Evaluation of Constrained Wireless
Nodes: A Cross Layer Approach
Department of Electrical Engineering
srikank@stanford.edu
Abstract
In
this report, preliminary results of our evaluation of new wireless sensors
called Micaz [1] and Telos
motes [2] that are based on CC2420 [3] radio are presented. A flexible
evaluation tool was developed with which different experiments with these motes
under different settings were possible. 30 Micaz
motes were evaluated in a complex indoor environment on a testbed
provided by Intel Research at
1 Introduction
Cross
layer design and analysis [5, 6] is an emerging field and has gained
significant attention in recent years. The main criterion for cross layer
design approach is to optimize the design of different protocol layers such
that the overall objectives of the application(s) are efficiently met subject
to additional constraints on power, size, computational capability, etc.
Wireless
sensor network (WSN) is a special kind of network in which the overall
objective is to sense some physical phenomena of interest and efficiently
deliver the collected data to a data collection node (generally referred to as
a Sink) in a timely fashion with some special requirements as dictated by the
application. The nodes used in a WSN, generally referred to as wireless sensors
(also referred as motes) are usually operated using batteries and are very
limited in their computational capabilities. For some applications such as
environmental monitoring, these nodes are expected to last for months without
battery replacement. A cross layer design strategy seems appropriate for such
networks where the nodes that are constrained in their capabilities have to
meet some overall application objectives. A thorough understanding of
capabilities of some of the commercially-off-the-shelf (COTS) wireless radios
that are built for WSN applications would be a good starting point for such a
cross-layer design and analysis of WSNs for different
applications [7, 8].
In
this report we present preliminary results of the evaluation of Micaz and Telos motes that are
based on the new radio chip called CC2420 from Chipcon.
CC2420 is based on IEEE 802.15.4 [9] which is a standard for wireless sensors.
For this reason it is worth a thorough evaluation. The rest of the report is
organized in to following sections. Section 2 introduces the evaluation tool
that was developed in nesC and Java. Section 3
presents the detailed experimental setups of the evaluations. In section 4 we
present the experimental results and observations. We provide the summary of
our results and conclusion in section 5 followed by future work in section 6.
2 Evaluation Tool
One
of the main contributions of this project is the development of a flexible
evaluation tool. The tool has 3 parts. The first part is the nesC code [10] to be loaded into the motes with TinyOS [11, 12] so that they can accept commands, send
packets, receive packets, record important statistics of any received packet
and send collected statistics. The second part is the java code that runs on a
laptop or a PC. This code implements commands such that the motes can be
instructed to send packets at a given rate to a specified receiver mote. It can
also instruct the receiver mote to send the statistics of all the packets it
received. The third part is the script that runs on the PC or laptop that
automates the evaluation. It takes care of programming all the motes with
appropriate codes, runs the java code for different node pairs and stores the
data retrieved by the java code in a data file. It also takes care of sending
the commands over either the serial port (Telos motes
have their USB ports mapped to serial COM ports) or over Ethernet (Micaz motes are connected using Ethernet to the PC/Laptop).
The third part is especially very useful when a large number of nodes are
involved and when the same experiment has to be repeated for different
evaluation parameters.
Fig. 1.
Evaluation Tool
Fig.
1 shows the functionalities of 3 parts of the evaluation tool. The script that
runs on the PC or the laptop starts with uploading the appropriate nesC code into different motes. It also assigns unique mote
IDs to different motes. Once the upload is complete for all the motes it
triggers the java code and specifies the node pair (sender mote ‘m’ and
receiver mote ‘n’) for which the evaluation has to be done. The java code then
commands the specified mote ‘m’ using “Send” command to send ‘p’ number of
packets at the rate of ‘r’ packets/second to mote ‘n.’ Upon completion of
sending ‘x’ packets, mote ‘m’ sends “Send Complete” to the java module of the
PC. The java module then sends “Retrieve” command to mote ‘n’ so that the
statistics of the received packets can be collected. Mote ‘n’ then starts
sending the statistics to the java module. Upon completion of sending all the
statistics of the packets it received from mote ‘m,’ mote ‘n’ sends “Retrieve
Complete” to the java module. In the meantime, as the statistics packets are
received from mote ‘n,’ the java module sends this information to the script
module which opens a data file corresponding to the node pair and stores this
data for future analysis. This process is repeated for different node pairs,
different power levels and different channels. In this tool the variables m, n.
p, r and x can be easily changed. The power levels and channels are changed by
uploading different compiled codes into the motes which is also trivial. This
compilation has to be done only once. The script will recompile it for
different motes with different mote IDs and also upload the code to the
corresponding mote.
