ENERGY/LATENCY TRADE-OFFS IN GEOGRAPHIC ROUTING

FOR ULTRAWIDEBAND WIRELESS SENSOR NETWORKS

Samuela Persia

1,3

and Dajana Cassioli

2,3

Fondazione Ugo Bordoni, Via del Policlinico 147, 00161, Rome, Italy

Computer Science Department, University of L’Aquila, Via Vetoio 1, 67010, Coppito, AQ, Italy

RADIOLABS Associated Laboratory, University of L’Aquila, Via G. Gronchi 18 Zona Ind.le di Pile, 67100, L’Aquila, Italy

Keywords:

Energy-aware Wireless Sensor Networks, Geographic Routing, Latency, Location Tracking UWB Networks.

Abstract:

Wireless Sensor Networks (WSNs) may exploit accurate localization capabilities of ultrawideband (UWB)

transceivers to improve performance of high layer protocols. We analyze power consumption of a WSN

accommodating both communication and positioning into the same UWB transceiver and implementing a ge-

ographic routing algorithm, either the Greedy Perimeter Stateless Routing (GPSR) or its energy aware version,

the e-GPSR. Power consumption depends on beacon rate for positions’ updates, the number of hops to reach

the destination and the number of neighbors per node. On the other hand, the beacon rate impacts the reliabil-

ity of the neighbor lists; the number of hops impacts on the end-to-end latency; the number of neighbors, i.e.

the network connectivity, impacts on routing performance. The presented analysis assesses, by means of both

theoretical investigations and simulation results, the main trade-offs between power consumption and latency

that can be applied to obtain the best achievable performance.

1 INTRODUCTION

WSNs have attracted great interest in both research

and commercial communities for their wide applica-

tion range and versatility. Typical applications require

nodes with average power consumption substantially

lower than for other wireless networks. In recent

years the research focused on how to overcome the

power issue by introducing energy-aware functional-

ities in WSNs design, even though a common criteria

for all applications cannot be established.

UWB technology enables low power transmis-

sions, thanks to the efﬁcient use of spectrum and low

duty-cycle signals, and the easy integration of local-

ization functionality. The use of power-awareness and

location aided algorithms at network (NWK) layer ap-

pears as the best choice. Performance of such algo-

rithms is strictly related to the knowledge of neigh-

bors information (Stojmenovic, 2002).

In this paper we consider the Greedy Perimeter

Stateless Routing (GPSR) algorithm (Karp and Kung,

2000) and the energy-aware version, namely the e-

GPSR, we recently proposed in (Persia and Cassioli,

2007). Our analysis shows the trade-offs that should

be applied when designing this type of networks, be-

cause the main factors impacting energy consump-

tion also impact latency and routing performance. We

show that the e-GPSR guarantees low power con-

sumption and latency, thanks to its intrinsic forward-

ing criteria.

2 PROBLEM STATEMENT

Basic requirements of WSNs are low power consump-

tion (uniformly distributed among network nodes)

and low latency (especially for mission-critical appli-

cations). For low data rate location/tracking (LDR-

L/T) UWB WSNs exploiting geographic routing, a

further requirement is given by low beacon trafﬁc,

aiming to the best balance between positioning and

communication tasks and to the optimal beacon up-

date rate to maintain good performance of geographic

forwarding. The power consumption of each node

depends on its own activities, while the latency de-

pends on node’s connectivity and the reliability of

geographic routing depends on the beacon rate and

nodes’ mobility. In the following, we formulate the

problem statement through theoretical investigations,

showing the interrelations between power consump-

tion, latency and neighbor lists updating.

115

Persia S. and Cassioli D..

ENERGY/LATENCY TRADE-OFFS IN GEOGRAPHIC ROUTING FOR ULTRAWIDEBAND WIRELESS SENSOR NETWORKS.

