Content uploaded by F. Cuomo
Author content
All content in this area was uploaded by F. Cuomo on Jan 19, 2013
Content may be subject to copyright.
1722 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 20, NO. 9, DECEMBER 2002
Radio Resource Sharing for Ad Hoc Networking
With UWB
Francesca Cuomo, Cristina Martello, Andrea Baiocchi, Member, IEEE, and Fabrizio Capriotti
Abstract—Ultra-wideband (UWB) radio is becoming a
promising field for new generation’s digital communication
systems. This technique, based mainly on the impulse radio
paradigm, offers great flexibility and shows enormous potential in
view of a future broadband wireless access. In this paper, we aim
at presenting the main principles to design a multiaccess scheme
based on UWB. The potential of UWB is exploited within a dis-
tributed ad hoc wireless system, where we describe the principles
for the definition of a medium-access control (MAC) for mobile
computing applications and we analyze the main performance
results derived from simulations. A general framework for radio
resource sharing is outlined for classes of traffic requiring both
elastic-dynamic and guaranteed-reserved bandwidth. Then,
we discuss the issue of supporting the proposed radio resource
sharing scheme by means of a distributed MAC protocol.
Index Terms—Ad hoc networks, medium access control (MAC)
protocols, power control, radio resource sharing.
I. INTRODUCTION
ULTRA-WIDEBAND (UWB) technology is an emerging
paradigm both in the field of radar applications and
digital communications. UWB systems are mostly based on
impulse radio (IR) technology, which has recently reached
an appreciable degree of development so as to be able to
support high data rates with low power consumption and low
complexity in terms of transmission/reception operations [1],
[2]. By combining a transmission over a wide radio spectrum
band with lower power and pulsed data, UWB causes less
interference than conventional narrowband radio and offers
potential to hit the market in unlicensed bandwidths.
Today it is clear that UWB is a promising field to create small,
high bit rate transceivers that could be used for a wide set of ap-
plications, from wireless local area networks (LANs) to ad hoc
networks, from IP mobile-computing to multimedia-centric ap-
plications. In this context, a consistent amount of literature has
been dedicated to the analysis of UWB transmission/reception
principles and of its relevant performance [3]–[7]. Besides these
key issues, the challenge in using UWB technology in wireless
communication systems lies in the development of multiple ac-
cess techniques and radio resource sharing schemes.
The aim of this paper is to outline a framework for the
adoption of UWB in a wireless ad hoc system, where the main
Manuscript received December 14, 2001; revised June 24, 2002. This work
was supported in part by the European IST Programme within the framework
of the WHYLESS.COM project IST-2000-25197.
The authors are with the University of Rome “La Sapienza,” Infocom Depart-
ment, 00184 Rome, Italy (e-mail: cuomo@infocom.uniroma1.it; martello@in-
focom.uniroma1.it; baiocchi@infocom.uniroma1.it; fcapriotti@libero.it)
Digital Object Identifier 10.1109/JSAC.2002.805309
potential of this technique can be exploited. We concentrate on
the definition of radio resource sharing principles that could be
applied to support IP pure “best effort” traffic (e-mail, WEB
browsing, file transfer, etc.) or data services with specific
quality-of-service (QoS) requirements. The paper is organized
as follows. Section II describes the reference architectural
scenario. Section III introduces the model considered for the
definition of the radio resource sharing scheme. The joint
power and rate assignment paradigm for radio resource sharing
is given in Section IV, whereas Section V presents simple,
suboptimal algorithms and procedures that apply the resource
sharing principles identified in Section IV. In Section VI, we
illustrate performance results while supporting elastic-dynamic
bandwidth class of traffic. Finally, Section VII concludes the
paper and discusses lines for future work.
II. REFERENCE ARCHITECTURAL SCENARIO
A. UWB Multiple Access
IR transmits extremely short pulses (0.1 to 1.5 ns) giving
rise to wide spectral occupation in the frequency domain (band-
width from near dc to a few gigahertz). The typical pulse, named
“monocycle,” is the building block for data transfer that is com-
monly obtained by using the pulse position modulation (PPM).
In a binary context, a logical “zero” is transmitted by one mono-
cycle centered at time whereas a logical “one” is transmitted
by the monocycle shifted by seconds (centered at ). In
order to allow several users to share the same radio resource si-
multaneously, the time-hopping (TH) code is added [2]. Below,
we consider the use of pseudorandom TH codes.
