Wireless Networks 10, 29–42, 2004

 2003 Kluwer Academic Publishers. Manufactured in The Netherlands.

Distributed Power Control for Energy Efficient Routing

in Ad Hoc Networks
PIERPAOLO BERGAMO, ALESSANDRA GIOVANARDI ∗ , ANDREA TRAVASONI, DANIELA MANIEZZO,
GIANLUCA MAZZINI and MICHELE ZORZI
DI, University of Ferrara, via Saragat 1, 44100 Ferrara, Italy

Abstract. In this paper, distributed power control is proposed as a means to improve the energy efficiency of routing algorithms in ad hoc
networks. Each node in the network estimates the power necessary to reach its own neighbors, and this power estimate is used both for
tuning the transmit power (thereby reducing interference and energy consumption) and as the link cost for minimum energy routing. With
reference to classic routing algorithms, such as Dijkstra and Link State, as well as more recently proposed ad hoc routing schemes, such
as AODV, we demonstrate by extensive simulations that in many cases of interest our scheme provides substantial transmit energy savings
while introducing limited degradation in terms of throughput and delay.
Keywords: ad hoc networks, energy efficiency, power control

1. Introduction works, namely energy consumption [6,7], selecting the short-

est path is not the best choice in general, since a number of
In the recent literature, ad hoc networks have gained much short hops usually results in less energy consumption com-
attention, due to the convenience of building mobile wireless pared to fewer longer hops. A few papers have recently ap-
networks without any need for a pre-existing infrastructure. peared, in which various routing strategies are compared in
An ad hoc network is a collection of wireless mobile hosts terms of energy consumption, i.e., based on the amount of
which are able to cooperatively establish communications, us- energy which is spent in correctly delivering a packet to its
ing no fixed infrastructure or centralized administration. In final destination. A simulation study in [8] compares several
such an environment, each host acts as a router and forwards routing protocols with respect to energy consumption, with-
packets to the next hop in order to reach, through multiple out proposing any improvement to those protocols to achieve
hops, the final destination. better energy efficiency. Another study of the energy con-
Many issues need to be addressed in ad hoc networking, sumption of traditional ad hoc routing protocols has been
one of the most relevant being the packet route selection [1]. done in [9]. In [10], two algorithms for routing in energy-
A number of algorithms which can be used to find convenient constrained, ad hoc, wireless networks are presented, which
routes have been presented in the past literature, e.g., in [2–5]. work on top of existing on-demand ad hoc routing protocols.
In proactive routing protocols [2,4], paths towards all destina- Singh et al. [11] and Subbarao [12] propose new metrics to
tions are periodically refreshed even if not used. Normally, be minimized in finding the routing path. Those metrics are
these protocols require nodes to broadcast information about power-aware, that is their target is to maximize energy effi-
their neighbors, and, based on this information, each node ciency (e.g., by finding the lowest energy routing path) and/or
in the network computes the minimum path to every possi- the lifetime of the whole network (e.g., by balancing traffic).
ble destination. In reactive routing protocols [3,5] the path to Chang and Tassiulas [13] propose a new path cost for a sta-
reach a destination is discovered only when needed by means tic network which can be used with any existing algorithm.
of a procedure called route discovery. With this procedure, The objective is to find the link cost function which will lead
the source finds more than one path and selects the “shortest” to the maximization of the system lifetime. Energy efficient
location based routing is studied in [14].
(usually in terms of number of hops). Note that routing algo-
As an alternative (or in addition) to the minimization of the
rithms need to know the network topology, so a given amount
energy cost of delivering a packet, the energy status of inter-
of control traffic overhead is always present and should be
mediate nodes is also a concern, as it directly affects the net-
minimized in order to increase the useful data throughput and
work lifetime. Several papers have considered routing metrics
to decrease the packet delivery time and the energy consump-
which explicitly include residual node energy [15–17]. Spe-
tion.
cific battery characteristics are used to devise smart routing
Most of these proposals are based on the intuitive goal
strategies in [18].
of choosing the shortest paths, i.e., minimizing the number
In order to achieve optimal network connectivity, very re-
of hops. In view of another primary concern in ad hoc net- cent papers propose to use power adjustments at some or all
∗ Corresponding author. nodes. Several definitions of optimal network connectivity
E-mail: [email protected] can be adopted, i.e., the one that minimizes the energy used
30 BERGAMO ET AL.

in delivering packets, or the connectivity which guarantees a mizations, types of traffic and traffic intensity. Our results
non-partitioned network. Ramanathan and Rosales-Hain [19] show that the relative gains may be significantly affected by
consider the problem of adjusting transmit power levels to the specific scenario considered, but they lead to the conclu-
achieve a desired degree of connectivity in the network, while sion that in many cases of interest the proposed DPC scheme
using the minimum transmit power. ElBatt et al. [20] pro- makes it possible to gain significantly in terms of energy ef-
pose to use power to maintain a certain degree of connectivity ficiency while paying a small price in terms of performance
in terms of number of reachable neighbors. Algorithms for degradation.
the generation of energy-efficient multicast trees have been It should be remarked at this point that our approach
proposed in [21,22]. Energy-efficient multicast is considered implicitly assumes that the transmit power is the dominant
in [23] as well. Additional papers on energy-efficient routing source of energy consumption. Some very recent papers point
include [24–30]. out that in some practical scenarios this may not be true, and
In many of these studies, new metrics directly related to different solutions must be sought in which nodes are aggres-
the transmit power and/or to the battery status of the nodes in sively put to sleep, as an idle listening transceiver may con-
the path are considered, but an exact knowledge of the topol- sume as much as a transmitting radio [33–35]. Our results
ogy is always assumed; therefore, they do not consider the may still be applicable to environments where sleep modes
signaling overhead necessary to update the topology knowl- are used at the MAC layer to avoid idle listening [36].
edge, which on the other hand consumes energy and increases The paper is structured as follows. In section 2 the DPC
network congestion. mechanism is described in detail. In section 3 the network
The main contribution of the present paper is the proposal architecture and the simulation setup are explained. In sec-
of a novel strategy, called Distributed Power Control (DPC), tion 4 extensive simulation results are presented. Section 5
which acts in combination with the routing layer. This is real- concludes the paper.
ized by means of a mechanism which estimates the amount of
power which is needed for reliable communications over any
link.1 This power is then used both to transmit a packet over 2. DPC mechanism
the link, and as the link weight in a minimum-weight path
search algorithm. In this way, transmit power can be tuned The proposed distributed power control strategy works at two
in order to build the desired connectivity diagram. In addi- different levels: hop-by-hop and end-to-end. In particular,
tion, the transmit power information is used to privilege lower it is based on the preliminary selection, hop-by-hop, of a
energy paths when looking for a packet route. Existing rout- suitable transmit power level, with the aim to reduce the en-
ing protocols, such as proactive and reactive protocols, can ergy consumption and to increase the overall network perfor-
be modified in order to incorporate this power control feature mance. Furthermore, this transmit power level is used as the
which tries to jointly minimize the interference in the network link cost function in the path discovery and selection.
and the energy consumption of multihop operation. This method requires that each node can record in a suit-
In order to assess the advantages of the proposed approach, able packet format field the power level, PTX , used to transmit
we have performed extensive simulations under various oper- that packet. Furthermore, it requires that the radio-transceiver
ating conditions. We consider first the classic Dijkstra routing can estimate the received power, PRX (many drivers of prod-
algorithm, in which the network topology is assumed known ucts based on the IEEE 802.11 standard provide this informa-
without overhead. While unrealistic, this case represents a tion).
performance bound and can help us understand the potential With the knowledge of PTX and PRX , the generic node is
of our DPC idea. Next, we examine the Link State routing al- able estimate the link attenuation. In particular, when a station
gorithm, which explicitly accounts for all the signaling over- receives a packet from a neighbor, the channel attenuation is
head needed in order to gain knowledge about the network simply computed as the difference (in dB) of the transmitted
topology. We remark that this additional traffic must be con- power PTX and the received power PRX . For the simple case
sidered, since it consumes energy while not being directly re- of a symmetric channel, where we neglect possible channel
lated to the delivery of useful traffic. Finally, as a realistic time fluctuations and we assume the same interference power
example of routing schemes specifically designed for ad hoc level, the attenuation affecting the transmission of that station
networks, we consider the Ad hoc On-demand Distance Vec- towards that neighbor would be the same as measured. Thus,
tor (AODV) algorithm [5]. For all three cases, we evaluate the a good choice for the transmit power P
TX could be:
performance in terms of both energy efficiency (which is our
P
TX = PTX − PRX + SR + Secth (1)
main goal here) and quality of service (which should never
be excessively compromised). Relevant tradeoffs and com- where SR is the minimum power level required for correct
parisons are highlighted in many different cases, including packet reception and Secth (Security Threshold) is a power
propagation effects, node mobility, protocol parameter opti- margin introduced to take into account channel and interfer-
1 Very similar work recently appeared in [31]. Only after submission of this ence power level fluctuations, i.e., to make the transmission
manuscript did we become aware of that work, which on the other hand was more reliable in view of the fact that the channel is not sym-
published several months after the conference presentation of the present metric. We assume that Secth is the same for each terminal
contribution [32]. and its value should be properly set as a function of network
DISTRIBUTED POWER CONTROL FOR ENERGY EFFICIENT ROUTING 31

