Cognitive Radio Networks: Realistic or Not?

Konstantinos Pelechrinis Prashant Krishnamurthy

Martin Weiss
Taieb Znati
University of Pittsburgh University of Pittsburgh University of Pittsburgh University of Pittsburgh
[email protected]
[email protected]
[email protected]
[email protected]
This article is an editorial note submitted to CCR. It has NOT been peer reviewed.
The authors take full responsibility for this articles technical content. Comments can be posted through CCR Online.

ABSTRACT

1.

A large volume of research has been conducted in the cognitive radio (CR) area the last decade. However, the deployment of a commercial CR network is yet to emerge. A
large portion of the existing literature does not build on real
world scenarios, hence, neglecting various important aspects
of commercial telecommunication networks. For instance, a
lot of attention has been paid to spectrum sensing as the
front line functionality that needs to be completed in an efficient and accurate manner to enable an opportunistic CR
network architecture. While on the one hand it is necessary
to detect the existence of spectrum holes, on the other hand,
simply sensing (cooperatively or not) the energy emitted
from a primary transmitter cannot enable correct dynamic
spectrum access. For example, the presence of a primary
transmitters signal does not mean that CR network users
cannot access the spectrum since there might not be any
primary receiver in the vicinity. Despite the existing solutions to the DSA problem no robust, implementable scheme
has emerged. The set of assumptions that these schemes are
built upon do not always hold in realistic, wireless environments. Specific settings are assumed, which differ significantly from how existing telecommunication networks work.
In this paper, we challenge the basic premises of the proposed schemes. We further argue that addressing the technical challenges we face in deploying robust CR networks
can only be achieved if we radically change the way we design their basic functionalities. In support of our argument,
we present a set of real-world scenarios, inspired by realistic settings in commercial telecommunications networks,
namely TV and cellular, focusing on spectrum sensing as
a basic and critical functionality in the deployment of CRs.
We use these scenarios to show why existing DSA paradigms
are not amenable to realistic deployment in complex wireless
environments. The proposed study extends beyond cognitive radio networks, and further highlights the often existing gap between research and commercialization, paving the
way to new thinking about how to accelerate commercialization and adoption of new networking technologies and
services.
Categories and Subject Descriptors: C.2 [Computer Communication Networks], C.2.0 [General]: Data Communications, C.2.1 [Network Architecture and Design]: Wireless Communication
General Terms: Design, Management
Keywords: Cognitive radio networks, Spectrum sensing,
TV bands, Cellular networks

The increasing demand for wireless connectivity has shifted

the attention and efforts of many researchers all over the
globe towards the opportunistic dynamic spectrum access
paradigm. This concept is not new, and was first introduced
by Mitola [1]. In brief, the idea is that licensed spectrum can
be made accessible to unlicensed users (secondary) when
the licensed (primary) entities are absent. This absence
of a primary user from a specific frequency band at a given
point in time and space is referred to as spectrum hole [2]; a
reserved portion of the spectrum that is not in use. In other
words, a spectrum hole is a function of frequency, time and
location.
During the last years there have been significant advancements in hardware technology, enabling engineers to build
radios that can understand their environment and dynamically alter their transmission parameters (e.g., transmission
frequency, modulation type etc.). One would have expected
that such developments would have lead to large scale cognitive radio network deployments. However, this is not the
case, and even a prototype large-scale deployment is yet
to appear. In this challenge paper, we argue that this is
largely due to the specific mindset we have when we consider
the research and design of protocols for CR networks. The
majority of the work in this area is theoretical and makes
a number of assumptions that may not hold in practice.
Even though these studies are arguably important and can
provide scientific insights, they are not the best avenues to
drive practical implementation and eventually commercial
adoption and success. As we will elaborate in the following
sections, specific assumptions that are prevalent in the literature can either expose the primary receiver to harmful
interference or limit the performance of a CR network.
Much of the existing literature focuses on the detection of
signals at a given location from a single primary transmitter
using threshold schemes. If the received signal from this primary transmitter is below a predefined threshold, the band
is declared to be vacant. More often than not, a single transmitter/receiver is assumed as the secondary network. As
we will see in the following section the vast majority of the
studies in this area, are minor variations of the above idea.
The main objective of the different schemes is to reduce the
sensitivity requirement at the cognitive radio sensor that
determines the absence of primary transmissions.
Furthermore, another feature of the existing literature is
that it tends to treat the entire frequency spectrum in a unified way. That is, solutions are not different depending on
the frequency bands to be used by the CR network. How-

