Professional Documents
Culture Documents
Transmission Power Control Techniques For Wireless
Transmission Power Control Techniques For Wireless
Transmission Power Control Techniques For Wireless
net/publication/222831574
CITATIONS READS
135 823
5 authors, including:
Some of the authors of this publication are also working on these related projects:
Developing Next Generation Smart Services and Applications for Smart Cities View project
All content following this page was uploaded by Luiz H. A. Correia on 05 October 2019.
Received 28 November 2006; received in revised form 28 March 2007; accepted 18 July 2007
Available online 7 August 2007
Abstract
Communication is usually the most energy-consuming event in Wireless Sensor Networks (WSNs). One way to signif-
icantly reduce energy consumption is applying transmission power control (TPC) techniques to dynamically adjust the
transmission power. This article presents two new TPC techniques for WSNs. The experimental evaluation compares
the performance of the TCP techniques with B-MAC, the standard MAC protocol of the Mica 2 platform. These exper-
iments take into account different distances among nodes, concurrent transmissions and node mobility. The new transmis-
sion power control techniques decrease energy consumption by up to 57% over B-MAC while maintaining the reliability of
the channel. Under a low mobility scenario, the proposed protocols delivered up to 95% of the packets, showing that such
methods are able to cope with node movement. We also show that the contention caused by higher transmission levels
might be lower than analytical models suggest, due to the action of the capture effect.
2007 Elsevier B.V. All rights reserved.
Keywords: Wireless sensor networks; Transmission power control; Medium access control; Energy consumption
1. Introduction
1389-1286/$ - see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.comnet.2007.07.008
4766 L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779
in general, it is not possible to replace or recharge AEWMA, employs calculations to determine the
batteries due to the number of nodes deployed or ideal transmission power based on the reception
inhospitable environmental conditions. Hence, power, transmission power and average noise. Due
energy conservation is a critical factor in WSNs. to their simplicity, those protocols can also be
Severe hardware and energy constraints preclude applied to more resourceful networks. Experiments
the use of protocols developed for MANETs, which carried out in the Mica 2 platform show the effi-
comparatively possess more resources. The strict ciency of the protocols, considering parameters such
requirements of WSNs force networking protocols as energy and throughput. The experiments took
to be as much energy-efficient as possible. Medium place in indoor and outdoor environments.
access control (MAC) protocols, for example, mod- This article is organized as follows. Section 2
ify transceiver parameters and the topology of the summarizes the previous research on TPC tech-
network in order to reduce energy consumption. niques. The proposed methods are detailed in Sec-
One of the transceiver’s parameters is the transmis- tion 3. The implementation of each techniques on
sion power. the Mica 2 platform is discussed in Section 4. Sec-
Transmission Power Control (TPC) techniques tion 5 describes the evaluated scenarios and presents
improve the performance of the network in several the achieved results. Finally, Section 6 draws the
aspects. First, power control techniques improve conclusions and future work.
the reliability of a link. Upon detecting that link
reliability is below a certain threshold, the MAC 2. Related work
protocol increases the transmission power, improv-
ing the probability of successful data transmissions Gomez and Campbell analyzed the benefits of
[2–5]. Second, only nodes which really must share transmission power control in wireless multi-hop
the same space will contend to access the medium, networks [2]. The authors showed that per-link
decreasing the amount of collisions in the network. range adjustments outperform global range trans-
This enhances network utilization, lowers latency mission adjustments by 50%. Thus, instead of glob-
times and reduces the probability of hidden and ally defining a transmission range that keeps the
exposed terminals [5]. Finally, by using a higher network connected, wireless networks should adjust
transmission power, the physical layer can use mod- transmission ranges on each link. They also demon-
ulation and coding schemes with a higher bit/baud strated that the average traffic capacity per node is
ratio [6,7], increasing the bandwidth in the presence constant on a TPC-aware network even if more
of heavy workloads, or decreasing it to maximize nodes are added to a fixed-size area. This is not true,
energy savings. however, if the transmission range is kept fixed. For
Although being an effective mechanism to reduce such networks, the traffic capacity decreases when
energy consumption, TPC is not implemented in more nodes are added, due to increased interference.