Fig. 1 reports an example of transmission by two users,
each characterized by a TH code word. Whereas the first
user uses the TH code , the second uses the
word . Each code word element corresponds to
one of the possible time shifts in the period, that is,
the pulse repetition time (typically a hundred or a thousand
times the monocycle width). Each is divided into time
bins of period . consecutive pulses transmitted at the
pulse repetition time are dedicated to the transmission of 1-b
(symbol). The bit rate associated to one code word is then
. The period of a code word is denoted as
and generally .
In Fig. 1, we assumed the two users were synchronous both
with the and with each other: whereas the first assumption
can be easily satisfied, the second one is more unrealistic. How-
ever, though the users are not synchronized with each other and
the TH codes are chosen in a pseudorandom way, catastrophic
0733-8716/02$17.00 © 2002 IEEE
CUOMO et al.: RADIO RESOURCE SHARING FOR AD HOC NETWORKING WITH UWB 1723
Fig. 1. UWB multiple access scheme.
collisions are not very likely to occur while just mutual inter-
ference arises and can be compensated by transmitting several
monocycles for the same bit [2].
B. Ad Hoc Concept Based on UWB
Recent research efforts have been dedicated to ad hoc net-
works. Until recent years, these radio networks were considered
mainly for military applications, as their distributed architec-
ture offered a fundamental, operative advantage. As regards the
commercial sector, ad hoc networks are expected to enhance the
networking world, by providing access where no infrastructure
is available, or by quickly and cheaply extending existing cov-
erage areas [8], [9].
Several technologies are emerging for ad hoc networking
[10], [11]. UWB appears competitive in this field and could be
exploited as a promising and flexible transmission technology
that brings some specifically suitable advantages, thanks to a
few UWB features matching exactly the requirements needed
to design an ad hoc network.
First of all, UWB can provide high data rates in indoor,
dense multipath environments [12], [13], as is expected of the
technologies of future generation radio systems. An additional
feature of UWB is its flexibility in the reconfiguration process
of data rate and power, due to the availability of a number of
transmission parameters, which can be tuned to better match the
requirements of a data flow. As far as radio-terminal equipment
is concerned, this is generally cheaper than the equipment for
traditional technologies, as the structure of the receiver is
extremely simple due also to the absence of a carrier.
Moreover, IR calls for the synchronization of transmitting-
receiving pairs (communicating through a link), but works
efficiently even though different links in the network are
asynchronous; this feature is particularly suitable in an ad
hoc network, where the absence of an infrastructure implies a
highly complex synchronization of all the network terminals.
In our work, we aim at defining a multiple access control pro-
tocol that can be applied in different architectural scenarios (a
wireless, local area access or a pure ad hoc network) to support
mobile computing applications [14].
C. Reference Architectural Model
The reference architectural model is reported in Fig. 2. The
model includes radio terminals (RT) and access points (AP).
We selected a distributed mechanism to handle radio resource
sharing that could be used both in an infrastructure network,
(where APs interconnect the RTs to the fixed network) and in a
pure ad hoc network, supporting peer-to-peer communications
between RTs. Fig. 2 highlights the presence of an medium-
access control (MAC) domain, where access to the radio
resource is controlled. Every MAC domain refers to the area
where transmission of an RT (or AP) has an impact on the
transmission/reception of other RTs(or APs). As a consequence,
the multiple access control function will operate in every MAC
domain, in order to share the capacity among the RTs/APs
belongingto that domain. Mutualinterference among competing
RTs/APs plays a fundamental role in the control of access to
the radio part and power control is seen as a mechanism to
be used jointly within MAC procedures in order to increase
use of the radio trunk [15], [16].
The radio resource sharing mechanism we are going to pro-
pose jointly manages both powers and data-rates in order to sup-
port two different classes of traffic. A first class of traffic named
reserved bandwidth (RB), requires a QoS expressed as a given
amount of bandwidth negotiated at the beginning of a session;
this class typically accommodates time-constrained data flows.
A second class of traffic named dynamic bandwidth (DB), can
elastically adapt the bandwidth to the varying system condi-
tions; this class can be used to map the classic “best effort” ser-
vice of the IP networks. The name DB refers to the fact that
the MAC can dynamically reconfigure the amount of bandwidth
into a time scale of the packet duration. This reconfiguration is
needed in order to counteract changes in interference conditions
and to use the radio resource efficiently.
III. RADIO RESOURCE SHARING MODEL
UWB transmission was analyzed in [1] and [2] according to
the standard, additive white Gaussian noise (AWGN) hypoth-
esis, with independent users exploiting pseudorandom codes.
