Robust Sensing and Fusion
over Disadvantaged Networks
Sensors/estimators
can generate state estimates of dynamic targets and send them to the fusion
center. The state estimate from an individual sensor is prone to error and can
be biased due to the limitation of the sensing device and the
noise/interference incurred to the sensing signals. In addition, the estimation errors from
multiple sensors could be correlated. The fusion center combines individual
estimates to generate combined estimates in order to achieve performance
superior to individual sensor estimates. During the fusion process, the state estimates
within a window W from a sensor will be considered, and the fusion will be
carried over M sensors. A larger W could potentially improve fusion quality, at
the cost of potentially a higher processing complexity and a larger initial
detection delay.
In some
application scenarios, the sensors could be far away from each other and far
from the fusion center. The communications have to be carried over long-haul
links such as satellite, fiber or the mix of the two. The long-haul communications could lead to significant
transmission delay (in the order of W), delay jitter, and loss. This is especially the case with satellite
links. Although fiber transmission has much lower loss and error rate per link,
a transmission over multiple fiber links through a wide area network could also
experience packet droppings. TCP has been commonly used in network
transmissions to ensure reliable end-to-end transmissions. However, the
retransmission through TCP could significantly increase delay and jitter, and
the problem will be even more severe over satellite links where link loss is a
norm. As a result of TCP congestion control procedures, it may even throttle
the transmission. These transmission problems would significantly impact the
fusion quality.
The objective
of this work is to investigate techniques that could provide timely and
effective state estimates in such a wide-area transmission scenario. This
includes the investigation of network techniques that improve transmission quality
and robust fusion methods. To focus on the problem due to long-haul
transmission, the investigation will start with single target or fixed number
of targets. Some potential directions
that may be explored are as follows.