ACM SIGCOMM Computer Communication Review

44

INTRODUCTION

Volume 43, Number 2, April 2013

ever, the nature of the licensed technology in different parts

of the spectrum can be significantly different, therefore, requiring very different considerations. For instance, television is a broadcast system with passive receivers, while cellular networks typically include bi-directional asymmetric1
transmissions. Such very different types of primary services
create very different kinds of spectrum holes as well, which
require different treatment.
A recent study by Weiss et al. [3] has classified the resources for a CR network, that is, spectrum holes, based on
their characteristics both in space and time. Across time,
spectrum holes can be static, periodic or stochastic, while
across space they can be contiguous or non-contiguous. The
most simple type of spectrum holes that can be used is the
temporarily static and spatially contiguous spectrum holes.
TV white spaces fall into this category. In particular, after the migration from analog to digital TV, there are unused bands (frequencies 54-698 MHz, which correspond to
TV Channels 2-51), and the first coordinated effort to utilize these frequencies for cognitive radio communications has
emerged through various proposals (e.g., IEEE 802.22 and
the White Spaces Coalition [4]). Even though these bands
can still be used by unlicensed but authorized users (e.g.,
microphones and medical telemetry equipment), a temporal
static, spatially contiguous spectrum hole can be discovered
fairly easily. In particular, a database that includes all the
information for unused analog TV bands across various locations suffices. In this way, the discovery process takes
place offline and does not require the deployment of complex online sensing techniques. It is interesting to see that
the one and only effort to commercialize CR networks presented above, i.e., 802.22, does not rely on any spectrum
sensing scheme proposed in the research literature. On the
contrary, it utilizes a simple, centralized solution. At first
glance, it might not appear as the most appealing approach
(e.g., from the perspective of its research novelty). However,
it is one that can be easily applied and can drive a successful
deployment.
Using two representative examples of primary user technologies, that of television and of a cellular network, we argue that current research proposals fail to address system
level questions; are the spectrum holes we can identify
with existing algorithms really useful? Why does the
presence of a primary signal necessarily render the
frequency unusable? How can we identify the regions
where passive receptions of a broadcast system exist?
How can we take advantage of the different uplink and
downlink frequencies in asymmetric systems? Similar
questions are many times ignored, regardless of their importance towards the realization of commercial CR networks.
The answers to some of these questions might be simpler
than we think (e.g., the database solution presented above)
and/or require research directions that are substantially different from the current literature. We argue through our examples that if we change our mindset and align our thinking
more with the way commercial telecommunication networks
operate and less in terms of mathematical modeling and
protocol design, some of these questions may be answered
leading to more rapid development of CR systems.
The contribution of this paper is twofold. Firstly, it brings
up to the surface, aspects that have been traditionally loosely
1

or inadequately treated in the current literature of cognitive

radio networks. However, we argue that these issues are
rather important if we envision the deployment of such systems. Unless we start considering them, practical solutions
will not be quickly viable. Secondly, and more important,
we manifest the disjoint paths that research and industry
have followed in the course of the years. Research should
be able to drive commercial deployments. Nevertheless, this
is not the case in the majority of the instances. The focus
of this paper - CR networks - is possibly the most glaring
example.
The rest of the paper is organized as follows. Section 2
briefly discusses representative literature on spectrum sensing. Section 3 introduces the two examples we will build our
arguments on. We further elaborate on them in Sections 4
(broadcast TV system) and 5 (cellular network). Finally,
Section 6 discusses the scope of and concludes our study.

2.

In this section we will briefly describe the approaches that

are currently proposed for spectrum sensing2 . Despite the
fact that efficient spectrum sensing has been identified as key
to the success of cognitive radio networks very early [6], we
still lack a satisfactory approach to perform this task. Existing literature can be broadly categorized into three classes:
(i) non-cooperative transmitter detection, (ii) co-operative
transmitter detection and (iii) interference-based detection.
Non-cooperative detection: This is the most basic
form of spectrum sensing where the secondary transmitter
tries to decide whether there is a primary transmitter using the spectrum or not. The detection problem is formulated using hypothesis testing, where the null hypothesis is
the absence of the primary transmitter and the alternate
hypothesis indicates its presence [7]. The baseline of this
single-user sensing scheme is that of energy detection. The
secondary transmitter measures the average energy on a specific channel and the decision is based on a threshold comparison (e.g., [8] [9]). However, more accurate detection
is possible if the secondary user has information about the
primary users signals. In this case, matched filter detection
can maximize the SNR and perform optimally [6]. Nevertheless, given that a priori knowledge of the primary signal
is not always available, other approaches such as cyclostationary feature detection (e.g., [10] [11] [12]) and eigenvalue
based detection (e.g., [13] [14]) are proposed. In all of these
schemes, every secondary transmitter operates in isolation
and takes his decision based on his own measurements. Noncooperative spectrum sensing has been criticized because of
its inherent stringent requirement for high reception sensitivity; secondary transmitters need to be able to detect
primary signals at their circuitry with the received signal
strength (RSS) as low as the noise level. Hence, a second
class of algorithms used to identify spectrum holes includes
cooperation among many secondary users as discussed below.
Cooperative detection:
Non-cooperative spectrum
sensing approaches are subject to high uncertainties due to
wireless propagation effects. For instance, fading can cause
2
The list of papers presented in this section is by no means
exhaustive. We are only interested in presenting the general
approaches used in spectrum sensing in order to build on
them our discussion in the following sections. The interested
reader can see [2] and [5] for additional references.

Different frequencies are used for uplink and downlink.

