Download Cognitive Radio

A Proposed Means With Which to Address the Hidden Node Problem and Address Security
Issues in Cognitive Radio
Cognitive Radio (CR) is a radio technology that refers to the ability of a radio device to
sense/learn the communication parameters of its environment and to adapt its transmissions
accordingly. The ability of a radio device to do this with sufficient sophistication will enable
such devices to transmit in underutilized licensed bands without affecting the communications of
the licensee. The Shared Spectrum Company, who famously measured over 80% underutilization
of the spectrum (30-2900MHz) in New York during the Republican Convention in 2004 and
dispelled the myth that the spectrum is ‘crowded’, has measured similar underutilization in other
locations since then.
The value of this technology is enormous since it opens up large portions of the valuable
communications spectrum for use by such devices. CR technology will enable the continuing
worldwide exponential growth and innovation in wireless communications. It is a technology
currently in its nascent stage with the FCC very recently making available a 16MHz portion of
the analogue TV spectrum at 700Mz for, among other purposes, CR. Though there is currently a
growing body of literature on the topic it is currently speculative and consensus on how
technichal challenges relating to the deployment of CR ought to be addressed has not been
reached though the current draft IEEE 802.11 WRAN standard has served to bring these issues
into focus.
There are however specific technical challenges that must be overcome. Chief among these is the
ability of the cognitive radio to adapt its transmissions with sufficient agility so as not to
adversely affect the communications if the incumbent licensees or Primary Users (PUs). There
are a number of proposals which aim to detect PU transmission even at very low power and on
that basis avoid (CR) transmission in that band (matched filter detection, cyclostationary feature
detection etc.). However, apart from adding complexity to the cognitive device none have solved
the ‘hidden node’ problem which is the presence of a PU receiver in the region covered by the
CR transmission, albeit the CR not having detected a PU signal because of its low power. The
‘hidden node’ problem must be overcome to guarantee non-interference with PU signals.
Furthermore, the chosen link must be stable for effective CR communications. This requires
means with which to predict link stability based on spectrum usage patterns such that the
scenario of the CR initiating transmission only to be shortly afterward forced to terminate due to
the sudden presence of a PU signal is avoided. It is proposed to address the ‘hidden node’
problem through REM.
‘The REM is an abstraction of real-world radio scenarios; it characterizes the radio environment
of cognitive radios in multiple domains, such as geographical features, regulation, policy, radio
equipment capability profile and radio frequency emissions’ [1] – in short it is a map of ‘what’s
going on’ in the radio environment and it is by means of reference to this map that CR would
base its decision on whether or not to transmit. The REM would be hosted in a powerful server.
The effect of CR transmissions can be predicted by the REM and so resolve the ‘hidden node’
problem. Spectrum usage patterns can be used by the REM to predict link stability. This
information can be communicated to the CR over a pilot channel. The radio emissions portion of
the REM is initially based on signal strength predictions and updated with spectrum readings
from CR devices and/or monitoring stations so that it provides a real-time map in space and
frequency. The REM offers other advantages too. Relevant portions of the REM (based on GPS
position) can be periodically downloaded greatly relaxing the technical requirements for
sophisticated spectrum sensing capability and computational resources in the CR making these
devices much cheaper. The REM would act as broker in spectrum trading transactions and
enforce regional radio emissions regulations. The REM is an attractive solution on which to base
CR on technical, regulatory and commercial grounds but crucially also in addressing the
‘Hidden-Node’ problem.
To address the ‘Hidden-Node’ problem a highly accurate propagation predictor can be used
which would run on the REM server ( - be it global or local) or partly on the REM server and
partly on the CR device. by accuratley predicting those regions where the PU signal is likley to