DOI: 10.5220/0003834601150118

In Proceedings of the 1st International Conference on Sensor Networks (SENSORNETS-2012), pages 115-118

ISBN: 978-989-8565-01-3

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

The total energy consumed by a WSN of N nodes to

transmit M packets can be expressed as (Sozer et al.,

2000; Wang et al., 2006):

tot

∑

j=1

∑

i=1

i, j

A(r

) (1)

where n

is the number of hops needed by the i−th

source node to reach the destination, E

i, j

is the min-

imum energy necessary for the packet j transmitted

by the i−source to be received by the next hop, r

its

coverage range and A(r

) ∝ r

loss

is the path loss at-

tenuation with 3 ≤ α

loss

≤ 4 for harsh environments.

Let’s assume for sake of simplicity that the farthest

node from the destination, i.e. at a distance d, needs

N hops to transmit one packet and that the coverage

range is r

= r = d/N for all nodes, thus E

i, j

= E

Its total consumed energy is:

hop

= E

) + E

)2. .. + E

= E

)

N(N + 1)

(2)

For a number of hops N

< N, we have A(

) <<

) and





N(N + 1)

< E





+ 1)

, (3)

thus paths with a large number of hops consume less

energy than paths involving a small number of hops.

On the other hand, large numbers of involved hops

imply increased latency.

The latency in a network is commonly measured

in terms of number of hops required to relay a packet

from the source (S) to the destination (D) located at

a distance d. With reference to Fig. 1, applying the

Figure 1: Forwarding mechanism.

analysis in (Cheng and Robertazzi, 1989), the ex-

pected value of the vacant distance r

for single-hop

distance is:

E[r

] =

1 − e

−µ(d−r

)

(1 + µ(d − r

i−1

))

µ(1 − e

−µ(d−r

ei−1

)

(4)

where r

and r

are depicted in Fig. 1, with r

= d

and µ is the density of nodes in the area. The expected

value of the r

is E[r

] = d − E[r

] and the average

number of hops l needed to transmit a packet from S

to D is:

E[l] = d

E[r

]

e (5)

where d·e is the ceiling operator and r is the nodes’

coverage radius. Evaluating (5) for different cover-

age radii in a WSN of N = 40 nodes and d =10 m,

in a sensing area of 25 × 25 m

, we obtain that the

average number of hops is around 2 for r =6 m,

whereas it drastically increases for a small coverage

radius. Hence, large nodes’ coverage range is needed

to achieve low latency, with a consequent increase of

power consumption. Accurate analysis of power con-

sumption in LDR-L/T UWB sensor networks, which

usually produce low data trafﬁc, cannot leave out the

energy fraction spent for the positioning beacon trans-

missions, exploited at the NWK layer by location

aided routing. Frequent beacon transmissions pro-

vide up-to-date positioning in mobile scenarios that

guarantees the highest level of delivery success rate of

geographic routing algorithms. The drawback is that

frequent beacon transmissions in a uncoordinated ac-

cess scheme (aloha) increase the collision probability,

with consequent packet loss.

Performance analyses of geographic routing that

assume a perfect knowledge of neighbors positions

available at any time without control trafﬁc are valid

for static scenarios, but are not realistic for mobile

scenarios. Our analysis uses this ideal case as a

benchmark of achievable performance (CASE: Ideal),

whereas two other realistic scenarios are deﬁned to

investigate the two effects of the beacon trafﬁc: an

increase of the overall trafﬁc load, which is managed

ﬂatly by the aloha MAC, and a potential loss of po-

sition updates. The ﬁrst scenario models the bea-

con trafﬁc load as a trafﬁc source that adds up to the

data trafﬁc, but the perfect knowledge of exact neigh-

bors position at any time is still assumed. The aim

of this test case (CASE: Beacon) is to ﬁnd the max-

imum beacon repetition rate tolerable by the WSN

such that packet collisions do not cause the collapse

of the WSN. An even more realistic scenario (CASE:

POS+Beacon), where each node periodically broad-

casts beacons and periodically acquires neighbors in-

formation at each beacon interval T , is used to evalu-

ate the impact of loss of position updates.

3 PERFORMANCE ANALYSIS

We consider a UWB sensor network of M nodes lo-

cated in random positions in an area of 25 m × 25 m.

Anchor nodes are static, the others mobile. All nodes

SENSORNETS 2012 - International Conference on Sensor Networks

116

within the communication range r of a node are con-

sidered as its neighbors for the GPSR, whereas a sub-

set is selected for the e-GPSR according to its selec-

tion criteria explained in the following.