ACM SIGCOMM Computer Communication Review

RELATED STUDIES

45

Volume 43, Number 2, April 2013

large degradations at the received signal strength from a

single radio. Obtaining more measurements from a larger
number of co-located radios can provide us with a more robust decision. This is the goal of a cooperative spectrum
sensing scheme. There are two basic approaches followed
in cooperative detection, namely (a) soft combining and (b)
hard combining (e.g., [15] [16] [17]). In soft combining systems, CRs report their actual signal strength measurements
to a fusion center, while in the latter case they report their
binary decision. The fusion algorithm can be either a logic
AND, a logic OR or a majority vote based scheme. There
are different angles from which one can see the benefits of
collaborative sensing. For instance, collaboration can reduce
the sensitivity requirements for individual radios [18]. It can
also be viewed as means for reducing the time required for
sensing or reducing the false alarms rate [19] [20].
Regardless of the improvements over non-cooperative schemes,
the above systems still suffer from some major limitations.
There is still lack of information with regards to the location
of the primary receiver. Even though a primary transmitters signal can be detected with larger accuracy, interference
avoidance is still not guaranteed at the primary receiver.
Furthermore, spectrum sensors need to be co-located in an
fairly small area, in order to assure that the path loss component of the received signal from the primary transmitter
is the same for all sensors. If this is not the case, then the
performance of the cooperative schemes will be poor. If sensors are distributed over a larger area the different path loss
components of the signal strength will (legitimately) yield
many inconsistencies among the measurements of the single
radios. This falls back to one of the questions posted in our
introduction. What is the value of the spectrum hole identified from a scheme with these requirements? We will further
elaborate on this in this paper later.
Interference-based detection: Both cooperative and
non-cooperative detection do not take into consideration the
presence of a primary receiver. However, since interference
takes place at the receiver, identifying its presence and/or
position is important. The idea of Interference temperature
(IT) has been proposed by the FCC [21], to capture the additional interference (above the noise floor) that a primary
receiver can tolerate. This means, that using the interference temperature model, we can have simultaneous transmissions from a primary and secondary user, as long as the
interference from the latter is beyond the IT of the primary
receiver. Even though as a concept this is appealing, there
are many practical issues related with this approach. For
instance, how is it possible to measure the interference at
the primary receiver without prior knowledge of its position
relative to the secondary transmitter [22]? An interesting
approach was presented by Wild and Ramchandran [23].
Low cost sensor nodes are mounted on primary receivers to
detect the local oscillator leakage power emitted by the RF
front-end of the primary receiver. This can essentially detect
the presence of the latter. This information is consequently
sent as feedback to the secondary users. However, such approaches have received low attention from the community
and have been practically abandoned as unrealistic. The
main argument against such proposals is that they require
large scale infrastructural upgrades. Nevertheless, as we will
argue in Section 4, if we want to have a real-world cognitive
radio network deployment, this is a promising direction that
needs revisiting.

ACM SIGCOMM Computer Communication Review

46

3.

TOPOLOGIES FOR CR NETWORKS

In this paper, we will use two specific topologies and corresponding scenarios to illustrate the problems with existing
approaches and the need for a completely new thinking for
the success of CR networks. These topologies are shown in
Figure 1.
Primary
transmitter

Primary
receiver in FDD system
can deal with
small interference

-116 dBm contour

Secondary
link

Secondary
link uses
uplink
frequencies

Primary
transmitter
Primary
receiver
faces interference
(a)

(b)

Figure 1: Two Topologies and Scenarios for Investigating Challenges with CR Networks

3.1

TV White Spaces

As our first example, we consider the most widely discussed spectrum for usage in a CR network setting which is
that of broadcast television white spaces. The corresponding topology and scenario is shown in Figure 1(a). The
IEEE 802.22 standard states that any sensing algorithm
used should be able to detect digital television signals at -116
dBm, with probabilities of false alarm and mis-detection,
both equal to 0.1. Based on this requirement, if no signal
is detected the spectrum is declared as free and secondary
transmissions can be scheduled.
This causes two problems that we outline below. First,
the signals from the digital TV transmitter can be received
with high strength (i.e., RSS >> - 116dBm) at distances of
several miles [24]3 and TV signals with very low RSS at the
range of -116 dBm will only be received in areas much further from the TV tower. This observation is important, since
it practically restricts the cognitive radio network to be in locations that are very far from residential areas. The question
that comes to mind is whether there are commercially viable
applications that could benefit from a CR network located
in an area devoid of human presence. After some thought,
one can conclude that there might exist applications that
can take benefit from a cognitive radio network architecture
in such places (e.g., road safety applications with road side
units operating in these bands or sensor networks in forest
areas). However, one could also counter-argue that in these
areas, unlicensed bands can be as effective as opportunistic
usage of licensed bands as the demand is likely to be low in
such locations.
Moreover, except from the case of carrier sensing based
wireless networks, where a transmitter has to back-off when
it senses a competing transmission, interference affects only
the reception of the signal in other communication scenarios.
This causes the second problem with TV bands. As shown
in Figure 1(a), it is possible that the -116 dBm contour is
not circular due to radio propagation vagaries. Although
3
The actual distance is determined by many factors such
as the TV tower location and transmission power, the TV
receiver/antenna location, etc. but 70 miles is a typical
distance.

