Technical Challenges for Cognitive Radio in the

TV White Space Spectrum

Stephen J. Shellhammer, Ahmed K. Sadek and Wenyi Zhang
Corporate Research and Development
Qualcomm Incorporated, San Diego, CA 92121, USA.
AbstractThe FCC recently issued the regulatory rules for
cognitive radio use of the TV white space spectrum. These new
rules provide an opportunity but they also introduce a number
of technical challenges. The challenges require development of
cognitive radio technologies like spectrum sensing as well as new
wireless PHY and MAC layer designs. These challenges include
spectrum sensing of both TV signals and wireless microphone
signals, frequency agile operation, geo-location, stringent spectral
mask requirements, and of course the ability to provide reliable
service in unlicensed and dynamically changing spectrum. After
describing these various challenges we will describe some of the
possible methods for meeting these challenges.
I. INTRODUCTION
Recently the Federal Communications Commission (FCC)
in the United States issued a report and order (R&O) which
permits cognitive use of the TV white space spectrum. White
space is the term used by the FCC for unused TV spectrum.
These new regulatory rules open up an opportunity to develop
new wireless networks to utilize this spectrum. This VHF
and UHF spectrum provides superior propagation and building
penetration compared to other unlicensed spectrum in other
bands like the 2.4 and 5 GHz bands. However, access to this
new spectrum also comes with some technical challenges. The
FCC rules specify a number of requirements on these cognitive
wireless network. The focus of this paper is to describe those
challenges and discuss some possible methods of meeting
those challenges.
A summary of the FCC rules is provided in Section II.
One of the issues that comes up when considering the use
of this spectrum is, How much spectrum is actually avail-
able? This question is addressed in Section III. One of the
cognitive radio technologies that is required to utilize this
spectrum is spectrum sensing, where the cognitive radio makes
observations of the RF spectrum and from those observa-
tions determines which of the TV channels is occupied and
which is unoccupied. The unoccupied channels represent white
space that can be used by the cognitive radio network. The
challenges for spectrum sensing are covered in Section IV.
Operating in a new frequency band with dynamic frequency
access requirements has an impact on the RF architecture.
These RF challenges are described in Section V. The second
cognitive capability required to operate in this white space
spectrum is location awareness, which is referred to as the
geo-location capability, and is described in Section VI. To
minimize any interference to licensed transmission on adjacent
channels and channels beyond the adjacent channel, the FCC
rules provide limits on out-of-band (OOB) emissions which
impact the spectral mask of any cognitive radio network using
this spectrum. The challenges of this strict spectral mask are
described in Section VII. Finally, though not a strict FCC rule,
it is important to provide reliable operation in these channels
given the unlicensed nature of these cognitive radio devices.
The ability to provide reliable operation is critical to the
success of any wireless network. The challenges of providing
reliable operation in this dynamic spectrum is described in
Section VIII.
II. FCC RULES
In November 2008 the United States FCC issued a R&O on
the unlicensed use of TV white space spectrum [1]. A number
of the requirements to operate in TV white space are based on
cognitive radio technology including location awareness and
spectrum sensing. There are a number of other requirements
also, that are intended to provide protection for the licensed
services that operate in the TV bands. These requirements
impose technical challenges for the design of devices operating
in TV white space spectrum.
The devices operating according to these rules are referred
to as TV band devices (TVBDs) by the FCC. There are two
classes of TV band devices: xed and personal/portable. To
simplify the terminology we will use the shorter term portable
for the personal/portable devices. The portable devices are
further divided into Mode I and Mode II devices.
Fixed devices are permitted to transmit up to 30 dBm (1
watt) with up to 6 dBi antenna gain, while portable devices
are permitted to transmit up to 20 dBm (100 mw) with no
antenna gain. Fixed devices are permitted to use a higher gain
antenna as long as the transmit power is decreased dB-for-dB
for any antenna gain above 6 dBi.
The TV channels include the very high frequency (VHF)
channels 2-13 and the ultra high frequency (UHF) channels
14-51. However, there are restrictions on which channels are
permissable for use by TVBDs. Fixed devices are permitted
in the VHF channels except channels 3-4 and on the UHF
channels except channels 36-38. Portable devices are not
permitted in the VHF band. Portable devices are permitted
on the UHF channels except 14-20 and channel 37. The
exclusion for channels 3-4 is to prevent interference with
external devices (e.g. DVD players) which are often connected
to a TV utilizing either channel 3 or 4. Portable devices are
not permitted on channels 14-20 since in 13 metropolitan areas
some of those channels are used for public safety applications.
Finally, Channel 37 is a protected channel, used for radio
astronomy measurements.
Television broadcast signals are protected with a protection
contour. The FCC rules provide distances that a TVBD must
be outside the protected contour for it to transmit. Within the
protected contour there are special rules for operation on a TV
channel adjacent to the TV broadcast channel. Fixed TVBDs
are not permitted to operate on channels adjacent to the TV
broadcast channel. Portable devices are permitted to operate
on an adjacent TV channel; however, when operating on
an adjacent TV channel, the maximum allowed transmission
power is 16 dBm (4 dB lower than on non-adjacent channels).
There are also requirements on antenna height for xed
devices. The sensing antenna must be mounted outside and
must be at least 10 meters above ground. The transmit antenna
must be outdoors and be mounted no more than 30 meters
above ground. So if the same antenna is used for both sensing,
receiving and transmitting, then that antenna must be between
10 and 30 meters above ground.
The FCC has strict out-of-band emission requirements to
prevent interference with licensed transmissions in other chan-
nels. A detailed description of these out-of-band emission
requirements and their impact on the transmission spectral
mask for TVBDs is provided in Section VII.
There are several interference avoidance mechanisms that
must be supported in these TVBDs. These mechanisms are
the cognitive radio capabilities provided in these devices. The
rst requirement is that these TVBDs must be location aware,
which in the FCC document is referred to as having a geo-
location capability. This location awareness is a cognitive
radio capability frequently mentioned in the literature. This
location awareness is coupled with Internet access capability
for both xed TVBDs and Mode II portable devices, and
must be accurate to within 50 meters. This Internet access
is utilized to obtain access to a database containing informa-
tion about licensed transmission in the various TV channels.
These licensed transmissions include ATSC (Digital TV) high-
power broadcasts, ATSC and NTSC (Analog TV) low-power
transmitter, and wireless microphones used by the broadcast
industry. The second interference avoidance mechanism is
spectrum sensing in which the TVBD must observe the various
TV channels and determine if these channels are occupied
by any licensed transmission. More will be said about the
challenges of spectrum sensing in Section IV.
III. WHITE SPACE AVAILABILITY
Fixed and portable devices have different requirements
among which is the separation distance from digital and analog
TV protected contours. In particular, xed devices are not
allowed to operate on rst adjacent channels to a TV station.
On the other hand, portable devices are allowed to operate
on rst adjacent channels subject to lower maximum transmit
power constraints (cf. Section II). This translates to different
white space availability for xed and portable devices.
Both xed and Mode II TVBDs are required to access
a TV band database in order to determine the permissible
set of operating channels. This step is followed by spectrum
sensing to conrm the emptiness of these channels. According
to the FCC rules, the TV database should contain the follow-
ing information on full-power television stations, digital and
analog Class A stations, low-power television stations (LPTV),
television translator stations, and television booster stations
transmitter coordinates (latitude and longitude),
effective radiated power (ERP),
height above average terrain of the transmitter (HAAT),
horizontal transmit antenna pattern (if the antenna is
directional),
channel number,
station call sign.
In this paper, white space availability is characterized based
on the information available on the Media Bureaus Consol-
idated Data Base System (CDBS)
1
for post-DTV transition
full-power television stations. The white space availability
in the fty most populous cities in the United States
2
is
characterized. In particular the following are calculated:
the histogram for the number of white space channels
available for xed device operation in the UHF band,
Figure 1, and
the histogram for the number of white space channels
available for portable device operation, Figure 2.
From Figures 1 and 2, the average number of channels
available for xed devices is ten, while the average number of
channels available for portable devices is twenty. On the low
side, there are four cities with only two channels available
for xed device operation while there are three cities with
six channels available for portable device operation. We note
that the calculation of white space availability in this paper is
based only on full-service TV stations database. In general
the presence of other primary incumbents such as LPTV
or wireless microphones can reduce the available spectrum.
The white space availability is different from one location
to another and hence frequency agile operation is desired to
efciently utilize the white space. Frequency agile operation
and the impact of different RF architectures on the available
spectrum is discussed in more details in Section V.
IV. SPECTRUM SENSING
A. Single-Node Sensing
In the FCC rules, the main requirement on spectrum sensing
is that a TVBD should be able to detect the presence of (digital
and analog) TV signals and wireless microphone signals at a
received power level 114 dBm. To understand this sensitivity
requirement, we note that the noise power level of a 6 MHz
TV channel is typically around 100 dBm (including a 6 dB
noise gure), and therefore the 114 dBm received power
level translates into a SNR of around 15 dB. Furthermore,
since for lower UHF TV band the effective antenna size is
reduced compared with higher RF frequencies, a phone-form
factor receive antenna will have a low antenna gain, say, 5
3 dB. Consequently, the spectrum sensing problem for each
1
http://www.fcc.gov/mb/cdbs.html
2
http://en.wikipedia.org/wiki/List of United States cities by population
Fig. 1. White Space availability for xed devices.
0 5 10 15 20 25 30
0
1
2
3
4
5
6
Histogram for White Space Availability for Portable Devices
Number of White Spaces for Portable Devices
N
u
m
b
e
r

