Nemo - Computer Science and Engineering

Report
Nemo: A High-fidelity Noninvasive
Power Meter System for
Wireless Sensor Networks
Ruogu Zhou, Guoliang Xing
Department of Computer Science and Engineering,
Michigan State University
Wireless Sensor Networks Platforms
• Microscopic and inexpensive devices
– Densely deployed to increase sensing fidelity
• Ad-hoc deployment
– Powered by battery; transmit wirelessly
• Various form factors
2
Scarcity of Power
• Small energy reservoir on node
– Usually 2 AA batteries
Node1
x
• Energy-efficiency is crucial for WSN
– Many energy-efficient protocols are proposed
– Their effectiveness is hard to verify
x
x
• Power outages are common in deployment
Node2
Node3
– Greatly impair sensing fidelity
– Exact reasons are usually unknown
Node4
Node5
Base Station
3
In-situ WSN Power Meters
• SPOT[IPSN’07], iCount[IPSN’08]
• Low sampling rate/resolution
– Cannot capture sleep power consumption or power transients
SPOT mounts on MicaZ
iCount with Telos
4
In-situ WSN Power Meters
• SPOT[IPSN’07], iCount[IPSN’08]
• Low sampling rate/resolution
– Cannot capture sleep power consumption or power transients
• Invasive to host node
– Require host CPU, RAM , I/O and timer
– Installation requires wiring and soldering
5
Nemo: Noninvasive High Fidelity
Power Meter
• Retrofit with after-market platforms w/ power metering
• Noninvasive to host node
– Standalone meter, plug &play, work with virtually any platform
• High measurement fidelity
– 2uA-200mA dynamic range, >5 KHz sampling rate, <1uA resolution
TelosB node
+
• Real-time communication with host
– Enable real-time monitoring and energy-aware runtime adaptation
Nemo
6
Challenges
• Noninvasiveness and real-time communication?
– Only connection b/w meter and host is power rail
– No dedicated data wires
• Only
Highconnection
fidelity and low power consumption?
– High fidelity usually results in high power consumption
– Ex: ADC w/ high dynamic range consumes > 10 mA current
7
Outline
• Motivation
• Challenges and System design
– Host-meter Communication
– High Fidelity Measurement
• System evaluation
• Case study
• Conclusion
8
Voltage Modulation (Meter->Host)
• Modulate supply voltage of host to transmit measurements
– Modulator: A Schottky diode controlled by a switch
• Host decodes by sampling supply voltage
– Most built-in ADCs can be programmed to measure supply voltage
• Host cannot modulate supply voltage
– Cannot be applied to host-> meter link
Voltage
Modulator
Diode
ADC
Sensor
010011011100
Switch
Power Positive+
Modulation Control
Power Ground
9
Current Modulation (Host->Meter)
• Modulate own current draw to transmit data to meter
– Modulator: Any component that can be switched fast, e.g. LED
• Meter decodes by measuring host current draw
Currrent
Modulator
010011011100
Nemo
Sensor
Power Ground
Current
Measurement
Power Positive+
10
Outline
• Motivation
• Challenges and System design
– Host-meter Communication
– High Fidelity Measurement
• System evaluation
• Case study
• Conclusion
11
Fidelity Requirements
• Wide dynamic range
– Sleep (~2uA) to Active (~200mA), 5 orders of difference
• High sampling rate
– > 5kHz to capture power transients
• High resolution
– Monitor sleep power (< 1uA) which determines system life
Power transients
caused by radio
on/off
30
Current (mA)
25
20
15
10
5
0
-2
0
2
4
6
8
Time (ms)
10
12
14
16
12
Current Measurement 101
• Shunt resistor (current sensing resistor)
– Convert current intensity to voltage signal
• Pre-amplifier
– Amplify voltage signal to a proper level
• ADC
– Convert analog signal to digital signal
– 2uA to 200mA dynamic range and < 1uA resolution  18-bit ADC
Power Positive+
ADC
Sensor
Power Ground
13
Current Measurement 101
• Shunt resistor (current sensing resistor)
– Convert current intensity to voltage signal
• Pre-amplifier
– Amplify voltage signal to a proper level
•
High
dynamic
range
ADCs
are
expensive
ADC
andtopower
– Convert analog signal
digital signalhungry!
– 2uA to 200mA dynamic range requires an 18-bit ADC
14
Solution: Auto-ranging
• High resolution needed only when measuring small current
– Small current does not need 0-200mA dynamic range
• Wide dynamic range needed only when measuring large current
– Large current does not need <1uA resolution
• Adjust measurement range and resolution dynamically
– Large current -> use wide measurement range, low resolution
– Small current -> use narrow measurement range, high resolution
15
Implementation of Auto-ranging
• Adjust shunt resistor to change measurement range& resolution
– Wide (narrow) range, low (high) resolution -> small (large) shunt resistor
• Use low dynamic range low power ADC
– Adjust measurement range according to ADC reading
• Shunt resistor switch
– A series of electrically controlled shunt resistors
– Adjust resistance by shorting one or more resistors
Resistors
Pwr Positive+
Input
Output
ADC
Sensor
Switches
Pwr Ground
16
Outline
•
•
•
•
•
Motivation
Challenges and System design
System evaluation
Case study
Conclusion
17
Implementation & Experiment Setup
•
•
•
•
PCB area 1.5 inch by 2.5 inch
System software implemented in C and assembly
Nemo is calibrated using a set of resistors
Agilent 34410A Bench-top digital multi-meter as reference
Agilent 34410A
benchtop DMM
Extech handheld DMM
Agilent DSO2024A Oscilloscope
TelosB
Nemo with battery
18
Measurement Fidelity (I)
• TelosB mote running a sense-and-send app as load
Current (mA)
60
Radio on
ADC on
Match groundtruth closely
Ground-truth
40
Measurement
Radio off
ADC off
20
Radio RX on
0
0
2
4
Radio TX on
6
8
10
12
14
Time (ms)
CDF
100
50
Average Error: 2.09%
CDF
0
0
1
2
3
4
5
6
Relative Error (Percent)
7
8
9
10
19
Measurement Fidelity (II)
• Sampling rate: constant 8.192 KHz
• Dynamic range: 0.8 uA to 202 mA
• Resolution < 1 uA when current is less than 2.5 mA
2
10
Resolution (uA)
resolution
48 uA
1
10
Dynamic range
0.8 uA to 202 mA
0
6.6 uA
10
Resolution < 1uA
-1
0.68 uA
10
0.069 uA
0.013 uA
-2
10 -3
10
-2
10
-1
10
0
10
1
10
2
10
Current (mA)
20
Case Study: Sleep Power of Mote
• 3 randomly selected TelosB motes running Null app
• Nemo is attached as power meter
• Surface of mote is heated to 80oC, then cooled down to 0oC
60
Mote1
Mote2
Mote3
Sleep Current (uA)
50
40
5X difference
Difference
<1uA
30
20
10
0
20
30
40
50
60
Temperature (Degrees Celsius)
70
80
21
Conclusions
• A noninvasive in-situ power meter for WSN
– Plug and play, high measurement fidelity
• Novel communication scheme for host-meter comm.
– Voltage&current modulation for communication over power rails
• Auto-ranging technique for high measurement fidelity
– Dynamically configure meter according to measurement requirements
• Evaluation in real experiments
– High dynamic range, high sampling rate, high resolution, low error
22
Q/A
23

similar documents