density, terminal speed and channel conditions. Note that the In the DPC realization, a limited range for the transmit
P
TX update can be performed also sniffing packets directed to power has been introduced, i.e., minimum and maximum val-
other terminals, so that the transmit power information can be ues for the transmit power, PTXmin and PTXmax (established
refreshed more often. in the start up phase) have been set. Note that in the classic
In our DPC proposal this hop-by-hop power level selec- version of the algorithms the transmit power is always fixed
tion is also used to select the path guaranteeing low energy to a given value that, in our implementation, is equal to the
consumption. In fact, the P
TX values associated to all links maximum power PTXmax .
are considered as the cost functions used by the routing al- In the following a general overview of classic Dij, LS and
gorithms to select the packet path, thus implementing energy AODV are given, with a description of the changes required
saving at the end-to-end level. to implement DPC.
The proposed DPC can be applied to many routing algo-
rithms and for each routing scheme different implementations 2.1. Dijkstra
are possible. In this paper we focus on three different routing
protocols: Dijkstra (Dij) [37], Link State (LS) [38,39] and Ad The classic Dijkstra (Dij) algorithm [37,38] has the aim to
hoc On Demand Distance Vector (AODV) [5,40]. build a path between source (So) and final target (T a), by
Dij and LS are classic routing protocols, conceived for minimizing the number of hops. The link between any pair of
wireline networks, while AODV has been designed specif- nodes a and b has unit cost, la,b = 1, if the involved nodes
ically for ad hoc networks. In order to know the network are connected, and la,b = ∞ otherwise.
topology in every moment, these routing schemes are in gen- Let Dv be the cost function relative to the link between the
eral based on signaling traffic exchange between nodes. The Soth and the vth generic station and let F be the set including
signaling packets contain information about the neighbors of the nodes to be passed to reach the target T a; the algorithm is
a given node and/or about the whole network topology; on therefore:
the basis of this knowledge a given source can select the path 1. F = {So} and Dv = lSo,v ∀v ∈
/ F;
to the target. Note that in order to maximize the final perfor-
mance (in terms of percentage of correctly delivered packets, 2. select w
∈ F | ∀z
∈ F , w
= z, Dw
Dz ;
end-to-end delivery time, and energy efficiency) the signaling 3. if w = T a or Dw = ∞ then exit;
traffic is usually to be minimized.
In this paper, the LS and AODV signaling overhead has 4. F = F ∪ {w};
been explicitly simulated to have a realistic performance in- 5. ∀z
∈ F Dz = min[Dz , Dw + lw,z ];
vestigation. On the other hand, no signaling traffic has been
considered for Dij, where the station positions are always 6. go to step 2;
assumed known. This provides upper bound performance We assume the node locations are always known, even in
which can be useful for a comparison with the LS and AODV a node mobility situation. Furthermore, in order to have the
algorithms. final paths always updated, the algorithm reported above is
The LS routing protocol falls in the class of proactive rout- continuously recomputed by each node.
ing protocols, while AODV is part of the reactive protocol In the DPC implementation the algorithm is the same, but
class. In Proactive protocols the topology update and the sig- if the nodes are in visibility the cost function la,b is equal to
naling traffic exchange is performed periodically, while in the the transmit power, chosen following equation (1). In both
reactive algorithms it is performed when strictly necessary, classic and DPC schemes no signaling traffic is considered.
i.e., when a packet must be sent and no useful topology in-
formation is available. So, in general, reactive protocols limit 2.2. Link State
the signaling traffic overhead but result in a larger latency in
packet delivery. The Link State (LS) routing protocol [38,39] is based on the
Note that in their classic definition these algorithms try to exchange of suitable signaling packets, in order to discover
minimize the number of hops to reach a destination. In the the network topology and its changes as a consequence of the
proposed power controlled version these schemes minimize node mobility. It is an example of a proactive protocol and it
the total energy necessary to deliver a packet (which does not works as follows. Each node shall:
necessarily result in the smallest number of hops). For exam- (1) Discover its one hop neighbors and the status of each dis-
ple, in the classic Dij routing scheme the cost function associ- covered link (symmetric, i.e., bi-directional, or asymmet-
ated to each link is always 1; in the power controlled version, ric). It accomplishes this goal by periodically sending a
the cost function of the generic link between two nodes is the broadcast HELLO packet with Time to Live (TTL) equal
power needed to transmit a packet from a node to the other, to 1. A HELLO message contains:
estimated as in equation (1). So, in the power controlled Dij
version, the path is chosen by selecting the route which min- • the source address,
imizes the sum of the powers needed on each link. The same • the list of addresses of the node neighbors towards
considerations may be applied to LS and AODV. which a valid bidirectional link exists,
32 BERGAMO ET AL.