1724 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 20, NO. 9, DECEMBER 2002
Fig. 2. Reference architectural scenario and MAC domains.
The receiver is assumed to be synchronized exactly on the in-
tended pulse stream. The bit error probability is evaluated by
assuming that the background, noise and UWB self interference
are both gaussian and are, therefore, a function of the signal-to-
noise ratio (SNR).
We consider pairs of communicating UWB terminals, each
pair consisting of one transmitter and one receiver and using one
pseudorandom code. In the following, we refer only to RTs even
if the model applies also to communications involving also APs.
Then, links are active and the SNR at the th link’s receiver is
(1)
where we use the following definitions:
binary bit rate of the th link;
average power emitted by the th link’s transmitter;
path gain from the th link’s transmitter to the th link’s
receiver;
background noise energy plus interference from other
non-UWB systems;
an adimensional parameter depending on the shape of
the monocycle.
Typicalvalues of the above parameters are as follows [2] : the
pulse duration is 0.75 ns and ns; 10 ;
10 V s.
Equation (1) is structurally the same as the one found
for asynchronous wideband code-division multiple access
(WCDMA). Apart from the value of key parameters (e.g., the
mutual interference “weight” ), the major points that are
specific to IR UWB are those showing how power and rate can
be adjusted.
The rate is , so that it can be modified
by changing either the basic chip time (that appears to be
quite a technological challenge), or the number of chip times
per pulse repetition frame, or the number of pulses corre-
sponding to an information bit. The latter is also the easiest way:
it is similar to the WCDMA variable spreading factor, but differs
mainly in its fine adjustment, which, in the case of WCDMA,
is geometric. Furthermore, as opposed to WCDMA, to adjust
does not imply a change in code, since it can be varied by
keeping the code length constant, instead simply the number
of pulses to be integrated at the receiver changes.
As regards power, this is , where is
the pulse energy. Hence, the power can be modified by changing
either the pulse energy ,or ,or ; these latter two are
specific in the case of UWB IR. Adjusting is an extremely
simple way of adapting the power level, even though it trans-
lates into quantized variations because is an integer. In the
remainder of this paper, we deal with and as though they
were continuous variables. This is justified by the typical values
of and that imply a negligible quantization effect on
power and rate values.
Moreover, in our model we assume that path gains remain
fixed; path gains do, in fact, fluctuate due to changing multipath
patterns and ultimately to terminal mobility. It is a customary
assumption to deal with path gains as constants, which is jus-
tified when RT mobility is significant on a time-scale which is
at least one size larger than the time-scale for the MAC layer
connection adaptation (e.g., the time needed for power and rate
assignment, channel measurement, signaling). This is indeed a
reasonable assumption in a mobile computing context. A sim-
ilar modeling to the one assumed here is typical of most works
dealing with ad hoc network algorithms (e.g., routing) [17] or
power control schemes [18].
IV. JOINT POWER AND RATE ASSIGNMENT AS AN
OPTIMIZATION PROBLEM
A basic requirement of the physical layer is to offer bit
transmission with an error probability no greater than a given
threshold value. That in turn means that the SNR of the th
link must not be maintained below a specified threshold .
Moreover, an upper constraint is enforced on the average value
of the transmission power level; this value is referred to as
. Therefore, from (1), we infer that the power levels and
bit rates should meet the following constraints:
.(2)
It is natural to pose the power and rate assignment as an op-
timization problem, and to aim at optimizing some throughput,
delay or energy consumption metric under the constraints in (2)
CUOMO et al.: RADIO RESOURCE SHARING FOR AD HOC NETWORKING WITH UWB 1725
[18]–[20]. The assignment problem splits into two subproblems,
depending on the required service guarantees.
In the case of RB flows, the target value of comes as a re-
quirement from the network layer; hence, the point is to check
whether feasible power levels can be set in all transmitters, so
that the required bit rates are supported. That in turn is equiv-
alent to the problem of the existence of a solution for the in-
equalities system (2) with a given set of values for bit rates. In
addition, the minimal power values are searched for.
The RB traffic problem can be specifically stated as follows
as shown in (3), at the bottom of this page, where and are
the -dimensional vectors of rates and powers, respectively.
