Title: Network architecture and processing for positioning and distributed estimation
1Network architecture and processing for
positioning and distributed estimation
- Stefano Tennina, Fortunato Santucci
HYCON WP4d meeting KTH - March 14th, 2008
2Outline
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Positioning service
- Voice over WSN
- Long term platform
- Remarks and further issues
3Background and motivation
- Existing mining ventilation controls are poor
- Continuous monitoring of air quality absent
- Air pumped in the rooms manually controlled with
huge safety margins - Communication capability simply by voice over
walkie talkie.
- Automatic control solutions for safety and energy
optimization. - Low Latency and High Reliability of measurements
4Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Remarks and further issues
5Communication architectures
- Networked sensors in the vertical ventilation
shaft, in access tunnels and extraction rooms - Sensor nodes in extraction rooms are wireless due
to blasting activities - Basic configuration
- Uniform radio network where nodes in tunnels and
extraction room are wireless (e.g. IEEE 802.15.4)
6Communication architectures
- Hybrid wired-wireless configuration
- Presence of cabling leads to use of power line
communication (PLC) devices or simply 802.3 - Gateway GW1 PLC/IEEE 802.15.4 at tunnel entrance
- Nodes N1 monitor the shaft and are powered by the
line - Static ad hoc network and multi-hop routing
- Nodes N2 in tunnel and extraction rooms powered
by batteries - Mainly static network and multi-hop routing
- Low power strategies, especially for nodes close
to GW.
7Communication architectures
- With some mobile nodes
-
- Nodes N3 are mobile (on trucks or hand-held)
power supply is not an issue - Cluster tree topology with dynamic backbone
- Nodes N3 act as cluster heads to save nodes N2
energy - Nodes N2 dynamically associate with Nodes N3 or
acts as cluster heads forming a static backbone.
8Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Remarks and further issues
9Latency and Energy models
- Three classical methodologies of latency analysis
are currently under investigation - Analytical high level end-to-end delay by
resorting to an abstraction of the communication
protocol - Analytical Network Calculus which iteratively
sums per-hop delays () - Simulations with discrete event network
simulators for validation and synthesis.
() Source A. Koubaa, M. Alves, E. Tovar,
"Modelling and Worst-Case Dimensioning of
Cluster-Tree Wireless Sensor Networks",
technical report, Polytechnic Institute of Porto
(ISEP-IPP), http//www.hurray.isep.ipp.pt
10Latency and Energy models
- Energy constraints on battery powered sensor
nodes involves proper communication protocols and
algorithm design at all layers - Distributed Source Coding techniques for
compressing data, exploiting their redundancy - Energy-Aware Routing protocols for taking into
account energy costs in the path selection
metric - Joint Routing MAC designs like the SERAN
clustered protocol, which supports - Data aggregation mechanisms
- Robustness against node failures and clock drifts
11Latency and Energy models
Sleep 0.020mAh
Simulated energy depletion over a tunnel with an
EAR algorithm
Idle 0.426mAh
RX / Carrier Sense 19.7mAh
TX 17.4mAh
CC2420 radio states on CrossBows MICAz node
12Network Lifetime example
Once assumed a measurements reporting period
(e.g. 60s) and a battery discharge model
Network Lifetime
Per node estimated Battery Consumptions
13Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Remarks and further issues
14Distributed estimation
- Gas concentration dynamics
- incoming airflow from tube
- gas emission-consumption of trucks
- outgoing airflow through room exit.
- Reliability of measurements implies the need for
a method to reconstruct an expected profile from
distributed sensors
15Distributed estimation
- Assumptions
- Gas stratification in a room is a function of the
height of the room - Gas concentration is constant over the plane at
each fixed height - No fixed sensor in the room due to blasting
activities. - Goal
- Reconstruct the profile with a grid of H x W
fixed points at each room entrance, minimizing
the number of points of H and W. - Note
- The profile is obtained by averaging the W noisy
sensor measurements for each point of H.
16Distributed estimation
- A practical example
- Measure a known profile (of light) by means of a
very dense network of MICAz nodes - Downsample the measurement points while taking
into account the performance of reconstruction.
17Distributed estimation
18Distributed estimation
- Parameters
- h, w integer decimation factors
- D interpolation factor
- M (h,w,i) linear interpolation of the measured
under-sampled curve, i1,,N - Mref (i) the reference curve profile.
- Performance indexes
- Goodness Of Fit (GOF)
- Error Root Mean Square (RMS)
- Area Ratio
- Normalized Error.
19Distributed estimation
20Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Positioning service
- Voice over WSN
- Long term platform
- Remarks and further issues
21Positioning in mines
Gold mine "CANMET" (Canadian Center for Minerals
and Energy Technology) Val dOr, 700km north of
Montreal (Quebec), Canada.