Volume 43, Number 2, April 2013

a secondary transmitter may, with quite complex processing, detect the signal level from the primary transmitter to
be below this value, it has no knowledge of the location or
distance of a passive primary receiver (a television). Thus,
a secondary transmission can cause significant interference
to a receiver. On the contrary, a secondary user within
the contour will sense energy from the primary broadcast
transmitter. Nevertheless, this does not necessarily mean
he cannot use the band, since there might not be any primary receiver to be interfered. Unless we can detect a primary receiver that is actively receiving the broadcast signal,
secondary transmissions may be either harmful to the primary system or lead to many missed opportunities for the
secondary users. We investigate this further in Section 4.

3.2

dynamic and unpredictable topology of commercial wireless

networks, broadcast systems (such as TV) exhibit a particular characteristic that will render any current (collaborative
or not) sensing algorithm ineffective. In brief, a broadcast
systems receivers are passive, that is, there are no signs
of their presence. Any appropriate receiver that is in range
of the broadcast transmitter can receive the signal; if the
secondary user is under the coverage of the primary transmitter he will sense the latters signals. Thus, it appears
that a cognitive radio network utilizing the TV spectrum is
not possible in urban, residential areas, which have strong
TV coverage.
However, as previously mentioned, despite the presence of
a primary broadcast signal, if no primary receiver is in the
near vicinity (in our case if no TV receiver is turned on),
the secondary user can be allowed to occupy the spectrum
without harm to the primary system. Therefore, in broadcast systems like the TV, current sensing algorithms will
result in an extremely large number of missed opportunities
for the secondary user. To assess the viability of this claim,
we have conducted a large scale user study to identify the
usual times that people have their TV receivers turned on
and off and hence, identify periods that opportunistic access
is possible. We would like to reiterate here that during the
off periods, even though a cognitive sensor would sense
broadcast primary signals, secondary transmissions would
be possible since the receiver is not powered on and thus
not harmed.
We have conducted two different surveys to identify whether
spectrum opportunities exist. For the first survey, we have
utilized the Amazons Mechanical Turk platform to recruit
people willing to participate in an anonymous survey for a
small amount of money. The purpose of this survey is to
obtain a general view of the usage patterns of people (e.g.,
times that the receivers are off, type of TV service, etc.).
The second survey is of smaller scale and was conducted
with a paper questionnaire distributed to the households of
a single apartment building. The purpose of this study is
to identify whether similar patterns of TV usage exist in
a concentrated residential area. Since in reality there are
many receivers within a geographical cluster, we want to
study the cumulative activity patterns of co-located users.
This will reveal any opportunities for secondary user access
in the vicinity of these users. We would like to emphasize
here on the fact that as with any interview-based conclusions, they might exhibit biases such as those introduced by
the difference between the actual and the declared behavior
of the subjects. Nevertheless, we believe that these biases
in our case are not significant.
Large Scale Activity Pattern Survey: The main
results from the first survey are presented in Figures 2 and
3. We see that on weekdays people use their TV receivers
mainly in the evening hours (e.g., after 6:30 p.m.). Late
nights and morning times are fairly idle, hence, allowing opportunities for secondary access to spectrum. However, this
pattern changes on the weekends to a more uniform distribution of the active periods, allowing for less secondary
spectrum access opportunities. It is interesting to note the
presence of many periods with low (if at all) activity. The
period between 4:30 a.m. and 6:30 a.m. is completely inactive even during weekends. Delay tolerant applications could
take advantage of these patterns and operate during the inactive periods, with high data rates. For instance, back up

A Cellular FDD Network

As the second example, we consider a cellular network

that employs frequency division duplexing (i.e., there are
separate frequencies for the uplink and downlink). In the
US, the uplink (824-849 MHz) and downlink frequencies
(869-894 MHz) are separated by 45 MHz in one commonly
used slice of licensed spectrum. In a given cell, on the uplink
frequencies, only the base station is the receiver as shown in
Figure 1(b). Channels that are 1.25 MHz (5 MHz) wide are
employed in 3G systems based on the CDMA200 (UMTS)
standard.
It is also well known that the load in cells varies significantly over time and day of week (e.g., [25] [26]). Since
CDMA systems are interference limited, a secondary transmission that employs only the uplink frequencies may be able
to operate without causing any harm to the primary system
[27]. Especially if the secondary transmissions are employing low transmit power for short range links, there may be
negligible interference or harm. This is further helped by
the fact that the path loss for short distances usually has a
smaller exponent compared to large distances. Thus, while
the signal attenuates significantly and has very small power
by the time it reaches the primary receiver (the base station), it is sufficient to perhaps support high data rates at
shorter distances.
However the cellular spectrum cannot be used for secondary applications as of now. If the spectrum is sensed, it
will be discovered to be occupied by the primary users. The
transmitters in this case are the mobile phones which will
all employ the same frequency in a CDMA system. Thus,
a very small number of mobile phones transmitting on the
uplink would still tag the spectrum as occupied making it
unusable by a secondary system even when it is unlikely to
cause any harm at the primary receiver. Of course, in this
case the quality of the secondary link might be impacted
by the interference from primary transmissions. We further
consider this scenario in Section 5.