The first inequalities can be compactly stated in matrix
form as , where , is a diagonal
matrix whose th diagonal element is , is a
nonnegative matrix with nil diagonal elements and off-diagonal
elements equal to and is an -dimensional vector with
elements equal to . A positive vector , satisfying the
first inequalities of (3), exists if the matrix is such that
the spectral radius of is less than one. Remember that
the spectral radius of a matrix is the maximum of the modulus
of its eigenvalues. A sufficient condition for this and for
to exist and to be a nonnegative matrix, is that the matrix is
strictly (or irreducibly) dominant diagonally. In that case, the
power assignment is Pareto optimal, in the sense
that any other power assignment satisfying (3) is such that
. This power assignment is optimal both from the point
of view of energy consumption and because—according to the
hypothesis of the reconfigurability of powers—it minimizes
the blocking probability of new links (i.e., the probability of
the event that a new link request is offered and cannot be
assigned a feasible power level according to (2), given a set
of already established links).
In the case of DB traffic, a strict requirement on the bit
rate does not exist; as a consequence, it is clear from the
form of (2) that a feasible solution for power levels always
exists, provided that the bit rates are sufficiently small. The
DB traffic optimization problem for joint power and rate
assignment is defined in this work by the following target
function representing the overall system net throughput:
(4)
Due to the constraints (2), it can be verified that for any fixed
,wehave for all feasible with
(5)
Moreover, we have when coincides with
the set of the maximum rates satisfying (2). Hence, the optimiza-
tion (i.e., maximization) of agrees with that of
in the hypercube . The main result for DB traffic is
that a dyadic form turns out to be the optimum solution, i.e.,
either the new link is shut off, or it is admitted and then trans-
mission is performed at peak power. In other words, all active
DB traffic links have to use peak power to optimize the overall
throughput according to (5). In fact, can be demonstrated
to be convex with respect to each variable in the hypercube
. Thus, we can write
(6)
where the last max operator is carried over the entire permuta-
tion of binary values belonging to . According to (6),
the maximum of is achieved in one of the vertices of the
hypercube . This corresponds to the extreme choice
of either zero or maximum power level transmission of the
pairs of RTs.
We shall show that this dyadic optimum solution entails an
essential unfairness in resource sharing, which can be mitigated
by a possible change in radio channel path gains (e.g., because
of mobility), if mobility is on a smaller time scale than connec-
tion/session lifetime. However, there is a basic conflict between
system efficiency [optimization of (4)] and fairness.
In Section V, we present a practically feasible, distributed,
(sub)optimal solution of the optimization problem mentioned
in this section for RB traffic (power minimization) and for DB
traffic (throughput maximization). Furthermore, we discuss the
fairness issue for DB traffic.
V. A PROTOCOL FOR RADIO RESOURCE SHARING
This section exploits the models defined in Sections III and
IV in the context outlined in Section II, i.e., UWB ad hoc
networking. Section V-A outlines an MAC protocol for the
RB class with the objective of maintaining QoS requirements
by simple, distributed actions of RTs. In brief, since the
implementation of the minimum power solution would require
a complete reconfiguration of powers to adapt to every network
change due to new accesses or releases, a different, suboptimal,
solution is proposed, which requires the acquisition of a
margin with respect to the minimum power, and which aims at
avoiding reconfigurations, in order to be easier implemented in
the distributed environment under consideration. Section V-B
approaches the issue of defining a practical solution for the
optimization problem for the DB class, that lends itself to a
distributed implementation in the ad hoc environment; here
given find the minimum such that (3)
1726 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 20, NO. 9, DECEMBER 2002
“practical” refers essentially to computational simplicity and
low signaling and measurement complexity. Moreover, an
extension is proposed to mitigate the potential unfairness of the
optimization approach for DB traffic.
In the case of both DB and RB traffic, the proposed subop-
timal solutions are based on local measurements and signaling,
achieving a tradeoff between signaling load and accuracyof the
optimization. In this work in particular, we adopt astep-by-step
approach: we assume links are active and a new link requests
to join in. Then, the aforementioned problems are solved in
order to decide whether the new link is admissible and which
is the (best) bit rate and/or power level for it.
A. A Simple Suboptimal Distributed Algorithm for Power and
Rate Assignment: RB Case
We discuss the case of a UWB terminal pair wishing to start
a new RB link. Let this pair be labeled zero. We assume that the
constraints (2) are met for on-going RB traffic links. The rate
associated to the new link is constrained by QoS require-
ments (e.g., limited transfer delay for real time services, time
deadlines for given blocks of data, a target SNR ).
