Http:// Ruzzelli, Cotan O’Hare, Tynan, Havinga Protocol assessment issues in low duty cycle sensor...
-
date post
21-Dec-2015 -
Category
Documents
-
view
217 -
download
0
Transcript of Http:// Ruzzelli, Cotan O’Hare, Tynan, Havinga Protocol assessment issues in low duty cycle sensor...
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Protocol assessment issues in low duty cycle sensor networks:
The switching energy
A.G. Ruzzelli, P. Cotan*, G.M.P. O’Hare, R. Tynan, and P.J.M Havinga**
Adaptive Information Cluster (AIC) group @ PRISM Laboratory
School of Computer Science and Informatics, University College Dublin (UCD), Ireland.
*Department of Electronic Engineering, Technical University of Catalonia, Spain.
**Department of Computer Science, University of Twente,The Netherlands.
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Summary
• Generality on protocol energy assessment• The low duty Cycle through the wake up concept• Switching between transceiver states
• Phase1: Board measurements– The sensor node– The experimental approach– The measured results
• Phase2: Switching energy assessment– The S-MAC protocol– Performance evaluation– Simulation setup– Simulated results
• Considerations• Conclusions
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Generality on protocol energy assessment
• Energy consumption mainly due to the transceiver activity;
• Protocol energy assessment based on transceiver states:– Transmit time;– Receive time;– Idle time (Sleeping time in sensor-nets);– Switching time (USUALLY NOT ASSESSED);
• Switching energy negligible in ad-hoc wireless network protocol assessment (e.g. WiFi);
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Switching in standard wireless networks• Is defined as the transition time that elapses between the end of a transceiver
state and the beginning of the following one;
• Possible switching states consist of:– RX/TX and TX/RX– TX/Sleep and Sleep/TX– RX/Sleep and Sleep/RX
• State transition is fast little amount of energy is consumed.
• Switching energy is much smaller than total energy spent.
• Transceiver data sheets report average switching time but not the energy spent.
• Related work show that assessment of novel protocol architectures for WSNs inherited the switching energy negligibility.
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Sensor network characteristics
• Energy consumption: primary objective • The wake-up concept • Very low duty cycle (even less than 5%) • Small packets smaller than in ad-hoc networks (e.g.
temperature data is few bytes)• Low data traffic per node
Can we consider switching energy still negligible for low duty cycle sensor networks?
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Phase 1: The experimental model
••Analysis conducted on different EYES sensor node prototypes
•Prototypes mounted different off-the shelf transceiver for sensor networks;
•Investigation of Tr1001, CC1000 and CC1010 transceivers;
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Phase 1:The experimental approach• The voltage drop is gauged across high-side series resistor placed between the battery (+
terminal) and the input power connector;
• Current consumption, power and energy consumption derived from the voltage.
• Hardware connected to an oscilloscope.
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
The measuring circuit• Based on INA110 instrumentation amplifier
fast settling time and high slew rate device.
• Two resistors used: “low power mode” and “Tx/Rx mode”.
• Test performed by a square waveform of 1 kHz and of 5 V amplitude at the input of the INA 110 connected through an attenuating resistive divider circuit.
• Good precision and low distortion for conducting measurements at the
edges.
INA110 main characteristics
Bias 50 pA max
Settling time (Vout 20V) 3 us to 0.1 %
CMRR 106 dB min
Gain 1, 10, 100, 200, 500
Input impedance 5x10^12 ohm || 6pF
Slew Rate 17 V/us
Small signal BW 470 kHz (Gain = 100)
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Preliminary notes:
• The CC1010 had a processor built in and therefore the CPU on that board could be put into sleep mode.
• CC chipcon class presented higher sensitivity than TR1001.
• CC1000 and CC1010 boards were configured with the oscillator ON in low power mode shorter switching time (2ms activation if OFF)
Board measurement results
Boards current and power consumption
Current [mA] Power [mW]
SL RX TX SL RX TX
TR1001 0.005 4.8 12 0.015 14.4 36
CC1000 0.11 10 11 0.33 30 33
CC1010 0.15 26 26 0.45 78 78 Boards switching energy
Boards switching times
Switching Energy [uJ]
SL to RX SL to TX RX to SL TX to SL RX to TX TX to RX
TR1001 8.82 25.2 0.116 2.83 25.2 8.85
CC1000 19 20.5 0.7 0.75 22.4 21.4
CC1010 45.8 47.75 1.83 1.93 61.43 61.61
Switching Times [us]
SL to RX SL to TX RX to SL TX to SL RX to TX TX to RX
TR1001 700 700 10 10 700 700
CC1000 850 850 10 10 850 850
CC1010 1600 1600 10 10 850 850
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Phase 2: Switching energy assessment
• The values obtained are applied to the SMAC protocol;
• SMAC is normally used as benchmark against other novel architectures;
• Results obtained by using the OmNet++ simulator
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
The SMAC protocol
• SMAC divides time in two periods: active time and sleeping time;
• Active period = SYNC period for node sync update, Request To Send (RTS), Clear to Send (CTS).
• Communication establishing: – neighboring nodes synchronize to the start of
the active period then local broadcast of SYNC packets.