4.

THE CASE OF PASSIVE RECEIVER

In the previous section we have discussed scenarios where

spectrum sensing algorithms as envisaged in the literature
fail to address the real problem; either the secondary user
will miss opportunities since the interference to the primary
receiver is not harmful (FDD network or no receivers exist
in the area) or will cause harmful interference since the detection of a primary transmitter does not reveal anything for
the location of the primary receiver (TV broadcasts). Even
though both of the problems are related to the complex,

ACM SIGCOMM Computer Communication Review

47

Volume 43, Number 2, April 2013

1
Weekdays

0.8
0.6
0.4
0.2
0

1am

5am

10am

3pm

8pm

Weekend
0.8
0.6
0.4
0.2
0

Time

Figure 2: TV receiver activity

in bins of 30 minutes for a typical weekday.

Frequency of TV turned on

Frequency of TV turned on

Frequency of TV turned on

1am

5am

10am

3pm

0.6
0.4
0.2
0

8pm

Figure 3: TV receiver activity

in bins of 30 minutes for a typical weekend.

5am

10am

3pm

8pm

Figure 4:
Activity for a
receiver using the broadcast
through the air TV signals.

Weekdays
[7pm-9:30pm]

Weekends
[9am-noon], [4pm-10pm]

Table 1: TV receiver activity for the users of broadcast TV service of a specific block apartment building.

will have spectrum opportunities during this time frame. On

the contrary, if this 20% is scattered across the country then
the locations that will have these chances are fewer.
Thus the cumulative activity of primary co-located users
is an important feature. To assess this quantitatively, we
have performed the same survey with the residents of a block
of apartments in a building consisting of 50 apartments in
Pittsburgh, PA. Figure 5 depicts our results.
As we can see there are some clearly zero activity periods
for the residents of this building, especially on weekdays.
This cumulative activity of people allows many opportunities for a cognitive radio network to operate, even when
current sensing algorithms would declare the presence of a
primary transmitter. These opportunities are even larger,
if we again consider only the activities of the people that
actually use the broadcast TV service, rather than cable,
satellite, or fiber. Of the residents of the building under examination, only 4% utilize broadcast TV service and Table
1 shows the union of the active periods for these users.
Proposed Solutions:
The passive nature of the receiver in many broadcast systems like television, can have
a negative effect on the successful deployment of a CR network. Ideally the solution to the above problem would be to
transform the passive receiver to an active one. In this way,
we could sense the air for an active reception and decide to
either access the medium or not. It is widely accepted in the
research community that one of the properties that a proposed solution needs to satisfy is to not involve large scale
infrastructural changes. This is certainly a nice feature, but
this is not the way that commercial operators always work.
For example, very recently, a top US cable operator has required its customers to use specific hardware, delivered to
them for free, in order to be able to continue receiving their
service [28] as it transitioned from analog to digital service.
This is certainly something that no research study would
suggest. Nevertheless, this is the actual way that commercial operators work.
Considering this, we ask why not simply equip every receiver with a box that informs the CRs in its vicinity when

4
For instance, there might be a TV show that is of extreme
interest for these people and thus, it does not reflect the
whole population in other areas of the country or even of
the same city.

ACM SIGCOMM Computer Communication Review

1am

Time

Time

services on the cloud could run on scheduled times (e.g., every morning between 5 a.m. and 6 a.m.). Note here that
even such applications would not be possible with the current spectrum sensing algorithms; primary signals would be
sensed and no access would be granted to secondary users
applications.
Some people might argue that the above levels of activity are still large. Hence, from their perspective, even with
an optimal spectrum sensing algorithm that is able to identify active TV receivers, the secondary user would not have
many chances to access the spectrum; the claim is that there
are almost always, even small in numbers, some active receivers in a region. Thus, in our survey, we also asked the
participants the type of service they are using for their TV
reception. Surprisingly or not, only 12% of them actually
use broadcast TV. The majority of the users - 77% - use
cable (and some use fiber and satellite). Since people that
utilize cable and other technologies are unharmed by secondary spectrum access, in Figure 4 we present the activity
levels only for the users that actually receive the stations
signals over the air. The results clearly show that there
are many more opportunities for secondary access to the
spectrum (especially on weekdays). This might have been
expected, knowing that this class of users is a very small
portion of the total population. Note also that the broadcast users (even though we do not have demographics from
the survey we conducted) might be located in areas where
there is no cable or fiber service yet (or satellite is still expensive). These areas may correspond to isolated locations
that might as well not be of interest for deploying a CR network (recall our earlier discussion related to the value of a
spectrum hole).
Apartment Block Case Study:
The results from
the Mechanical Turk survey reflect general patterns of users
that may or may not be located in the proximity of each
other. In order for a secondary link to be activated in a given
location, all the primary receivers in the vicinity should be
not active. Even a single active primary device will force
the secondary user to backoff. For instance, if all 20% of the
users that have their receivers turned on between 1 a.m. and
1:30 a.m. (see Figure 2) are spatially located close to each
other (e.g., a neighborhood of a specific city, or an area of a
specific state4 ), then CR networks deployed in other areas