any existing MAC protocol for WSNs. This occurs Ammari and Das developed analytical models to
due to the highly imprecise nature of readings pro- evaluate how the transmission power affects latency
vided by the transceiver, and also due to the and energy consumption in WSNs [12]. The authors
restricted resources found in current nodes. Previ- showed that, by increasing the distance traveled at
ous works are based on models and simulations that each hop, the end-to-end latency decreases at the
assume precise readings and that the transmission cost of a higher energy consumption. For a small
range can be assigned to an arbitrary value, which distance per hop, however, less energy is consumed,
is not the case for real hardware. Our analysis, on but the latency increases as more hops must be tra-
the other hand, is based on experiments with pro- versed. The authors then proposed the creation of
duction hardware employed by a number of quality of service (QoS) classes with different latency
research groups [8–11]. and energy guarantees based on the transmission
In this article we propose two transmission power employed on the communication. The ana-
power control protocols for WSNs, which can be lytical models, however, do not consider the effects
embedded into any existing MAC protocol. The of collisions on the communication, a critical issue
first, called Hybrid, calculates the ideal transmission as the transmission power increases.
power using a closed control loop that iterates over PCMA (Power Controlled Multiple Access) is a
the available transmissions powers in order to main- MAC protocol that allows communication at mini-
tain a target link quality. The second, called mum propagation ranges, allowing spatial reuse of
L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779 4767
the medium [5]. Agarwal et al. proposed a distrib- The use of TPC techniques in the MAC layer
uted power control algorithm for MANETs [13]. alone is not sufficient, since the communication in
Pires et al. extended this algorithm by adding a table MANETs is usually multi-hop. Thus, if the routing
in each node, which stores the transmission power protocol does not take into account the differences
used on previous transmissions [14]. Jung and Vai- of transmission power in each link, routes can be
dya proposed the adjustment of the transmission energy-inefficient. Based on this issue, Kawadia
power for each frame in order to mitigate asymmet- and Kumar studied in [16] whether TPC techniques
ric links caused by transmission power variation [3]. should be implemented in the MAC or in the rout-
The existing TPC methods for MANETs usually ing layer. According to the authors, TPC techniques
depend on calculations involving floating-point should be coded in the routing layer in order to
computations. Due to the severe hardware con- maintain modularity. Based on this assumption,
straints of sensor nodes, the existing algorithms they proposed four TPC-aware routing protocols
must be simplified or new algorithms tailored to to minimize the energy consumption in MANETs.
restricted environments must be developed. In order to define the ideal transmission power for
Lin et al. proposed a closed-loop TPC protocol each link of the routes, the protocols execute several
for WSNs that approximates the ideal transmission instances of a routing establishment algorithm, one
power using linear equations [15]. Based on empiri- for each transmission power available. Each
cal data, the authors show that link quality can be instance periodically sends HELLO packets to iden-
roughly approximated by the received signal tify the neighbors of the node, hence determining
strength using a linear relation. This lead to the cre- the nodes reachable by each transmission power.
ation of a two-phase protocol. On the initialization Packets are sent through the routes built with the
phase, nodes send probe packets to all their neigh- smallest transmission power.
bors, storing RSSI readings (rtij , which is the However, the protocols developed by Kawadia
received signal strength on node j for the ith itera- and Kumar spend a high amount of energy to build
tion on transmission power t) for all the available their routes due to the execution of several instances
powers (pt) on the radio. Using this information, of the routing algorithm. As energy consumption is
the protocol finds aj and bj for every link j such that critical in WSNs, we advocate that the transmission
rtij ¼ aj pt þ bj using the least square approxima- power must be calculated by the MAC layer due to
tion. On the operational phase, the protocol period- its higher energy-efficiency. TPC-aware routing
ically sends probe packets to update both algorithms would query the MAC protocol to check
parameters. If the perceived link quality drops the transmission power or reliability of each avail-
below a certain threshold, the receiver notifies the able link.
sender, which in turn recalculates aj and bj for the A problem related to transmission power control
link. is the propagation on low power wireless radios.