o
f

C
i
t
i
e
s
Fig. 2. White Space availability for portable devices.
single TVBD becomes detecting the presence or absence of
certain target signals (digital/analog TV, wireless microphone)
across a bandwidth of 6 MHz, and at a SNR of 20 18
dB.
Before describing the challenges with spectrum sensing, it is
useful to briey introduce the three signal types to be sensed,
namely, digital TV, analog TV, and wireless microphone. The
digital TV in the United States follows the ATSC standard.
TV programs are modulated using 8-level vestigial sideband
modulation (8VSB) and the modulated signal occupies almost
the entire 6 MHz TV channel uniformly, with a pilot carrier
which contains approximately 7% of the total signal power
and is located at approximately 310 kHz above the lower
edge of the channel. Figure 3 displays the received power
spectral density (PSD) of a typical ATSC signal. The analog
TV in the United States follows the NTSC standard, and
may still be used for low-power local broadcasting after the
DTV transition. The luminance part of signal is amplitude-
3 2 1 0 1 2 3
10
0
10
20
30
40
50
P
S
D
(
d
B
)
f(MHz)
Power Spectral Density
Fig. 3. Received PSD of a typical ATSC signal. The scale of the y-axis is
not normalized. The pilot carrier exhibits like a narrow spike near the lower
edge of the TV channel.
3 2 1 0 1 2 3
20
10
0
10
20
30
40
50
60
P
S
D
(
d
B
)
f(MHz)
Power Spectral Density
Fig. 4. Received PSD of a typical NTSC signal. The scale of the y-axis is
not normalized. The video, color, and audio carriers are evidently noticeable
at their corresponding frequency locations.
modulated, and its (video) carrier, containing more than half
of the total signal power, is located at 1.25 MHz above
the lower edge of the channel. The color part of signal is
quadrature-amplitude-modulated with suppressed carrier, and
its subcarrier is approximately 3.58 MHz above the video
carrier. The audio part of signal is frequency-modulated, and
its subcarrier is 4.5 MHz above the video carrier. Figure 4
displays the received PSD of a typical NTSC signal.
Wireless microphones, as the most common Part 74 devices,
are allowed by the FCC [2] to transmit in TV channels which
are not occupied by ATSC/NTSC signals, provided that their
transmit power and bandwidth meet certain requirements. The
maximum bandwidth of a wireless microphone is 200 kHz.
Therefore, multiple such devices may operate within a single
TV channel, and their carrier frequency locations are not a pri-
ori xed, unlike carriers of ATSC/NTSC signals as described
in the previous paragraph. There is no specic requirement on
the modulation scheme for wireless microphones, whereas in
the TV band, analog frequency modulation (FM) is the most
common practice. Figure 5 displays the received PSD of a
typical wireless microphone signal within a TV channel.
3 2 1 0 1 2 3
30
20
10
0
10
20
30
40
50
60
P
S
D
(
d
B
)
f(MHz)
Power Spectral Density
Fig. 5. Received PSD of a typical wireless microphone signal. The scale
of the y-axis is not normalized. Note that the carrier frequency location may
change within the 6 MHz TV channel.
Now return to the spectrum sensing challenges. At SNR
as low as 20 dB, a number of challenges emerge. On
the one hand, it becomes unreliable and even impractical to
utilize coherent reception techniques and implement signature
sequence acquisition. On the other hand, energy detection,
which seeks to detect signal presence through comparing the
received power against a certain threshold, is known to be
fundamentally awed at low SNR, under uncertainty with
noise power level and distribution [3]. Alternative techniques
based on spectral correlation analysis [4] also have limited
applicability here. First, their performance rapidly degrades as
SNR decreases. Second, for FM wireless microphone signals,
the cyclostationarity is actually too weak to be exploited.
For ATSC/NTSC signals, spectrum sensing is a simpler task
compared with wireless microphone signals, which will be
elaborated shortly. The key observation is that, such signals
statistically do not resemble white Gaussian noise, rather they
exhibit narrowband features through the ATSC pilot, the NTSC
video carrier, etc. Furthermore, the frequency locations of
those narrowband features are xed, and the local SNR
in their vicinity can be boosted through appropriate ltering.
The spectrum sensing problem hence can be posed as a
binary hypothesis test between pure noise and a carrier with
unknown phase in noise; or, specically for NTSC, as a binary
hypothesis test between pure noise and a stochastic signal in
noise. Both of the two hypothesis test problems have been
well understood (see, e.g., [5]), and have been demonstrated
as effective using real signals collected in laboratory. As an
example, Figure 6 displays the performance of ATSC sensing,
using laboratory collected ATSC signal (as clean) from dual
antennas and passing through different channel models as
indicated in the gure. The false alarm rate (1%) is obtained
with respect to pure white Gaussian noise.
The main challenge in spectrum sensing is for wireless
microphone signals. Theoretically speaking, even though the
carrier frequency location of a wireless microphone signal is
unknown in advance, it is still possible to distinguish between
such a signal and pure noise, because as illustrated in Figure
5 the PSD of wireless microphone signals exhibits a narrow
28 27 26 25 24 23 22 21 20 19
10
5
10
4
10
3
10
2
10
1
10
0
ATSC Detection Performance
(4x6ms, Dual antenna correlation = 0.9, False alarm rate = 0.01)
SNR (dB)
P
m
is
s