As
mentioned above, our evaluation tool has been created such that many parameters
such as channel, transmission power level, sender mote, receiver mote, number
of packets and packet rate can be easily changed. We will include packet size
as one of the parameters that can be changed in the future. Changing any of
these parameters and then carrying out the evaluation can be quite useful. For
example evaluation carried out for different power levels can give insight into
the design of power adaptation schemes such that latency can be minimized or
collisions can be minimized. Evaluation carried out for different packet rates
can help in deciding the suitability of motes for applications requiring either
less frequent or more frequent samples of some physical phenomenon. Evaluation
for different channels can give insight into which channels experience high
interference from other sources operating in the same band or can give insight
into the design of frequency agile schemes.
3 Experimental Setup
We
did preliminary evaluation of Micaz [1] and Telos [2] motes. We carried out the evaluation of Micaz motes on the mirage testbed
provided by Intel Research at
The
evaluation of Telos motes was carried out in the
corridor of
Fig. 2.
Experimental Setup
We
had set the packet rate ‘r’ to 1 packet every 10 milliseconds and did the
evaluation in channel 11 for 5 different power levels (0, -3, -7, -15 and -25 dBm) for both the setups. However, the number of packets
‘p’ sent by motes was different for the two setups. For the mirage setup we had
p = 200 and for the linear array setup we had p = 160 due to RAM space concerns
with the Telos motes. For the mirage setup we were
able to carry out more evaluations than the linear array as it didn’t need any
supervision from our end. For the linear array since we had placed them in the
corridor we had to be always there to make sure nobody disturbs them or takes
them away. So, for the mirage setup we conducted the same experiment (in
channel 11) again at a different time to see the temporal variations of
asymmetry. This is discussed in detail in the next section. We also did the
same experiment in channel 19 (for the mirage set up). However we have not
discussed the results for channel 19 here in this report for brevity. Please
note that in this report when we talk about Micaz
motes we imply the mirage setup and when we talk about the Telos
motes we imply the linear array setup.
Fig. 3.
Motes on the mirage testbed each grid is 3’x3’.
4 Preliminary Results and
Observations
In
this section we first define a few terms, briefly discuss prior work and then
discuss results and observations from our work.
4.1 Terminology
(i)
Packet Success
Rate: This is the ratio between the number of packets received successfully and
the total number of packets sent.
(ii)
Packet Loss Rate:
This is simply 1 – Packet Success Rate.
(iii)
Asymmetry: There
is said to be link asymmetry between two motes ‘m’ and ‘n’ if the packet loss
rate in one link (from m to n) is significantly different from the packet loss
rate in the other (from n to m).
(iv)
Signal Strength
Indicator (SSI) [3]: It is the 8 bit value (RSSI_VAL) given by CC2420 chip that
gives the average SSI value over 8 symbol periods. The usual range for SSI is
between 200 and 250 with 250 being the highest SSI. The actual power at the
CC2420 RF pin is calculated using P = RSSI_VAL + RSSI_OFFSET [dBm]. RSSI_OFFSET is usually -45.
(v)
Link Quality
Indicator (LQI) [3]: CC2420 gives a 7 bit correlation value based on first 8
symbols after Start Frame Delimiter (SFD) which is converted to an LQI value
that ranges usually between 50 and 110 with 110 indicating the maximum quality.