Although computationally expensive, the authors Lal et al. characterized the reliability of wireless links
argue that, as the noise level on the medium varies and proposed algorithms to approximate it on real-
at most 3 dBm per hour, one might send one probe time [17]. Reijers et al. [18] studied the effect of obsta-
packet per hour to properly adjust the transmission cles and environmental changes on link quality.
power. The most significant drawback of the Zhou et al. [19] developed a new propagation model
method is its enormous memory consumption, due that closely resembles the results obtained from
to the huge amount of RSSI readings that must be experimental data. Son et al. studied the effects of
cached to calculate the parameters of the curve. concurrent transmissions on signal to noise ratio
The authors show that, for 20 neighbors, their and on link reliability on WSNs [20]. They identified
method consumed 50% of the total memory. Our that collisions occur when the signal strength of two
method is much lighter, allowing the transmission or more consecutive transmissions are separated by a
power to be updated for every transmitted packet, certain threshold. For values outside this area, the
and consumes only 10% of the available memory.1 weaker transmissions are ignored. De Couto et al.
proposed an extension to Dynamic Source Routing
(DSR), where the route calculation takes into
1
A direct comparison is possible because both nodes use the
account the reliability of the links, greatly reducing
same operating system and have the same amount of memory, the amount of retransmissions [21]. Woo et al. pro-
differing only by the employed radio. posed mechanisms to assess link reliability on WSNs
4768 L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779
[10,22]. They model the reception of packets as a the Mica 2 platform, for instance, there are 22 differ-
Bernoulli stochastic process and apply signal filters ent levels, separated by roughly 1 dBm [23]. Fur-
to determine the instantaneous packet reliability. ther, as the power switching operation is fast (it
takes 20 ls on Mica 2 nodes), it is possible to iterate
3. TPC techniques over the available power levels, increasing or
decreasing the transmission power when necessary.
In this section we describe the new TPC tech- Algorithm shows the operation of the algorithm,
niques. The first, called Hybrid, calculates the ideal which is described below. It operates in two phases,
transmission power using a closed control loop that where the first phase (lines 8–14 of the algorithm)
iterates over the available transmissions powers in determines the ideal transmission power, while the
order to maintain a target link quality. The second, second phase copes with medium changes. The algo-
called AEWMA, employs calculations to determine rithm works as follows. The first phase starts with
the ideal transmission power based on the reception the transmission power set to the maximum value
power, transmission power and average noise. allowed by the transceiver. A node wishing to deter-
These techniques can be deployed on any existing mine the ideal transmission power sends a power
MAC, since they are simple enough to execute in query message (MPQ) piggy-backed in data packets
restricted environments. In the following, the term at the ‘‘current’’ ideal transmission power and waits
‘‘ideal transmission power’’ is defined as the lowest for an acknowledgment (ACK) packet. If the recep-
power level able to successfully transmit messages tion is confirmed, the transmitter decreases the ideal
from a node to another. transmission power by one level, and sends another
MPQ message on the next data packet. When a MPQ
3.1. The hybrid method message is not acknowledged (lines 22–28), the
transmitter assumes that the ideal transmission
The ideal transmission power can be determined power was found, triggering the second phase.
using successive refinements. Transceivers have a The second phase of the algorithm (lines 14–21
limited number of allowed transmission powers. In and 22–28) uses ACKs to adjust the transmission
20: end if
21: end if
Require: timeout (ACK, dst) //packet sent to dst timed out
22: phasedst 2ndphase;
23: lostdst lostdst + 1;
24: receiveddst 0;
25: if lostdst = LI then
26: txPowerdst txPowerdst + 1;
27: lostdst 0;
28: end if
29: end procedure
power as follows. If a number of consecutive trans- of signal attenuation, and must satisfy the following
missions are not acknowledged (this number is conditions2 [5]. The formulas refer to a sender, node
called the increase threshold level, or LI), the ideal A, and to a receiver, node B.
transmission power is increased by one level. Since
the noise can also decrease due to environmental • The transmission power must lie within the oper-
changes, communication can also improve, thus ational limits of the transceiver:
the transmission power is lowered if a certain num- P TXlower 6 P TXmin 6 P TXupper : ð1Þ
ber of consecutive messages are successfully received
(the decrease threshold level, or LD). The values of • The transmission power must compensate the
LI and LD must be set up according to the through- attenuation imposed by the propagation of the
put of the application, avoiding a late reaction when signal from the sender to the receiver, guarantee-
the throughput is low and an early reaction when ing that the received signal strength is higher than
the throughput is high. The algorithm does not dis- the minimum sensitivity of the radio (RXthreshold).
