| Summary: | 碩士 === 國立清華大學 === 資訊工程學系 === 102 === Existing context-aware mobile applications directly control sensors in the
mobile devices in an uncoordinated and non-optimized manner, which leads
to redundant sensor activations and energy waste. Optimal and coordinated
sensor usage dictates a comprehensive mobile middleware solution with sensor
scheduling on single device to bring together the information from all
applications/sensors and intelligently select the best set of sensors to activate.
While the widespread use of smartphones, we cooperate the sensors on multiple
smartphones and infrastructure sensors to build a novel crowdsensing
system.
In Chap. 3, we design, implement, and evaluate a novel green sensor management
middleware for single device that rigorously makes tradeoffs between
energy consumption of sensors and accuracy of inferred contexts. The
problem is formulated rigorously as mathematical optimization problems that
(i) minimize the total energy consumption while achieving the required accuracy
and (ii) maximize the overall accuracy under a given energy budget. Two
optimal algorithms for these two optimization problems are proposed, which
provide the performance bounds. As they may lead to prohibitively long running
time, two efficient heuristic algorithms are then presented, which run in
real-time. Extensive trace-driven simulations are conducted using traces from
real Android users to evaluate the performance of the proposed middleware
and algorithms. The simulation results indicate that the heuristic algorithms:
(i) always terminate in real-time, (ii) result in small optimization gap of up
to ∼ 2%, and (iii) lead to better performance for larger problems. We also
implement and evaluate the proposed middleware and algorithms on real Android
smartphones, showing their practicality and efficiency.
For the extension, we consider the sensor scheduling on multiple smartphones
and infrastructure sensors in Chap. 4. We apply the extensive consideration
to crowdsensing system. We present a Smartphone Augmented
Infrastructure Sensing (SAIS) system that offers better situation awareness to
officials and civilians for minimizing the amount of generated carbon dioxide.
The SAIS system minimizes the carbon footprint by solving the task
assignment problem. We mathematically formulate the problems and optimally
solve it using optimization problem solvers, and we also proposed an
efficient task assignment algorithm (ETA) for lower running time. Our tracedriven
simulations show the results of our efficient algorithm: (i) saves up to
364 times in carbon footprint, (ii) outperforms by up to 8 times in responding
time, and (iii) achieves a small optimization gap of ∼ 2%.
|
|---|