Weekdays
Weekends

0.8

48

Volume 43, Number 2, April 2013

the former is active? This could be perhaps financed for from

the revenues from freeing the spectrum. Also note that only
TV receivers that use the air need to be equipped (hence,
even systems such as [23] if not carefully deployed would not
work efficiently since they do not distinguish between a cable
TV or a broadcast receiver from the air). Further, the percentage of TVs that are equipped with some form of wireless
interface (e.g., WiFi, Bluetooth etc.) has dramatically increased during the last years. We asked the participants in
our surveys if their TV receiver is equipped with some type
of wireless interface; 24% of them answered Yes. These
interfaces could be utilized for our purposes, reducing the
cost for distributing the special boxes (just similar to some
receivers that were able to keep receiving digital TV signals
without any modification and extra hardware). This would
be a different type of sensing where specific messages in an
unlicensed band would inform CRs about the existence of
passive receivers.
A more drastic solution (which however requires many
parties to negotiate and agree to corresponding service level
agreements or SLAs e.g., FCC, providers etc.), is to completely abandon the broadcast TV service. Since only a
small fraction of people utilize it, they could be served with
other technologies for free (of course with the corresponding
SLAs in place). This would free up the whole TV spectrum
for usage by alternative wireless technologies. We recognize
that both of the above solution sketches are radical and extreme, but in the authors opinion, only with drastic solutions
will the vision of CR networks become viable.
1

Weekend
Frequency of TV turned on

Frequency of TV turned on

Weekdays
0.8

0.6

0.4

0.2

through the use of spread spectrum. The maximum number of active mobiles on a channel Mmax can be roughly
determined by setting Eb /I in this expression to the minimum acceptable value. If we use some sample numbers
such as Gp = 3.84M cps/12.2kbps = 315, vf = 0.4, c = 0.8,
f = 0.75, and the minimum Eb /I = 3.2 (5 dB), Mmax = 113
per channel. As long as the Eb /I value is maintained at the
base station, and no mobile is denied service, there is no
harm to the primary system. Service providers typically design their network to support peak capacity, which occurs
infrequently, at specific hours/days of the week. At other
times, the load is significantly smaller [25] [26]. The operator can move users to other channels when a given channel
reaches capacity.
The question that arises is whether a secondary CR network can operate in the uplink frequencies when the cellular
network is not at capacity. By looking at the simple expression for Eb /I, this is indeed the case when M < Mmax
provided the interference caused by the secondary CR
transmitter does not exceed the interference that may
have been caused by the Mmax M active mobile stations if they were present. A CR transmitter however
cannot determine the corresponding transmit power by simply sensing the spectrum (even mobiles that are in idle mode
occasionally transmit control data that may indicate that
the spectrum is occupied although they have low duty cycles). A CR transmitter thus needs some information from
the primary system (referred to as collaboration in [27]) or
has to use a worst case approximation. Also, it is not subject
to power control and causes a constant interference at the
base station if the CR transmitter is stationary. If the CR
network has mobile transmitters and receivers, the situation
becomes more complicated. Further the secondary receiver
faces interference from the primary active mobile phones.
Proposed Solutions: The secondary network should use
a transmission scheme that is based on spread spectrum
and follows the 3G standards. It can operate by picking
spreading codes fairly independently (since the uplink in
CDMA systems use Gold or Kasami-like sequences to separate transmissions). With smartphones that can perhaps
operate in an ad hoc mode and inexpensive hardware (similar to femtocells) that operate using 3G standards, a secondary network will be viable in this spectrum.

1am

5am

10am

3pm

8pm

0.8

0.6

0.4

0.2

1am

5am

10am

(a) Weekdays

3pm

8pm

Time

Time

(b) Weekends

Figure 5: TV receiver activity in bins of 30

minutes for a specific block apartment building.

5.

THE CASE OF FDD SYSTEM

In this section we undertake a simplified analysis of the

FDD uplink in a UMTS cell to demonstrate the feasibility
of secondary transmissions even when the spectrum is presumably occupied. In CDMA systems [29], the energy per
bit to interference ratio (Eb /I) at the base station receiver
on a given 1.25 MHz (cdma2000) or 5 MHz (UMTS) channel needs to be between 5 and 7 dB and it is maintained
nearly the same for all mobile stations communicating with
the base station through strict power control. The value
of Eb /I on the uplink depends on the number of active
phones M in the cell, their voice activity vf (what fraction of time the user speaks), the power control accuracy c
and the fraction of interference from neighboring cells f . A
Gp c
where
simple expression for Eb /I is Eb /I = vf (M 1)(1+f
)
Gp is the processing gain (chip rate/bit rate) in CDMA

ACM SIGCOMM Computer Communication Review

Figure 6: Eb /I at the secondary receiver for different

numbers of active primary mobile stations in a cell.
The operating constraints of these networks are not very
clear at this point. We can think of a CR network operating