The power level is chosen so that a margin is acquired over
the minimum required SNR . The margin, denoted as max-
imum sustainable interference (MSI), represents the amount of
additional UWB interference that can be tolerated while main-
taining SNR to below Formally, the th link can be charac-
terized by a margin derived as
(7)
where .
A number of constraints must be met: i) the MSI value must
be nonnegative; ii) the interference due to the new RB link on
the on-going RB links must be limited within their MSIs; iii) the
power level cannot exceed .
The constraints listed above are all met iff
with (8)
The larger the acquired margin, the more probable the ac-
commodating new links without having to rearrange the power
levels, as assumed here. On the other hand, large margins also
imply an inefficient use of the spectrum and of RT energy. A
tradeoff exists between the minimum power solution defined
in Section IV, that implies power level rearrangements at any
change in the air interface, and a high margin approach with
static power levels, chosen once and for all on activation of the
RB link and kept constant throughout the link lifetime.
1) MAC Procedures and Implementation Issues in the RB
Case: In this section, we give a description of a distributed pro-
tocol applying the defined algorithm for the access of RB flows.
We assume that RTs wishing to communicate with QoS require-
ments must initiate an access procedure at the beginning of the
data session to establish a link.
As said previously, the access procedure includes both mea-
surements and signaling operations in order to achieve a good
tradeoff between slight signaling and sufficientnetwork knowl-
Fig. 3. Signaling procedure for the access of an RB flow.
edge. More in detail, the access rule expressed by (8) can be
evaluated on the basis of quantities which can be discoveredby
means of local measurements (e.g., ), with the exception of
, which requires explicit signaling of the receiving
RTs. To be more precise, every receiver must signal its amount
of MSI, whose value cannot be captured by means of measure-
ments, and at the same time, every new transmitter must listen
to this signaling in order to compute .
A scheme describing the complete procedure is shown in
Fig. 3. The procedure involves the transmitter for the new link
(the RT ) and receiver (the RT ), as well as ’s neighboring
receivers (e.g., the RT ), and consists of the following steps.
Step 1) RT contacts to notify its intention to start a
communication at a given rate, this signaling is also
heard by ’s neighboring receiver which in turn
is triggered to signal its MSI; this represents the
signaling phase. Obviously, if other receivers are in
’s MAC domain, they are also requested to answer
giving their MSI.
Step 2) RT measures the perceived interference and noti-
fies the relevant result to ; this is a measurement
phase.
Step 3) RT has now acquired all the information needed to
check the access rule given by (8), and can, thus, per-
form the access check phase. In short, computes
on the basis of the MSI values and of the
interference. The fulfillment of condition (8) means
that a transmission power exists which is compliant
with the requirements of the other links (i.e., with
their MSIs) and simultaneously allows the new link
to acquire a positive MSI, given the required rate and
the target SNR.
Step 4) The final phase is the link activation, during which
a confirmation handshake is performed by and .
This acknowledgment implies also the update of the
MSIs of the already active links; this is performed on
the basis of new measurements. As the transmission
from to starts, , as a new receiver, also com-
putes the amount of its MSI.
CUOMO et al.: RADIO RESOURCE SHARING FOR AD HOC NETWORKING WITH UWB 1727
With reference to the access procedure, a specificimplemen-
tation issue is concerned with the channels supporting signaling.
First of all, the signaling procedure considers multichannel air
interface to be a fact, since it must support concurrent active
links which will interfere each other.
A broadcast channel must be provided that will carry the
access signaling exchange (see the access request). In our
specific context, this implies devoting an a-priori known TH
code heard by everybody. Of course, multiple, concurrent ac-
cesses are possible and, thus, this broadcast channel could be
used simultaneously by several RTs; however, the asynchro-
nism among different transmissions ensures that, even though
reciprocal time shifts are very short (in the order of ns), no
catastrophic collisions occur, but just multiuser interference
arises.
Furthermore, a critical point concerns the signaling of MSIs
by several RTs, which are all triggered by the same event, i.e.,
the access request; the relevant messages must also becollected
by one RT ( in the example). The adoption of UWB facili-
tates the solution of this problem, since all MSI report messages
could be brought onto the same channel, i.e., the same TH code;
the various transmissions—starting with different delays—par-
tially overlap in time, but can still be separated if a number of
receivers are provided, each of which is trying to capture one
transmission.
