| |
Researchers: Mohammad A. Kanso, M. Eng Student, Prof. Michael Rabbat
Problem:
Security and safety personnel need intelligent infrastructure to monitor environments
for detecting and locating assets. The ability to detect the location of these objects in a timely and efficient manner allows quick response from
security personnel directing evacuation. RF tomography is an approach that uses wireless sensors, as in Fig 1 below, to locate assets in an environment.
It essentially infers characteristics about the medium by analyzing wireless RF signals that traverse that medium.
In previous works, sensor nodes were used in a centralized fashion to perform RSS measurements throughout the environment to discover its contents.
Since sensor networks have limited battery support, energy efficiency is important. Moreover, decentralized schemes provide robustness to failure of the network's fusion center.
Figure 1: Wireless links among nodes in a sensor network used to monitor the environment
Approach:
In this project, we introduce Compressed RF Tomography, which combines advantages of compressed sensing theory and RF tomography.
Compressed sensing relies on minimizing L1 norms to recover sparse data from a set of linear measurements. If we assume that few changes occur in the environment in a certain period, then compressed sensing may efficienctly discover those changes using only a few number of RSS measurements.
Few changes reflect the fact that there are few movements in the monitored environment. Since less RSS measurements are needed, less sensors are required to be turned on, hence saving battery power.
Moreover, we introduce decentralized mechanisms to perform monitoring cooperatively among the sensor nodes. An optimization problem is distributed among
the sensor nodes which iterate over the solution until convergence. Techniques such as incremental subgradient and POCS optimization methods were employed.
Computer and real simulation results illustrate the efficiecy of our approach compared to previous simpler approaches.
Simulations:
Simulations were performed for both centralized and decentralized scenarios. In a centralized approach, all data is continuously sent over to the fusion center.
The fusion center performs compressed sensing reconstruction techniques to locate obstructions within the environment. Fig 2 below shows results from real sensor data using our compressed
approach and the original least squares minimization approach. Due to the sparsity assumption, monitoring can be done more accurately.
|
|
| (a)
|
(b) |
Fig 2: Comparison of results obtained from real simulations. Compressed RF Tomography results in (a) clearly porduce more accurate results than simpler L2 approaches in (b)
Incremental subgradient and Projection Onto Convex Sets (POCS) methods were employed to perform decentralized processing of RSS measurements. This can be done in determinstic and randomized settings.
In a deterministic setting, sensor nodes perform updates in a cycle, requiring overhead of setting up this cycle. In a randomized approach,
a node is randomly chosen for an update. This may be implemented by setting up timers with random timeouts on each sensor. The deterministic approach performs better
as Fig 3 below demonstrates.
|
|
(a)
|
|
| (b) |
Fig 3: Decay of cost in (a) and MSE in (b) versus the number of iterations in decentralized RF tomography
Publications:
M. A. Kanso and M. G. Rabbat, "Compressed RF Tomography for Wireless Sensor Networks: Centralized and Decentralized Approaches", to appear in the Proceedings of the 5th IEEE International Conference on Distributed Computing in Sensor Systems (DCOSS), Marina Del Rey, CA, June 2009.
M. A. Kanso and M. G. Rabbat, "Efficient Detection and Localization of Assets in Emergency Situations," in the Proceedings of the Third International Symposium on Medical Information & Communication Technology (ISMICT), Montreal, QC, Canada, Feb. 2009.
|
Return to RF Tomography Projects