ITU PedA
ITU PedB
IEEE 802.22
AWGN
Fig. 6. Spectrum sensing performance of ATSC signals.
3 2 1 0 1 2 3
15
20
25
30
35
40
45
50
P
S
D
(
d
B
)
f(MHz)
Power Spectral Density
Fig. 7. Received PSD of a typical TV channel with wireless microphone
signal and narrowband interference signals. The scale of the y-axis is not
normalized. The wireless microphone signal is at 1 MHz above the lower
edge of the TV channel, and the other spikes are from unknown sources.
spike which, with high probability, would not be produced
by white Gaussian noise. Indeed, laboratory tests have shown
that sensing techniques based on this idea can reliably detect
the presence of wireless microphone signals even as low as
20 dB SNR. This fact is not entirely surprising, because
a wireless microphone signal can at most occupy 200 kHz
within a 6 MHz TV channel, and therefore within the 200 kHz
bandwidth the SNR can at least be boosted by
610
6
20010
3
15
dB. The challenge, however, is that such narrowband spikes
are also observed in received signals without the presence of
wireless microphones. In other words, the spectrum sensing
problem is not the binary hypothesis between pure noise
and wireless microphone signal in noise, because the pure
noise hypothesis indeed contains narrowband interference.
To illustrate this, Figure 7 displays the received PSD of a
TV channel, which contains one wireless microphone signal
(located at 1 MHz above the lower edge, and with power
level approximately 105 dBm) and a number of narrowband
interference signals from unknown sources.
There are different possible sources of narrowband inter-
ference. Spurious emissions and unintentional transmissions
are universal from all types of electronic devices, and some
reside within the TV band. The FCC rules do pose stringent
limits on spurious emissions and unintentional transmissions
[6] such that they would not interfere communications. For
spectrum sensing, however, these interference signals turn out
to have strengths similar to the signals to be sensed, say,
wireless microphone signals below the noise oor. In addition,
other sources of narrowband interference include leakages
from adjacent TV channels, quantization noise from analog-to-
digital conversion (ADC), and RF impairments like nonlinear
inter-modulation (see, e.g., [7]).
From the preceding discussion, it is evident that spectrum
sensing of wireless microphone signals mandates a classica-
tion procedure, in order to expurgate narrowband interference.
Without doing so, it is likely that most of TV channels would
be decided to be occupied and consequently no TVBD would
be allowed to operate. There are two main technical hurdles
in the classication problem. First, there lacks a technical
standard among different wireless microphone manufacturers.
All design parameters (e.g., operational frequencies, FM fre-
quency deviation, side-tone placement, etc.) can drastically
vary among manufacturers and device models. Therefore, it
is difcult to abstract features that are stable and common
for all wireless microphones. Second, analog FM signals, in
their general form, is nothing but a carrier wave (CW) with
a gradually changing phase. If the phase changes too slowly
with time, say, when the input to a wireless microphone is
silent, then the resulting wireless microphone signal would
not be much different than CW-like spurious emissions.
Addressing the technical hurdles outlined in the previous
paragraph is both interesting and useful research. The spectrum
sensing problem may be most effectively solved from a pattern
recognition perspective; alternatively, the theory of robust
detection [8] may prove to be another useful tool. It is still
unclear how the challenge will be settled eventually, or if there
exist any fundamental performance limits. Nevertheless, the
bottom line is that, for spectrum sensing at low SNR, the
problem need be posed carefully in a way that does not over-
simplify the reality.
B. Networking Issues
In the previous subsection, we considered spectrum sensing
for individual TVBDs. When multiple TVBDs are connected
to form a network or several networks, networking issues of
spectrum sensing arise. One such issue, collaborative sensing,
has already received heightened interest recently; see, e.g.,
[9]. Roughly speaking, collaborative sensing seeks to apply
ideas from distributed detection and data fusion to jointly
process the spectrum sensing statistics from multiple TVBDs.
The inherent diversity stemming from distributed observations,
when exploited appropriately, leads to more efcient spectrum
sensing schemes compared with single-node sensing.
We briey outline another networking issue in the following.
Recall from the previous subsection that the sensitivity require-
ment of spectrum sensing is 114 dBm, or, 20 18 dB
SNR for typical phone-form factor TVBDs. Such a demanding
performance specication thus can only be met when the
co-channel interference from peer TVBDs is minimal. As
we have seen from the previous subsection, narrowband,
spike-like interferences pose the fundamental challenge for
identifying unoccupied TV channels. Furthermore, even if
TVBDs signals all have AWGN-like PSD, their existence may
still noticeably raise the effective noise oor. For a sensing
TVBD, if its co-channel peer TVBDs transmit thus producing
an interference power level comparable with the thermal noise
oor, then the overall interference-plus-noise would be raised
by approximately 3 dB, which in turn would degrade the
sensing sensitivity accordingly.
One possible way of handling the aforementioned issue is a
network quieting protocol, that mandates all the TVBDs in a
geographic area to simultaneously turn off their transmit cir-
cuitry, so as to quiet the TV channel for a certain time period
to sense the TV/wireless microphone signals. This idea bears
some similarity with the well-known request-to-send/clear-to-
