Low-Power Wireless Sensor Networks

양유진

1. INTRODUCTION

2. NODE ARCHITECTURE CONSIDERATIONS

a. Computation and Dynamic Voltage Scaling

b. Radio Communication Hardware

3. ENERGY EFFICIENT NETWORKS

a. Signal Processing in the Network

b. System Partitioning

c. Energy Efficient Link Layer

4. POWER AWARE SOFTWARE

a. Energy Efficient Node Operating Systems

b. Energy Scalable Node Software

c. Applications Programming Interface(API)

5. CONCLUSION

1.Introduction

• A long node lifetime under diverse operating conditions demands power-aware system design.

• Computation and communication are partitioned and bal-anced for minimum energy consumption. Software that un-derstands the energy-quality tradeoff collaborates with hardware that scales its own energy consumption accord-ingly.

• Using the MIT μAMPS project as an exam-ple, this paper surveys techniques for sys-tem-level power-awareness.

2.Node Architecture Considera-tions

< Variables >• energy consumption at each ar-

chitectural block• leakage currents in the inte-

grated circuits• output quality and latency re-

quirements of the end user to the Duty cycles of radio transmis-sion.

2.a. Computation and Dynamic Voltage Scaling

• Energy consumption in a static CMOS-based processor can be classified into switching and leakage components.

C_tot : total capacitance switched by the computa-tionV_dd : supply voltage

V_th : device threshold voltageV_T : thermal voltage

2.a. Computation and Dynamic Voltage Scaling

• Sufficiently low duty cycles or high supply voltages, leakage energy can exceed switching energy

• we can reduce Vdd and the processor clock frequency together to trade off latency for energy savings.

• Reduction in clock frequency allows the processor to run at lower voltage.

• The quadratic dependence of switching energy on supply volt-age is evident, and for a fixed voltage, the leakage energy per operation increases as the operations occur over a longer clock period.

2.a. Computation and Dynamic Voltage Scaling

• sacrifice filter quality, the processor can run at a lower clock speed and thus operate at a lower voltage

2.a. Computation and Dynamic Voltage Scaling

In each example, our DVS-based implementation of energy-quality tradeoffs consumes up to 60% less energy than a fixed-voltage pro-cessor

2.B. Radio Communication Hard-ware

• : Power consumption of the transceiver• : transmit/receive on-time• : start-up time of the transceiver• : output transmit power : duty cycle of the receiver

The average energy consumption for a sensor radio (Figure 5) when sending a burst packet

2.B. Radio Communication Hard-ware

As the start-up time in-creases, the radio en-ergy becomes domi-nated by the start-up transient rather than the active transmit time.

2.B. Radio Communication Hard-ware

3. Energy Efficient Net-works

Sensor collaboration is important for two reasons1. Data collected from multiple sensors can of-

fer valuable inferences about the environ-ment

2. sensor collaboration can provide tradeoffs in communication versus computation energy

a. Signal Processing in the Network

3.a. Signal Processing in the Net-work

The energy-efficient network protocol LEACH (Low Energy Adap-tive Clustering Hierarchy) utilizes clustering techniques that greatly reduce the energy dissipated by a sensor system

3.b. System Partitioning

• FFT : The FFT results are phase shifted and summed in a fre-quency-domain beamformer to calculate signal energies in 12 uni-form directions

• LOB : estimated as the direction with the most signal energy

( Figure8.b ) Since the 7 FFTs are done in parallel, we can reduce the supply voltage and frequency without sacrificing latency.

3.c. Energy Efficient Link Layer

• Error control can be provided by various algorithms and techniques, such as convolutional coding, BCH coding, and turbo coding.

4. Power Aware Software

a. Energy Efficient Node Operating Systems - If the overheads in transitioning to sleep states were negli-gible, then a simple greedy algorithm could makes the system go into the deepest sleep state as soon as it is idle

- But in reality, transitioning to a sleep state and waking up has a latency and energy overhead.

- implementing the right policy for transitioning to the avail-able sleep states is critical.

4.a. Energy Efficient Node Operation Systems

• By our definition of node sleep states

• implementing a hierarchical node shutdown policy based on thresholds and statistical event prediction

4.a. Energy Efficient Node Operation Systems

4.b. Energy Scalable Node Soft-ware

• Transforming software such that most significant computations are accomplished first improves the energy-quality scalability can be improved. (EX : FIR firtering operation)

• If the energy availability to the node were re-duced, we may want to terminate the algorithm early to reduce computational energy

• In an unscalable software implementation, this would result in severe quality degradation.

• Figure 11 demonstrates the improved energy-quality characteristics of an energy-scalable im-plementation of the FIR filtering operation

4.b. Energy Scalable Node Soft-ware

4.c. Applications Programing Inter-face(API)

• An application programming interface is an ab-straction that hides the underlying complexity of the system from the end user.

• By defining high level objects, a functional inter-face and the associated semantics, APIs make the task of application development significantly eas-ier.

• An API consists of a functional interface, object abstractions, and detailed behavioral semantics.

4.c. Applications Programing Inter-face(API)

The functional interface itself is divided into the following

Functions that gather the state

(of the nodes, part of a network, a link between two nodes

etc.) Functions that set the state

(of the nodes, of a cluster or the behavior of a protocol) Functions that allow data exchange between nodes and the

basestation Functions that capture the desired operating point from the

user at the basestation Functions that help visualize the current network state Functions that allow users to incorporate their own models

(for energy, delay etc.)

5. CONCLUSION

Distributed sensor networks designed with built-in power awareness and scalable energy consumption will achieve maximal system lifetime in the most challenging and diverse envi-ronments.