tinguish node and transmission failures due to the If the reception power is below RXthreshold, nodes
use of ACKs to assess link reliability. Broadcast will be unable to decode the message. The atten-
packets are always transmitted at a fixed power uation suffered by the signal (GA!B) is asymmet-
since they are not acknowledged. ric, and is inferred by the transmitted (PTX) and
Initial experiments showed that the method pre- the received (PRX) power:
sented fluctuations on the transmitted power [24]. P RX
Nodes used a transmission power too close to the GA!B ¼ ð2Þ
P TX
average noise, thus reducing the probability of a
correct reception. In order to avoid this, we apply Thus, the ideal transmission power must satisfy
Eq. (4) of the AEWMA method, shown in Section the equation:
3.2, to guarantee a transmission power that is RX threshold
always above the average noise by a certain thresh- RTXmin P ð3Þ
GA!B
old. Whenever a packet arrives at a reception power
too close to the average noise, the receiver notifies • Another factor that influences the communica-
the sender that the transmission power must be tion is the background noise (also referred as
increased by one level (lines 6 and 7) using a special noise), caused by signals naturally present in the
bit on the ACK. Upon the arrival of the ACK, the environment [17,18]. Thus, in order to differenti-
sender increases the transmission power by one level ate data from noise, the reception power of the
(lines 8–11). Due to the use of both iterative and data must be higher than the power level of the
mathematical techniques to assess the ideal trans- noise (NB) by a certain threshold, called
mission power, we call this method hybrid. SNRthreshold, as described in the following
equation:
SNRthreshold N B
3.2. The AEWMA method P TXmin P ð4Þ
GA!B
The second method takes into account that ideal
2
transmission power can be calculated as a function For clarity, the relations are expressed in mW.
4770 L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779
In short, the ideal transmission power must sat- the dashed curve. In order to have a stable transmis-
isfy Eqs. (3) and (4) at the same time, as ex- sion power, we must apply signal filtering tech-
pressed by the following equation: niques to the result of Eq. (5) (the solid curve
shows the filtered output). In order to do so, we
RX threshold SNRthreshold N B
P TXideal ¼ max ; employ an EWMA (Exponentially Weighted Mov-
GA!B GA!B
ing-Average) function.
ð5Þ The EWMA function is a weighted average that
assigns exponentially decreasing weights to old
The AEWMA method uses Eq. (5) to calculate data. Suppose we take a sequence of readings (I)
the ideal transmission power of the next transmis- as input and output an instantaneous ‘‘average’’
sion and works as follows. Nodes periodically sam- for every new reading (O). This average is based
ple the signal strength when no transmissions occur on a factor a, where 0 < a < 1. The output value
in order to determine the base noise (NB). If node A in iteration i is given by Oi = Oi1 · (1 a) + Ii · a.
wishes to communicate with node B, it transmits the The equation ensures that, in a given iteration i, the
first packet to B at the current transmission power (i k)th element of the input sequence will contrib-
(PTX). The packet includes PTX on its headers. ute with weight a · (1 a)ik to the current output
When B receives the packet, it determines the value. Another property of this calculation is that,
received signal strength (PRX, or reception by decreasing the value of a, more importance is
power) and calculates the ideal transmission power given to past inputs over the recent ones. The value
(PTXideal) from A to B using Eq. (5). Next, B sends of a must be carefully chosen, since a high value
the calculated power to A piggy-backed on the leads to constant variations on the output, while a
acknowledgment. The received value is stored on a low value leads to very slow changes. To avoid this
table to be used on subsequent communications effect on our experiments, Section 5.1 evaluates sev-
with node B. eral values of a, for both indoor and outdoor
AEWMA assumes that the signal attenuation is environments.
symmetric. This is a reasonable assumption, since The choice of the averaging function took two
most MAC protocols in WSNs rely on some sort factors into consideration. First, the memory foot-
of acknowledgment messages, which requires reli- print of the method should be as low as possible.
able links on both directions. The values of Operations such as maximum, minimum, median
RXthreshold and SNRthreshold vary for each trans- and average were discarded, since the node must
ceiver, and thus must be empirically determined. maintain a list of the last transmission power values
Initial testing showed that the output of the cal- for each link. Second, the implementation should be
culation fluctuates, since the input values are always fast and simple, avoid floating-point variables, and
changing due to small variations in the environment using few or no divisions3. Thus, functions relying
and the quality of the batteries [24]. The variation of on differential equations were discarded.