send (RTS/CTS) mechanism in 802.11 networks, whereas the
purpose of RTS/CTS is to avoid interference-induced collision
in transmission. Since the tolerable interference power level
(received by sensing TVBDs) for ensuring reliable spectrum
sensing may need to be even lower than the noise oor, as
discussed in the previous paragraph, the cleared geographic
area provided by RTS/CTS may not be large enough.
The idea of quieting a geographic area further leads to
an important question, that is, how to dene a geographic
area. If a large number of TVBDs extend hop-by-hop to
cover a long strip of area (for example, Manhattan Island)
such that all TVBDs are connected, do we still need to
quiet all of them simultaneously? Quieting a large geographic
area containing many TVBDs possibly belonging to different
heterogeneous networks also requires a standardized protocol
to enable interoperability among heterogeneous networks, as
well as efcient and robust techniques for distributed network
synchronization.
V. RF ARCHITECTURE CHALLENGES
One of the challenging features of the white space is its
variation across space and time. More specically the available
channels are not contiguous and vary from one location to
another. In addition the white space available in a given
location can vary with time if one or more of the TV band
primary users start/stop operation. This requires frequency ag-
ile architectures to map to the available white space spectrum,
retune to a new operating channel, or tune-away to perform
sensing measurements. The requirements are more challenging
for frequency division duplex (FDD) networks that need two
separate channels for operation. In particular, some of the RF
challenges that need to be solved for FDD networks include
independent tuning of transmitter and receiver,
providing RF isolation (in the order of 50dB) between
transmitter and receiver for variable transmitter and re-
ceiver frequencies,
developing highly linear receivers over a wide dynamic
range to handle in-band high power TV broadcasts.
Among the possible approaches to implement the RF front
end for a frequency agile transceiver are: duplexers, switching
RF lter banks, and tunable lters. It is important to note
that white space availability may change according to the
duplexing method used: time division duplexing (TDD) or
FDD. In addition, different FDD RF architectures can result in
different white space availability. In the following we present
white space availability results for some case studies. Three
white space availability metrics are analyzed.
Maximum number of independent networks that can uti-
lize the white space without requiring co-channel sharing.
The minimum number of TV channels occupied by other
networks before co-channel sharing is required.
The average number of TV channels occupied by other
networks before co-channel sharing is required. The ran-
domness here is generated by assuming that any channel
of the white space can be lost with equal probability.
A. Time Division Duplexing
The computations for TDD is straightforward because each
white space represents an opportunity. Therefore, the three
metrics dened above can be actually reduced to only one
metric which is the number of available channels. This is not
true however for FDD networks as discussed in the following
sections.
B. Frequency Division Duplexing
Three different RF architectures for the FDD network are
considered.
The single duplexer approach in which the whole band
is divided into two subbands.
The dual duplexer approach in which the whole band is
divided into four subbands.
The tunable lter approach.
1) Single Duplexer Approach: FDD systems commonly use
a duplexer to isolate the transmitter band and the receiver
band. A typical single duplexer will divide the UHF spectrum
equally into two bands with few channels left as a band gap.
For example, band A contains channels from 14 to 31, band
B contains channels from 34 to 51 and channels 32 and 33
are left to serve as a duplexer band gap. If the white space
distribution in the UHF band was uniform, this design would
be optimal in terms of white space availability.
However, the white space distribution is in general nonuni-
form. For example, in the Bay area most of the white space
is in the lower part of the spectrum. Accordingly a uniform
single duplexer which divides the UHF band into two equal
bands will in general result in limited white space availability
especially in highly congested areas as the Bay area. In
general, an optimal single duplexer should divide the white
space equally in the two bands. Since the distribution of white
space differs from one location to another, it is not possible
to nd one optimal design that leads to the best results in all
locations. To overcome this problem, we study the tradeoff
that different duplexer designs present. A good design should
increase the white space availability in areas with very few
white spaces as the Bay area while not reducing white space
availability in other locations signicantly. To capture this
SD
0
Band A (14 to 31) and Band B (34 to 51)
SD
1
Band A (14 to 30) and Band B (33 to 51)
SD
2
Band A (14 to 29) and Band B (32 to 51)
SD
3
Band A (14 to 28) and Band B (31 to 51)
SD
4
Band A (14 to 27) and Band B (30 to 51)
SD
5
Band A (14 to 26) and Band B (29 to 51)
SD
6
Band A (14 to 25) and Band B (28 to 51)
TABLE I
DIFFERENT SINGLE DUPLEXER DESIGNS.
Design Mean: p = 1 p = 3 Minimum: p
SD
0
10.2 2.53 1
SD
1
10.3 2.54 1
SD
2
10.3 4.86 2
SD
3
10.1 6.66 3
SD
4
9.56 6.64 3
SD
5
9.12 6.54 3
SD
6
8.52 7.28 4
TABLE II
ANALYTICAL RESULTS FOR DIFFERENT SINGLE DUPLEXER DESIGNS.
tradeoff, the generalized average of the number of independent
networks in the fty largest cities in the United States is used
as a metric to compare different single duplexer designs. The
generalized average can be dened as follows
GA
p
=