Source C. Nerguizian, C. Despins and S. Affès,
Geolocation in Mines With an Impulse Response
Fingerprinting Technique and Neural Networks",
IEEE Transactions on Wireless Communications, VOL
. 5, NO. 3, MARCH 2006
Source A. Chehri, P. Fortier, P.-M. Tardif,
Frequency Domain Analysis of UWB Channel
Propagation in Underground Mines, Vehicular
Technology Conference, 2006. VTC-2006 Fall. 2006
IEEE 64th Sept. 2006 Page(s)1 - 5
22Positioning
- In a 2D scenario a node can estimate its position
if its distance from at least three reference
nodes is available, along with the position of
these nodes in a Common Reference System
R2
BS
BS
MS
R1
R3
BS
23Position RefinementAlgorithm
Steepest descent algorithm
Error function
Update algorithm for a node i Require
, where al AccuracyLevel Ensure a new
estimation of 1 for each node i do 2 3
4 , rp RangingPenalty 5 Update
estimation and accuracy level
24MICAz RSSI Calibration
- Received Signal Strength Indicator (RSSI) imposes
a characterization of the propagation
environment, by means of the estimation of two
parameters - Parameter A defined as the absolute value of the
average power in dBm received at a close-in
reference distance of one meter from the
transmitter, assuming an omni-directional
radiation pattern - Parameter n defined as the path loss exponent
that describes the rate at which the signal power
decays with increasing distance from the
transmitter. This decay is proportional to d(n),
being d the distance between transmitter and
receiver.
25MICAz RSSI Calibration
- The parameters A and n can be estimated
empirically by collecting RSSI data (thus path
loss data) for which the distances between the
transmitting and receiving devices are known. - A least-squares best-fit line is used to glean
the specific values of A and n for the
environment in which the data were measured - A is the y-intercept of the line, and
- n is the slope of that line.
For that environment - A 59.6dBm - n 1.84
26Geometric Dilution Of Precision
- Mean and Standard Deviation of positioning error
decrease as the angle under which a node sees
references increases - This results to a constraint on anchors positions
for obtaining an acceptable level of the
positioning error.
27Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Positioning service
- Voice over WSN
- Long term platform
- Remarks and further issues
28Voice over WSN
- Advantages
- More pervasiveness
- Reduced power consumption
- Capability to perform local signal processing
- Improve communication quality exploiting
distributed source coding. - Disadvantage
- Limited bandwidth (802.15.4 250Kbps).
- Cooperation between nodes distributed in the
environment.
29Voice over WSN
FireFly node
A network of 42 nodes in the National Institute
for Occupational Safety and Health (NIOSH)
experimental coal mine in Pennsylvania Appropriat
e choices on time synchronization, MAC and
Routing allow to carry a streaming audio signal
from a mobile node to an external gateway over a
multihop path
Source R. Mangharam, A. Rowe and R.
Rajkumar, "Voice over Sensor Networks, 27th IEEE
Real-Time Systems Symposium (RTSS), Rio de
Janeiro, Brazil, December 2006.
30Voice over WSN
- Our prototype
- A MICAz node samples an audio signal with 8,6KHz
frequency - It applies an Adaptive Differential Pulse Code
Modulation (ADPCM) compression and send packets
over the radio - A Stargate gateway receives packets and decodes
audio signal - Audio can be reproduced on a PC wired (via e.g.
LAN) or wireless (via e.g. WLAN) linked to
gateway.
31Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Positioning service
- Voice over WSN
- Long term platform
- Remarks and further issues
32Advanced services and Long term platform
Ventilation Control
Security
Video
Phone Calls
Applications
Safety
Middleware of Services on the mobile gateways
Data Fusion
Protocol Adaptation
Positioning
IEEE 802.15.4
IEEE 802.11
IEEE 802.3
Power Line
Heterogeneous Networks
33Reminder
- Background and motivation
- Communication architectures
- Latency and energy models
- Distributed estimation
- Advanced Services
- Remarks and further issues
34Remarks and further issues
- Need for characterization of propagation channel
a ray tracing simulation tool for tunnels is
available at UAQ, along with a kit for UWB
channel measurements and a coverage planning tool - Mobile gateways are under development at UAQ by
resorting to Software Defined Radio (SDR)
technologies - A biometric badge for physical and logical access
control has been developed at UAQ - A hybrid (symmetric/asymmetric) key management
scheme has been proposed and is under development
in cooperation with UCB, along with an Intrusion
Detection System (IDS) that can be deployed also
over tiny nodes.