slides - University of Wisconsin–Madison

Download Report

Transcript slides - University of Wisconsin–Madison

A Wireless Spectrum Analyzer in Your Pocket
Tan Zhang, Ashish Patro, Ning Leng, Suman Banerjee
University of Wisconsin-Madison
Tan Zhang / Snoopy / HotMobile 2015
1
What is Spectrum Sensing?
-50
-60
-70
Power (dBm)
-80
-90
-100
-110
-120
-130
-140
-150
500
520
540
560
580
600
620
640
660
680
700
Frequency (MHz)
Tan Zhang / Snoopy / HotMobile 2015
2
Application of Spectrum Sensing
Whitespace spectrum
-50
-60
-70
Power (dBm)
-80
-90
-100
-110
-120
-130
-140
-150
500
520
540
560
580
600
620
640
660
680
700
Frequency (MHz)
Spectrum Vacancy
Tan Zhang / Snoopy / HotMobile 2015
Network Diagnosis
Device Management
3
Problem of A Cathedral Approach
Spectrum Analyzer
Cumbersome
Expensive
Sophisticated
(up to 30kg)
(10k – 50k)
(RF/IF gain, filter bw)
Tan Zhang / Snoopy / HotMobile 2015
4
Opportunity of Mobile Phone Sensing
Can we use smartphones and tablets for spectrum sensing?
Compact
Tan Zhang / Snoopy / HotMobile 2015
Cheap
Easy to use
5
Opportunity of Mobile Phone Sensing
Spatial
Distribution
Androidat
and
IPhonescale
devices
Enable
spectrumofanalytics
massive
New York
Tan Zhang / Snoopy / HotMobile 2015
Los Angeles
6
Opportunity of Mobile Phone Sensing
Airshark
WiFi
Chipset
Power
• Approach 1 – leverage subcarrier energy sample from
built-in WiFi chipsets, e.g., Atheros 92xx
Subcarrier
Frequency
Tan Zhang / Snoopy / HotMobile 2015
7
Opportunity of Mobile Phone Sensing
• Approach 2 – attach customized analyzer dongle
WiSpy (WiFi band)
RTL-SDR (TV band)
Tan Zhang / Snoopy / HotMobile 2015
8
Limitation of Existing Mobile Phone Sensing
• Narrow frequency range
– WiFi or TV band
• Low spectrum resolution
– 64 subcarriers over a 20MHz band
– 1000 times worse than spectrum analyzer
Tan Zhang / Snoopy / HotMobile 2015
9
Goal
Develop hardware attachment and signal processing
technique to enhance spectrum sensing on mobile devices
Spectrum Knowledge out of Your Pocket
Snoopy
Tan Zhang / Snoopy / HotMobile 2015
10
Outline
• Smartphone based sensing platform – Snoopy
– Frequency translator
– Spectrum sensing on WiFi cards
– Statistical feature based signal detection
• Implementation
• Evaluation
• Future work
Tan Zhang / Snoopy / HotMobile 2015
11
Snoopy Overview
Power
Power
Power
WiFi band
FFTs
Frequency
(2.4GHz)
Antenna
Tan Zhang / Snoopy / HotMobile 2015
Signal Type, Power
Frequency
Translator
12
Frequency Translator
Tune
to let
Input
Output
Frequency Mixer
Frequency
Synthesizer
Tan Zhang / Snoopy / HotMobile 2015
13
Spectrum Collection on WiFi Cards
• Recent 802.11 WiFi chipsets expose subcarrier
energy samples
– e.g., Intel 5300, Atheros 92xx and 93xx
Metric
Performance
Capture bandwidth
20/40MHz
Capture delay
120us
Spectrum resolution
312KHz
Challenging to determine vacant spectrum
Tan Zhang / Snoopy / HotMobile 2015
14
Challenge of Determining Whitespaces
from WiFi Spectrum Scan
Pilot
-120
-130
-140
-150
578 579 580 581 582 583 584
-100
-110
-120
-130
-80
-100
-110
-120
-130
-140
-150
518 519 520 521 522 523 524
-150
626 627 628 629 630 631 632
Frequency (MHz)
Frequency (MHz)
Noise fluctuation reduce
peak detection accuracy
WiFi Scan
Pilot
-90
-100
-60
Power (dBm)
Power (dBm)
-70
-90
-140
Frequency (MHz)
-60
Noise (whitespace)
Need to detect TV Tone
and wireless microphone
-110
FFT
Microphone
Power (dBm)
-100
-90
-70
-80
-60
Tone
-90
-100
Power (dBm)
Power (dBm)
-90
Spectrum Analyzer
Power (dBm)
-90dBm power
FFT TV
-70
-80
-90
-100
-110
-110