4.2 Prior Work
In
[13, 14 and 15] the evaluations were conducted for motes based on older radios
such as CC1000 and TR1000 RF chips. In [13] the authors carried out an
extensive evaluation of one of the older motes and presented analysis related
to different protocol layers (mainly physical (PHY) and media access control
(MAC)). In [14] the authors found some link asymmetries and speculated that the
asymmetries could be due to hardware miscalibration
in the radios. In [15] the authors confirmed that these asymmetries were indeed
due to hardware miscalibration and not due to the
channel. In [16] the authors presented a link estimation framework based on
experimentation with older motes. Our work presented here is different from
these prior works in that we are carrying out the evaluation of a new class of
motes that are based on CC2420 radios. In [17] the authors had some preliminary
results of Telos motes. They also indicated that LQI
would be a good indicator of the link quality. However this work did not
include an extensive evaluation of Telos motes. Our
work does include preliminary evaluation of Telos and
Micaz motes, and some of our findings are different
from what was shown in [17]. In [18] the authors have done extensive study of
how orientation of the motes affects SSI. They had done this evaluation using Micaz motes. However they did not measure parameters such
as LQI and packet loss rate. Their work was primarily focused on showing SSI
may not be a good indicator of distance.
4.3 Results and Observations
The
focus of our evaluation is to try to answer some of the following questions:
(i)
Do link
asymmetries exist in motes (Micaz and Telos) based on the latest radio?
(ii)
If link
asymmetries do exist then what might cause them: hardware miscalibration
or some other phenomenon?
(iii)
How does the
packet loss rate look like for various links, and how is it related to SSI and
LQI of that link?
(iv)
How does the
packet loss rate look like over distance?
4.3.1 Link Asymmetry
As
mentioned in the previous section we conducted the evaluation of Micaz for different power levels in channel 11. We
calculated packet loss rates for different node pairs and for links in both
directions for every node pair. We also calculated asymmetry by looking at the
difference in packet loss rates between two motes in both directions. We say
that link m->n has asymmetry if its packet loss
rate is greater than 40% that of link n->m. We plot such asymmetries as
shown in Fig. 4. The plots simply have motes numbered as N1 (indicating mote 1)
and different motes are plotted as if they form a circle. This is done just for
clarity purpose. This was not how the motes were actually placed on the testbed. The plots are also color coded to show the
directionality of the asymmetry and the level of asymmetry involved. If the asymmetry
is in link m->n then the line from Nm starts with blue and gradually becomes
magenta, green, yellow, orange, red, light brown, dark brown, etc. The level of
asymmetry is based on how many of these colors in that order we see in that
line for a link. For example, in Fig. 4a(i) let us look at the line from N17 to N4. It starts from
N17 (dark blue at N17) and ends at N4 with dark brown with all the colors in
between (magenta, green, yellow, orange, red and light brown) indicating very
high asymmetry in the link (N17 -> N4) where mote 17 (N17) was the
transmitter and mote 4 (N4) was the receiver. On the other hand we can look at
the line from N8 to N15. The line starts from N8 (dark blue at N8) and ends at
N15 with magenta (no other colors) indicating not as bad an asymmetry as it is
in N17->N4.
In
our preliminary plots we found that there were a few asymmetric links at
different power levels. We did the same experiments at same power levels with
same motes at a different time to see if the motes involved in the asymmetr
From
these plots we see that very few links (maximum 3 at power level -3dBm) have
asymmetry at the same power level in both the experiments. All other
asymmetries involve different links (and so different motes). This makes one
speculate that the asymmetry is not due to hardware miscalibration
as we do not see the same motes involved in asymmetries at different times
under same experimental settings. This then raises the question (ii) asked in
section 4.3: what causes these temporal asymmetries then? We speculate that
these temporal asymmetries are due to some local interference possibly from
WLAN (802.11b/g) network (which operates in the same 2.4 GHz ISM band as Micaz and Telos motes).
|
|
|
|
a(i) |
b(i) |
|
|
|
|
a(ii) |
b(ii) |
|
|
|
|
a(iii) |
b(iii) |
|
|
|
|
a(iv) |
b(iv) |
|
|
|
|
a(v) |
b(v) |
Fig. 4.
Asymmetry Plots for Micaz
This
result is quite different from what was observed in prior works for older
motes. This could be due to the fact that our motes are based on newer radios that
have more advanced architecture than the older radios.