the result in time is shown graphically in Fig. 1 by Although EWMA seems to require floating-point
operations, by carefully choosing a values in the
form 1 21n , we were able to implement an integer
-6 version of the calculation using only three shifts
Calculated TX power
Calculated transmission power (dBm)
3.3. Complexity analysis consumption, as it stores only the past output value
and presents an adjustable standard deviation. The
This section evaluates the memory and CPU last filter tested was the PID filter, which obtained
requirements of the AEWMA method. The first standard variations from the order of 0.85 up to
constraint imposed on the TPC calculation is time- 0.43. From this study we discarded the moving aver-
based. Since the ACK carries the calculated trans- age filter due to their inferior result when compared
mission power, the TPC algorithm must complete to the EWMA and PID filters.
before the ACK is sent. In Mica 2 nodes, this time Next, we ran both EWMA and PID filters in an
is equivalent to 600 ls. emulator for the Mica2 processor. First, in order to
Eq. (5) demands division operations, which fit the algorithms in a sensor nodes, we made an
require several cycles to complete. To make the implementation based only on integers to avoid
implementation more efficient, we coded the opera- floating-point emulation. A test on a PC using real
tions using measures in dBm, thus requiring only traces showed that our implementation deviated at
additions and subtractions. We evaluated the run- most 0.5 dBm from the result of their floating-point
ning time of the calculation on an emulator of the counter-part. Next, we ran the code on the emulator
Mica 2 processor to ensure that the method would to identify their CPU consumption. As we could not
not delay the transmissions [24]. After running inject traces into the emulator to provide realistic
20,480 iterations of the AEWMA method, we found data, we ran the PID and EWMA algorithms
that the code compiled with no optimizations 320,000 times. The EWMA demanded, on average,
demands, on average 834.17 CPU cycles (208.54 20 instructions to complete, while the PID
ls), which is approximately the time required by the demanded 80. From this test we conclude that both
radio to transmit one byte4. Hence, the AEWMA PID and EWMA filters are suitable to WSNs. How-
method uses one third of the maximum computa- ever, for simplicity reasons, we chose the EWMA
tional ‘‘budget’’. over the PID because it demands only one parame-
The choice of the filter function was based on ter to be optimized, while the PID demands three.
both CPU and memory consumption. The Kalman
filter [25] was discarded due to its high complexity as 4. Implementation of the TPC techniques
it demands matrix calculations. Thus, we were
restricted to simpler filters, such as EWMA, PID We implemented the TPC methods on the Mica 2
and moving average. platform running the TinyOS operating system [26].
Before implementing the filters in hardware, we We chose this environment since the TinyOS and
tested them over a real trace of the transmission the Mica 2 architectures are de facto standards for
power calculation, in order to evaluate their effec- commercial and academic WSNs. Furthermore,
tiveness. We measured their performance based on the hardware on Mica 2 represents the typical band-
their standard deviation for 400 runs. The threshold width, memory capacity and processor of other
values of PID and the alpha value of EWMA were popular hardware choices.
empirically varied in order to find their smallest The TPC techniques were implemented inside the
standard deviation. standard MAC protocol provided in TinyOS, called
The moving average filter showed a high stan- B-MAC [8]. The B-MAC protocol is a CSMA/CA
dard deviation, varying from 1.13 for the average protocol without channel reservation (RTS/CTS
of the last two packets up to 0.9 for the average messages). The use of acknowledgment (ACK) is
of the last five packets. The EWMA, in contrast, optional, being defined by the application. The pro-
decreased the standard deviation for increasing val- tocol does random back-offs to avoid collisions and
ues of a. For 0.5, for example, the deviation was keeps retransmitting packets until they are correctly
equal to 1, decreasing to 0.85, 0.76 and 0.62 for a received (when ACKs are activated).
equal to 0.7, 0.8 and 0.9, respectively. Clearly, the Since the memory is a scarce resource in WSNs,
EWMA function is a much better choice when com- we measured the amount of memory used by each
pared to the moving average in terms of memory method, including the underlying MAC protocol.
B-MAC took up 241 bytes (out of 4KB), the Hybrid
method required 353 bytes and the AEWMA
4
Assuming the typical bandwidth on Mica 2, which is method required 393 bytes, respectively. Most of
38.4 kbps. the increase in memory consumption was due to
4772 L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779
the neighbor table that stores the information 5.1. Alpha in EWMA
required to calculate the transmission power for
each node. This table is composed of 20 entries, Since the performance of the AEWMA method is
allowing each node to communicate with 20 of their heavily dependent on the selection of a, we per-
neighbors at a time. The table size is fixed due to formed a set of experiments on indoor and outdoor
limitations of the language used to program the environments to evaluate the most suitable value for
nodes, which does not allow dynamic memory allo- each environment. In order to speed up the compu-
cation. Since the number of nodes on the table is tation, we showed in Section 3.2 that the value of a
modifiable at compile time, WSN designers can must be in the form of 1 21n . We used the values
adjust its value according to the needs of each net- 0.03125, 0.125, 0.25 and 0.5. Note that higher values
work. Although each entry requires 4 bytes for the of a will increase the variation of the signal, as the
most memory-hungry method, the transmission most recent measurements have a higher weight
power table accounted for an increase of RAM on the calculation. On the other hand, when a is
usage of 63%. too small, the time required to converge is high
Because the transmission of data spends more and the output takes a long time to react to changes.