be below the detection threshold of CR devices. In this regard the ‘Hidden-Node’ problem may
be resolved by deploying sensors in appropriate locations in these regions which would relay
aquired data back to the REM server. It is in this sence that proposed solutions such as matchedfilter detection and cyclostationary feature detection could come into play rather than equipping
every CR with such technology.
The second, but equally crucial facet in addressing the ‘Hidden-Node’ problem is determining
location-based CR transmission power limitations which is why the prediction tool used must not
only be very accurate but also extremely fast to facilitate reasonable call set-up times.
Since the successful deployment and operation of cellular telephone networks in the eighties and
nineties in liscenced bands propagation studies have fallen out of favor with the research
community with cellular operators settling largely on empirical path-loss models and/or basic
diffraction models with which to plan their networks. The reasons for this are that the state of the
art at the time were either not easily implemented or computationally expensive and/or
demanded expensive topological databases. Planning gains achieved by using more sophisticated
algorithms for these networks are generally considered marginal.
Since it is unfeasable to drive-test all areas due to economic constraints and due to the fact that it
can only be performed on roads (or in limited off-road conditions), deterministic propagation
techniques which accurately predict the large-scale fading signal must be the mainstay for the
purpose of addressing the ‘Hidden-Node’ problem as outlined here. The impact of clutter is most
probably best addressed using statistical techniques with determining the Ricean K-Factor or
Nakagami m-Parameter of prime importance here. It is of note that this is also achievable for
broadband signals [ ]. The purpose here is not only to address the ‘Hidden Node’problem but to
do so in a quatifiable manner, establishing precise error margins based on worst-case propagation
conditions of the PU signal, for the purposes of allaying the legitimate concerns of incumbents
and regulators on the deployment of CR technology.
Integral Equation (IE) methods are a promising approach in this regard and have received some
attention in the ninties.
Electromagnetic field coverage is expressed exactly as an IE. Elements of this equation can, a
priori, be eliminated by virtue of their negligible contribution. It is this feature of the IE
formulation that makes it a suitable environment for finding optimum computational methods.
However IEs have a well-documented complexity with a quadratic dependence on the total
number of integration intervals. It is well-known [3] that these intervals need to be subwavelength in dimension to ensure convergence which results in prohibitively large computation
times. For this reason IE methods have been largely confined to electrically small and medium
sized bodies.
Subsequent development in propagation modeling using integral equations has yielded a robust,
highly accurate, rapid and easy-to-implement means and has negligible setup-times and memory
requirements (other than topological data) by which to obtain the large-scale-fading signal over
electrically massive scatterers [1]. A standard deviation of ~9dB from measurements taken over
11Km of semi-rural terrain was achieved in 0.05sec using a standard desktop and a 50m
resolution database.
Coupled with the now readily available highly resolute digital terrain databases, greater
processor speeds and clutter and permittivity data, this work is ideally placed for development to
effectively address the ‘Hidden-Node’ problem as outlined above.
The method in question is very briefly outlined here and the reader is referred to [] for further
details. For simplicity the problem is treated as two-dimensional TMz, the surface is taken to be
a perfect electrical conductor (PEC) and forward scattering is assumed – that is all radiation is
taken to propagatate away from the transmitter. The latter two assumptions are justifiable in the
case of grazing incidence of transmitter radiation which is predomininantly the case for the
Danish profiles examined here.
The surface is impinged by a monochromatic TM Z polarized cylindrical wave of wave number
 emanating from an infinite, unit current carrying wire of negligible cross-section, placed a
distance above and transverse to the terrain profile. A time variation of e jt is assumed and
suppressed.
An electric current J is induced on the surface, which satisfies the EFIE [18], [19]:
E (r ) 