• the list of the addresses of all nodes, i.e., the nodes the network in order to reach the destination; when the desti-
which are heard by this node (i.e., from which a nation is reached, it sends back a control packet over the path
HELLO has been received) but whose link is not yet from which it has received the request.
validated as symmetric. The AODV protocol is mainly based on four procedures:
Upon reception of a HELLO message, the node updates • Route Discovery: starting phase for discovering a path;
the neighbor entry corresponding to the sender node, and • Forward Path Setup: used to inform the source and the
sets the link status to symmetric if it finds its own address relay nodes about a certain path;
in the HELLO payload.
• Route Maintenance: used to manage the time validity of
(2) Send a broadcast packet (LS_PKT) with highest TTL, a path and the link failures due to mobility, channel or traf-
containing the list of neighbors with symmetric links just fic conditions;
discovered, and the weight of the corresponding links. • Local Connectivity Management: used to have recent
information on the neighboring nodes
(3) With the information received with the LS_PKT packet,
build a network graph in which only symmetric links are Every time a source needs a route to a destination, it starts
present. Dij’s algorithm can be used to find the minimum a Route Discovery process [40,43]. This process is based
cost path to all possible destinations and to build the rout- on the transmission of broadcast packets, called Route Re-
ing table. quests (RREQ), each characterized by an identity number
(ID), which is rebroadcast through the network until the des-
Note that, as in the Dij scheme, we have assumed for clas- tination is reached.
sic LS a link cost function equal to 1 if the nodes are in visibil- For each new RREQ received, an intermediate node cre-
ity, and equal to ∞ otherwise. In the power controlled version ates a table entry. If the node receives the same RREQ (i.e.,
of LS, the final path is selected by minimizing the sum of the with the same ID) in a given interval, called BROADCAST
transmit powers on each link, computed as in equation (1). RECORD TIME [40], it does not rebroadcast it.
Note that in both classic and DPC versions the HELLO and The Forward Path Setup procedure is the following.
LS_PKT packets are sent at the maximum transmit power When the destination node receives a RREQ, it replies to the
PTXmax . The transmission rates of the HELLO and LS_PKT source with a unicast packet, which is sent using the same
packets have been set according to [2]. path followed by the arrived RREQ. The unicast packet is
In the DPC scheme, upon reception of a HELLO message, called Route Reply (RREP). When a RREP is received, every
the generic node fixes the cost of the link to the power: intermediate node refreshes the route table information and
P TX = PTXmax − PRX + SR + Secth . forwards the RREP packet until the source is reached. The
information in the table of each node is deleted if it not re-
P TX is also the transmit power used by the receiving node freshed within a fixed time. Note that, when the destination
in a successive transmission over this link, and is inserted in receives multiple RREQs with the same ID and relative to
the next HELLO packet as the cost function associated to that the same source-destination pair (necessarily from different
link. Furthermore, for every received data packet, the receiv- neighboring nodes), it does not reply to them for the BROAD-
ing node updates the cost of established links in its routing CAST RECORD TIME. After that, it starts again to reply to
table with the same procedure used for HELLO packet but the RREQs for that node pair.
the maximum power PTXmax is replaced with the transmission Furthermore, to limit the signaling traffic, if an intermedi-
power of the received packet, which may be expected to lead ate node receiving a RREQ knows the route to the final desti-
to a reduction of the energy spent in transmission. nation, it directly replies to the source with a RREP. Also in
this case, if this node receives other RREQs relative to that
2.3. AODV same pair of nodes, it does not send any further RREP.
The packets to be sent are stored by the source node in a
AODV [40,41] is an on-demand, reactive, single path, loop buffer until the RREP is received.
free, distance vector protocol. Classic AODV uses routing ta- The Route Maintenance procedure is used to manage the
bles characterized by only one entry per destination, so that, time validity of a path and the link failures due to mobility,
with respect to the Dynamic Source Routing (DSR) [3] pro- channel or traffic conditions. When a target becomes unreach-
tocol, no multiple paths per destination are present. able, the relay node sends broadcast packets, called Route Er-
Even though many variations to AODV have been pro- ror (RERR), to inform the source and the other intermediate
posed in literature (for instance, AOMDV: On Demand Mul- nodes about the link failure.
tiple Distance Vector [42]), in this paper we focus on the clas- Regarding the Local Connectivity Management, each
sic protocol by reviewing the basic features and by describing node sends periodically a broadcast packet with TTL = 1,
the changes needed in order to implement distributed power called HELLO packet, to inform all neighbors that it is
control. “alive”.
AODV is a reactive protocol, i.e., it finds the path to a des- Finally, to determine freshness of routing information and
tination only when needed. The generic procedure is to flood to prevent routing loops, a mechanism based on a sequence
DISTRIBUTED POWER CONTROL FOR ENERGY EFFICIENT ROUTING 33

number is used. More in details, each node of the network is rival times and service times. The isochronous traffic is a
characterized by a number called sequence number which is classic CBR traffic used to simulate multimedia stream-
increased every time its neighborhood topology changes. ing. The Mac module, described below, treats asynchro-
To apply the power control scheme to AODV we have nous and isochronous packets in a different manner: for
changed some features of the protocol. In particular, each instance, if the IEEE 802.11 DCF is used, the isochronous
node of the network records in its route table the total power packets are sent without RTS/CTS and ACK handshake.
to reach a destination and the power to reach the next hop in • Channel: simulates radio propagation. Path loss, shadow-
the path, computed as in equation (1). The information re- ing and Rayleigh or Rice fading via Jakes’ simulator [44]
quired to estimate the transmit power as in equation (1), such are implemented.
as the link attenuation, is collected by using the RREQ and
• Mob: simulates the terminal mobility. Pseudo-linear mo-
the RREP packets that carry the knowledge of the transmit
bility is implemented: new directions and speed of termi-
powers PTX as explained in [32]. To refresh transmit power
nals are recomputed at constant time intervals. The speed
information more frequently, data packets are also used.
is chosen as a Gaussian random variable with a given mean
Even if in the classic AODV a node does not reply to mul-
value and the direction is computed as an uniform random
tiple RREQs received within the BROADCAST RECORD
variable with a mean value that is the most recent direc-
TIME, in our DPC scheme it does so. This is introduced to
tion.
support the power control scheme, since we need all the pos-
sible routes that a packet could use to reach the final desti- • Radio: realizes the transmission and the reception of
nation, even those characterized by a large number of hops, packets. The packets are transmitted with a power level
which may correspond to lower energy consumption. decided by the upper layers following the mechanism de-
Therefore, when a node receives a RREQ already received, scribed in Section 2. Furthermore, it estimates the received
it only checks if the corresponding route is more energy effi- power and it simulates the capture effect if present, i.e.,
cient than the route in its routing table. If the route is really the ability to correctly receive a packet even if collided
more efficient, the node broadcasts the RREQ or sends back depending on the Signal to Interference and Noise Ratio
a RREP (if the node is the destination or knows the route to (SINR). The use of spread spectrum and channel coding
the target). can also be accounted for.
Note that this modification increases the routing traffic. On • Mac: simulates the channel access mechanism. An IEEE
the other hand, the changes we applied permit us to find a 802.11 DCF protocol [45] has been included into this
route with lower power consumption. So, the system is ex- module so far.
pected to introduce an energy gain if the energy saving in- • Route: simulates the Network layer. It generates routing
troduced in the packet delivery is greater than the additional packets to discover neighbors and to compute the path to
energy loss due to the routing signaling procedure. reach the destination. LS [39], Dij [38] and AODV [5,40]
have been implemented.
3. Network architecture and SAM structure • Statistics: collects results. The performance indices we
have identified are: success probability, Psucc (percent-
To evaluate the DPC effectiveness in a realistic environment a age of delivered packets), normalized average end-to-end
general network architecture, integrating traffic features, ter- packet delivery time, Tacc (the average time from a packet
minal mobility, channel behavior, medium access scheme and arrival until its correct delivery to the final destination di-
routing protocols has been considered. vided by the average time to transmit a packet over the
This architecture has been simulated, by integrating all air interface), average energy spent per packet, Ep (ratio
characteristics typical of the Physical, Data Link and Network between the total energy spent during simulation and the
layers of the ISO/OSI model into a general Simple Ad hoc number of correctly delivered packets), throughput S (ra-
siMulator (SAM). tio between the useful time and the total time of simula-
SAM is a discrete event simulator, composed by many tion).
modules or entities: Traffic, dealing with the traffic genera-
tion; Channel, simulating the radio propagation; Mob, deal-
ing with the terminal mobility; Radio, simulating the hard- 4. Simulation results
ware transceiver; Mac realizing the Data Link Layer; Route,
In this Section some numerical results regarding the Dij, LS
implementing the Network layer; and Statistics, collecting
and AODV routing algorithms with and without DPC are
statistics results.
shown. Note that in the classic versions the transmit power
The entity structures and the protocols currently available
P
TX is constant and equal to PTXmax , while in the DPC ver-
are the following:
sions it is variable in the range [PTXmin , PTXmax ], according to
• Traffic: generates the network traffic. Two different traf- the attenuation and the variation rules described above.
fic types have been implemented: asynchronous traffic In Dij we assume that each node always knows the net-
and isochronous traffic. The asynchronous traffic is re- work topology without signaling packet exchange (perfor-
alized by considering a Poisson distribution for both ar- mance upper bound), and that an intermediate node, when
34 BERGAMO ET AL.