energy than the computation, we opted to minimize Due to those considerations, we do not employ val-
the number of bytes transmitted in detriment of ues over 0.5 or lower than 0.03.
memory and processing efficiency. Thus, power con- Outdoor scenario: In this scenario we evaluate
trol information is piggy-backed in data and four different values of alpha for distances among
acknowledgment (ACK) packets. The additional nodes varying from 5 up to 20 on increments of
headers increased the data packet size by three bytes 5 m. Results are averaged over four independent
(one byte for the transmitted power and two bytes experiments with a confidence interval of 95%. In
for the sender address). ACK packets are increased terms of average delivery rate, most configurations
by five bytes (two bytes for the sender address, performed similarly, delivering as much as 97% of
another two bytes for the receiver address and one the packets. For an a = 0.031, however, the delivery
byte for the ideal transmission power). rate was up to 3% smaller than the others due to the
The parameters of each method were defined long time necessary to respond to changes in the
after a series of experiments (see [24] for more transmission power. The average transmission
details). We determined the SNRthreshold as power is shown in Fig. 2. The conservative value
10 dBm and RXthreshold as 85 dBm. For the of 0.031 performed poorly, employing 5 dBm. We
Hybrid method, LI was set to one and LD to eight. attribute this to the long convergence time of the
The evaluation of a is shown in Section 5.1. estimator, making it more conservative. For a high
value of a, such as 0.5, the filter gives too much
5. Evaluation importance to the recent calculation, thus incorpo-
rating spikes in the transmission power. Those vari-
This section evaluates the implementation of the
proposed TPC techniques on the Mica 2 platform.
We evaluate two metrics, the average transmission 10
power and the average delivery rate. Their values
are taken from the mean of five independent exper- 5
Transmission Power (dBm)
values due to the time-consuming nature of the Fig. 2. Average transmission power for some values of a in an
experiments. outdoor scenario.
L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779 4773
6 100
AEWMA(0.031)
AEWMA(0.125)
4 AEWMA(0.25)
2
60
0
40
–2
5m
20 10 m
–4 15 m
20 m
0
B–MAC B– Hybrid AEWM AEWM
–6 (5 dBm MAC (0 dBm A(0,25
)
A(0,12
5)
) )
2.0
The outdoor experiments were performed in an
open field, avoiding thus any source of interference. 0.0
0.125. –12.0
Hybrid AEWMA(0,25) AEWMA(0,125)
The average delivery rate is shown in Fig. 4. B-
MAC at 5 dBm keeps an astounding 97% of packet Fig. 5. Average transmission power in the outdoor scenario.
4774 L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779
or larger than 15 m, all TPC protocols employed a tively. For the AEWMA protocol, as shown in Sec-
transmission power over 0 dBm. Hence, the poor tion 5.1, the value of a on the indoor environment
performance of B-MAC using 0 dBm in Fig. 4 cor- makes almost no difference to its performance due
responds to the use of an insufficient transmission to the mild environment.
power. For transmission powers of 15 m and The last protocol evaluated was B-MAC, trans-
20 m, the TPC techniques had to transmit at around mitting at 5 dBm. Since it used a higher transmis-
3 and 5 dBm, thus B-MAC at 0 dBm should have an sion power than the TPC protocols, B-MAC
insignificant delivery rate. achieved a delivery rate of 97.9%, which is around
Converting the achieved transmission power to 2% superior to the best results achieved with trans-
the energy consumed by the radio, we found that mission power control, though at the cost of a
the AEWMA method consumed 57.5% and 43% higher transmission power.
less than B-MAC in 5 m and 10 m, respectively,
for a transmission power of 5 dBm. Comparing to
5.4. Spatial reuse
B-MAC at 0 dBm, the AEWMA method consumed
35% and 13.5% less for the same distances. The
The next experiment evaluates medium reuse and
Hybrid method consumed up to 13% more energy
interference. We positioned four nodes in line
than the AEWMA method, since it was more con-
directly on the ground in an empty corridor, as
servative when decreasing the transmission power.
depicted in Fig. 6 (arrows denote data transmis-
Overall, AEWMA slightly outperformed the Hybrid
sion). The nodes were deployed very close to each
method. All TPC methods delivered as much pack-
other because the transmission range is significantly
ets as B-MAC and consumed less energy for 5 and
lowered when nodes are placed on the ground.
