Title: ENERGY: THE ROOT OF ALL PERVASIVENESS
1ENERGY THE ROOT OF ALLPERVASIVENESS
- Anthony Ephremides
- University of Maryland
- April 29, 2004
2PERVASIVE NETWORKING
- Ability to access the network
- (Anywhere, Anytime, Anyone)
- Focus on wireless
- More specifically Ad Hoc
- (multihop)
3KEY ELEMENTS
- Wireless Channel
- fading
- interference
- SINR
- Portable Energy Supply
- efficiency
- vs.
- limited
- SOFT LINK GRAPHS
- CROSS-LAYER COUPLING
gt
lt
4TWO ILLUSTRATIONS
- MAC/ROUTING
- with
- Energy Metrics
- Processing vs. Transmission
- in
- Sensor Networks
5MAC/ROUTING
- Routing algorithm flows on each link
- MAC assigns resources to competing flows
- Actual link throughput depends on MAC
- New link quality metric values
- Routing Algorithm new flows
6MULTI-HOP AD HOCNETWORK
- Single channel slotted time
- Separate control channel
- Single transceiver per node
- Power control - Pmax
- regulate interference
- save energy
- Simple attenuation model
- free-space, distance based
- SINR
-
gt
lt
7SCHEDULING
- Scheduling Rules
- A node can only be associated with one active
link at a time. - SIR requirements are satisfied.
- The link with the lowest metric has the top
priority.
8Scheduling Algorithms 1. Power is preset. Links
are added (if SIRs are satisfied) in the order of
link metric. Easy for distributed
implementation. 2. With iterative power control.
Links are added (if SIRs are satisfied) in the
order of link metric. Difficult for distributed
implementation. 3. First find maximal number of
links that can coexist, then run iterative power
control. Remove links until SIRs are satisfied.
Difficult for distributed implementation.
9Simulation Results No rerouting.
Throughput and Delay for different scheduling
algorithms.
10JOINT SCHEDULING AND ROUTING
Routing Bellman-Ford algorithm with routing
distance
11CONTROL OFSENSOR NETWORKS
- Application Major Driver
- But, in all cases Longevity
- (energy efficiency)
- Major Challenges
- MAC
- Routing
- (map application-related objective function to
link metric or MAC priority)
12SENSOR NETWORK FOR DETECTION
control node
13MODEL
control center
Each Node Collects Independently T
independent Binary Measurements
14MODEL (cont.)
- Three Operating Options
- - Centralized
- All data transmitted to CC
- - Distributed
- Each node decides transmits its decision
to CC
- - Quantized
- Each node sends a quantized M-bit quantity
to CC - where
15ENERGY CONSUMPTION ANALYSIS
- - Energy for Data Processing
- - based on of comparisons
- - represents the energy consumed for
one comparison - - is the of comparisons
- - Energy for Transmission
- - based on the distance from sensor nodes to
control center and of bits - transmitted
- - represents the energy consumed for
transmitting one bit data - over a unit distance, for a fixed
communication system - - represents the distance from sensor
nodes to control center - - is the of bits transmitted
- - Total Energy
16ENERGY CONSUMPTIONANALYSIS (cont.)
- - Energy Consumption per Node for Three Options
- - Centralized Option
- - option 1 transmit all observations to CC
- - option 2 transmit of 1 out of T
observations to CC - - Distributed Option
- - Quantized Option (suboptimal solution)
- where represents the expected of
comparisons needed for the suboptimal solution,
which is a function of
17Energy ConsumptionAnalysis (cont.)
- - Energy Consumption vs. Accuracy
- fix
vary - example 1
- example 2
18NEXT STEPS
- Spatial/Temporal Correlation
- Routing (map objective function to link metric)
- Broader measurement model
- MORE FUNDAMENTALLY
- Couple processing energy (dictated by the chosen
SP algorithm) to the embedded system design.
(memory management, signal flow graphs for
software vs. hardware split, computing fabric) - Trade-off transmission to processing under such
INTERACTIVE design (ultimate cross-layering)
19CONCLUSIONS
- Holistic cross-layer design from energy point of
view - Application dependency/exploitation
- It takes courage to succeed
-
- It takes energy to be pervasive