receiving a packet to be forwarded, re-computes the path to

the final target (since node mobility can change the network
topology). On the other hand, in the LS and AODV simu-
lations the signaling packet exchange is explicitly taken into
consideration.
At the MAC level, the IEEE 802.11 DCF protocol has been
considered. In the DPC scheme, the MAC signaling packets,
such as RTS, CTS and ACK, are sent with the same power as
the data packets. On the other hand, at the network layer, the
signaling packets, if present, are sent at the maximum power,
PTXmax , so as to increase the number of neighbors able to hear
these packets, thereby decreasing the traffic overhead needed
to build the whole network topology.
Regarding the physical layer, a channel behavior with path
loss and log-normal shadowing has been taken into account.
The impact on the system performance of more complex Figure 1. Psucc and Tacc vs. Ep for Dij, with fixed and mobile nodes. Path
loss only, Poisson traffic and PTXmax = −12 dBm.
channel models, such as Rayleigh or Rice fading is currently
under investigation. E = (EpC − EpDPC )/EpC · 100, P = (Psucc DPC − P C )/P C ·
succ succ
The cases with and without mobility, i.e., with nodes in 100, where the superscripts C and DPC refer to classic and
pseudo-linear movement and in fixed positions, are discussed, DPC schemes, respectively. For the delivery time we con-
by varying some system parameters. sider: T = TaccDPC /T C . For the first two parameters, posi-
acc
Simulation results for both Poisson and CBR traffic are tive values denote a performance improvement obtained with
shown, with an arrival rate of λ packets/s per node. λ is rela- DPC. For the third parameter, the improvement is shown
tive to the data traffic only, i.e., it does not include the signal- whenever T < 1.
ing traffic generated at the Network layer.
The system parameters used in the simulations are the fol- 4.1. Poisson traffic
lowing: number of nodes, N = 30, located in a square room
of size 10×10 m; power decay law with the distance, β = 2.5; In this Section some results are presented in order to test
log-normal shadowing deviation σ = 6 dB; minimum power the effectiveness of DPC, for the case of Poisson traffic.
necessary to correctly receive a packet before de-spreading, The SAM Poisson traffic implementation is based on service
SR = −76 dBm; capture threshold active, simulation time, and inter-packet times having exponential distribution and
500 s; coding gain 2 dB (used only for the routing packets); the generic source-destination pair is independently chosen
spread spectrum gain, 10.41 dB; minimum transmit power, at random, with uniform distribution on all stations.
PTXmin = −44 dBm; mobility refresh time, tmob = 0.3 s; av- Due to the specific type of traffic considered here and to
erage speed of the generic node, vmob = 0.75 m/s; speed de- the reactive nature of AODV, a route discovery would need
viation, σmob = 0.3 m/s; maximum mobility angle, θ = 180◦. to be performed for every packet. This of course would pro-
Furthermore, we have considered two different values for duce too much signaling traffic, which results in extremely
the maximum transmit power, PTXmax = −22 dBm and poor performance. For this reason, in this section on Poisson
−12 dBm, to simulate two multihop behaviors: the first with traffic, no results will be presented for AODV.
higher number of hops in the generic path, the second with For these simulations we have considered Dij and LS,
lower number of hops in the generic path. All simulations by setting: channel data rate, 1 Mbit/s; average arrival rate,
have been performed with the security threshold Secth span- λ = 1 packet/s, average service rate, µ = 128 packets/s, cor-
ning the range from 0 to 15 dB and from 0 to 10 dB in the responding to a packet size of 1024 bytes; channel behavior
PTXmax = −12 dBm and PTXmax = −22 dBm cases, respec- with path loss only.
tively.
The node locations are chosen randomly in the selected 4.1.1. Poisson traffic and Dij
area, with uniform distribution. In particular, with fixed termi- In figure 1, a performance comparison between the classic
nals, we have focused the attention on connected topologies. and DPC version of Dij algorithm is shown, for PTXmax =
This choice is because the system performance is evaluated −12 dBm, by considering two cases: (i) fixed nodes and
by averaging over all nodes in the network, so that isolated (ii) all nodes moving at pedestrian speed (MOB). In partic-
nodes would negatively bias the performance results. ular, Tacc and Psucc are reported as a function of Ep in order
As performance metrics we have considered: average to highlight the relevant tradeoffs between quality of service
success probability, Psucc , average delivery time, Tacc , en- and energy efficiency. The classic versions are represented
ergy spent per packet, Ep , and throughput S. Finally, in by two points, while the DPC versions by curves obtained
many cases a comparison between DPC and classic schemes varying Secth in the range from 0 to 15 dB. Note that when
will be performed by reporting the performance gaps. These Secth = 15 dB, the DPC scheme achieves roughly the same
gaps are computed, for Ep and Psucc , as the percentages: Psucc and the same average delivery time, with slightly lower
DISTRIBUTED POWER CONTROL FOR ENERGY EFFICIENT ROUTING 35

Figure 2. Psucc and Tacc vs. Ep for Dij, with fixed and mobile nodes. Path Figure 3. Psucc and Tacc vs. Ep for LS, with fixed and mobile nodes. Path
loss only, Poisson traffic and PTXmax = −22 dBm. loss only, Poisson traffic and PTXmax = −22 dBm.

energy consumption (a gap E = 5.5% can be seen). The can achieve at Secth = 0 dB a lower performance degradation,
curves with and without mobility are basically the same, since but also a lower energy gain with respect to the case with
the Dij algorithm recomputes the path at each intermediate PTXmax = −12 dBm.
node (to counteract possible network topology changes) and This lower energy gain is probably due to a lower gap
the transmit power is sufficiently high to avoid link failure as between the transmit power used in the DPC case and that
a consequence of mobility. used in the classic scheme, with respect to the case with
If we consider Secth = 0 dB, the system performance de- PTXmax = −12 dBm. This behavior is present both with fixed
creases, since no power margin is used, but very large energy nodes and with mobile nodes. On the other hand, the case
gains can be achieved. In fact, Psucc decreases (P = −10%) with mobile nodes shows lower Psucc (−22%) and lower Tacc
and Tacc increases (T = 4), with an energy gain of E = 87% (−21%) with respect to the case of fixed nodes in both the
(i.e., energy consumption is reduced by a factor of 8!). By classic and DPC schemes. This is a consequence of the lim-
considering intermediate values of Secth , we can trade-off ited transmit power that is not sufficient to counteract the link
performance degradation for energy savings. failure due to the node mobility even if the path is updated
In the classic case the average number of hops in the path node-by-node. This phenomenon determines a decrease of
is 1.3, while in the DPC case, it spans from 4.4 to 1.3 for the packet delivery probability and reduces the delivery time
Secth from 0 dB to 15 dB. So, the effect of power control is to since shorter paths are more reliable. Another possible inter-
increase the number of hops, especially with low Secth , with pretation is related to the occurrence, when nodes are moving,
consequent degradation of the final delivery time. of poorly connected network conditions, i.e., with isolated
Regarding the curve trends, Psucc slightly increases with nodes or subnetworks, which degrade the success probabil-
Secth because of the increase of the transmit power, which ity.
results in a smaller number of packet relays (i.e., of hops) In the system situations of figure 2, the classic case is char-
and then limits the collision probability. Note also that for acterized by an average number of hops in the network equal
the selected system parameters Psucc is spanning quite high to 3.2 for fixed nodes and 2.9 for mobile nodes, i.e., higher
values, close to 1. Tacc has a trend with a maximum: the than in the PTXmax = −12 dBm case, as expected, while in
initial power increase is not sufficient to decrease the num- the DPC case it is spanning from 4.6 to 3.0 when Secth ranges
ber of hops, and results in higher interference, more packet from 0 dB to 10 dB. Note also that the maximum Psucc is not
collisions and more packet retransmissions, all of which neg- close to 1, as in figure 1, as a consequence of a higher av-
atively impact on the delivery time. When the power is suf- erage number of hops and then of a higher packet collision
ficient to decrease the number of hops, the delivery time de- probability.
creases. So, in the Dij case the DPC effectiveness, in terms of en-
The trends of Psucc and Tacc as a function of Ep for Dij, for ergy savings, is present both with and without terminal mo-
PTXmax = −22 dBm, by considering both fixed nodes and all bility and is quite relevant if some limited performance degra-
nodes in movement, are reported in figure 2, where Secth is dation can be tolerated.
varied in the range from 0 to 10 dB. In both cases of mobility
and no mobility, in the same performance conditions the DPC 4.1.2. Poisson Traffic and LS
scheme achieves a E = 3.0% energy gain with respect to the The same performance indices of figure 1, with PTXmax =
classic scheme. If Secth = 0 dB a Psucc reduction (P = −22 dBm, are shown in figure 3 for the LS protocol, where
−5.0%) and a Tacc increment (T = 1.4) are seen, with an both fixed and mobile nodes are considered. With respect
energy gain of E = 50%. So, when PTXmax = −22 dBm, we to the Dij algorithm (figure 2) the energy spent per packet,
36 BERGAMO ET AL.

