Professional Documents
Culture Documents
The MIT Faculty has made this article openly available. Please share
how this access benefits you. Your story matters.
AbstractWe examine the problem of collecting data from an underwater vehicle (AUV) equipped with an acoustic modem
underwater sensor network using an autonomous underwater to gather data from the sensors [4]. In the applications of
vehicle (AUV). The sensors in the network are equipped with interest, sensors are deployed for long-term monitoring and
acoustic modems that provide noisy, range-limited communica-
tion to the AUV. One challenge in this scenario is to plan paths are fixed to the ocean floor.1 Hence, we have a robotic sensor
that maximize the information collected and minimize travel network that includes stationary measurement nodes and an
time. While executing a path, the AUV can improve performance AUV that gathers data from these nodes. The problem now
by communicating with multiple nodes in the network at once. becomes one of planning the AUVs path to minimize its travel
Such multi-node communication requires a scheduling protocol time and maximize information gathered. We will refer to this
that is robust to channel variations and interference. To solve
this problem, we develop and test a multiple access control as the Communication-Constrained Data Collection Problem
protocol for the underwater data collection scenario. We perform (CC-DCP).
simulated experiments that utilize a realistic model of acoustic In our prior work, we showed that the CC-DCP is closely re-
communication taken from experimental test data. These simula- lated to the classical Traveling Salesperson Problem (TSP) [5].
tions demonstrate that properly designed scheduling protocols are The key difference is that information is gathered from sensors
essential for choosing the appropriate path planning algorithms
for data collection. through a noisy channel, whose reliability decreases with
distance and can be modeled probabilistically. We previously
Index Termspath planning algorithms, acoustic communica- showed that the CC-DCP can be modeled as a TSP with
tion, underwater robotics, sensor networks
probabilistic neighborhoods, and we provided algorithms that
solve the problem approximately [6].
I. I NTRODUCTION Related problems have been studied in the context of robotic
gain [dB]
SeaBED-class AUV [13] operated by the NOAA Northwest
Fisheries Science Center, is equipped with a WHOI Micro-
15
Modem and 12.5 kHz ITC-3013 hemispherical transducer
for acoustic communications [14]. In September of 2010,
Lucille assisted in mapping the submerged portion of the 10
San Andreas Fault off Northern California, at approximately
39 500 N, 124 W. During this survey, the AUVs onboard
5
networking stack, capable of handling data fragmentation and 100 200 300 400 500 600 700 800 900 1000
distance [m]
image compression [15], transmitted one three-second packet
every five seconds. These packets were encoded using both Fig. 1. Gain (normalized) vs. transmission distance. Dots show measured
Frequency-Hopping Frequency Shift Keying (FH-FSK) and values; solid curve shows an estimated trend (a first-order logarithmic-scale
Phase Shift Keying (PSK), and transmitted using 4-5 kHz polynomial fit to the ensemble mean at each distance yields k0 =1.9).
bandwidth around a center frequency of 10 kHz.
Throughout the course of the dive, the vehicle maintained
a constant altitude above the seafloor of 3 m, at a depth of (ii) the random component obeys a Gaussian distribution, y
approximately 130 m. The surface ship, the R/V Pacific Storm, N (0, 2 ).
varied in slant range from 200 m to 1 km from the vehicle. The Figure 1 summarizes the recorded values (from the deploy-
surface ship remained underway with the hydraulics running ment described above) of the gain as a function of distance.
during this experiment, resulting in significant noise being The solid curve represents the log-distance model (3), whose
generated across all frequencies, including those used for parameters g0 and k0 were obtained by first-order polynomial
communication. These conditions are typically experienced by fitting.3 We emphasize again that the model parameters will
AUVs operating from near-shore vessels on the continental in general depend on the operational conditions, i.e. that the
shelf. values indicated in the figure are representative of the 8-12
kHz acoustic band and transmission distances on the order of
several hundreds of meters.
B. Acoustic Channel Model Shown in Figure 2 is the histogram of the random compo-
To specify a propagation model, we represent the gain as nent y = gg. This figure motivates our second conjecture, i.e.
the Gaussian model for y. The variance 2 is calculated from
g(d, t) = g(d) + y(t), (2) the data at hand. We note that its value appears to be invariant
where g(d) is the mean value of the gain at a distance for the range of distances considered, although greater distance
d,2 and y(t) is a random process. In this model, the gain spans could require sectioning. We also note that the variance
g(d) represents the expected communication quality C(xv , xn ) will depend on the bandwidth, decreasing as the bandwidth
when d = D(xv , xn ) (see Section II). We do not consider increases. Similar conclusions have been found using different
changes in water pressure with depth, which would affect the data sets [16].
propagation speed. Such changes could be accounted for in
the signal processing layer by inserting guard time slots to C. Packet Error Modeling
account for slight variation in propagation speed.
We utilize an underwater acoustic noise model developed
We now proceed to establish two models based on our
in prior work [2], [11]. This model accounts for noise factors
experimental data: one that relates the mean value g to the
in the environment, such as wind and shipping activity, as
distance d, and another that specifies the probability distribu-
well as thermal noise and turbulence. We also assume a block
tion function (pdf) of the random component y. We utilize
log-normal fading model for the received signal-to-noise ratio
a channel model similar to prior work [3] that identifies
(SNR) based on section IV-B. Let PS be the probability of
log-distance parameters. We also add an additional random
symbol error averaged over the SNRs. For a packet with Q
component, and we specify the overall power loss, including
symbols encoded with a code of rate r, the average packet-
all frequencies and all propagation paths. These models will be
error-rate is given by
valid for the chosen operating conditions (frequency band and
transmission distances). Specifically, we make the following PD = 1 (1 PS )rQ . (4)
conjectures:
(i) the mean value obeys a log-distance model There is no known simple approximation to PS when SNR is
log-normally distributed, and we employ Monte-Carlo meth-
g(d) = g0 k0 10 log d (3) ods to perform simulations. In this model, the packet success
2 The distance is varying with time, i.e. d = d(t). 3 Logarithms are taken with base 10.
0.16
executed in the data transfer phase would also be upper
experiment bounded.