10 m.
Nodes transmit at a rate of 12.5 packets per second,
which is the maximum rate at which nodes are able
5.3. Indoor environments to log the packets for later analysis. This is a limita-
tion of the OS, as the effective bandwidth utilization
In this scenario, a sender transmits 1440 packets of 12.5 pps is far from the maximum supported by
during a period of 6 min. The receiver node was the radio.
3.9 m apart from the sender, in a computer lab with In order to justify this scenario, we recur to a
people moving around. This test aims to identify simple queuing theory model to identify the proba-
how protocols behave in an indoor environment bility of finding a busy medium. The radio transmits
with various obstacles and sources of interference. 38,400 bps in manchester encoding, and a packet is
The first set of experiments evaluates the best val- composed of 46 bytes, including the preamble, syn-
ues for the parameters of each protocol. We varied chronization bytes and data. Assuming that packets
the values of LI (number of consecutive packets not are produced of a rate of 25 pps (the aggregate
acknowledged to increase the transmission power) throughput of both transmitting nodes), the proba-
for the Hybrid protocol, and a for the AEWMA bility of a node finding at least one client on the sys-
protocol. In contrast to the outdoor environment, tem, which is equivalent to having someone already
in the indoor environment the likelihood of two transmitting, is equal to 23.95%.
packet failures in a row is slightly higher. Hence, The aim of this scenario is to measure the conten-
when increasing LI from one to two, we were able tion on the medium, as well as the amount of colli-
to achieve the delivery rate found in the outdoor sions. We use those metrics to verify whether the use
scenario (95%). The average transmission power of transmission power control allows more concur-
when LI = 2 is around 16% smaller when LI = 1, rent transmissions at the same time, as indicated
(3.7 ± 0.55) dBm and (4.3 ± 0.65) dBm, respec- by previous analytic results [2]. According to those
0
-2
95 -4
90
Average delivery rate (%)
-6
85
-8
80
-10
75
-12
70
65 -14
-16 AEWMA (0.25)
60
Distancing AEWMA (0.0625)
55 -18 Hybrid (LD = 8)
Approaching -20
50
Combined 0 10 20 30 40 50 60
45
B- M
A
AEW
MA(0
Hybr
id L Time (s)
C (5 .25) D =4
dBm
)
Fig. 11. Average transmission power for a mobile node
Fig. 10. Average delivery rates for a mobile node (indoor). (outdoor).
L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779 4777
-2 in future works.
-4
-6
-8 References
-10
-12
[1] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci, A
-14
survey on sensor networks, IEEE Communications 40 (8)
-16 AEWMA (0.25) (2002) 102–114.
-18 Hybrid (LD = 4)
[2] J. Gomez, A.T. Campbell, A case for variable-range trans-
-20
0 10 20 30 40 50 60 mission power control in wireless multihop networks, in:
Time (s) Proceedings of the IEEE INFOCOM, vol. 2, 2004, pp. 1425–
1436.
Fig. 12. Average transmission power for a mobile node (indoor). [3] E.-S. Jung, N.H. Vaidya, A power control MAC protocol
for ad hoc networks, in: Proceedings of the Eighth Annual
International Conference on Mobile Computing and Net-
the differences among the TPC methods are less working (MobiCom), ACM Press, 2002, pp. 36–47.
pronounced. [4] P. Karn, A new channel access protocol for packet radio, in:
American Radio Relay League Ninth Computer Networking
Conference, 1990.
6. Conclusions and future work [5] J.P. Monks, Transmission power control for enhancing the
performance of wireless packet data networks, Doctor of
Wireless sensor networks must be designed with Philosophy, University of Illinois at Urbana-Champaign,
energy-efficiency in mind. The adjustment of the 2001.