• Data message exchanges follow the RTS/CTS/DATA/ACK; nodes switch between different states
periodically.
RTS
CTS
Data
Transmitter Receiver
time
ACK
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Simulation setup• Three nodes and one gateway in a line
– Node 3 = Source; Node1 & Node2 = Forwarder; Gateway = Destination
• nodes communicate with direct neighbours only.
• Results averaged between node2 and node1 values (higher node switching activity)
• 13 independent simulations of 20 minutes each. • 10 independent random seeds for clock skew and offset inaccuracies.
• Traffic load regulated by Node 3– 16 bytes packet– Generation rate: 60s(low traffic) and 2s (high traffic).
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Performance evaluation metrics
• Energy TX %: spent by per node per bit transmitted;
• Energy Switch %: spent per node for the total number of transitions of two consecutive states;
• Energy Sleep %: energy spent by one node during the time of inactivity referred to as the sleeping state;
• Total consumption per node: all previous metrics plus RX energy and idle listening.
• Duty cycle changed by varying the node active period
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Simulated results (1): Total consumption
Total Energy consumption 2s msg generation period
0
2
4
6
8
10
12
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty cycle [%]
Tot
al e
nerg
y [J
] 9
TR1001
CC1000
CC1010
Total Energy consumption 60s msg generation period
0
2
4
6
8
10
12
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty cycle [%]
Tot
al e
nerg
y [J
] 9
TR1001
CC1000
CC1010
Low traffic
High traffic
• The simulations ended after 50 packets were correctly relayed from source to destination;
• The results show only a little increase of consumption in high data traffic conditions;
• The CC family present higher energy consumption profile than Tr1001 due to:
– The processor built;– The oscillator left ON in low power
mode (oscillator OFF → >5mA current consumption to wake-up)
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Simulated results (2): Low traffic condition
• For all transceivers and duty cycles, switching energy is between the sleeping energy and energy TX;
• Switching energy can be higher than the
energy TX.
• Lower bound of 1.7% for the duty cycle due to an intrinsic operational limit of SMAC.
• Other existing protocols that can work below 1% duty cycle (e.g. BMAC)
CC1000
0
2
4
6
8
10
12
14
16
18
20
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty Cycle [%]
% of t
he to
tal e
ner
gy i
Energy TX
Energy Switch
Energy Sleep
TR1001
0
1
2
3
4
5
6
7
8
9
10
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty Cycle [%]
% of t
he to
tal e
ner
gy f Energy Switch
Energy Sleep
Energy TX
CC1010
0
2
4
6
8
10
12
14
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty Cycle [%]
% of t
he to
tal e
ner
gy i
Energy TX
Energy Switch
Energy Sleep
Switching energy as percentage of the total consumption
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Simulated results (3): High traffic condition
• Maximum switching value above 6% for 1.7% duty;
• Oscillator ON causes higher sleeping energy of CC family thanTR1001.
• Expected higher % of switching energy for duty cycle lower than 1.7%
CC1010
0
2
4
6
8
10
12
14
16
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty Cycle [%]
% of t
he to
tal e
ner
gy i
Energy TX
Energy Switch
Energy Sleep
CC1000
0
2
4
6
8
10
12
14
16
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty Cycle [%]
% of t
he to
tal e
ner
gy i
Energy TX
Energy Switch
Energy Sleep
TR1001
0
5
10
15
20
25
10 9.6 9.3 8.8 8.3 7.7 7 6.3 5.6 4.8 3.6 2.7 1.7
Duty Cycle [%]
% of t
he to
tal e
ner
gy i
EnSleep
Energy TX
Energy Switch
Switching energy as percentage of the total consumption
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Considerations and guidelinesConsidering 5% as the lower bound of energy consumption significance:
– For TR1001 and CC1010, the switching energy needs to be computed if the node duty cycle is equal to or less than 3% and 3.6% respectively;
– For CC1000, the switching energy needs to be computed if the node duty cycle is equal to or less than 2.7% and 3.6% respectively;
– Sleeping energy consumption of TR1001 can be neglected in any case simulated as less than 2%;
– For CC1000 and CC1010 in low traffic load conditions, the transmitting energy becomes significant at 2.5% duty cycle or lower.
• Although similar, total energy consumptions might greatly differ in their inner energy usage composition The choice of a protocol to use is not only based on the application but also on the radio on board
http://www.cs.ucd.ie/csprism/Ruzzelli, Cotan O’Hare, Tynan, Havinga
Conclusion
• Values of switching energy have been obtained by direct measurements on different boards;
• The measurements have been applied to the SMAC protocol;
• Considerations and protocol assessment guidelines have been derived;
• In low duty cycle sensor-nets, the switching energy should be computed together with transmitting, receiving and sleeping energies;
• The obtained results help improve the MAC protocol evaluation process and empowers decisions relating to the judicious protocol/hardware choice for an specific set of WSN applications;
• Switching energy is expected to account for an even more significant percentage of the total power consumed as the duty cycle get closer to 1% such as in BMAC;
• Future work activities include the investigation of TDMA protocols that allow lower node duty cycle and more complex topologies.