0.14 Gaussian
1.9
7500
1.8
1.6
6500
1.5
1.4 6000
Contours (p = 0.1) Contours (p = 0.1)
1.3 Contours (p = 0.3) Contours (p = 0.3)
Contours (p = 0.5) 5500 Contours (p = 0.5)
1.2 Contours (p = 0.7) Contours (p = 0.7)
Contours (p = 0.9) Contours (p = 0.9)
1.1 5000
0 5 10 15 20 0 5 10 15 20
ARQs ARQs
4
x 10
3 2
2.9 1.9
Gain/Cost Ratio (packets/sec)
2.8 1.8
2.6 1.6
2.5 1.5
2.4 1.4
Contours (p = 0.1) Contours (p = 0.1)
2.3 Contours (p = 0.3) 1.3 Contours (p = 0.3)
Contours (p = 0.5) Contours (p = 0.5)
2.2 Contours (p = 0.7) 1.2 Contours (p = 0.7)
Contours (p = 0.9) Contours (p = 0.9)
2.1 1.1
0 5 10 15 20 5000 5500 6000 6500 7000 7500 8000
ARQs Mission Time (sec)
Fig. 3. Simulations of an AUV collecting data from an underwater sensor network. Averages are over 100 random deployments in a 1 km 1 km area
with 100 nodes. The AUV executes a data collection tour found using a TSP with neighborhoods. The simulations are performed with varying neighborhood
size and number of ARQs.
we are in the process of deriving equations to calculate infor- [5] D. L. Applegate, R. E. Bixby, V. Chvatal, and W. J. Cook, The Traveling
mation gain when correlations exist between sensors, which Salesman Problem: A Computational Study. Princeton Univ. Press,
2006.
causes the the value of information to become subadditive. [6] G. Hollinger, U. Mitra, and G. Sukhatme, Mobile underwater data
collection using acoustic communication, in Proc. IEEE/RSJ Int. Conf.
ACKNOWLEDGMENT Intelligent Robots and Systems, 2011.
[7] D. Bhadauria and V. Isler, Data gathering tours for mobile robots, in
The authors gratefully acknowledge Jonathan Binney, IEEE Int. Conf. Intelligent Robots and Systems, 2009, pp. 38683873.
[8] A. Krause and C. Guestrin, Near-optimal nonmyopic value of informa-
Arvind Pereira, Hordur Heidarsson, and Srinivas Yerramalli tion in graphical models, in Proc. Uncertainty in Artificial Intelligence,
at the University of Southern California for their insightful 2005.
comments. Thanks also to Chris Goldfinger of Oregon State [9] A. Krause, C. Guestrin, A. Gupta, and J. Kleinberg, Near-optimal
sensor placements: Maximizing information while minimizing commu-
University, the captain and crew of the R/V Pacific Storm, and nication cost, in Proc. Information Processing in Sensor Networks,
Elizabeth Clarke of the NOAA Northwest Fisheries Science 2006, pp. 210.
Center for their support of this work. [10] A. Dumitrescu and J. Mitchell, Approximation algorithms for TSP with
neighborhoods in the plane, J. Algorithms, vol. 48, no. 1, pp. 135159,
2003.
R EFERENCES [11] L. Berkhovskikh and Y. Lysanov, Fundamentals of Ocean Acoustics.
Springer, 1982.
[1] R. N. Smith, Y. Chao, P. P. Li, D. A. Caron, B. H. Jones, and G. S.
[12] M.Porter, Bellhop code, available online at http://oalib.hlsresearch.
Sukhatme, Planning and implementing trajectories for autonomous un-
com/Rays/index.html.
derwater vehicles to track evolving ocean processes based on predictions
[13] H. Singh, A. Can, R. Eustice, S. Lerner, N. McPhee, O. Pizarro,
from a regional ocean model, Int. J. Robotics Research, vol. 29, no. 12,
and C. Roman, Seabed AUV offers new platform for high-resolution
pp. 14751497, 2010.
imaging, EOS Trans. AGU, vol. 85, no. 31, 2004.
[2] G. Hollinger, S. Yerramalli, S. Singh, U. Mitra, and G. S. Sukhatme,
[14] L. Freitag, M. Grund, S. Singh, J. Partan, P. Koski, and K. Ball,
Distributed coordination and data fusion for underwater search, in
The WHOI micro-modem: An acoustic communcations and navigation
Proc. IEEE Conf. Robotics and Automation, 2011, pp. 349355.
system for multiple platforms, in Proc. IEEE Oceans Conf., 2005, pp.
[3] M. Stojanovic, On the relationship between capacity and distance in
10861092.
an underwater acoustic communication channel, ACM SIGMOBILE
[15] C. Murphy and H. Singh, Wavelet compression with set partitioning
Mobile Computing and Communications Review, vol. 11, no. 4, pp. 34
for low bandwidth telemetry from AUVs, in Proc. ACM Int. Wkshp.
43, 2007.
UnderWater Networks, 2010.
[4] I. Vasilescu, K. Kotay, D. Rus, M. Dunbabin, and P. Corke, Data
[16] P. Qarabaqi and M. Stojanovic, Adaptive power control for underwater
collection, storage, and retrieval with an underwater sensor network, in
acoustic channels, in Proc. IEEE Oceans Conf., 2011.
Proc. Int. Conf. Embedded Networked Sensor Systems, 2005, pp. 154
165.