1
n
n

i=1
x
p
i
1
p
. (1)
Negative values of p in (1) assigns more weight to smaller
values of x
i
. Different single duplexer designs are described in
Table I. The white space availability results for these designs
averaged over the largest fty cities in the Unites States are
shown in the Table II. From these results, SD
3
( Band A (14 to
28) and Band B (31 to 51)) captures a good tradeoff between
the minimum number of available channels and the average
white space availability.
2) Dual Duplexer Approach: A dual duplexer can be used
to increase white space availability for FDD networks. The
dual duplexer approach divides the whole band into four
sub-bands which results in a more exible RF architecture
compared to the single duplexer. One of the disadvantages of
the dual duplexer design is the increased band gaps required
to separate the four bands, therefore reducing the total white
space available. The design of the dual duplexer can be
optimized to increase white space availability and overcome
this disadvantage. One possible approach is to switch between
a set of single duplexers.
3) Tunable Filter Approach: Tunable lters are the most
exible RF architecture for an FDD network since the trans-
mitter and receiver can be independency tuned to different
channels. In order to achieve the required RF isolation between
the transmitter and receiver, a tunable LC lter with a high
inductor Q factor is required. High inductor Q factors can
be achieved, for example, through Q-enhancement techniques
[11]. For this case study, we assume a tunable lter with Q
equals 400 which represents an ideal scenario. To achieve the
required RF isolation between the transmitter and receiver, the
selected channels should be separated by at least 5 channels.
C. White Space Availability for Different RF Architectures
We analyzed the white space availability for three different
locations: San Diego, Dallas and the Bay area. Figure 8
compares the number of networks that can be constructed
using the different approaches. Figures 9 and 10 compare the
worst case and average white space availability results for the
different duplexing and RF architecture approaches.
The results show that TDD provides the most exible
duplexing approach in terms of utilizing the available white
spaces especially in congested areas as in the Bay area.
For areas with plenty of white spaces as in San Diego, the
performance of the single duplexer, dual duplexer and tunable
lter FDD designs are comparable. For such cases the single
duplexer would strike a good tradeoff between performance
and complexity.
VI. GEO-LOCATION
The FCC rules require both xed TVBDs and per-
sonal/portable TVBDs in operating Mode II be directly con-
nected through the Internet to incumbent databases. The prin-
cipal purpose of this requirement is to provide a mechanism
to inform TVBDs about their neighboring TV/wireless micro-
phone signals and peer TVBDs. Indeed, all TV broadcasting
stations (including low-power TV, translators, boosters, etc.)
are required to be archived in incumbent databases, and such
information essentially overrides the outcomes of spectrum
sensing for TV signals. That is, for a TVBD located within
the contour of a TV station as indicated in the geo-location
database, even if spectrum sensing reports that the TV channel
is unoccupied (say, due to shadowing), the TVBD still needs
to view this channel as occupied by TV.
Providing incumbent databases requires knowledge of the
locations of TVBDs themselves. The FCC rules specify a pre-
cision of 50 meters for TVBDs locations. For xed TVBDs,
their locations are manually set when they are installed. Since
their installation is thoroughly planned and performed by
professionals, obtaining their locations is by no means a
technical challenge. For personal/portable TVBDs, if global
positioning service (GPS) is equipped and TVBDs are outdoor,
obtaining their geo-locations may still be less a technical
challenge. If no GPS is available or if TVBDs are indoor, then
obtaining geo-locations becomes a challenging task. Reference
anchors, e.g., peer TVBDs with known geo-locations, may
not be accessible. Even if reference anchors are accessible,
distance ranging using physical-layer measurements may be
inaccurate, due to inaccuracy in anchors geo-locations as well
as channel effects such as shadowing and multipath fading, and
may be insecure, due to spoong and other possible attacks.
VII. SPECTRAL MASK REQUIREMENTS
In [1] the FCC has provided a number of out-of-band
emission requirements that impact the spectral mask of a
Number of networks without requiring co-channel sharing
0
5
10
15
20
25
30
35
Bay Area Dallas San Diego
TDD
FDD - Single Duplexer
FDD - Dual Duplexer
FDD - Tunable Filter
Fig. 8. Number of networks without requiring co-channel sharing.
Minimum Number of TV channels occupied by other White
Space Networks before co-channel sharing is required
0
5
10
15
20
25
30
35
Bay Area Dallas San Diego
TDD
FDD - Single Duplexer
FDD - Dual Duplexer
FDD - Tunable Filter
Fig. 9. Minimum number of TV channels occupied by other networks before
co-channel sharing is required.
Average Number of TV channels occupied by other White
Space Networks before co-channel sharing is required
0
5
10
15
20
25
30
35
Bay Area Dallas San Diego
TDD
FDD - Single Duplexer
FDD - Dual Duplexer
FDD - Tunable Filter
Fig. 10. Average number of TV channels occupied by other networks before
co-channel sharing is required.
TVBD. There are three sets of rules that impact the spectral
mask: adjacent channel rules, beyond adjacent channel rules,
and rules for Channels 36-38. Some of these rules are stated
in terms of spectral masks and other are in terms of the
electromagnetic eld strength measured at a specied distance
within a specied bandwidth. These eld strength rules must
be translated into spectral mask requirements.
The FCC rules state that In the 6 MHz channels adjacent to
the operating channel, emissions from TVBD devices shall be
at least 55 dB below the highest average power in the band.
These measurements are made using a minimum resolution
bandwidth of 100 kHz. This indicates that a spectral mask in
the adjacent TV channel is required to be at least 55 dB below
the maximum in-band signal.
The out-of-band emission requirements for beyond the
adjacent channel must satisfy FCC Section 15.209. In that
section of Part 15 it states that in the UHF frequency band the
electromagnetic eld strength measured in 120 kHz bandwidth
at a distance of 3 meters from the transmitter must be below
200 microvolts/meter. Since this is a xed eld strength
measurement in order to translate this into a spectral mask
requirement we will need to consider the transmission power
and antenna gain of the TVBD. First we must be able to
convert between electromagnetic eld strength and transmit
power.
To do this we begin with the formula for the eld strength
in terms of the transmit power. The eld strength at d meters
from transmitter, assuming free space path loss, is given by
[10],
FS = P + G+ 104.77 20 log(d) (2)
The eld strength (FS) is in dB microvolts/meter (dBu),
the transmit power (P) is in dBm, the antenna gain (G) is
in dB and the distance (d) is in meters. Once we calculate
the out-of-band transmit power in 120 kHz we must relate
that to the total transmit power to determine a spectral mask
requirement. If we let P
TX
be the total transmit power and
BW the bandwidth which captures the majority of the signal
power, then the spectral mask requirement can be written as,
=