Ep , is higher, since LS implies considerable signaling traffic,

which is absent in Dij. The presence of this signaling traffic
also justifies a lower Psucc (especially with terminal mobility)
and a higher Tacc with respect to the Dij case, due to higher
network load.
The average number of hops with fixed nodes is the same
as for the Dij scheme. Furthermore, with fixed nodes, when
Secth = 0 dB, i.e., no transmit power margin, the Psucc gap
with respect to the classic case is about P = −13%, the
Tacc ratio is about T = 1.3 and the energy gain is about
E = 30%. So, with respect to Dij a lower energy gain is
achieved with higher performance decrease in terms of suc-
cess probability, due to the signaling traffic overhead. In the
case of mobile nodes the maximum energy gain (E = 4%) of
the DPC scheme is present when Secth = 10 dB which is the
margin value guaranteeing also the same Psucc and Tacc of the Figure 4. Psucc and Tacc vs. Ep for Dij, with fixed nodes. Path loss and
classic algorithm. In fact, by decreasing Secth the system per- shadowing, CBR traffic and PTXmax = −22 dBm.
formance decreases not only in terms of success probability
and of access time but also in terms of energy gain. This be-
havior could be related to a low margin to counteract mobility,
which determines a poor performance behavior and a higher
energy consumption since more retransmissions and signal-
ing traffic are present. Furthermore, Psucc is very low (rang-
ing from 0.2 to 0.4), as a consequence of the lateness in the
topology update with respect to the terminal mobility. Note
that on average the packets experience a low Tacc with respect
to the case without mobility, probably because only packets
sent over paths characterized by a small number of hops are
correctly delivered, which results in an optimistically biased
delay average. In fact the average number of hops is ranging
from 2.4 to 2.8, while in the case of fixed nodes it was ranging
from 3.3 to 4.5. Finally, Ep is higher in the case with mobil-
ity, due to the higher number of retransmissions, consequence
of a generally higher link failure probability. Figure 5. Psucc and Tacc vs. Ep for LS, with fixed nodes. Path loss and
So, in the LS case the DPC effectiveness, in terms of en- shadowing, CBR traffic and PTXmax = −22 dBm.
ergy savings, is present especially without terminal mobility
(in this case if some limited performance degradation can be ing to a packet size of 512 bytes, channel behavior with path
tolerated, the energy gain is considerable). However, with re- loss and shadowing.
spect to Dij, the energy gain is lower and the performance In figures 4 and 5 the same performance indices of fig-
degradation is higher, as expected, as a consequence of the ures 1–3 are reported for Dij and LS, respectively, by com-
signaling traffic. On the other hand, by considering all nodes paring the cases with path loss only (PL) and with path loss
mobile with pedestrian speed, the DPC LS gives only limited and shadowing (Sh). We have set λ = 2 packets/s per user
energy gains, as a consequence of the lateness in the topology and PTXmax = −22 dBm.
update and of the presence of signaling traffic.
4.2.1. CBR with Dij and Fixed nodes
4.2. CBR traffic Figure 4, relative to Dij, shows that shadowing determines a
Psucc decrease with respect to the case of path loss only, both
In this section we investigate the case of CBR traffic, which in classic and DPC schemes. In the PL case, Psucc is always
is the traffic considered in almost all papers on ad hoc rout- close to 1; in the Sh case, Psucc is 0.92 for classic and DPC
ing, by comparing Dij, LS and AODV both in classic and with Secth = 10 dB, and it decreases by decreasing the Secth
DPC cases. CBR traffic is characterized by constant service margin. Regarding Tacc , it is about the same for PL and Sh in
and inter-packet times. Furthermore, the source-destination the classic version.
pairs are chosen randomly (with uniform distribution over all Note that for a given topology which in the PL scenario
nodes) in the start-up phase and they remain the same for all corresponds to a connected network, there is no guarantee that
packets of each connection. For these simulations we have the network will also be connected in the presence of shadow-
considered Dij, LS and AODV, by setting: channel data rate, ing. If a node is unreachable from the others, Dijkstra simply
2Mbit/s; average service rate, µ = 256 packets/s, correspond- drops the packet at its source, since it cannot find a path to-
DISTRIBUTED POWER CONTROL FOR ENERGY EFFICIENT ROUTING 37

wards the destination. The result is a lower energy consumed

to deliver a packet because many packets are not transmit-
ted at all and they are not lost for collisions. Moreover, the
time required to deliver a packet is also smaller than in the
Path Loss scenario due to the fact that longer routes are more
likely to be partitioned. The main conclusion is that there is
a considerable energy saving also when DPC is used in shad-
owing environments, at the price of somewhat higher packet
loss rate.
Regarding the energy efficiency of DPC, in the PL situa-
tion it can achieve high gains (E = 58%) with the same Psucc
and T = 2.5, at Secth = 0 dB. In the Sh case, the maximum
energy gain (E = 76%) is achieved for Secth = 0 dB, but
some performance degradation (P = −11% and T = 10)
must be tolerated in this case. It is sufficient to increase Secth
to 8 dB to have no Psucc degradation and T = 1.5, while still Figure 6. Psucc vs. Ep for Dij and LS, with fixed nodes and increasing
maintaining a sizable energy gain of E = 25%. number of mobile nodes (2 through 10). Path loss only, CBR traffic and
PTXmax = −22 dBm.
So, with CBR traffic, fixed nodes and path loss DPC Dij
shows very good performance and high energy saving. In the Table 1
presence of shadowing an interesting trade-off between per- Performance gaps P, T and E, for Dij and LS. Path loss only, λ = 2 pack-
formance degradation and energy gain can be achieved, by ets/s per user, for an increasing number, #, of mobile nodes.
suitably setting the value of Secth . # P T E Secth

4.2.2. CBR with LS and Fixed nodes Dij 0 −0.7 2.5 58 0

Dij 2 0.25 2.2 61 0
Figure 5, relative to LS, shows a lower sensitivity of this pro-
Dij 5 −1.5 2 58 0
tocol to shadowing with respect to Dij. In fact, the Psucc trend Dij 10 −0.65 2 52 0
is quite the same in PL and Sh, in both the DPC and the clas- Dij 20 −0.25 2 52 0
sic version. The difference in terms of Tacc between PL and Dij 30 0.6 1.9 54 0
Sh seen for DPC Dij is not observed here. LS 0 −10 2 41 0
Furthermore, Tacc is lower with Sh. A brief data analysis LS 2 −21 2.4 40 0
can justify such a behavior: shadowing can allow some links LS 5 −48 1.7 16 0
which are not present in the case of path loss only, so, in this LS 10 −60 1 −1 0
LS 20 −72 1.2 −36 0
simulation topology, the mean number of hops to reach the LS 30 −84 0.8 −129 0
destination may be lower with Sh than PL. Regarding the en- LS 0 −2 1 0.3 10
ergy efficiency of DPC LS, in the PL situation it can achieve LS 2 −3.9 1 4 10
high energy gains (E = 41%) with P = −10% and T = 2, at LS 5 −3.3 1.2 2.5 10
Secth = 0 dB. LS 10 3.6 1 5.2 10
LS 20 6 0.8 5.7 10
By increasing Secth the performance improves and the en- LS 30 −7 0.9 3.5 10
ergy gain decreases. In the Sh case, the maximum energy
gain (E = 50%) is for Secth = 0 dB with P = −13.5% and
T = 1.5. It is sufficient to increase Secth to 10 dB to have Each mobile node moves with pseudo-linear movement and
P = −3.5%, T = 1 and E = 14%. average speed 0.75 m/s. Note that Dij performance is not
So, with CBR traffic, fixed nodes, path loss and shadow- affected by mobility, since the topology is always assumed
ing, DPC LS shows energy saving effects, even if less rel- known. On the other hand, LS suffers from node mobility,
evant than in the Dij case, due to the presence of signaling both in classic and DPC versions. Psucc progressively de-
traffic. Both in PL and Sh cases the maximum energy gain grades by increasing the number of mobile nodes. Regarding
(E = 40/50%) can be achieved at Secth = 0 dB with a per- DPC the loss of Psucc is more evident if no security margin is
formance gap P = −10/–13%. By increasing the security considered, as expected.
margin the performance gap reduces, by maintaining a lower, To summarize the performance behavior with mobility and
but still non-trivial, energy gain (E = 14%). path loss, in table 1 the performance gaps P, T and E are re-
ported for both Dij and LS, by considering λ = 2 packets/s
4.2.3. CBR with Dij and LS, Mobile nodes per user and an increasing number, #, of mobile nodes. Note
To investigate the effects of node mobility, we have consid- that without security margin DPC Dij achieves an energy gain
ered different network scenarios by increasing the number of E > 52% with quite the same Psucc and Tacc about 2 times that
mobile nodes in the network from 0 to 10. Figure 6 shows of the classic case. This behavior is maintained also when all
Psucc as a function of Ep for Dij and LS in the PL case, by 30 nodes move. Regarding LS, if no security margin is con-
considering a number of mobile nodes equal to 0, 2, 5 and 10. sidered, we can note that the P negative gap increases with
38 BERGAMO ET AL.

