surroundings. These networks are schedules are required among nodes resulting
employed for critical applications like surveillance and remote increased message overhead, energy consumption
monitoring. The basic operations of such a network are and implementation complexity.
periodic sensing, data gathering and data transmission by 2) Sometimes, the slots remain unused leading to the
individual sensor nodes to the sink, which acts as the final loss of bandwidth.
data aggregation point, via some intermediate nodes. The 3) Clocks can drift and tight synchronization might
sensor nodes have limited battery-power and they mainly incur too much overhead.
deplete energy in the process of data communication. However, a guaranteed packet delivery and bounded
Therefore, it is very important to use energy-efficient latency are highly desirable in critical applications, which can
protocols at each level of operation. be ensured in TDMA.
Moreover, in wireless sensor networks, nodes share the In multihop TDMA scheduling, spatial reuse of time slots
communication medium. It is the responsibility of the MAC is possible, i.e. more than one node can transmit at the same
protocol to control the media access. These protocols can be time slot if their receivers are beyond the interference range of
either contention-based or schedule-based. each other. There are two types of conflicts, namely primary
In contention-based scheme i.e. in CSMA (carrier-sense conflict and secondary conflict. Primary conflict occurs when
multiple access), nodes check the channel before transmission a node transmits and receives at the same time slot or receives
and if the channel is busy, they randomly back off for a short more than one transmission, destined to it at the same time
time and try again. It is better for networks, where the slot. Secondary conflict occurs when a node, an intended
contention is low and burst traffic is expected. receiver of a particular transmission, is also within the
In schedule-based protocol like TDMA (Time Division transmission range of another transmission, intended for other
Multiple Access), time is split into equal time intervals, called nodes.
time frames and each time frame is further divided into time- So far, a lot of research has been done for solving the
slots of equal length such that individual nodes are assigned at routing problem in sensor networks in energy-efficient way

