0. Towards Commoditized Real-time Spectrum Monitoring
Ana Nika, Zengbin Zhang, Xia Zhou*,
Ben Y. Zhao and Haitao Zheng
Department of Computer Science, UC Santa Barbara
*Department of Computer Science, Dartmouth College

1. Spectrum as a Valuable Resource
Billions of $ spent on spectrum auctions
Efficient utilization is critical
Malicious users can “misuse” spectrum without authorization
Increasingly feasible via cheaper, smarter hardware
Active, comprehensive monitoring a necessity and challenge
Spectrum usage density will continue to grow  current monitoring tools do not scale
Spectrum enforcement: how do we detect and locate unauthorized users?

2. Challenges in Spectrum Enforcement
Coverage
Large and growing deployments, small/fixed measurement area
Abstract models impractical in outdoor settings
Responsiveness requires “real-time” measurements
Periodic spectrum scans?
Offline data processing likely insufficient
Infrastructure cost and availability
State of art: bulky, expensive spectrum analyzers
Alternative: USRP GNU radios

3. Our Approach: Real-time, Crowdsourced Spectrum Monitoring
Crowdsourcing measurement platform
Scales up in coverage and measurement frequency
Scales with demand/impact
Higher density usage areas ->
Low-cost commoditized platform
Explore replacement of specialized H/W with commody
Reduced cost, availability (integrated w/ next gen phones?)
Compensate for lower accuracy with redundancy

4. Outline Introduction Spectrum Monitoring System Crowdsourced Framework
Commoditized Platform
Feasibility Results
Additional Challenges

5. Crowdsourced Measurement Framework
Approach
Individual users monitor and collect spectrum activities in local neighborhood
Submit real-time results in to (centralized) spectrum monitoring agency
Agency aggregates/disambiguates consensus monitoring results

6. Commoditized Measurement Platform
Two hardware components
Commodity mobile device (smartphone)
Cheap & portable Realtek Software Defined Radio (RTL-SDR)
RTL-SDR as “spectrum analyzer”
DVB-T USB-connected dongle
Frequency range: MHz
Max sample rate: 2.4MHz
Cheap: <$20 per device
Mobile host serves as “data processor”
Translates raw data into data stream

7. Key goal: Evaluate feasibility of SDR platform
Sensing sensitivity
8-bit I/Q samples (vs.  Missing weak signals
How significant are errors (relative to alternatives)
Net impact on event detection?
Sensing bandwidth
Up to 2.4MHz bandwidth (vs. 20MHz)
Must sweep wider bands sequentially
Max frequency of sensing operation?

8. Impact of Sensing Sensitivity

9. Noise Measurements
RTL-SDR based platforms report higher noise variance
With sensing duration ≥1ms, RTL-SDR based platforms perform similarly to USRP

10. Signal Measurements
RTL-SDR platforms report lower SNR values compared to USRP platform
Smartphone’s microUSB interface does not provide enough power to RTL- SDR radio

11. Impact on Spectrum Monitoring
Signal detection:
USRP platform, SNR ≥ -2dB
RTL-SDR/laptop, SNR ≥ 7dB
RTL-SDR/smartphone, SNR ≥ 10dB
For 1512MHz band, 12dB difference  ~50% loss in distance

12. Addressing Sensitivity Issues
Deploy many monitoring devices with crowdsourcing
Redundant sensors increases probability of nearby sensor to target transmitters
Look at specific signal features
Pilot tones
Cyclostationary features
Pro: more reliable than energy readings
Con: additional complexity on mobile sensing devices

13. Impact of Sensing Bandwidth

14. Scanning Delay
RTL-SDR scan delay is two times higher than USRP (2.4MHz) because its frequency switching delay is higher
RTL-SDR radios can finish scanning a 240MHz band within 2s

15. Impact on Spectrum Monitoring
RTL-SDR/smartphone achieves <10% detection error (for 24MHz band)
As the band becomes wider (120MHz), error rate can reach 35%

16. Overcoming Bandwidth Limitation
Leverage crowdsourcing
either divide wide-band into narrow-bands and assign users to specific narrow-bands
aggregate results from multiple users w/asynchronous scans
Use novel sensing techniques
QuickSense
BigBand
Challenge: how to realize these sophisticated algorithms on RTL-SDR/smartphone devices

17. Remaining Challenges Coverage
Solution
Passive measurements from wireless service provider’s own user population
On-demand measurements from users of other networks
Leverage incentives and on-demand crowdsourcing model

18. Remaining Challenges Measurement Overhead
Spectrum monitoring overhead
Energy consumption
Bandwidth usage
Solution
Energy consumption: schedule measurements based on user context, e.g. location, device placement, movement, etc.
Bandwidth: secure in-network data aggregation and compression

19. Remaining Challenges Measurement Noise
Accuracy of spectrum monitoring affected by
Noise into monitoring data
Potential human operation errors
Solution
Expect/model noisy data
Use models for signal estimation: Gaussian process, Bayesian and Kalman filters

20. Thank you! Questions?