Table 2
Performance gaps P, T and E, for Dij and LS. Path loss and shadowing,
λ = 2 packets/s per user, for an increasing number, #, of mobile nodes.

# P T E Secth

Dij 0 −11 10 76 0
Dij 2 −11 1.4 79 0
Dij 5 −10 2 75 0
Dij 10 −13 2.2 77 0
Dij 20 −21 3.5 79 0
Dij 30 −21 2.7 71 0
Dij 0 −0.6 1.5 25 8
Dij 2 0 1.4 0 8
Dij 5 −0.85 1.1 17 8
Dij 10 −2 1.2 31 8
Dij 20 −1.6 1.6 34 8
Dij 30 −2.7 1.3 24 8
LS 0 −13 1.5 50 0 Figure 7. Psucc vs. Ep for Dij and LS, with fixed nodes and increasing
LS 2 −38 1.4 27 0 packet rate (1 through 4 packets/s). Path loss only, CBR traffic and PTXmax =
LS 5 −45 2 7 0 −22 dBm.
LS 10 −59 1.4 −21 0
LS 20 −68 1.3 −69 0
LS 30 −88 1 −381 0 4.2.4. CBR with Dij and LS, network load
LS 0 −3.4 1.2 14 10 Other investigations have been performed to test the impact
LS 2 −3.2 1 11 10
of the network load on the final performance. In a fixed node
LS 5 −8 1.2 −0.17 10
LS 10 −11 1.2 2 10 scenario, we have varied the packet rate per user, λ, from 1
LS 20 −7 1.2 11 10 to 4 packets/s. In figure 7 Psucc as a function of Ep is shown
LS 30 −26 1.2 −18 10 for Dij and LS, by considering no mobility and path loss only.
As expected, the percentage of delivered packet decreases by
increasing the network load, and this trend is more evident in
LS where lower Psucc than for Dij is achieved. The same trend
the increase of the number of mobile nodes and also the en- is seen with shadowing, where, as in the mobility investiga-
ergy gain vanishes for #  10. By considering Secth = 10 tion case, Dij shows lower performance if no security margin
dB, the P gap reduces considerably, but the energy gain is is present.
quite limited. So, the DPC applied to LS is deeply affected To summarize the performance behavior by changing the
by node mobility, as a consequence of the joint effect of the network load, in table 3 the performance gaps P, T and E are
lateness in the topology update with respect to the node speed reported for both Dij and LS, in PL and Sh cases, by consid-
and of the reduced radio coverage (causing a more rapid link ering fixed nodes and an increasing network load, N (pack-
failure with mobility with respect to the classic case, where ets/s). In both the Dij and LS cases no relevant performance
the radio coverage is higher). gap changes are observed by varying the network load within
Regarding the impact of node mobility on Dij and LS in the specified range.
the presence of shadowing, the Psucc trend as a function of Ep
is similar to that of figure 6 where only path loss was consid- 4.2.5. CBR and AODV
ered. The only difference is relative to the Dij case where if Regarding AODV, in figure 8 Tacc and Ep are reported ver-
a low Secth is considered, a Psucc degradation is observed (as sus Secth , with fixed nodes, path loss only and packet rate
expected since channel fluctuations are not taken into con- λ = 2 packets/s. In this case Psucc is very close to 1 for both
sideration in the transmit power selection). This Dij trend DPC and classic cases and constant for all values of Secth .
becomes more evident by increasing the node mobility. The main result shown here is that, with stationary nodes and
To summarize the performance behavior with mobility and Secth = 0 dB, DPC AODV consumes about 40% less than
shadowing, in table 2 the performance gaps P, T and E are classic AODV with the same number of successfully deliv-
reported for both Dij and LS, by considering λ = 2 packets/s ered packets and a negligible delivery time degradation. Note
per user and an increasing number, #, of mobile nodes. Note that in this case the access time suffers very little when Secth
that, as anticipated above, with shadowing DPC Dij needs the increases, while for LS and Dij the opposite is true, due to the
presence of a security margin (Secth = 8 dB) to maintain the large number of routing control packets that are generated to
same performance as in the classic scheme (in the PL case this find an energy-efficient path and to keep it alive. For higher
was achieved without margin), and that this increased Secth Secth , more nodes are blocked, and the delivery time is conse-
implies a lower energy gain (ranging from 17 to 34%, ex- quently degraded, even though the gap is at most about 10%.
cept the situation of 2 mobile nodes, probably due to a poorly The minimum of the access time curve for Secth = 2 dB (and
connected network topology). As far as LS is concerned, the not for 0 dB) is mainly due to a negligible number of lost
behavior is similar to that of table 1 with path loss only. packets, i.e., simulation fluctuations.
DISTRIBUTED POWER CONTROL FOR ENERGY EFFICIENT ROUTING 39

Table 3
Performance gaps P, T and E, for Dij and LS, for the PL and Sh cases and
fixed nodes, and for increasing network load, N (packets/s).

N P T E Secth

Dij PL 1 −0.26 2.2 59 0

Dij PL 2 −0.7 2.5 58 0
Dij PL 3 −4 3.3 58 0
Dij PL 4 −7 106 60 0
Dij Sh 1 −10.68 2.6 76 0
Dij Sh 2 −11.06 10 76 0
Dij Sh 3 −14.34 587 75 0
Dij Sh 4 −8.04 3000 77 0
Dij Sh 1 −0.6 1.5 25 8
Dij Sh 2 −0.6 1.5 25 8
Dij Sh 3 −1.5 1.3 25 8
Dij Sh 4 1 1.5 25 8
LS PL 1 −5.6 1.7 35 0 Figure 9. Tacc and Ep vs. the packet rate λ for AODV with fixed nodes, path
LS PL 2 −10 2 41 0 loss only, PTXmax = −22 dBm, and Secth = 0 dB for DPC.
LS PL 3 −9 2 47 0
LS PL 4 −14 3.6 48 0
LS Sh 1 −17 1.5 29 0
LS Sh 2 −14 1.5 50 0
LS Sh 3 −18 1.9 56 0
LS Sh 4 −11 1.3 67 0
LS Sh 1 −0.9 1.2 10 10
LS Sh 2 −3.4 1.1 14 10
LS Sh 3 −4.6 1 13 10
LS Sh 4 −3.7 1 18 10

Figure 10. Throughput vs. packet rate per user, for Dij, LS and AODV, with
CBR traffic, path loss only and fixed nodes.

close to 50% for an arrival rate of 4 packets per second per

node.
Our implementation of DPC AODV can achieve a very
high energy saving. Unfortunately, the best results are ob-
tained with very low values of Secth . When nodes are mobile,
Figure 8. Tacc and Ep vs. Secth for AODV with fixed nodes, packet rate
even moderate mobility may compromise the correct trans-
λ = 2 packets/s, path loss only, and PTXmax = −22 dBm. mission and the performance suffers. Nevertheless, we are
investigating the possibility to realize a new version of DPC
AODV that addresses this problem. The main goal is to re-
Shadowing does not impact the performance compared to alize a dynamic protocol able to be aware of mobility and to
the PL environment, and the corresponding performance re- automatically set Secth in order to get the best performance in
sults are not shown here, as they are very similar to those every situation.
shown in figure 8.
In figure 9, access time and energy per packet versus the 4.2.6. CBR performance comparison with Dij, LS and AODV
packet arrival rate are shown for both versions of AODV (with In figure 10 the throughput (ratio between the useful time and
Secth = 0 dB for DPC). The delay time of the DPC version the total time of simulation) as a function of the packet rate
is very close to the classic case even though the packet ar- per user, for Dij, LS and AODV, with CBR traffic, path loss
rival rate increases, a first interesting result. Moreover, DPC and fixed nodes is reported, for Secth = 0 dB in the DPC
AODV shows a very interesting result that proves the effec- schemes. As expected, the best channel utilization is relative
tiveness of our proposal; the energy gain grows when the to Dij (due to the fact that Dijkstra does not generate control
packet arrival rate increases. The energy reduction is very packets) which shows an always increasing throughput with
40 BERGAMO ET AL.