As
for the Telos motes we found no asymmetries at high
power levels (0, -3 and -7 dBm) and so we haven’t
shown them. At lower power levels we saw only a few asymmetric links and
primarily from the last mote (N17). In Fig. 5 we have shown asymmetries at
power levels -15 dBm and -25 dBm
only.
|
|
|
Fig. 5.
Asymmetry Plots for Telos
4.3.2 Packet Loss Rate
Packet
loss rate is a very important metric for an application. It is also an important
metric for routing algorithms. It would be interesting to see if packet loss
rate has any correlation with SSI or LQI or both. We have plotted packet loss
rate against SSI and LQI in Fig. 6 for Micaz motes.
We have only shown the plots for one experiment as it looks very similar to the
plots of the same experiment carried out at a different time. The overall
observation is the same for both the experiments.
|
|
|
|
a(i) |
b(i) |
|
|
|
|
a(ii) |
b(ii) |
|
|
|
|
a(iii) |
b(iii) |
|
|
|
|
a(iv) |
b(iv) |
|
|
|
|
a(v) |
b(v) |
Fig. 6.
Packet Loss Rate vs. SSI and LQI for Micaz
Fig.
6 has plots of packet loss rate against SSI on the left and plots of packet loss
rate against LQI on the right for different power levels. For the packet loss
rate vs. SSI plots we take a link (m -> n), and plot its packet loss rate
and average SSI. We then plot the distribution (mean – standard deviation and
mean + standard deviation) of SSI corresponding to that link and for that
packet loss rate. We repeat the same for every link (note that link n -> m
is different from link m -> n). We do the same for the plots for packet loss
rate vs. LQI.
From
the plots it is clear that the SSI has very small variance (variance is simply
the square of standard deviation) compared to LQI for any link. This means that
the SSI we observe in a single packet is a good estimate of the average SSI
over many packets in that link. Moreover from all these plots it is clear that
generally for SSI values greater than 215 the packet loss rate less than 15%
(except for one or two glitches we see in Fig.6a(i)
which we will ignore for now). But for SSI values less than 215 we enter a grey
region where the packet loss rate varies rather unpredictably. On the other
hand if we look at the plots for LQI, the mean LQI (the dots in the middle of
the horizontal lines) looks promising as a good link indicator compared to SSI.
However due it its variance we should be careful not to use instantaneous LQI
values (calculated over a single packet) and rather use the mean LQI (over many
packets). To summarize we believe that the instantaneous SSI (over a single
packet) can be a very good indicator of link if its value is above 215. If it
is below 215 then we believe that the mean LQI would be a good link estimator.
As part of the future work we will come up with a more mathematical approach of
such a link estimator. We will also investigate how many packets should we use
to compute the mean LQI. This will also depend on how fast the channel is
varying. This observation is quite interesting as it is somewhat different from
the general belief that the LQI alone is a good estimator. Instead we have
shown that SSI is indeed a promising indicator if its value is above 215.
In
Fig. 7 we have plots of SSI and LQI for only three power levels (0, -7 and -25 dBm) for Micaz motes. It is clear
that at 0 and -7 dBm power levels we see higher
values of SSI and at lower power levels we only see lower values of SSI.
However it is interesting to note that we see high and low values of LQI at all
power levels. It is interesting to note that for SSI values approximately
greater than 215 almost all of them map to the same range of LQI values (between
100 and 110). Note that this is the value we picked as a cut off for using SSI
as a good indicator for packet loss rate. Below 215 a particular SSI value maps
to a wide range of LQIs (from 50 to 105). This might
suggest that LQI may be a good candidate as link quality indicator for SSI
below 215. This observation matches our observation based on packet loss rate
vs. SSI and LQI plots. However, we speculated that the average LQI (over many
packets) would be a better link quality indicator than instantaneous LQI (over
a single packet).
|
|
|
|
(i) |
(ii) |
|
|
|
|
(iii) |
|
Fig. 7.