[6] A. Ghosh, D.R. Wolter, J.G. Andrews, R. Chen, Broadband
transmission power, performed by transmission
wireless access with wimax/802.16: current performance
power control (TPC) protocols, is a technique to benchmarks and future potential, IEEE Communications
diminish energy consumption in the communication. Magazine 43 (2) (2005) 129–136.
Due to the limited resources of wireless sensor net- [7] V. Kawadia, P.R. Kumar, A cautionary perspective on cross
works (WSNs), the TPC techniques proposed for layer design, IEEE Wireless Communications 12 (1) (2005)
3–11.
mobile ad hoc networks (MANETs) are not applica-
[8] J. Polastre, J. Hill, D. Culler, Versatile low power media
ble to WSNs. This article introduced and evaluated access for wireless sensor networks, in: Proceedings of the
through experiments two new TPC techniques spe- Second International Conference on Embedded Networked
cifically designed to WSNs. Sensor Systems, ACM Press, 2004, pp. 95–107.
The protocols were evaluated in diverse scenarios, [9] A. Woo, D. E. Culler, A transmission control scheme for
media access in sensor networks, in: Mobile Computing and
varying parameters such as the environment type, the
Networking, 2001, pp. 221–235.
distance among nodes, multi-hop transmissions and [10] A. Woo, T. Tong, D. Culler, Taming the underlying
mobility. Results showed that our enhancements challenges of reliable multihop routing in sensor networks,
enabled energy savings of up to 57% over fixed trans- in: Proceedings of the First International Conference on
mission power communications. Further, our evalu- Embedded Networked Sensor Systems, ACM Press, 2003,
pp. 14–27.
ation showed that TPC techniques increase the
[11] W. Ye, J. Heidemann, D. Estrin, An energy-efficient mac
throughput due to an increased spatial reuse, as a protocol for wireless sensor networks, in: Proceedings of the
lower transmission power diminished the amount IEEE INFOCOM, New York, NY, USA, 2002, pp. 1567–
of back-offs before nodes send their packets. We 1576.
found that the amount of packet drops also depends [12] H.M. Ammari, S.K. Das, Trade-off between energy savings
and source-to-sink delay in data dissemination for wireless
on the capture effect, which is not considered on
sensor networks, in: Proceedings of the Eighth ACM
existing analytical models. Finally, TPC protocols International Symposium on Modeling, Analysis and Sim-
must be adjusted to accommodate mobile nodes, ulation of Wireless and Mobile Systems (MSWiM), ACM
since in such situations their parameters must be Press, New York, NY, USA, 2005, pp. 126–133.
tuned to provide a faster response. [13] S. Agarwal, S. Krishnamurthy, R. Katz, S. Dao, Distributed
power control in ad hoc wireless networks, in: Personal and
TPC techniques can achieve higher gains when
Indoor Mobile Radio Communication (PIMRC), vol. 2,
tightly integrated with communication protocols. IEEE, 2001, pp. 59–66.
Routing protocols, for example, could be modified [14] A.A. Pires, J. F. de Rezende, C. Cordeiro, ALCA: a new
to use the transmission power as a routing metric, scheme for power control on 802.11 ad hoc networks, in:
4778 L.H.A. Correia et al. / Computer Networks 51 (2007) 4765–4779
and Performance Modeling. His main research areas are mobile Canada (1988/1989) and passed a sabbatical year at Univ. Paris 6
Computing, distributed algorithms, and network management. and U. Evry, France (2004/2005). He headed the Department of
Computer Science at the UFMG (1998/2000). He heads the
computer network group at UFMG and has been the technical
José Marcos Nogueira is an Associate coordinator of the SIS project. He has served in various roles,
Professor of Computer Science at the including general chair (1985) of SBRC and LANOMS, TPC
Federal University of Minas Gerais chair or co-chair of SBRC, WCS/Globecom 2006, ICC 2006. He
(UFMG), Brazil. His areas of interest has been TPC Member in IEEE/IFIP NOMS, IEEE/IFIP IM,
and research include computer networks, IEEE/IFIP MMNS, IPOM, SBRC IEEE/IFIP DSOM 2003/
telecommunications and computer net- 2005. He is member of the Brazilian Computer Society (SBC) and
works management, and wireless sensor IEEE CommSoc, and secretary of IEEE/CommSoc TCII.
networks. He received B.S. and M.S.
degrees from UFMG and Ph.D. degree
in Electrical Engineering from University
of Campinas, Brazil (1985). He held a
Post-Doctoral position at the University of British Columbia,