the offered load. The worst performance is that of LS, whose [7] V. Rodoplu and T.H. Meng, Minimum energy mobile wireless network,
throughput is always less than 0.2. AODV shows an inter- IEEE Journal on Selected Areas in Communications 17(8) (1999).
mediate behavior with a curve which tends to saturate at 0.4. [8] J.C. Cano and P. Manzoni, A performance comparison of energy con-
sumption for mobile ad hoc network routing protocols, in: Proc. of
Furthermore, DPC performance follows the trend of the clas- IEEE MASCOTS 2000 (September 2000) pp. 57–64.
sic schemes but is characterized by lower throughput. This [9] L. Feeney, An energy consumption model for performance analysis of
behavior is very clear for AODV where the routing signal- routing protocols for mobile ad hoc networks, ACM Journal of Mobile
ing traffic of DPC is always greater than in the classic case. Networks and Applications 6 (2001) pp. 239–249.
Similar considerations can be made when shadowing is con- [10] Y. Xu, J. Heidemann and D. Estrin, Adaptive energy-conserving routing
for multihop ad hoc networks, Research Report527, USC/Information
sidered. Sciences Institute, October 2000.
[11] S. Singh, M. Woo and C.S. Raghavendra, Power-aware routing in mo-
bile ad hoc networks, in: Proc. of ACM/IEEE MobiCom98 (October
5. Conclusion 1998).
[12] M.W. Subbarao, Dynamic power-conscious routing for MANETs: an
In this paper, we have proposed a distributed power control initial approach, in: Proc. of IEEE VTC Fall 1999, Amsterdam, The
scheme as a means to improve the energy efficiency of rout- Netherlands (September 1999).
ing algorithms in ad hoc networks. Each node in the network [13] J.H. Chang and L. Tassiulas, Energy conserving routing in wireless
ad-hoc networks, in: Proc. of IEEE INFOCOM’00, Tel-Aviv (March
estimates the power necessary to reach its own neighbors, and 2000).
this power estimate is used both for tuning the transmit power [14] Y. Xue and B. Li, A location-aided power-aware routingprotocol in mo-
(thereby reducing interference and energy consumption) and bile ad hoc networks, in: Proc. of GLOBECOM’01, San Antonio, TX
as the link cost for minimum energy routing. We have in- (November 2001) pp. 2837–2841.
[15] S. Agarwal, A. Ahuja, J.P. Singh and R. Shorey, Route-lifetime as-
cluded this technique in three routing protocols, namely the
sessment based routing (RABR) protocol for mobile ad-hocnetworks,
classic Dijkstra and Link State, as well as AODV, specifi- in: Proc. of IEEE International Conference on Communications, ICC
cally designed for ad hoc networks. It has been shown how 2000, New Orleans, LA (June 2000) pp. 1697–1701.
some limited degradation of the quality of service (in terms of [16] S. Tragoudas and S. Dimitrova, Routing with energyconsiderations in
throughput and delay) can effectively be traded off for energy mobile ad-hoc networks, in: Proc. of 2000 IEEE Wireless Communi-
cations and Networking Conference, Chicago, IL (September 2000)
savings that in some cases can be very significant.
pp. 1258–1261.
Further investigations include applying DPC to other ad [17] W. Kyungtae, Y. Chansu, L. Dongman, Y.Y. Hee and B. Lee, Non-
hoc routing protocols such as DSR and Optimized Link State blocking, localized routing algorithm for balanced energy consump-
Routing (OLSR). Also, we are investigating in more detail tionin mobile ad hoc networks, in: MASCOTS 2001, Proceedings of
the effect of node mobility on the performance of the various Ninth International Symposium on Modeling, Analysis and Simulation
of Computer and Telecommunication Systems, Cincinnati, OH (August
protocols. Preliminary results for AODV show that some crit-
2001) pp. 117–124.
ical situations may arise in this case, and some initial ideas [18] C.F. Chiasserini and R.R. Rao, Routing protocols to maximizebattery
on how to solve this problem by suitably defining the power efficiency, in: Proceedings of IEEE Military Communications Confer-
control mechanism are currently being tested. ence (MILCOM’00), Los Angeles, CA (October 2000) pp. 496–500.
[19] R. Ramanathan and R. Rosales-Hain, Topology control of multihop
wireless networks using transmit power adjustment, in: Proc. of IEEE
Acknowledgements INFOCOM 2000 (March 2000).
[20] T.A. ElBatt, S.V. Krishnamurthy, D. Connors and S. Dao, Power man-
agement for throughput enhancement in wireless ad-hoc networks, in:
This work was supported by ALCATEL and by the European
Proc. of IEEE ICC’00, New Orleans, LA (June 2000).
Commission under the EYES project (IST-2001-34734). [21] L. Badia and M. Zorzi, On the construction of broadcast and multicast
trees in wireless networks – global vs. local energy efficiency, in: Proc.
of IEEE GLOBECOM2002, Taiwan (November 2002).
References [22] A. Michail and A. Ephremides, “Energy-efficient routing for
connection-oriented traffic in ad-hoc wireless networks, in: Proc. of
[1] E.M. Royer and C.K. Toh, A review of current routing protocols for ad PIMRC’00, Vol. 2, London (March 2000) pp. 762–766.
hoc mobile wireless networks, IEEE Personal Communications (April [23] I. Kang and R. Poovendran, On the lifetime extension ofenergy-
1999). efficient multihop broadcast networks”, in: Proceedings of the 2002
[2] P. Jacquet, P. Muhlethaler and A. Qayyum, Optimized link state routing International Joint Conference on Neural Networks, IJCNN’02, Vol. 1,
protocol, IETF Draft, http://www.ietf.org/ID.html (Febru- Honolulu, HI (May 2002) pp. 365–370.
ary 2000). [24] S. Banerjee and A. Misra, Energy awareness and power control: min-
[3] D.B. Johnson and D.A. Maltz, The dynamic source routing protocol imum energy paths for reliable communication in multi-hop wireless
for mobile ad hoc networks, IETF Draft, http://www.ietf.org/ networks, in: Proc. of the Third ACM International Symposium on Mo-
ID.html (October 1999). bile Ad Hoc Networking and Computing (June 2002).
[4] C.E. Perkins and P. Bhagwat, Highly dynamic destination-sequenced [25] J. Gomez, A.T. Campbell, M. Naghshineh and C. Bisdikian, Conserv-
distance-vector routing (DSDV) for mobile computers, Comp. Comm. ing transmission power in wireless ad hoc networks, in: Proc. of Ninth
Rev. (October 1994). International Conference on Network Protocols, ICNP 2001, Riverside,
[5] C.E. Perkins, E.M. Royer and S.R. Das, Ad hoc on-demand distance CA (November 2001) pp. 24–34.
vector routing, IETF Draft, http://www.ietf.org/ID.html [26] Q. Li, J. Aslam and D. Rus, Online power-aware routing in wireless
(October 1999). ad-hoc networks, in: Proc. of ACM/IEEE MobiCom2001 (July 2001).
[6] C.E. Jones, K.M. Sivalingam, P. Agrawal and J.C. Chen, Survey of en- [27] X.Y. Li and P.J. Wan, Constructing minimum energy mobile wireless
ergy efficient network protocols for wireless networks, ACM Wireless networks, ACM Mobile Computing and Communications Review 5
Networks (September 2001). (October 2001).
DISTRIBUTED POWER CONTROL FOR ENERGY EFFICIENT ROUTING 41