P
TX
10 log

BW
0.12

P
oob
(3)
Now let us apply these equations to determine the spectral
mask beyond the rst adjacent channel. For d = 3 meters,
the 200 microvolts/meter give the following maximum out-of-
band transmission power, measured in 120 kHz, beyond the
rst adjacent channel,
P = 20 log(200) G104.77 + 20 log(3) (4)
= (G+ 49.21) (5)
To determine the spectral mask beyond the rst adjacent
channel we need to know the total transmit power, the antenna
gain, and the signal bandwidth. The transmit power and the
antenna gain are different for xed and portable devices. We
will use the maximum values allowed by the FCC. For xed
we will use a maximum transmit power of 30 dBm and an
antenna gain of 6 dBi. For portable devices we will use
Adjacent
Channel
Beyond
Adjacent
Channel 37
Fixed 55 dB 69 dB 95 dB
Portable 55 dB 53 dB 79 dB
TABLE III
SPECTRAL MASK REQUIREMENTS
maximum transmit power of 20 dBm and a 0 dBi antenna
gain. In both cases we will use a signal bandwidth of 5 MHz,
which is a reasonable choice since it uses the majority of the
6 MHz channel bandwidth but allows for some guard band.
For xed we have the out-of-band emissions are P
oob
=
(6 + 49.21) = 55.21dBm. Therefore, for a xed device
the spectral mask beyond the rst adjacent channel is,
=

30 10 log

5
0.12

(55.21) = 69 dB (6)
If we do the same calculation for portable devices we get
that,
=

20 10 log

5
0.12

(49.21) = 53 dB (7)
The third requirement is for the eld strength in Channel 37
and also the two channels adjacent to that channel, Channels
36 and 38. The eld strength limit is lowered from the bound-
ary between Channels 35 and 36 from 120 dBu measured at
1 meter, down to 30 dBu within Channel 37.
It is possible for a portable device to transmit in Channels 36
and 38 as long as they meet these eld strength requirements.
No matter which channel you operate in, the eld strength
from the out-of-band emissions in Channel 37 must be below
30 dBu, as measured in 120 kHz. The out-of-band transmit
power in Channel 37 is then,
P = 30 G104.77 + 20 log(1) = (G+ 74.77) (8)
For a xed device the spectral mask in Channel 37 is,
=

30 10 log

5
0.12

+ (6 + 74.77) = 94.6 dB (9)

For a portable device the spectral mask in Channel 37 is,
=

20 10 log

5
0.12

+ (74.77) = 78.6 dB (10)