978-1-4244-4067-2/09/$25.00 © 2009 IEEE 813 Wireless VITAE’09


[14], [17], [18]. Several criteria such as power control at each scheduling algorithm, section IV shows the simulation results
node, minimum power path from any node to the sink node, and section V concludes the paper.
load balancing in each node [12], or lifetime maximization
[15], [16] for the network as a whole, have been considered as II. PROBLEM DEFINITION
the basis for deciding the routes from individual nodes to the This paper assumes that a WSN consists of a set of n
sink node. Whatever be the criteria, if it is assumed that homogeneous static sensor nodes, V = {1, 2,…, n} and a sink
sensors basically collect data and forward it via multihop node (n+1) distributed over a 2-D region. The positions of the
paths to the sink node, applying data aggregation in the sensor nodes and also the sink node are fixed, but not known
intermediate nodes, determination of routes from each globally. Each sensor generates one data packet of fixed
individual node basically is mapped to the problem of finding length per round, to be sent to the sink node. Each sensor
an appropriate spanning tree in the WSN. However, all these starts with an initial energy E that is depleted at each time the
works assume a conflict-free transmission protocol in node transmits or receives. The sink has unlimited amount of
existence in the MAC layer. This paper addresses the problem energy.
that given a spanning tree for routing in a given sensor
network, how we can minimize the frame length such that A. Network Topology
each node is assigned at least one slot per frame for conflict- Let us consider a multihop WSN with a set of n
free transmission keeping the latency as small as possible. So, homogeneous nodes, V = {1, 2,…, n} and a sink node (n+1)
basically it is a cross-layer optimisation problem where distributed randomly over a 2-D region. Depending on its
depending on the routing tree, the MAC layer assigns the slots neighbourhood, each sensor-i can adjust its transmission
to individual nodes so as to minimize the maximum latency. It power Pi ≤ Pmax (the maximum transmission power possible
is obvious that in this case, the scheduling will be more for the sensors) that can reach a subset of nodes in single hop
efficient as it considers the scheduling problem on a reduced lying within its range Ri corresponding to Pi.
graph (routing tree) instead of the original topology graph.
Earlier works on TDMA scheduling in sensor networks Definition 1. Given a WSN with a set of n homogeneous static
considered the problem in the MAC layer only, ignoring the sensor nodes, V = {1, 2,…, n} and a sink node (n+1),
routing details. A synchronized link-based TDMA scheduling assuming that each node operates with same range R
is presented in [1] to maximize network throughput. A multi (corresponding to Pmax), the network topology is represented
objective particle swarm optimisation is developed in [2] to by a graph G (V, E), where E = {(i, j)| i, j ∈V}, if the distance
find a better trade off between delay and energy consumption. between nodes i and j d(i, j)≤ R. G (V, E) is defined as the
These algorithms do not focus on minimum latency. A topology graph of the given sensor network.
TDMA link scheduling, which ensures minimum delay, is
presented in [3], but the algorithm is a centralized one. Definition 2. The set of nodes Mk(i)⊂V is called the i-hop
Distributed scheduling algorithms are proposed in [4]-[6], but neighbour set of node k, if for each j∈Mk (i), p(j, k)<i, where
each requires some computations to be done in centralized p(j,k) is the minimum hop distance between nodes j and k.
way. Hence these are not truly distributed. Algorithm in [4] is
not applicable for general trees and there is chance of token Definition 3. The number of hops along the shortest path from
loss in [5]. Distributed TDMA slot assignment algorithms are any node i to the sink node (n+1) is termed as the hop-count
developed in [7], [8] assuming the network is fully connected (hi) of node i.
which is not realistic. Randomised slot assignment strategies
are proposed in [8]-[10], which are not always collision free. B. Routing Tree
An edge colouring based scheduling is presented in [11] Given the topology graph G(V,E), for data aggregation at
where the overall message and time complexity is too high. the sink node, each sensor node has to decide one of its 1-hop
In this paper, it is assumed that for a multi-hop WSN, a neighbours as its next node for data routing. Determination of
spanning routing tree is known for data gathering from routes from each individual node basically is mapped to the
individual nodes to the sink. Next a distributed TDMA problem of finding an appropriate spanning tree in the WSN.
scheduling algorithm is proposed, which determines conflict- A routing tree or spanning tree can be constructed from a
free assignment of slots to the nodes such that the maximum topology graph based on different criteria like minimizing
latency to reach data from the nodes to the sink node is energy consumption along each path, i.e., shortest power path,
minimized. No knowledge about the physical position of the or balancing load in each node, or maximizing network
nodes is required. It requires just a one-time computation lifetime [12], or minimizing transmission delay of different
during the initialisation of the network provided the nodes routing paths etc.
remain static. Simulation studies have been done to evaluate
the performance of the proposed algorithm with respect to Definition 4. Given a topology graph G (V, E), the routing
latency and duty cycle. tree is a spanning tree T (V, A), rooted at the sink node, i.e.,
The rest of the paper is organized as follows: section II A⊂ E corresponding to the active transmission links along
defines the problem, section III presents the distributed which sensed data from individual nodes are routed to the
sink node.

814
Definition 5. Given a routing tree T (V, A), the interference
graph is C (V, I), where I ⊂ V×V is the set of edges such that
(i,j)∈ I, if d (i, j) ≤ max. {Ri, Rj}, assuming that each node-i
adjusts its range Ri such that Ri = max. {d (i, k)| (i, k)∈ A}.

Definition 6. The Scheduling graph, corresponding to T (V, A)


and C (V, I), is called S(V, AI), where AI=A∪I.