The sink node (SN) is assumed to have a commu-

nication range R able to cover the entire area, such

that downlink communications are direct and without

errors. The SN queries to the network the position of

a randomly chosen node. Every node receiving the

query veriﬁes whether it is the right destination: if

not, it discards the packet. Otherwise, it sends back to

the SN the reply packet through a multi-hop path, re-

actively created by means of one of the two versions

of the geographic algorithm under test.

The physical layer is modeled by the cascade of

a SNR Evaluator and a Decider. The SNR Evaluator

computes the signal-to-noise ratio (SNR) associated

to the received packets and the Decider compares its

output with a threshold SNR

, discarding the packets

whose SNR is below.

The MAC layer implements a pure aloha scheme.

At the NWK layer we adopt a routing scheme

based on the GPSR (Karp and Kung, 2000). Two

versions of the GPSR are compared, the traditional

and the energy-aware version in (Persia and Cassi-

oli, 2007). The difference between these two ver-

sions stays in the selection criteria of the next hop

in the Greedy mode, whereas there is no difference

in the Perimeter mode, which is the recovery strat-

egy in case of failure of the Greedy approach. During

the Greedy mode the traditional GPSR selects the next

hop according to the MFR (Most Forward within Ra-

dius) approach, i.e. the neighbor geographically clos-

est to the packet destination is selected. The e-GPSR

adds to the MFR two other criteria in the Greedy mode

which take into account both energy and latency con-

straints. For each node j, the possible candidates for

the next hop constitute a subset, N

(j)

∩

(j)

, of the N

(j)

neighbors of the present hop node j, such that the fol-

lowing conditions are veriﬁed:

(j)

⊆ N

(j)

| ∀i ∈ N

(j)

, α

≤ Θ

(j)

(6)

(j)

⊆ N

(j)

| ∀k ∈

(j)

, E

≥ E

(j)

T H

(7)

where N

(j)

and

(j)

are the subsets of nodes that ver-

ify (6) and (7), respectively, α

is the angle between

the line that joins the node j with the i-th node and the

line that joins the node j with the destination (the SN

for our cases), and E

is the residual charge of the k-th

neighbor node. The condition (6) can be seen as a la-

tency constraint, whereas (7) is the power constraint.

Each forwarding node j adaptively sets the angle

(j)

as Θ

(j)

= α

(j)

+ x

· (90 − α

(j)

) and the thresh-

old E

(j)

T H

as E

(j)

T H

= (1 − x

) · E

(j)

charge

, where α

(j)

is the

Table 1: Simulations’ setup.

RANDOM WAY POINT Speed: 0.5 m/s

MOBILITY MODEL Pause time: 0.1 s

DATA RATE R

= 1 Mb/s

TRAFFIC MODEL CBR at 1 packet/s

84 bits per packet

BATTERY CHARGE Initial: E

= 10 mAh

minimum angle obtained by considering, among all

neighbors of the node j, the nearest node to the j − D

line, E

(j)

charge

is the maximum residual charge among

the residual charges E

of its neighbors, and x

∈ [0,1]

and x

∈ [0,1] are routing design parameters (Persia

and Cassioli, 2007), which modulate the number of

neighbors of each node that varies not only due to the

coverage range.

Performance analysis has been carried out by dy-

namic system simulations in Omnet++. The simula-

tion assumptions and parameters are listed in Table 1.

For each case under study, we average the statistics

over a set of 5 independent stochastic discrete-event

simulations, each having a duration of 100 s. Fig. 2

shows the average and standard deviation (STD) of

residual charge for all nodes. In the presence of bea-

con trafﬁc for the GPSR, the average level of resid-

ual charge increases with increasing beacon inter-

vals, whereas the STD decreases for large beacon in-

tervals, indicating a not uniform discharge through-

out the WSN. For the e-GPSR, the average resid-

ual charge slightly decreases in the presence of bea-

con trafﬁc, and its dependence on the beacon in-

terval is negligible. The STD is constant with re-

spect to the beacon interval and slightly greater than

the ideal case. Hence, nodes discharge quite uni-

formly over the whole network, increasing the net-

work lifetime and the reliability of links and position-

ing. Nevertheless, the above analysis does not take

Figure 2: Average (upper plot) and standard deviation

(lower plot) of the residual charge of network nodes.