49

Volume 43, Number 2, April 2013

at close to the minimum transmit power to avoid causing

interference at the base station irrespective of the location
of the secondary user on the least used channels. A study of
mobile phone transmit powers in UMTS provides extensive
data on the distribution of uplink transmit powers in a 3G
network [30] . The data show that active mobile transmit
powers can be as low as -50 dBm in certain scenarios (although they can be as high as 20 dBm). The average over
several different scenarios (locations, network load, with and
without Bluetooth) is 5.6 dBm (3.63 mW). This is however
likely to restrict the throughput of the secondary network.
If multiple secondary radio transmitters exist, the rates and
transmit powers they use will be smaller. However cooperation with the primary network (which can share information
about unused channels or channels with low loads) can allow
the secondary network to operate more efficiently. For this,
of course, either a market for secondary use or regulatory
changes will become necessary.
To assess whether operating at a very small transmit power
is viable for a secondary network, we perform a simple simulation. The transmit power of a primary mobile varies inversely as the path-loss because of strict power control. The
signal strength from every mobile phone is thus received at
approximately the same power S at the base station. The
closer a mobile is to the BS, the lower is its transmit power.
If the path loss exponent is = 4 and the transmit power of
a primary mobile at the cell edge (distance R from the base
station) is Pedge , the transmit power at distance r < R will
P

the edge of the cell, the secondary receiver is far away from
most of the active mobiles and sees lower interference. In
the mid-regions, the secondary is not far away from mobiles
that have a reasonably large transmit power and may not
be able to operate unless M is very small.
However, this simple simulation demonstrates that there
is spectrum access opportunity for secondary users in a cellular telephone network employing FDD under low loads,
but the current approaches to dynamic spectrum access are
not suitable for exploiting it.

6.

r4

. Each of the M active primary mobiles causes inbe edge

R4
terference at the secondary receiver. Let the transmit power
and distance of the i-th primary mobile from the secondary
receiver be Pi and di . Then the overall interference includM
X
Pi
ing a voice activity of uf will be uf
. Assuming that
d4
i=1 i
the secondary transmitter is at a distance of 1 km from the
secondary receiver and its transmit power is Ps , the Eb /I
observed at the secondary receiver will be5 :
Eb /I =

Gp P s
M
X
Pi

uf

i=1

(1)

d4i

We consider a cell of size R = 5 km and place M primary mobiles uniformly randomly in the cell. We assume
perfect power control with the transmit power of primary
mobiles at 5 km from the base station to be 20 dBm. We
place the secondary receiver at a distance of 1, 2, 3, 4, and
5 km from the base station. We let Ps = -10 dBm, a fairly
small number compared to the average transmit power of
5.6 dBm of primary mobiles in a cell reported in [30]. We
compute the average Eb /I at the secondary receiver over
1000 runs for different values of M . The results are shown
in Figure 6. We see that the secondary receiver has a reasonable Eb /I close to the base station and at the edge of
the cell, while it has poorer performance in between. This
can be attributed to the fact that close to the base station,
the interference comes from mobiles that are transmitting at
much lower powers. In fact, the secondary transmitter may
have to reduce its transmit power at such locations beyond
-10 dBm so as to not cause harm to the primary system. At
5

Acknowledgment
This work was supported in part by NSF Grant 1247546.

7.

REFERENCES

[1] J. Mitola III. Cognitive Radio for Flexible Mobile

Multimedia Communications. In MoMuC, 1999.
[2] R. Tandra, S.M. Mishra, and A. Sahai. What is a
spectrum hole and what does it take to recognize one?
In Proceedings of the IEEE, 2009.
[3] M.B.H. Weiss, M. Al-Tamaimi, and L. Cui. Dynamic
Geospatial Spectrum Modelling: Taxonomy, Options
and Consequences. In elecommunications Policy
Research Conference, 2010.

This simplified model ignores extra cell interference f .

ACM SIGCOMM Computer Communication Review

SCOPE OF OUR WORK & CONCLUSIONS

In this paper we have tried to shed some light on the

reasons behind the lack of commercial deployment of a cognitive radio network. In order to do so we focused on a
specific functionality, important for the operations of such a
network, that of spectrum sensing. However, we would like
to emphasize on the fact that similar problems exist with respect to other functionalities, such as spectrum sharing and
spectrum access. We emphasize that our work should not
be viewed only as a question about the realization of cognitive radio networks. It should be seen more broadly as a
challenge to the real world applicability of a large volume of
existing research. For example, the research community is
used to putting aside, in the majority of the cases, solutions
that require large scale infrastructural changes. Without
trying to argue in this paper whether this is correct or not,
sometimes the only feasible solution(s) is(are) accompanied
by this drawback. Thus, we should be more open to them
and less critical of similar proposals. After all this is often the way that the world of commercial network operators
functions.
To sum up, solutions to technical problems might be much
easier than we think. Coming back to the topic of this paper,
CR networks, the nature of the primary user and the actual
operations of a real communication system are usually neglected or made complicated or oversimplified. Therefore,
the solutions proposed do not drive an effective commercially viable network architecture. Drastic solutions often
need to be taken, but for this, a careful examination of the
real-world underlying tradeoffs need to be considered. For
instance, in our sketch solution, for eliminating broadcast
TV completely, the involved parties need to consider the
potential benefits from the capacity/revenues obtained by
freeing up the spectrum as well as the cost to serve people
through other technologies (potentially for free).

