Biologically-inspired Autonomic Wireless Sensor Networks. Haoliang Wang 12/07/2015

Transcription

1 Biologically-inspired Autonomic Wireless Sensor Networks Haoliang Wang 12/07/2015

2 Wireless Sensor Networks A collection of tiny and relatively cheap sensor nodes Low cost for large scale deployment Limited computation and communication capability Strict energy constraints Unattended operations and self-star properties are desirable Self-configuration to reduce management complexity for large scale network Self-optimization to deal with the dynamic environment Self-healing Maintenance are too costly or even impossible Self-protection from compromising by adversaries

3 Challenges for Autonomy Centralized controller approach Many solutions are for resource-rich server clusters Extremely limited computation resources and energy on a sensor node Communication is expensive Collecting and disseminating information from/to the entire network is inefficient, especially for large scale network Single point failure We need decentralized approach to achieve autonomic operations

4 Seek inspirations from nature Similar problems may have been dealt with by nature for quite some time Radar is inspired by how bats locate their pray using ultrasound and echolocation Natural / biological systems are composed of entities making its own decision Collectively, those living organisms are complex adaptive systems and able to show many desirable self-star properties If we can capture the governing dynamics behind its self-organization, communication and coordination, we may mimic its behavior in the design of wireless sensor networks and achieve autonomy

5 Bio-inspired Examples Swarm Intelligence Firefly Synchronization Activator-inhibitor Systems Epidemic Spreading Evolutionary Computing Artificial Immune Systems and many more

6 Swarm Intelligence Social insects interact locally with each other according to a set of simple rules They exhibit collective intelligence at the system level to solve complex problems Examples Ant foraging for food Fish schooling Bird flocking Bee dancing

7 Swarm Intelligence (cont. ) Food foraging in an ant colony Ants first randomly wander around for food On their trail, ants leave pheromones, which will evaporate and disappear over time Once the food is found, the ant returns to the hive using the same path, strengthening the pheromones Subsequent ants decide whether to follow based on the strength of the pheromone Finally they will converge to a near-optimal path

8 Swarm Intelligence (cont. ) Application in wireless sensor networks BiO4SeL: A Bio-Inspired routing algorithm for Sensor Network Lifetime Optimization by Ribeiro et al. Based on the ant colony optimization algorithm, they developed BiO4SeL to optimize connectivity, coverage and lifetime in WSNs through dynamic routing Each message sent to the sink is an ant searching for food The ant will probabilistically choose the next hop based on the intensity of the pheromone of each neighbors The pheromone evaporates based on the energy level of the node

9 Firefly Harmonic Flashing An individual firefly may solely decide its flashing rate When met with a group of fireflies, an individual firefly will see the flashing of others and adjust its own own flashing rate to harmonize with the group Example of distributed synchronization without centralized control

10 Firefly Synchronization (cont. ) Mathematical models are designed to capture the synchronization behavior as a set of pulse-coupled oscillators Each oscillator has a local variable with its value varying from 0 to 1. The oscillator fires when it reaches 1. Then the variable jumps back to 0 When an oscillator fires, its pulse will pull the variable of its neighbors by a fixed amount, or bring them to firing, whichever is less Finally, all the oscillators will fire synchronously for almost any initial conditions

11 Firefly Synchronization (cont. ) Application in wireless sensor networks Firefly-inspired sensor network synchronicity with realistic radio effects by Werner-Allen et. al. Centralized synchronization may not be reliable when nodes start to fail They designed a distributed synchronization protocol based on firefly flashing Each node periodically broadcasts a sync message and whenever a node receives such message, it calculates its increment value Increments are stored and used reset the oscillator Increment values calculation taking into account the transmission and processing delays

12 Activator-inhibitor System A. Turing (1952): Morphogenesis (pattern formation in nature) can be explained by reaction and diffusion of chemicals in an initially uniform state The change rate of X and Y depends not only the value of themselves, but also their neighbors Given different f, g, μ, ν, dramatically different patterns can be generated

13 Activator-inhibitor System (cont. ) Application in Wireless Sensor Networks Evaluating activator-inhibitor mechanisms for sensors coordination by Neglia et. al. Sensors sleep as much as possible to save energy while maintaining sensing coverage and the communication backbone require coordination Use the pattern generated from the reaction diffusion equations to regulate the on-off state of the sensors Each node periodically broadcasts its activator and inhibitor values and upon receiving values from others, updates its own value based on the equations Sensors with a activator value exceeding a given threshold will start sensing

14 Drawbacks Bio-inspired approaches cannot immediately operate at the ideal performance level. It requires several iterations of adaptation to slowly reach the expected performance level Many bio-inspired approaches involve randomness in their design, which will reduce the determinism and controllability of the system behavior Though bio-inspired approaches eventually achieves its goals, usually we have limited knowledge of when such goal will be achieved

15 Potential Improvement For the centralized controller approach, there is an overmind controls every aspect of the system to achieve the goals For the bio-inspired approaches, we set simple rules on the entities at first and let them evolve to finally achieve the goals One potential solution is a hybrid design, combining the advantages of bioinspired approach and the advantages of conventional controller-based systems to accelerate the convergence of the system

16 Questions