ENERGY/LATENCY TRADE-OFFS IN GEOGRAPHIC ROUTING FOR ULTRAWIDEBAND WIRELESS SENSOR

NETWORKS

117

into account the success delivery rate of the consid-

ered routing algorithms. Not all data packets are cor-

rectly received at the destination, because some pack-

ets are discarded somewhere, due to packet collisions,

thus their relaying does not consume the expected en-

ergy level. Since the GPSR sends more control pack-

ets than the e-GPSR, because its neighbor list includes

more nodes, the GPSR’s energy consumption shown

above is sligthly lower than it would be if a larger

number of packets reached the destination. We may

formulate this issue in terms of average energy efﬁ-

ciency of all nodes, deﬁned as:

∑

j=1

(1 − x

( j))E

( j)

(8)

where E

is the initial energy, x

( j) is the residual

fraction of E

of the j node at the end of simulation,

( j) is the number of correctly delivered packets.

Fig. 3 depicts the average energy efﬁciency ver-

sus the beacon interval, assuming an initial energy

= 15 dBm (compliant to UWB regulation) and a S-

D distance d ' 10 m. Although the theoretical model

(2) predicts an energy consumption below 15 dBm

only when 40 hops are used, the GPSR consumes

' 30 dBm and the e-GPSR ' 5 dBm for a beacon

interval of 3 s with a maximum of 6 hops, as shown

in Fig. 4. There, the bar plot of the latency factor,

Figure 3: Average energy efﬁciency needed to transmit a

packet from the farthest node to the sink node.

i.e. the average number of hops necessary to deliver

data packets from S to D, is depicted. The e-GPSR

exhibits a comparable behavior for the two coverage

radii. Hence, a suitable trade-off between power and

latency can be found. The GPSR shows a non uniform

distribution of latency factor. This attitude becomes

remarkable in the realistic scenario, where packets are

delivered towards the destination only when the actual

route is discovered; otherwise packets are forwarded

among useless hops, due to the non perfect knowledge

of nodes’ positions.

Figure 4: Latency factor for coverage radii of 4m and 6m,

assuming a beacon interval of 3s.

4 CONCLUSIONS

Power consumption and latency of GPSR and e-

GPSR for a UWB WSN have been evaluated and

compared, based on theoretical models at ﬁrst, then

through dynamic system simulations. In realistic con-

ditions, i.e. in the presence of control and data traf-

ﬁc in the network, the e-GPSR is able to provide a

low power consumption and latency. The simultane-

ous satisfaction of both constraints enable the use of

UWB WSNs for time-driven applications such as in-

dustrial chain monitoring or security/alarm scenarios.

ACKNOWLEDGEMENTS

This work was supported by the ERC Starting Inde-

pendent Researcher Grant VISION (Contr. 240555).

REFERENCES

Cheng, Y. and Robertazzi, T. (1989). Critical connectivity

phenomena in multihop radio models. IEEE Transac-

tions on Communications, 37:770–777.

Karp, B. and Kung, H. (2000). GPSR: Greedy Perimeter

Stateless Routing for Wireless Networks. In Proc. of

the 6th Annual Intern. Conf. on Mobile Comput. and

Network, pages 243–254, Boston, MA, USA.

Persia, S. and Cassioli, D. (2007). Routing Design for

UWB Sensor Network in Industrial and Logistic Sce-

narios. In Proc. of the IST MOBILE SUMMIT, Bu-

dapest, Hungary.

Sozer, E. M., Stojanovic, M., and Proakis, J. G. (2000). Un-

derwater acoustic networks. IEEE Journal of Oceanic

Engineering, 25(1).

Stojmenovic, I. (2002). Position-Based Routing in Ad-Hoc

Networks. IEEE Comm. Mag., 40(7):128–134.

Wang, Q., Hempstead, M., and Yang, W. (2006). A realis-

tic power consumption model for wireless sensor net-

work devices. In Sensor and Ad Hoc Communications

and Networks (SECON).

SENSORNETS 2012 - International Conference on Sensor Networks

118