4
 J (r ' ) H
S
( 2)
0
(  | r  r ' |)dr ' .
(1)
r and r ' are vectors whose end-points are respectively the scattering and receiving points s  S .
E (r ) is the source electric field incident on the surface at the point given by r.
 is the wave impedance of the medium through which the radiation propagates and H 0( 2) is a
zero order Hankel function of the second kind which is the Green’s function for the problem.
The surface is discretised into N equal sized sampling intervals of length s with centre-points
indicated by the vectors ri and r j depending on whether they are scattering or receiving
intervals respectively. Using the Method of Moments [20] with unit pulse basis functions and
Dirac-delta weighting functions we get the following matrix relation:
E  ZJ
(2a.)
where
Ei  E (ri )
Z ji  s

4
H 0(2) (  | rj  ri |)
 
 1.781s  
Z jj  s

1  j ln 
4 
 
4e

J j  J (rj ).
2
(2b.)
E and J are column vectors of length N . Z , known as the impedance matrix, is N  N and
symmetric. The elements in the strictly lower triangle of Z correspond to forward-scattering and
those in the strictly upper triangle to back-scattering. The diagonal elements correspond to the
self-interaction of the sampling intervals. On the assumption of forward scattering, which is
equivalent to setting the strictly upper triangular elements of Z to zero, J is determined by
forward substitution:
j i
Ei   J j Z ji
for i  1.....N .
(3)
j 1
The order of complexity of determining J is O( N 2 ) . The total field at points above above the
surface is then the sum of the field from the source and the field scattered by the surface [18].
Using the notation of [22] and [17], the surface is divided into groups each containing M sampling intervals. There
are then N / M such groups.
Under two central assumptions: 1) The induced surface surrent is sinusoidal with an a-priori known envelope (e.g.
Rayleigh distributed) and the phase shift for each group can be forced to zero () can be manipulated such that the
following equation ensues:
Equation (10) is rewritten:
El  K  J l ' Zl 'l  J l Zll ,
l ' l
(11)
where
K  1 
jGl
e
 j s j
Z jl
Z ll
(12)
is approximately constant
Gl and consequently needs only to be evaluated once thereby obviating the need for
time-consuming group-specific aggregation or disaggregation stages [14]-[16], [22]. Equation (11) takes the form of
(2) and use of the latter over the former results in a reduction in the complexity from
O( N 2 ) to O(( N / M ) 2 )
and a reduction in memory requirements from O(N ) to O ( N / M ) . XXX This represents a 10**XXX reduction
…. XXX
The total field at points
l ' t
t above the surface is determined using a similar analysis giving:
E ttotal  E t  K  J l ' Z l 't
l '1
for t  1,2,......,
N
.
M
(13)
Comments:
The reason 10m groupsizes were chosen here is to provide as accurate as possible comparison
with measurements which were taken every 10m in []. Groupsizes up to about 50m yield only a
minor reduction in accuracy for these types of profiles – gently undulating terrain.
Backscattering effects on these types of profiles are also small. Smaller groupsizes must be used
for more rugged profiles and the effects of backscattering in these types of profiles is current
work in progress though preliminary results would seem to indicate indicate that these are not
great and would be supported by the fact that there would be significant absorption in this case.
Sidescattering is however a different matter since the grazing incidence phenomenon again
comes into play and this effect must be investigated thoroughly before the desired (time-saving
over 3-D) 2-D algorithm implemented in radial (or partially radial fashion for directional
antennas) fashion about the transmitter can be relied upon.
The PEC model for terrain, which has been shown by [] to be a valid one for gently undulating
terrain at grazing incidence cannot be taken for granted when dealing with more rugged profiles.
For this reason the FEM is currently being applied to the CEFIE.
Estimation of K in partially occluded regions such as in partially wooded regions is problematic.
It may be possible to address this problem with good accuracy using a ‘stochastic’ value for the
electrical permittivity based on clutter data. If successful this should eliminate the need for
empirical results (for example: a 5dB loss due to a cluster of trees) which are difficult to apply
with sufficient accuracy in different situations.
More accurate measurements using high resolution topological databases. Clutter and
permittivity databases.
Future Research:
More accurate solutions, algorithm modification, Estimation of K, forward backward, impact of
side-scattering, estimation of K to determine the distribustion of the envelope by which further
accuracy can be achieved and by which more precise error bounds can be determined – this can
be done for broadband signals too[], FEM applied to CFIE.CEFIE/CEFIEA, frequency selective
effects, time variation, weather, tropospheric effects. Precice accuracy requirements. Confidence
intervals. Best and worst case propagation conditions. Determining M.