Plots of LQI vs SSI for Micaz
|
|
|
|
a(i) |
b(i) |
|
|
|
|
a(ii) |
b(ii) |
|
|
|
|
a(iii) |
b(iii) |
|
|
|
|
a(iv) |
b(iv) |
|
|
|
|
a(v) |
b(v) |
Fig. 8.
Packet Loss Rate vs SSI and LQI for Telos
The
plots of packet loss rate against SSI and LQI for Telos
motes in a linear array are given in Fig. 8. Our observation for Micaz on mirage testbed also
applies to Telos motes in the linear array. It should
be noted that our experiments with Telos motes were
at small distances and so we see very high SSI and LQI values at high
transmission power levels (0, -3 and -7 dBm). This is
why the packet loss rate is also very low (about 2%) for these power levels.
However for lower power levels (-15 and -25 dBm) we
also see low SSI (about 200) and low LQI values (about 60). Our packet loss
rates are also as high as 90% for the lower power levels. But the overall
observation for Micaz still holds for the linear
array experiment with Telos motes: the SSI value does not have a high variance
for any link and if the SSI value is above 215 then we see packet loss rates
less than 15%. The mean LQI seems to be a good indicator of the packet loss
rate for cases where SSI is below 215.
In
Fig. 9 we have plots of LQI and SSI for all the power levels for the linear
array experiment with Telos motes. These plots show
that at high power levels we see only high SSI and LQI values. But at low power
levels we see high and low values of SSI and LQI. These plots are very similar
compared to the LQI vs. SSI plots for Micaz.
|
|
|
|
(i) |
(ii) |
|
|
|
|
(iii) |
(iv) |
|
|
|
|
(v) |
|
Fig. 9.
Plots of LQI vs SSI for Telos
motes
4.3.3 Spatial Characteristics
We
have the distance information only for the linear array experiment. It is also
very complex to find the spatial characteristics for the experiments conducted
on mirage with Micaz motes since these motes were at
locations that did not follow a specific pattern. However for the linear array
experiment with Telos motes we had arranged these
motes linearly and at regular distances of 2 feet. This is why in the following
discussions we present are only the spatial characteristics corresponding to
the linear array experiment with Telos motes.
|
|
|
|
a(i) |
b(i) |
|
|
|
|
c(i) |
|
|
|
|
|
a(ii) |
b(ii) |
|
|
|
|
c(ii) |
|
|
|
|
|
a(iii) |
b(iii) |
|
|
|
|
c(iii) |
|
|
|
|
|
a(iv) |
b(iv) |
|
|
|
|
c(iv) |
|
|
|
|
|
a(v) |
b(v) |
|
|
|
|
c(v) |
|
Fig. 10.
Spatial Characteristics of Telos Motes in Linear
Array
Fig.
10 shows the plots of packet loss rate, average SSI and average LQI over distance
for different transmission power levels. As for the average SSI plots for
different power levels each dot represents the average SSI for a node pair
separated by that distance. For example take a look at the plot Fig. 10b(ii) for distance of 2 feet. This will have the average
SSI values for node pairs (1, 2), (2, 1), (2, 3), (3, 2), (3, 4), (4, 3), (4,
5) and so on. We see that the average SSI is not the same for different node
pairs separated by the same distance. In fact the values sometimes differ by
about 25. We speculate that this variation is primarily due to multipath effects in the channel. Note that the multipath components involved in the node pair (1, 2) could
be different from the multipath components involved
in the node pair (2, 3) due the difference in locations of the sender and
receiver motes. This can also explain an interesting pattern of the mean of the
average SSI (the blue lines) that is seen for different power levels (except
for the power level at -25 dBm). The blue lines in these
plots represent the mean of all the average SSI values for a given distance.
