Title: TrafficInformed Minimum Power Configuration in Wireless Sensor Networks
1Traffic-Informed Minimum Power Configuration in
Wireless Sensor Networks
- Guoliang Xing
- Host Dr. Ying Zhang
- Collaborator Dr. Qingfeng Huang
- PARC
2Motivation
- Many sensor networks require long lifetime
- Several months to years of operational time
- Habitat monitoring, civil structure monitoring,
surveillance - Energy is scarce
- Low cost energy supply, e.g., AA batteries
- Wireless communication is energy costly
3Related Work
- Topology control
- Reduce radio transmission power
- Duty cycle control
- A subset of nodes remain on all the time
- Other nodes only turn on radios when necessary
4Topology Control
- Radio transmission power Ptx Karb
- r is the transmission distance
- K,a and b are constants, 2b 4
- Ptx increases quickly with distance ? Relaying
through multiple hops is more power efficient - Maintain a power efficient topology
- Reduced transmission ranges
- Connected network
- Low dilation
5Examples - Topology Control
6Duty Cycle Control
- Maintain sufficient service using a small num
of active nodes - Connectivity, coverage, dilation
- All other nodes operate on low duty cycles
- Turn on radios sporadically
- Keep radios off most of the time
on
on
off
off
off
An example of radio duty cycle
7Examples Duty Cycle Control
SPAN
CCP
SPANCCP
8Understanding Radio Power Cost
- Real-world data
- Cabletron network card Tx 1400mW, Rx 1000mW,
Idle 830mW, sleeping 130mW - Berkeley mote Tx 5.3 - 26.7 mA, Rx/Idle 8.0mA,
sleeping 1uA - Idle mode consumes considerable power!
- Putting a radio to sleep is the most efficient
way to reducing power!
9Critiques
- Topology control
- Only reduce the transmission power
- Keep all radios on all the time
- Duty cycle control
- Does not optimize transmission power
- Keep the backbone nodes on all the time
10Our Approach Minimum Network Configuration (MPC)
- Fine-grained radio power model
- Incorporate different working modes idle, tx, rx
- Minimize power consumption
- Minimize num of active nodes according to
communication activities - Minimize transmission ranges for active nodes
- Put all other nodes to sleep
- Optimal solution depends on the network traffic
11An Example
- a sends to c at rate of R
- Scheme 1 a ? c, b sleeps
- Scheme 2 a ?b?c
B Node bandwidth
12Problem Formulation
- Given
- A network G(V,E), E(u,v) u can communicate
with v - Sink t and a set of source nodes Ssi
- si sends data to t at rate of Ri.
- Find a subgraph G(V,E) of G such that
- si and t are connected
- The total power cost is minimal
link (u,v) is on the shortest path from si to t
13Problem Transformation
t
Total power cost
- NP-hard
- Steiner tree problem is a special case
- Two polynomial-time special cases
- Pidle?0 or SV
S1 R1 bps
S2 R2 bps
14Approximate Algorithm IShortest-path Tree
Heuristic
- For each si
- Weight network G(V,E) as follows
- Find the shortest path from si to t.
- End
- Approximation ratio S
- Optimal when Pidle?0 or SV
15Example
P(s1?t) 0.1x(22) P(s2?t) 0.2x(14)
Total idle power 2x5 Ptotal P(s1?t) P(s2?t)
total idle power 11.4
16Reusing Active Paths
P(s1?t) 0.1x(22) P(s2?t) 0.2x(24) Total idle
power 2x4 Ptotal P(s1?t)P(s2?t)Pidle9.6lt11
.4
17Approximate Algorithm IIIncremental
Shortest-path Tree Heuristic
- Label all nodes to asleep
- For each si
- Assign link weights as follows
- w(u,v)(Ri/B)(Ptx(u,v)Prx-2Pidle)Pidle if u
or v is asleep - w(u,v)(Ri/B)(Ptx(u,v)Prx-2Pidle)
otherwise - Find the shortest path si?t
- Label all nodes on the found path to active
- End
18Performance of Incremental Shortest-path Tree
- Approximation ratio S
- Approximation ratio ? 2 when
- Optimal when Pidle?0 or SV
19Constant-ratio Approximation Algorithm
- Difficult to find constant-ratio heuristics
- Little correlation between two sub-objectives
- 1. shortest-path tree weighted by
Ptx(u,v)Prx-2Pidle - 2. Steiner tree weighted by Pidle
- Bound the transmission power
- Ptx(u,v)Prx-2Pidle aPidle
- Steiner tree algorithm
- Assign all link the same weight Pidle
- Find the minimal Steiner tree connecting si and
t - Approximation ratio (1a)ß, where ßis the best
approximation ratio of Steiner tree algorithms
1.5
20Extending to More General Scenarios
- Shortest-path based heuristics
- Multiple sinks
- Heterogeneous network different Ptx(u,v), Prx,
Pidle for each node - Open question
- Overloaded network
21Future Work
- Performance comparisons
- Different heuristics
- Other techniques topology control and duty cycle
control - More efficient approximation algorithms
- Distributed/learning heuristics
- Load balancing
22Acknowledgements
- ERA members Dr.Ying Zhang, Dr. Markus Fromherz ,
Dr. Haitham Hindi , Dr. Wheeler Ruml , Dr. Lara
Crawford - Collaborator Dr. Qingfeng Huang
- Fellow interns