Example 1: Fig. 1 shows a topology graph G(V,E) with 15


nodes assuming each node transmits with same power Pmax.
Fig. 2 shows a routing tree T(V,A), corresponding to G(V,E) Fig. 1. Topology Graph G(V, E)
and in Fig.3, a scheduling graph S(V, AI) is shown, where the
interference edges are represented by dashed lines.
C. Data Aggregation Model
Given the routing tree, data sensed by each node are
forwarded to the sink node along the spanning tree. A 100%
data aggregation [15] model is assumed here by which each
intermediate node receiving data packets from its children
nodes, encapsulates all information along with its own data
into a single packet and forwards it to its parent node that Fig. 2. Routing Tree T(V, A)
finally reaches the sink node. With this data aggregation
model, each node will require just a single slot in a TDMA
frame for data routing. The duration of a time slot is assumed
to be sufficient to transmit one data packet only.
D. Time-slot Assignment Problem
With the system model described above, given the
scheduling graph S for a WSN, the objective of this paper is to
determine the conflict-free time slots for each node in such a
way that the maximum latency to reach a packet from a node
to the sink is minimized.
In this paper, it has been assumed that the transmission and Fig. 3. Scheduling Graph S(V, AI)
interference range of any node are same, and hence no 2-hop
A. Message Types
neighbors of a node will have same slot in a time frame to
avoid collision. To ensure that the slots selected by each node based on its
The TDMA slot assignment problem can be modelled as a local information only, are always conflict-free, nodes need to
direct extension of finding worst-case chromatic number in exchange the following messages:
the scheduling graph with the restriction that no colour can be 1) ‘SELECT’: After selecting its own time slot, a
reused within distance 2 in the graph which is known to be node broadcasts ‘SELECT’ message to its 2-hop
NP-hard [13]. Here, the added restriction is that along any neighbours.
path of routing tree T embedded in S, the assigned time slots 2) ‘READY’: When a node either finds itself as a
should be in increasing order from leaf towards sink. leaf node (no child) or receives ‘SELECT’
message from all of its children, it broadcasts
III. DISTRIBUTED SCHEDULING ALGORITHM ‘READY’ message to its 1-hop neighbours.
This paper presents a distributed algorithm for TDMA slot 3) ‘PERMIT’: From the queue of ready nodes, a node
assignment for the nodes in a wireless sensor network selects the minimum id node and sends a
assuming that the routing tree (T) is already constructed. The ‘PERMIT’ message, if the wait list is empty.
algorithm is distributed by nature. For the execution, each 4) ‘TERMINATE’: On receiving ‘SELECT’
node only requires the information about its 2-hop neighbours. messages from all its 1-hop neighbours, sink
It is assumed that: broadcasts a ‘TERMINATE’ message.
1) Each sensor node has unique-id. B. Procedure Time Slot Selection
2) Each node knows its 2-hop neighbours.
For each node-i with hop-count hi:
3) Each node knows its parent and children in T. Input: Mi(2) the set of 2-hop neighbours; Mi(1): the set of 1-hop
Based on this information, each node finds its slot to keep neighbours; hop-count hi; hop-count hj for each j ∈ Mi(2);
the latency as small possible when its turn comes. Ns : set of children of i.