We can summarize these spectral mask requirements in
Table VII.
To appreciate these spectral mask numbers one must con-
sider the typical spectral mask for an unlicensed system like
Wi-Fi. The OFDM PHY in 802.11 has a spectral mask that is
20 dB down at the edge of the channel. It linearly decreases
to 28 dB down at the middle of the adjacent channel, and
then reaches 40 dB down at the edge between the rst and
second adjacent channel. This is dramatically different than the
spectral mask required for even a portable TVBD. Figure 11
illustrates the portable TVBD spectral mask and the IEEE
802.11 OFDM PHY which has been scaled from 20 MHz
to 5 MHz. In this gure we only show the upper portion of
the spectral mask, above the carrier frequency. The mask is
...
-55 dBr
-53 dBr
-69 dBr
3 MHz 9 MHz 15 MHz
Adjacent
Channel
Second Adjacent
Channel
Channel 36
-20 dBr
-28 dBr
-40 dBr
802.11 OFDM PHY (scaled to 5 MHz)
Spectral Mask
TV White Space Portable Device
Spectral Mask
0 dBr
Invisible Text
Channel 37
Fig. 11. Comparison of Spectral Mask for portable TVBD and 5 MHz
version of IEEE 802.11
...
-55 dBr
-69 dBr
-95 dBr
3 MHz 9 MHz 15 MHz
Adjacent
Channel
Second Adjacent
Channel
Channel 36
-20 dBr
-28 dBr
-40 dBr
802.11 OFDM PHY (scaled to 5 MHz)
Spectral Mask
TV White Space Fixed Device
Spectral Mask
0 dBr
Invisible Text
Channel 37
Fig. 12. Comparison of Spectral Mask for xed TVBD and 5 MHz version
of IEEE 802.11
symmetric about the carrier frequency. This type of scaling
can be accomplished in an OFDM system by reducing the
clocking rate by a factor of 4. We see that the 802.11 spectral
mask is not close to attaining the required TVBD spectral
mask. Similarly, in Figure 12 we show the spectral mask for
a xed TVBD along with the frequency scaled version of
the 802.11 OFDM PHY. We see that the discrepancy in two
spectral masks is even more pronounced in the case of xed
TVBDs.
There are number of straightforward techniques that can
be used to improve the spectral mask, like increasing the
power amplier (PA) back-off, which will improve the OFDM
spectral mask, but that alone will not meet these spectral mask
requirements. Also, since this is a frequency agile system if
one were to apply RF ltering to the signal after the PA, that
ltering would need to be both frequency agile but also very
linear.
VIII. SPECTRUM SHARING AND INTERFERENCE
MANAGEMENT
Beside satisfying the FCC requirements for cognitive op-
eration in the TV band, it is desirable to achieve reliable
communications and acceptable performance levels in the
white space and efciently utilize the available spectrum.
One of the major challenges that face reliable operation in
the white space is interference among peer TVBDs given
the unlicensed nature of operation in this band. Managing
interference between nodes in the same network is generally a
difcult problem, and the problem becomes more challenging
when these TVBDs belong to heterogeneous networks using
different air interfaces.
A. Spectrum Sharing Approaches
Spectrum sharing techniques can be divided into three main
categories as follows.
Non-cooperative techniques
Rule-based techniques
Message-based techniques
1) Non-cooperative Techniques: In non-cooperative spec-
trum sharing techniques each system tries to maximize its own
utilization of the spectrum while mitigating the effects of in-
terference from other networks. Examples for non-cooperative
techniques include
Dynamic frequency selection (DFS) in which nodes select
channels with least interference.
Multichannel-DFS (M-DFS): In case of a one-to-many
system, the access-point can select a subset of the avail-
able channels to minimize the overall outage probability
and improve system throughput. DFS is a special case
of M-DFS when the access-point is restricted to use one
channel.
Interference cancellation via multiple antenna receivers.
If the receiver is equipped with multiple-antennas, in-
terference cancellation algorithms can be utilized. For
example, an MMSE receiver can help canceling the
incoming interference signal through estimating the inter-
ference covariance matrix and nulling out the dominant
interference signal.
Successive interference cancellation (SIC) in which the
access-point decodes the dominant interference signal
generated by other networks and cancels it out from the
received signal. There are several technical challenges in
order to implement such a technique. First it requires
a complex receiver capable of decoding different wave-
forms and signals received from different technologies.
Second the networks sharing the channel are in general
asynchronous. Some of these networks can be frame
based as in cellular air interfaces while others can be
packet based as in Wi-Fi. The packet sizes can be varying
as in Wi-Fi for example. Channel estimation is required to
enable SIC and this requires the ability to estimate timing
and frequency offsets and capture the pilot signals from
other networks using different air interfaces.
2) Rule-based Techniques: In rule-based techniques the
TVBDs agree on a set of rules to implement, and do not use
a control channel to exchange information. Although there
is no explicit coordination between the nodes, the set of
rules should be designed to ensure fairness, efciency and
interference avoidance. Examples for rule-based techniques are
listen-before-talk and transmit power control.
3) Message-based Techniques: In message-based sharing
techniques, the TVBDs operating in the unlicensed band ex-
change messages (over-the-air or backhaul) to enable efcient
and fair spectrum sharing and limit the interference among
them. A message-based sharing technique should specify how
to implement the control channel required for coexistence
NetA
NetB
Fig. 13. Example of a coexistence scenario in which NetA and NetB are
deployed as micro-cellular networks.
message exchange, and what information to share with other
TVBDs. Among the techniques proposed in the literature for
message-based sharing are [12], [13] and [14]. In [12] a
common spectrum coordination channel (CSCC) is proposed
in which a common channel at the edge of the used band
is reserved for sending coexistence information such as the
node ID, center frequency, bandwidth, transmit power, data
rate, interference margin etc., which can be used by neigh-
boring nodes to coordinate and adapt their transmission for
better sharing. A regional spectrum broker (spectrum clearing
house) is proposed in [13], [14] in which a spectrum clearing
house coordinates spectrum allocation to the existing service
providers based on the bandwidth required by each one of
them.
The downside with message-based protocols is that they
require the systems to have a common air-interface capable
of decoding the common control message. Moreover, some of
the above protocols require synchronization between different
systems. This might not be feasible for heterogeneous TVBDs.
B. Coexistence Scenarios
Different network deployment models can occur in the
white space. One can generally divide the deployment models
into planned deployment and unplanned deployment. Planned
deployments may include xed nodes that represent access-
points, basestations or xed customer premise equipment
(CPE) and portable/personal devices. Examples for planned
deployment networks include micro-cellular networks and
IEEE 802.22 wireless regional area networks (WRANS).
Unplanned deployments generally consist of user deployed
networks as wireless local area networks (WLANs) and Femto
cells.
Figures 13 and 14 depict two possible coexistence scenarios
where two micro-cellular networks are sharing the spectrum
in Figure 13, and a micro-cellular network is sharing the
spectrum with a WLAN in Figure 14. Other coexistence
scenarios are also possible.
Next we present some simulation results for non-cooperative
sharing techniques applied to the coexistence scenario depicted
in Figure 13. NetA is assumed to use an FDD cellular air-
interface and NetB is using Wi-Fi. The interference channel
model for Net A uplink and downlink are shown in Figures
15(a) and 15(b), respectively.
Wi-Fi AP Wi-Fi Station
Fig. 14. Example of a coexistence scenario in which NetA is deployed as a
micro-cellular while NetB is deployed as a WLAN.
(a) UL Interference Channel
(b) DL Interference Channel
Fig. 15. Interference channel model for NetA.
Figure 16 depicts the cumulative distribution function
(CDF) of the DL throughput achieved by NetA assuming 5
channels are available. From the results, it is clear that both
DFS and M-DFS can achieve a performance that is very close
to the scenario when NetA exists alone, i.e. no interference
from NetB. The rationale for this good performance is the low
interference regime that can be achieved in a low frequency
reuse system, which is achieved here by the distributed DFS
and M-DFS techniques.
The problem is more challenging in the uplink since the
link gain between the interfering access-points from NetB to
NetA access points can be higher compared to the uplink gain
in NetA. This is due to the following. Fixed access points can
transmit at higher power levels compared to portable devices
as discussed before in Section II. The antenna heights for
the access-points are larger than those for portable devices
resulting in lower RF propagation losses. Finally portable
0.5 1 1.5 2 2.5 3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
C
D
F
Throughput (Mbps)