-110
-120
578 579 580 581 582 583 584
-120
518 519 520 521 522 523 524
-120
626 627 628 629 630 631 632
Frequency (MHz)
Tan Zhang / Snoopy / HotMobile 2015
Frequency (MHz)
Frequency (MHz)
15
Statistical Spectrum Feature
Apply Fourier Transform on spectrum to collect entire shape features
Co-efficient
Coefficients can help signal detection
FFT Index
FFT over FFT
Tan Zhang / Snoopy / HotMobile 2015
16
Statistical Spectrum Feature
-90dBm power
WiFi Scan
Noise (whitespace)
Microphone
-60
-60
-70
-70
-70
-80
-90
-100
Power (dBm)
-60
Power (dBm)
Power (dBm)
TV
-80
-90
-100
-110
-110
-120
578 579 580 581 582 583 584
-120
518 519 520 521 522 523 524
-80
-90
-100
-110
-120
626 627 628 629 630 631 632
Frequency (MHz)
Frequency (MHz)
Use SVM classifier to learn the
importance of individual coefficients
FFT over FFT
Frequency (MHz)
0.4
0.2
Dominant peak
0.3
0.2
0
0
0
2
4
6
8 10 12 14 16 18
FFT Index
Tan Zhang / Snoopy / HotMobile 2015
Constant offset
0.3
0.2
0.1
0.1
0.1
0.4
Co-efficient
Square envelop
0.3
Co-efficient
Co-efficient
0.4
0
0
2
4
6
8 10 12 14 16 18
FFT Index
0
2
4
6
8 10 12 14 16 18
FFT Index
17
Implementation
• Frequency translation hardware
– Wide band digital radio
RF
RF chain
chain 11
RF chain 2
• 30MHz – 7.5GHz frequency range
• 1ms frequency switching delay
Tan Zhang / Snoopy / HotMobile 2015
18
Implementation
• Software
– Patch Ath9k driver to enable spectrum scan in the 2.4GHz
– WiSense – Android based application for WiFi band sensing
Tan Zhang / Snoopy / HotMobile 2015
19
Experiment
• Experiment setup
– Used a Linux based router with an Atheros 9280 card for
spectrum sensing
– Connect Snoopy and a ThinkRF WSA4000 analyzer to the
same antenna
Linux
Router
Tan Zhang / Snoopy / HotMobile 2015
20
Accuracy in Detecting Primary Signals
• Collect spectrum data in 8 UHF channels
• Use a RF attenuator to capture TV and microphone
signals from -50dBm to -90dBm
Mis-detection Rate (%)
25
20
15
Snoopy(statistic)
Snoopy(peak)
Analyzer(statistic)
Analyzer(peak)
1-5% gain of
statistical feature
15%
<10% worse than ThinkRF
10
5
7%
3%
0
-70
-80
-90
Signal Power (dBm)
Tan Zhang / Snoopy / HotMobile 2015
21
Accuracy of Measuring Channel Power
Absolute Power Error (dB)
• Measure in-band power of 6MHz TV channels
• Calculate absolute power difference between Snoopy
and ThinkRF analyzer in each channel
14
12
10
<4dB median error in
lower UHF band
Higher error due to
translator distortion
8
6
4
2
0
26
Tan Zhang / Snoopy / HotMobile 2015
28
32
38
Channel Number
49
22
Challenges and Future Work
• Connect off-the-shelf mobile devices
– WiFi dongle with SMA connector
– WiFi repeater to relay translated signal
• Reduce size and cost of translator
– Translator for receiving only
– 0.1 – 4GHz frequency range and $48
TL-WN722N
EST-1W
ZX05 – U432H
• Improve measurement accuracy
– Collaborative sensing with longitudinal measurements
– Leverage phase information from I, Q samples
Tan Zhang / Snoopy / HotMobile 2015
23
Conclusion
 Designed a smartphone based spectrum sensing platform to
enable citizen-contributed spectrum analytics.
 Designed statistical spectrum features to improve signal
detection accuracy from low-resolution spectrum.
 Built driver hooks and an Android application to enable
spectrum sensing on smartphones and tablets.
WiSense: http://research.cs.wisc.edu/wings/projects/wisense/
Tan Zhang / Snoopy / HotMobile 2015
24
WiSense Demo
Tan Zhang / Snoopy / HotMobile 2015
25
Thanks for your attention!
[email protected]
Tan Zhang / Snoopy / HotMobile 2015
26