[28] J.H. Ryu and D.H. Cho, A new routing scheme concerning power- Alessandra Giovanardi received the Laurea De-
saving in mobile ad-hoc networks, in: Proc. of IEEE International Con- gree in electronic engineering from the University of
ference on Communications, New Orleans, LA (June 2000) pp.1719– Bologna, Italy, and the Ph.D. degree in information
1722. engineering from the University of Ferrara, Italy, in
[29] C.K. Toh, Maximum battery life routing to support ubiquitousmobile 1996 and 2000, respectively. Since February 2000
computing in wireless ad hoc networks, IEEE Communications Maga- she is a fellows of the Telecommunication Group of
zine 39(6) (2001) 138–147. the University of Ferrara, Italy, and since February
[30] K. Tsudaka, M. Kawahara, A. Matsumoto and H. Okada, Powercontrol 2001 she is a lecturer of Digital Telecommunication
routing for multi hop wireless ad-hoc network, in: Proc. of GLOBE- Networks in the same University. Her research in-
COM’01, San Antonio, TX (November 2001) pp. 2819–2824. terest include: multimedia sources characterization,
[31] S. Doshi, S. Bhandare and T.X Brown, An on-demand minimum energy protocols for wireless LAN, mobile IP, multicast, 3G cellular systems, ad hoc
routing protocol for a wireless ad hoc network, ACM Mobile Comput- networks.
ing and Communications Review 6(3) (2002). E-mail: [email protected]
[32] P. Bergamo, D. Maniezzo, A. Giovanardi, G. Mazzini and M. Zorzi,
Distributed power control for power-aware energy-efficient routing in
ad hoc networks, in: Proc. of EW2002, Florence, Italy (February 2002)
pp. 237–243.
[33] B. Chen, K. Jamieson, H. Balakrishnan and R. Morris, Span: an energy- Andrea Travasoni received the Laurea Degree in electronic engineering in
efficient coordination algorithm for topology maintenance in ad hoc 2002 from the Unversity of Ferrara, Italy. Since March 2002 to June 2002 he
wireless networks, in: ACM/IEEE MobiCom2001 (July 2001). has worked at the University of Ferrara with the staff of Prof. M. Zorzi and
[34] C. Schurgers, V. Tsiatsis, S. Ganeriwal and M. Srivastava, Optimiz- Prof. G. Mazzini. Since July 2002 he works in Telecom Italia Lab, Torino,
ing sensor networks in the energy-latency-density design space, IEEE Italy (ex C.S.E.L.T.). His research interests are on: ad hoc networks, process
Transactions on Mobile Computing 1 (January–March 2002) 70–80. design and process reengineering.
[35] Y. Xu, J. Heidemann and D. Estrin, Geography-informed energy con- E-mail: [email protected]
servation for ad hoc routing, in: Proc. of ACM/IEEE MobiCom2001
(July 2001).
[36] Y. Wei, J. Heidemann and D. Estrin, An energy-efficient MAC protocol
for wireless sensor networks, in: Proc. of INFOCOM 2002, New York,
NY (June 2002) pp. 1567–1576.
[37] D. Bertsekas and R.Gallager, Data Networks (Prentice-Hall Interna- Daniela Maniezzo received the Laurea Degree in
tional, 1992). electronic engineering from the University of Fer-
[38] A.S. Tanenbaum, Computer Network (Prentice-Hall, 1989). rara, Italy, in December 2000. Since January 2001
[39] R. Callon, A Comparison of ‘Link State’ and ‘Distance Vector’ routing she is a Ph.D. student in Telecommunication Engi-
algorithms (December 1987). neering at the University of Ferrara. In 2002 she
[40] C.E. Perkins, S.R. Das and E.M. Royer, Ad hoc on demand dis- joined the research team of Prof. Kung Yao at the
tance vector, IETF Draft, http://www.ietf.org/internet- Electric Engineering Department of UCLA (Univer-
drafts/draft-ietf-manet-aodv-08.txt (March 2001). sity of California Los Angeles) where she was in-
[41] C.E. Perkins, Ad Hoc Networking (Addison-Wesley, 2000) pp. 173– volved in a NASA project for the design and perfor-
220. mance evaluation of an unmanned aircrafts telemetry
[42] J. Broch, D.A. Maltz, D.B. Johnson, Y.C. Hu and J. Jetcheva, A perfor- system. She is currently PostDoc at UCLA Computer Science Department
mance comparison of multi-hop wireless ad hoc network routing pro- and is conducting research on Ad Hoc networks with Prof. Mario Gerla. Her
tocols, in: Proc. of the 4th Annual International Conference on Mobile current research interest include the design of MAC and Routing protocols
Computing and Networking (October 1998) pp. 85–97. for wireless ad hoc and sensor networks as well as power control problems,
[43] S.R. Das, C.E. Perkins and E.M. Royer, Performance comparison of radio channel model and propagation environments.
two on-demand routing protocols for ad hoc network, in: Proc. of IN- E-mail: [email protected]
FOCOM’00 Conference, Tel Aviv, Israel (March 2000).
[44] W.C. Jakes, Microwave Mobile Communications (IEEE Press, 1993).
[45] IEEE 802 Group-Part 11, Wireless LAN Medium Access Control (MAC)
and Physical (PHY) Layers Specifications (1999).
Gianluca Mazzini received the Laurea Degree in
electronic engineering and the Ph.D. in electronic en-
gineering and computer science from the University
of Bologna, Bologna, Italy, in 1992 and 1996, re-
spectively. In 1996 he joined the University of Fer-
rara, Italy, where he currently is an Associate Pro-
Pierpaolo Bergamo received the Laurea Degree in fessor of Telecommunications. He teaches Telecom-
electronic engineering in 2000 from the University munications Networks, Internet, UMTS and Digital
of Ferrara, Italy. Since January 2001 he is a Ph.D. Transmission Systems. His research interests are re-
student at the University of Ferrara with the staff of lated to: spread spectrum communications, applica-
Prof. M. Zorzi and Prof. G. Mazzini. He is involved tion of chaotic system to telecommunications, non-linear dynamical system
in 6NET (EU) project with the goal of deploying an modeling, next generation of cellular/ambient systems, wireless LAN archi-
European IPv6 network. In 2002, he temporarily tectures and protocols, sensor networks, Internet mobile computing, routing
joined the staff of professor K. Yao of EE Dep. in and security. He served as technical organizer for Nonlinear Dynamics of
UCLA (University of California Los Angeles) for a Electronic System 2000 Conference and he was an Editor of the IEEE Pro-
NASA project about aircraft telemetry. His research ceedings May 2002 Special Issue on Applications of Nonlinear Dynamics to
interests are on: ad hoc networks, sensor networks, localization, radio chan- Electronic and Information Engineering. He is author or co-author of about
nel models, cryptosystems based on chaos theory, IPv6. 110 papers. He is an IEEE Senior Member.
E-mail: [email protected] E-mail: [email protected]
42 BERGAMO ET AL.

Michele Zorzi received the Laurea Degree and the munications systems, random access in mobile radio networks, and energy
Ph.D. in electrical engineering from the University of constrained communications protocols. Dr. Zorzi currently has served on
Padova, Italy, in 1990 and 1994, respectively. Dur- the Editorial Boards of the IEEE Personal Communications Magazine, the
ing the Academic Year 1992/93, he was on leave at ACM/URSI/Baltzer Journal of Wireless Networks, the IEEE Transactions on
the University of California, San Diego (UCSD), at- Wireless Communications, and the IEEE Transactions on Communications,
tending graduate courses and doing research on mul- the IEEE Transactions on Mobile Computing, the Wiley Journal of Wireless
tiple access in mobile radio networks. In 1993, he Communications and Mobile Computing, and is the Editor-in-Chief of the
joined the faculty of the Dipartimento di Elettron- IEEE Wireless Communications Magazine. He is also guest editor for spe-
ica e Informazione, Politecnico di Milano, Italy. Af- cial issues in the IEEE Personal Communications Magazine (Energy Man-
ter spending three years with the Center for Wire- agement in Personal Communications Systems) and the IEEE Journal on
less Communications at UCSD, in 1998 he joined the School of Engineering Selected Areas in Communications (Multi-Media Network Radios).
of the Università di Ferrara, Italy, where he is currently a Professor. His E-mail: [email protected]
present research interests include performance evaluation in mobile com-