Throughput No WiFi
Throughput DFS
Throughput MDFS
Fig. 16. DL Throughput for Network A. 5 channels available.
0 0.5 1 1.5 2 2.5 3 3.5 4
x 10
6
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Throughput (Mbps)
C
D
F


No WiFi
MF
MMSE
MMSE+SIC for Dominant WiFi Int
MMSE+SIC for Dom Coc Users then Dom WiFi Int
MMSE +Genie SIC for Dominant WiFi Int
Fig. 17. UL Throughput for Network A. 5 channels available, 2Rx antennas.
devices can be indoor which results in additional building
penetration losses.
The uplink throughput performance for several interference
mitigation techniques are depicted in Figure 17 where two
receive antennas are assumed at the access-point: a matched
lter receiver, an MMSE receiver, a SIC receiver which tries
to decode dominant NetB interference rst then decode NetA
users, a receiver that rst decodes some of the NetA users with
good link qualities followed by decoding dominant NetB inter-
ferer. To be able to decode dominant interference from NetB
the channel capacity between the interferer and NetAs receiver
must be higher than the rate transmitted by the interferer which
depends on the link quality of the interference source to its
intended receiver. The probability of satisfying this condition
is generally low and therefore almost no gain is seen from
such technique. An upper bound on the performance of the
SIC receiver is depicted in which the dominant interference
signal is cancelled out from the NetA receiver path by a genie.
The gain that can be seen from such genie scheme conrms
that the minor gain achieved by SIC is because of the low
probability of the event that the NetB signal is decodable.
IX. CONCLUSIONS
The new FCC rules allowing the use of the TV white space
spectrum will become one of the key drivers in the devel-
opment of dynamic spectrum access, fostering revolutionary
wireless applications, and meanwhile driving novel technical
solutions. As we have seen in this paper, many of the technical
challenges are unconventional and interdisciplinary in nature.
For example, the development of spectrum sensing techniques
involves RF design, robust signal processing, pattern recogni-
tion, networking protocols, etc.. The choice of RF architecture
is no longer merely a hardware issue, but will directly affect
the upper layer performance. To meet the stringent spectral
mask requirements, novel signal processing designs such as
transmit precoding and receive equalization become possibly
indispensable tools. Spectrum sharing raises new challenges in
applying techniques from information theory to heterogeneous
networks. We expect that recognizing those challenges will
lead to a fruitful interaction among academia, industry, and
policy makers, and nally lead to the success of cognitive
radio in the TV white space spectrum.
REFERENCES
[1] Second Report and Order and Memorandum Opinion and Order In the
Matter of Unlicensed Operation in the TV Broadcast Bands, Additional
Spectrum for Unlicensed Devices Below 900 MHz and in the 3 GHz
Band, Federal Communication Commision, Document 08-260, Nov. 14,
2008.
[2] FCC 47 CFR, Part 74, Subpart H: Low Power Auxiliary Stations
[3] R. Tandra and A. Sahai, SNR Walls for Signal Detection, IEEE Journals
of Selected Topics in Signals Processing, Vol. 2, No. 1, pp. 4-17, Feb.
2008
[4] W. A. Gardner, Spectral Correlation of Modulated Signals: Part I -
Analog Modulation, IEEE Transactions on Communications, Vol. 35,
No. 6, pp. 584-594, Jun. 1987
[5] H. V. Poor, An Introduction to Signal Detection and Estimation, 2nd ed.,
Springer-Verlag, New York, 1994
[6] FCC 47 CFR, Part 15: Radio Frequency Devices
[7] B. Razavi, RF Microelectronics, Prentice Hall PTR, 1997
[8] S. A. Kassam and H. V. Poor, Robust Techniques for Signal Processing:
A Survey, Proceedings IEEE, Vol. 73, No. 3, pp. 433-481, Mar. 1985
[9] A. Ghasemi and E. S. Sousa, Collaborative Spectrum Sensing for Op-
portunistic Access in Fading Environments, Proceedings IEEE DySPAN,
Nov. 2005
[10] T. Rappaport, Wireless Communications: Principals and Practice, Pren-
tice Hall, 1996.
[11] P. Lu; C. Wu; M. Tsai VHF/UHF high-Q bandpass tunable lters
design usingCMOS-inverter-based transresistance ampliers, IEEE In-
ternational Symosium on Circuits and Systems, June 1994.
[12] X. Jing and D. Raychaudhuri, Spectrum Coexistence of IEEE 802.11b
and 802.16a networks using reactive and proactive etiquette policies,
Mobile Net. Appl., 2006.
[13] J. Acharya and R. D. Yates, A framework for dynamic spectrum sharing
between cognitive radios, ICC 2007.
[14] S. Gandhi, C. Buragohain, L. Cao, H. Zheng, and S. Suri, A general
framework for wireless spectrum auction, Dyspan 2007