50

Volume 43, Number 2, April 2013

[4] The White Space Coalition.

[17]
http://www.engadget.com/tag/white%20space%20coalition/.
[5] I.F. Akyildiz, W-Y. Lee, M.C. Vuran, and S. Mohanty.
NeXt generation/dynamic spectrum access/cognitive
radio wireless networks: A survey. In Elsevier
[18]
Computer Networks, 2006.
[6] A. Sahai, N. Hoven, and R. Tandra. Some
fundamental limits on cognitive radio. In 42nd
[19]
Allerton Conference on Communication, Control and
Computing, 2004.
[7] F.F. Digham, M-S. Alouini, and M.K. Simon. On the
Energy Detection of Unknown Signals over Fading
[20]
Channels. In IEEE Transactions on Communications,
Vol (55), Issue 1, 2007.
[8] Y. Li. Signal processing issues in Cognitive Radio. In
[21]
Proceedings of IEEE, 2008.
[9] R. Tandra and A. Sahai. SNR walls for signal
detection. In IEEE Journal on Selected Topics in
[22]
Signal Processing, Vol (2), pp. 4-17, 2008.
[10] D. Cabric, S.M. Mishra, and R.W. Brodersen.
[23]
Implementation issues in spectrum sensing for
cognitive radios. In 38th Asilomar Conference on
Signals, Systems and Computers, 2004.
[24]
[11] A. Fehske, J.D. Gaeddert, and J.H. Reed. A new
approach to signal classification using spectral
[25]
correlation and neural networks. In IEEE DySPAN,
2005.
[12] S. Haykin, J. Reed, and D. Thomson. Spectrum
[26]
sensing in Cognitive Radio. In Proceedings of the
IEEE, 2008.
[13] Y. Zeng and Y.C. Liang. Text on eigenvalue based
[27]
sensing - For Informative Annex on Sensing
Techniques. In IEEE 802.22 Meeting Documents, 2007.
[14] Y. Zeng and Y.C. Liang. Maximum-Minimum
[28]
Eigenvalue Detection for Cognitive Radios. In IEEE
PIMRC, 2007.
[29]
[15] G. Ganesan and Y. Li. Agility Improvement through
Cooperative Diversity in Cognitive Radio. In IEEE
GLOBECOM, 2005.
[30]
[16] C. Sun, W. Zhang, and K.B. Letaief. Cluster-Based
Cooperative Spectrum Sensing in Cognitive Radio
Systems. In IEEE ICC, 2007.

ACM SIGCOMM Computer Communication Review

51

E. Visotsky, S. Kuffner, and R. Peterson. On

Collaborative Detection of TV Transmissions in
Support of Dynamic Spectrum Sharing. In IEEE
DySPAN, 2005.
S.M. Mishra, A. Sahai, and R.W. Brodersen.
Cooperative Sensing among Cognitive Radios. In
IEEE ICC, 2006.
G. Ganesan and Y. Li. Cooperative spectrum sensing
in cognitive radio, part II: Multiuser networks. In
IEEE Transactions on Wireless Communications, pp.
2214-2222, 2007.
J. Ma and Y.G. Li. Soft combining and detection for
cooperative spectrum sensing in cognitive radio
networks. In IEEE GLOBECOM, 2007.
ET Docket No 03-237 FCC. Notice of inquiry and
notice of proposed Rulemaking. In ET Docket No
03-237, 2003.
T.X. Brown. An analysis of unlicensed device
operation in licensed broadcast service bands. In IEEE
DySPAN, 2005.
B. Wild and K. Ramchandran. Detecting Primary
Receivers for Cognitive Radio Applications. In IEEE
DySPAN, 2005.
FCC digital TV channels coverage maps.
http://transition.fcc.gov/dtv/markets/.
C. Peng, S.-B. Lee, S. Lu, H. Luo, and H. Li.
Traffic-Driven Power Saving in Operational 3G
Cellular Networks. In MobiCom, 20011.
D. Tipper, P. Krishnamurthy, A. Rezgui, and
P. Pacharintanakul. Dimming cellular networks. IEEE
Globecom, 2010.
R. Saruthirathanaworakun and J.M.Peha. Dynamic
primary-secondary spectrum sharing with cellular
systems. IEEE Crowncom, 2010.
Comcast Digital Box Deployment.
https://digitalnow.comcast.com/.
Vijay Garg. Wireless Network Evolution: 2G to 3G.
Pearson LLC, 2001.
Arjun Bhupathi Raju. Characterization of uplink
transmit power and talk time in wcdma networks. MS
Thesis in Electrical Engineering, Virginia Polytrechnic
Institute and State University, July 2008.

Volume 43, Number 2, April 2013