When the experiment corresponding to the power level -25 dBm
was going on there was some human activity in the corridor although it was
going on at about
As
for the packet loss rate over distance it is almost negligible at power levels
0, -3 and -7 dBm. But at power levels -15 dBm and -25 dBm we see packet
loss rates of even 100% at farther distances. We would like to again go back to
our observation in section 4.3.2 where we mentioned that for SSI values greater
than 215 we have the packet loss rate to be less than 15% but for SSI below 215
we enter a grey zone. Now if we look at the average SSI vs. distance plots for
different power levels we notice that we see some SSI values that fall below
215 at power levels -15 dBm and -25 dBm. The node pairs corresponding to such SSI values could
have contributed to packet loss rates greater than 15% at those distances. For
example if we look at the average SSI vs. distance plot for power level -15 dBm (Fig. 10b(iv)) we see at least
two node pairs that had SSI values below 215 separated by a distance of 16
feet. If we look for the same distance and for the same power level at the
packet loss rate vs. distance plot (Fig. 10a(iv)) we
see two dots: one at about 55% and another at about 100%. This is just another
way to look at our observation presented in section 4.3.2. Otherwise we do not
see any pattern in the packet loss rate vs. distance plots across different
power levels. This is also true for the average LQI vs. distance plots. This
suggests that the packet loss rate need not necessarily be low at smaller
distances. It depends rather on specific location of the mote, interference
from other sources, multipath etc.
V
Conclusion
One
of the main contributions of this project is the evaluation tool itself which
has three parts: nesC code, java code and the script that
automates the evaluation. This tool has been built such that it is very easy to
use and flexible to add additional features in the future. It is autonomous and
requires no human handling once the evaluation starts. The experiments can be
repeated for different node pairs for different channels and for different
power levels without stopping. The statistics can be collected and stored in
proper data files as well. All this is coordinated by the script code we have
written as part of this evaluation tool.
We
carried out two sets of experiments. First set of evaluations had 30 Micaz motes on the mirage testbed
provided by Intel Research. These motes were arbitrarily chosen from a 123’x
48’ office indoor facility. Second set of evaluations had 17 Telos motes that were placed in a linear array in the
corridor of
Our
preliminary analysis indicates that the link asymmetry could be due to
interference with wireless LAN (WLAN) and not due to hardware miscalibration. This result is different from similar prior
work on motes that were based on older radios (TR1000 and CC1000) operating at
a different frequency band than Micaz and Telos. This possibly suggests that the radio architecture for
wireless sensors has improved since then.
The
results also show that SSI does not vary significantly from packet to packet
and that at very high SSI values (>215) the packet loss rate is
significantly low indicating a very good link. However at SSI values lower than
215, it enters a grey zone where the packet loss rate differs significantly for
the same SSI value for different links. As for the LQI it varies from packet to
packet for a given link. But when the average is computed over many packets it seems
to be a good indicator of the link. This observation can be very useful in
developing an efficient link estimation scheme as part of the future work. This
estimation technique can take SSI as an initial indicator since it does not
change significantly from packet to packet and then compute average LQI over
many packets to have an accurate estimate of the link quality. We will also
take in to account the nature of the wireless channel while developing this
link estimator. So far LQI alone has been believed to be a better indicator of
the link and our results show that this is not necessarily always true.
As
for the spatial characteristics of packet loss rate, average SSI and average
LQI, we have presented results only for Telos motes
as we had distance measurements only for the linear array experiment. The
average SSI vs. distance plots revealed that the average SSI values were
different for different node pairs separated by the same distance. We
speculated that this was caused due to multipath in
the channel. This speculation was also supported by a pattern in the mean of
average SSI plot which repeated with a downward shift as the transmission power
level was decreased. The packet loss rate and average LQI plotted over
different distances did not have any patterns. This possibly suggests that the
packet loss rate is not necessarily a simple linear function of the distance
instead it also depends on factors such as exact location, multipath
and interference.
Our
preliminary evaluation results have helped us make interesting observations
some of which are quite different from what has generally been believed
especially the observations on link asymmetry and on LQI being a good link
estimator. Our results also helped us define the next steps necessary to
transform our speculations in to claims. The following section lists some of
such next steps we have identified so far.
VI Future Work
As
part of the future work we would like to carry out the evaluation at packet
rates other than 1 every 10 milliseconds. More specifically characterize the
behavior of motes when the packet rate is very low as would be in many WSN
applications. We would like to enhance our evaluation tool to include packet
size as one of the available parameters to change. In the future we will also
let the receiver motes store the statistics of the received packets in
non-volatile storage. This will allow us to conduct evaluations for longer
durations and even if the battery runs out the statistics will be safe. We will
also include another feature in the evaluation tool where the commands can be
sent over the radio instead of over USB or Ethernet. This will eliminate the
need for using USB connectors to every mote.