815
Output: time slot (i)
Each node-i Initialises time-slot (i) =-1; Q (i)=φ; wait=-1;
1. If Ns = φ broadcast a ‘READY’ to its 1-hop neighbours;
2. If receives ‘SELECT’ from j ∈ Mi(2) then
Store time-slot (j);
If j is a child of i, delete j from Ns;
If wait=j then wait=-1;
If Q (i)≠ φ, select minimum id k from Q(i) and send
‘PERMIT’ to k; wait=k;
3. If receives a ‘READY’ from j ∈ Mi(1) insert j in Q(i);
If wait=-1then select minimum id k from Q(i) and send
‘PERMIT’ to k; wait=k; Fig. 4. Time Slots Assigned to Each Node of the Scheduling Graph
4. If receives ‘PERMIT’ from all j ∈ Mi(1),
IV. PERFORMANCE EVALUATION
Find out time-slot (i) i.e. conflict-free slot in 2-hop
neighbourhood, which is higher than the slots of children; For simulation studies, a 100unit×100unit deployment
Broadcast ‘SELECT’ with time-slot (i) to all j ∈ Mi(2); region is considered. Number of nodes (n) has been varied
from 100 to 500. Nodes are deployed randomly. The sink
Example 2: The proposed time slot selection procedure node is located at the mean position of the deployment region.
executed on the scheduling graph S(V,AI), shown in Fig. 3, Fig. 5 shows the variation of frame length, varying the total
assigns the time slots as shown in Fig. 4, resulting a frame number of nodes, keeping R = 20 units, generated by four
length of 12. different routing algorithms, namely shortest-power path,
least-hop path, minimum spanning tree (MST) and load-
C. Correctness and Termination
balanced routing [12] algorithms. In shortest–power path and
It is to be noted that the proposed algorithm guarantees that least-hop path routing, at each round, data sensed by each
each node starts the slot selection procedure after all its node is routed to the sink via intermediate nodes through the
children nodes complete it and also no two nodes within shortest paths with respect to power consumption and hop
distance two can start the procedure simultaneously. distance from the sink respectively. In MST routing algorithm,
The procedure terminates when all the nodes-j with hj = 1 MST is computed such that total power consumption of the
completes the time-slot-selection procedure. On receiving all network is minimized. For each value of n, frame lengths of
the ‘SELECT’ messages from its 1-hop neighbours, the sink 20 different scheduling graphs are plotted for each routing
broadcasts a ‘TERMINATE’ message. All sensor nodes then technique. Frame-length is increasing with the increase in n,
start their normal functions, i.e., sensing, receiving messages as the number of 2-hop neighbours in the scheduling graph
from its children (in the routing tree T) in the respective slots, increases with n.
and aggregating the data with its own and forwarding the In general, for a given topology graph G, the frame length
packet in its assigned slot to its parent. During the slots when is upper bounded by the maximum number of 2-hop
it is neither to receive data, nor to transmit, it may go to the neighbours existing in G, since no slot can be reused within
sleep mode and thus may conserve appreciable amount of distance 2 in G. Fig. 6 shows that for a typical scheduling
energy by reducing the duty cycle. graph (500 nodes), generated by load-balanced routing [12],
the frame-length obtained by the proposed TDMA algorithm
Remark: From the algorithm it is evident that here the is much lower than the upper bound, and also than that
maximum latency is same as the frame length of the schedule required for scheduling in the original topology graph.
which is the worst-case latency for data from a leaf node to Duty cycle of a node is the ratio p/m, where m is the frame
reach the sink node. length of the schedule and p is the number of slots during
D. Cross-Layer Issues which a node is either receiving or transmitting, i.e., not in
sleep mode. Fig. 7 shows the variation of average duty cycle
It is to be noted that the construction of routing tree and the by varying the total number of nodes, keeping R=20units,
power control that is applied to construct the routing tree T generated by different routing algorithms. It has been found
and the corresponding scheduling graph S are to be performed that load-balanced routing performs better compared to other
in the routing layer. It also requires a communication platform ones.
offered by MAC layer that may be TDMA, or any other In Fig. 8, the same parameters are shown in load-balanced
efficient MAC protocol. However, once S is constructed, the routing varying the number of nodes for different values of R.
proposed procedure executed on S, assigns timeslots to the It has been found that average duty cycle varies in the range of
nodes for TDMA, and optimises the frame length for 1-7%, showing that nodes may conserve power appreciably in
minimum latency. In this way, here the solution achieved in dense networks.
the routing layer influences the media control of MAC layer,
and results a cross-layer protocol for energy efficiency in V. CONCLUSION
WSN. For multihop wireless sensor networks, this paper presents
a simple distributed algorithm for assigning conflict-free time

816
slots to the individual nodes for transmission, assuming that
the sensor nodes aggregate their data to the sink along a 0.08

spanning tree. No information about the location of the sensor 0.07


R=20

nodes is required. Based on the information about its 2-hop 0.06 R=25

Avg. Duty Cycle


R=30
neighbourhood only, each node selects its conflict-free slot so 0.05

that the latency at each stage is minimum. Performance 0.04

evaluation shows that the latency (in terms of frame length) is 0.03

smaller than the upper bound (the maximum number of 2-hop 0.02

neighbours in the graph). It reveals the fact that instead of 0.01

considering the original topology graph for TDMA scheduling, 0


the latency can be improved by far, if we consider the routing 100 200 300 400 500

tree using a cross-layer protocol. Also, the reduced duty cycle # Nodes
helps to enhance the lifetime of individual nodes.
140
Fig. 8. Variation of Average Duty Cycle with Transmission Range

120 Least Hop REFERENCES


Least Power
100 [1] W. Wang, Y. Wang, X. -Y. Li, W. -Z. Song, O. Frieder, “Efficient
Frame_Length

