- GNU Radio

Report
TED AND KARYN HUME CENTER FOR
NATIONAL SECURITY AND TECHNOLOGY
GREM:
A Radio Environment Map Implementation
September 16, 2014
Tim O’Shea
Member of Technical Staff, Hume
Center
[email protected]
1
Bob McGwier, PhD
Director of Research, Hume Center
[email protected]
http://www.hume.ictas.vt.edu
Outline
•
•
•
•
The concept of Radio Environment Map
The constituent components and architecture
Concepts of operations and operators derived
Use in Dynamic Spectrum Sharing with some
examples.
• Collection of data
• Conclusions and further work
2
The Concept of REM
• Origins
– Developed at Virginia Tech by Dr. Jeff Reed and his
graduate students: Zhao, Youping; Reed, Jeffrey H.; Mao, Shiwen; Bae, Kyung K.; ,
"Overhead Analysis for Radio Environment Mapenabled Cognitive Radio Networks," Networking
Technologies for Software Defined Radio Networks, 2006.
• It is a database which contains information on your
radio environment such as:
• What do you populate the database with initially?
• What do your radio sensors see in your environment?
• What regulation and policy is in effect in the area of
your sensor?
• What service is being provided by the signals?
3
The Concept of REM (2)
• Was initially proposed to be a tool to be used in TV
White Space Systems (hereinafter TVWS, Reed, et.
al.)
• Spectrum Sharing has “grown up a lot” since TVWS
and much more sophisticated systems are being
proposed to enable wireless mobile to share
spectrum rather than get exclusive licenses
– Federated Wireless, Google have shown much interest
in spectrum sharing proposed by the FCC in its NPRM on
wireless providers sharing with USG (see their
comments at FCC and subsequent slides)
4
Concept of REM (Final)
• It is storage of needed/desired information built as
a dumb database with various intelligent query and
intelligence (autonomous) manipulation tools
5
Would you buy a used car from this man?
6
YES!
7
Algorithmic Tools Needed
• Spectrum Analysis Tools
• Signal Identification Tools
– Detect presence
– Classify or Identify Signal
• Aided greatly by a prior knowledge in the REM database
– Reduced time, energy, computational complexity of the system
needed to detect/identify/classify signals
– Able to identify licensed or known users in very weak signal conditions
based on a very reduced set of computational tasks
• Known and new statistical methods for signal classification
– Cyclostationarity based signal classification tools
– New method for computing Kurtosis of the signal to aid classification
8
More Algorithmic Tools Needed
• Wideband great for energy search
• Narrowband great for complex algorithms to work
– Polyphase Filterbank Channelizer (Analysis)
– Polyphase Filterbank Synthersizer and Arbitrary Rate
Resampler (Synthesis of new channels: harris, Rondeau,
McGwier)
• GnuRadio blocks done by Rondeau starting in 2008
• FPGA code done by Thaddeus Koehn, discussed at 11 AM on
Wednesday’s session and soon to be released to GnuRadio,
Ettus, and others
• Combined Wideband and channelization ideal
9
Kurtosis and classification
• Headley, McGwier, and Reed have shown that Kurtosis is a
very powerful tool in signal classification
• Kurtosis is a statistical measure and is fourth moment divided
by the variance squared.
• Take a sampled digital communication signal as a time series
and compute the kurtosis of the samples.
– The sample kurtosis is known for many constellation types under
the assumption of random data
– It is mostly insensitive to carrier offset and timing offset
– It is impacted by channel but, again, kurtosis aids us. Shalvi and
Weinstein have shown that a stochastic gradient process
minimizing kurtosis in a feed forward linear equalizer is THE
GLOBALLY OPTIMUM linear equalizer and on static or nearly static
channels, will converge. The kurtosis is that of the emitted
stream with the channel impairments greatly reduced.
10
Kurtosis as a tool
• Headley produced a GnuRadio block doing these
computations.
• I’m optimizing and using GSL for many of the
computations done by Headley in C++ and
replacing those with the library code since it is
already included.
• Examples are done in tool Headley has created.
• To be released after these modifications are done
and checked to GnuRadio and assigned to FSF.
11
Cyclostationarity Tools
• Gardner introduced us all to use of
Cyclostationarity for Signal ID, classification, etc.
• Reed and his students at VT have published several
important papers on using these tools and possibly
the most important for us is Kim, Reed, et. al. in
IEEE Dyspan in 2007.
12
Signal ID and REM
• We use these tools just covered (superficially) to verify
the contents of our current REM and to add new
unknowns to the REM
• When we have new unknown signals in the REM an
autonomous system so directed by priority scheduler,
will take up the task of further identifying the found
signal
• These techniques require computation engines and we
at VT and we know at Federated and Google they are
building towards use of distributed databases and
existing cloud for doing these computations.
MINIMIZES THE NEEDED INFRASTRUCTURE.
13
Without major computation?
• In mobile, dynamic, shared spectrum the database
is good for a few minutes or even a few seconds
• Not covered here because it is under active
investigation and will be used in commercial
offerings are the amazing, large volume of data
already existing in the “cloud” of available data
• We need distributed databases, sensors, and all of
the “data in the cloud” to do a really good job of
maintaining these dynamic REM’s in applications of
use to wireless users and providers.
14
BUT! Distributed, Mobile, Shared, ARGGG
• The CAP Theorem applies! You desire consistency,
availability, and partition tolerance in your distributed
database for REM
– You cannot afford to ship your “global database” to every sensor
in your network of sensors and some sensors won’t allow this
anyway. To the ones you can, to aid in the local production of
updates to REM, you partition the REM to the “relevant parts
near the sensor”.
• The CAP theorem says you cannot simultaneously guarantee
consistency, availability, and partition tolerance!
• What can you do? Demand EVENTUAL consistency since the
others are “non-negotiable” and design to minimize the lag
to consistency and the inevitable clashes that will arise for
distributed temporal consistencies.
15
Using the REM and some GR tools
• FCC releases Notice of Proposed Rule Making for
Spectrum Sharing of 3.4 GHz bands now held
exclusively in the USA by USG and in widespread use
by DOD on airplanes, ships, etc.
• NTIA and FCC propose HUGE restriction zones where
no sharing is to be allowed around the coastlines of
USA (removing 90+% of the available market!)
• Google, Federated Wireless, (Tim and Bob) undertake a
measurement campaign to show the exclusion zones
are ridiculous and are now the basis for public
comments submitted to FCC and available for
download at FCC or Federated Wireless web site.
16
Interference Map
• To prove the exclusion was ridiculous we built a large
DATABASE (REM) of signal measurements.
• Tim and Bob measured the radar strength
• Google and VT (not including Tim and Bob) measured
impact on extremely expensive LTE equipment.
• Tim and Bob showed that for these purposes using a
USRP B210 was a good as, and produced results
consistent with, the expensive equipment for
measuring the impact on LTE!
• So our used car salesmen was telling us the TRUTH!
17

similar documents