We
will also carry out evaluations with more motes at larger distances and in
different environments (parking lots, parks, etc). We would also like to
develop a good link estimator based on the observations made in this report
which will serve as a good metric for routing algorithms (and other algorithms)
in WSNs. While coming up with this model for a good
link estimator we will also consider the nature of the wireless channel: how
fast does the channel vary over many packets as this will give us a bound on
how many packets we can use to compute the average LQI for a link. We would
also like to repeat experiments conducted in this project in frequency bands
(channels) that do not overlap with WLAN channels. This will indicate if
factors other than interference contribute to link asymmetry. In the future we
would also like to include other scenarios such as multihop
communication and multiple simultaneous transmissions. We would also like to
look at channel models so that different environments can be characterized.
This will help in the efficient deployment of wireless sensors.
Bibliography
[1] Crossbow Technologies:
http://xbow.com.
[2] Moteiv Corporation:
http://www.moteiv.com.
[3] Chipcon: www.chipcon.com.
[4]
http://mirage.berkeley.intel-research.net.
[5] S. Cui, “Cross-layer Optimization in Energy-constrained Networks”,
[6] R. Madan,
[7] K.
Srinivasan, M. Ndoh, H. Nie,
H. Xia, K. Kaluri, D. Ingraham,
“Wireless Technologies for Condition-Based Maintenance (CBM) in Petroleum
Plants,” IEEE International Conference on Distributed Computing in Sensor
Systems 2005 (DCOSS 2005), Marina del Rey, CA,
Jun. 2005.
[8] D. Ingraham, R. Beresford, M. Ndoh,
K. Srinivasan, K. Kaluri, “Wireless Sensors: Oyster Habitat Monitoring in the
Bras d'Or Lakes,” IEEE International Conference on
Distributed Computing in Sensor Systems 2005 (DCOSS 2005), Marina del Rey, CA, Jun. 2005.
[9] IEEE standard for
information technology - telecommunications and information exchange between
systems - local and metropolitan area networks specific requirements part 15.4:
wireless medium access control (MAC) and physical layer (PHY) specifications
for low-rate wireless personal area networks (LR-WPANs),
IEEE Std 802.15.4-2003, 2003 Page(s):0_1 – 670.
[10] D. Gay, P.
Levis, R. Behren, M. Welsh, E. Brewer, and D. Culler,
“The nesC
Language: A Holistic Approach to Networked Embedded Systems,” Proc.
of Programming Language Design and Implementation (PLDI) 2003, Jun. 2003.
[11] P. Levis,
[12] D. Gay, P.
Levis, and D. Culler, “Software Design Patterns for TinyOS,”
Proc. of the ACM SIGPLAN/SIGBED 2005 Conference on Languages, Compilers, and
Tools for Embedded Systems (LCTES '05), 2005.
[13] Y. J.
Zhao, and R. Govindan, “Understanding packet delivery performance in dense wireless sensor
networks,” Proc. of ACM Sensys 2003,
[14] D. Ganesan,
B. Krishnamachari, A. Woo, D. Culler, D. Estrin and S. Wicker, “Complex Behavior at Scale: An
Experimental Study of Low-Power Wireless Sensor Networks,” Technical Report
UCLA/CSD-TR 02-0013, 2002.
[15] A. Cerpa,
[16] A. Woo,
and D. Culler, “Evaluation of Efficient Link Reliability Estimators for
Low-Power Wireless Networks,” UCB Technical Report, Nov. 2003.
[17] J. Polastre, R. Szewczyk, and D.
Culler, “Telos:
Enabling Ultra-Low Power Wireless Research,” IPSN/SPOTS 2005, 2005.
[18] D. Lymberopoulos,
Q. Lindsey and A. Savvides, “An Empirical Analysis of Radio Signal Strength Variability in IEEE
802.15.4 Networks using Monopole Antennas,” ENALAB Technical Report 050501.