MST
Load Balanced Interference Aware TDMA Link Scheduling for Static Wireless
80 Networks”, MobiCom’06.
[2] T. Wang, W. Zhiming, J. Mao, “A New Method for Multi Objective
60
TDMA Scheduling in Wireless Sensor Networks using Pareto-Based
40 PSO and Fuzzy Comprehensive Judgment”, HPCC 2007, LNCS 4782,
pp. 144-155.
20 [3] S. Cui, R. Madan, A. J. Goldsmith, S. Lall, “Energy Minimization and
Delay Analysis in TDMA Based Sensor Networks”, IEEE Tran. on
0
100 200 300 400 500
wireless comm., 2004.
[4] S. Gandham, Y. Zhang, Q. Huang, “Distributed Minimal Time
# Nodes Convergecast Scheduling in Wireless Sensor Networks”, Proc.
ICDCS’06.
Fig. 5. Comparison of Frame Length for Different Routing Algorithms [5] S. C. Ergen, P. Varaiya, “TDMA Scheduling Algorithms for Sensor
Networks”, Dept. of Elec. Engg. & Comp. Sc., Univ. of California,
600
Berkeley, Tech. Rep.2005.
Frame Length [6] A. Sridharan, B. Krishnamachari, “Max-Min Fair Collision-free
500
Scheduling for Wireless Sensor Networks”, MWN'04.
Upper Bound of
Scheduling graph
[7] J. Degesys, I. Rose, A. Patel, R. Nagpal, “DESYNC: Self-Organizing
Desynchronisation and TDMA on Wireless Sensor Networks”,
Frame Length

400 Upper bound of


Topology graph IPSN’07.
300
[8] Y. Wang, I. Henning, “A Deterministic Distributed TDMA Scheduling
Algorithm for Wireless Sensor Networks”, Proc. WiCOM 2007, pp.
2759-2762.
200
[9] B. Hohit, L. Doherty, E. Brewer, “Flexible Power Scheduling for
Sensor Networks”, Proc. IPSN’04, pp. 205-214.
100
[10] I. Rhee, A. Warrier, J. Min, L. Xu, “DRAND: Distributed Randomized
TDMA Scheduling for Wireless Ad-hoc Networks”, Proc.
0
10 15 20 25 30 35 40 MobiHoc’06, pp. 190-201.
Transmission Range [11] S. Gandham, M. Dawande, R. Prakash, “Link Scheduling in Sensor
Networks: Distributed Edge Coloring Revisited”, Proc. INFOCOM
2005, pp. 2492-2501.
Fig. 6. Variation of Maximum Latency with Transmission Range in a Typical
[12] P. Chatterjee, N. Das, “A Distributed Algorithm for Load-Balanced
Network (# Nodes=500)
Routing in Multihop Wireless Sensor Networks”, Proc. ICDCN’08,
0.09
LNCS 4904, pp. 332-338.
[13] S. Ramanathan, “A Unified Framework and Algorithm for (T/F/C)
Least Hop
0.08
DMA Channel Assignment in Wireless Networks”, Proc. IEEE
Least Power
0.07 INFOCOM’97, pp. 900-907.
MST
[14] C. Wen-yu, J. Xin-yu, Z. Yu, C. Kang-sheng, “A Load-Balanced
Avg Duty Cycle

0.06 Load Balanced


Minimum Energy Routing Algorithm for Wireless Ad Hoc Sensor
0.05 Networks”, Journal of Zhejiang Univ.y SCIENCE A, 2006, pp. 502-506.
0.04 [15] S. Lindsey, C. S. Raghavendra, “PEGASIS: Power-Efficient Gathering
in Sensor Information Systems”, Proc. IEEE Aerospace Conf.’02,
0.03
pp.1-6.
0.02 [16] M. Cardei, D. _Z. Du, “Improving Wireless Sensor Network Lifetime
0.01 through Power Aware Organization”, ACM Wireless Networks, vol.11,
no.3, pp.333-340, May 2005.
0
100 200 300 400 500
[17] L. Lin, N. B. Shroff, R. Srikant, “Energy-Aware Routing in Sensor
# Nodes Networks: A Large Systems Approach”, Proc. WONS 2006, pp-159-
169.
[18] C. Schurgers, M. B. Srivastava, "Energy Efficient Routing in Wireless
Fig. 7. Comparison of Average Duty Cycle for Different Routing Algorithms
Sensor Networks", Proc. MILCOM'01, pp. 357-361.

817

Potrebbero piacerti anche