B. A Simple Suboptimal Distributed Algorithm for Power and
Rate Assignment: DB Case
Let us assume that DB links are active and a new link
wishes to join in. The SNR and peak power constraints are met
for the already established links. Then
(9)
If a new link joins in, the rate of the other DB links are ad-
justed so as to achieve the maximum throughput under the SNR
constraint, that is
(10)
where . As a result of (9) and (10), it
follows that:
(11)
The target function is convex and, hence, it is
maximized either at or at . The inequality
is equivalent to
(12)
This is effectively an admission rule for new DB links, aiming
at the optimization of the overall DB throughput. The rationale
behind (12) is as follows: if the new link starts transmitting, the
target function is increased by (gain), whereas each of
the bit rates of the ongoing DB links will be lowered because of
increased interference (cost). The balance between the gain and
the costs is positive iff (12) is met. As a result, a new DB link
can be stopped in the following cases: 1) ongoing RB links do
not have a sufficient MSI to overcome the increased interference
caused by the new DB link; and 2) the overall DB throughput
would be decreased because of the new DB link entrant.
In order to highlight how the proposed suboptimal solution
could be implemented in a distributed fashion, we remark that
the condition can be interpreted as the
sum of a “gain” term ( , the bit rate of the newly admitted
link, if it can be admitted) and of a “loss” term (the sum of the
amount by which the bit rates of every other interfered link is
decreased) which must result nonnegative. Formally:
(13)
where the component represents the decrease in the rate
of the th link, due to the new access. As a consequence, to be
able to check (13) requires knowledge of the components ,
.
1) The Fairness Issue for DB Traffic: The unfairness of DB
traffic optimization depends essentially on the fact thatthe target
function (4) under consideration aims at maximizing overall
system throughput, as well as on the instantaneous character of
the derived admission control rule (13). To induce a fair radio re-
source sharing, we require the history of each RT to be weighted
into the admission control, so that RTs that have been penalized
may be protected from greedy RTs that already have more than
their fair share of bandwidth.
The key idea is to redefine the target function as a weighted
sum of the bit rates, with time varying weights depending on the
amount of average throughput obtained by each RT up until that
time (we denote this weight as ). The optimal solution is again
either shut-off or transmission at peak power, but a link is ad-
mitted when it improves the “weighted” sum of bit rates, i.e., it
could not improve the current, overall throughput, nevertheless
its throughput has sufficient “merit” (large weight) to deserve
admission.
The fair admission rule is the following, instead of (13):
(14)
1728 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 20, NO. 9, DECEMBER 2002
Below, we identify the weight on the basis of the amount of
bits transmitted by means of a DB data burst. Assume
bursts of length have been transmitted by a given RT in a
time . The averaged time rate, obtained by
an RT for the previous bursts up to the time of admission control
based on (14), is named . The expression to compute after
seconds, since the first request for transmission of the th
burst was issued, is
(15)
where is the amount of bits transmitted of the th burst
within the time interval . The averaged rate (15)
can be evaluated simply by maintaining two state variables (e.g.,
the averaged rate of the first bursts and the overall amount
of time required for these bursts). In accordance with (15), the
weight assigned by the entering link (zero) to the th active link,
is defined as
(16)
The equations above mean that, in the case of RTs receiving
a lower average bit rate than the one received by the requesting
RT, the relevant penalties of their bit rate reductions are ampli-
fied by a weight greater than one and the amount of “unfairness”
(i.e., the difference between the bit rates obtained in their history
by the requesting RT and the disturbed one) rapidly increase.
Conversely, if the requesting RT has a lower average rate than
the disturbed RT, the weight is less than one and this decreases
rapidly as the gap between the two enlarges.
2) MAC Procedures and Implementation Issues in the DB
Case: We now give a description of a distributed protocol, rele-
vant to the algorithm proposed for the DB class of service. Here,
we assume that access is performed per burst, that is, each time
an RT has a new burst ready to transmit, it carries out the access
procedure. In addition, we account for both cases of access, that
is to say, access based only on the optimization rule [i.e., con-
dition (13)] and access based on throughput optimization and
fairness [i.e., condition (14)].
Again, the distributed nature of the algorithm requires both
local measurements (to assess interference and noise before and
after the onset of the new transmission) and signaling.
Fig. 4 represents a scheme of the access procedure in the case
of a DB burst. We continue to refer to as the new transmitting
RT, to as the new receiving RT, and to as a RT neighbor
to .
Step 1) RT contacts by transmitting the access request
message at its maximum power ; this signaling
is also heard by ’s neighboring receiver which
is triggered to adapt the rate to the new, increased
interference and to signal its decrease of rate .
In the case in which fairness is accounted for [see
admission rule (14)], the average rate of is also
updated and signaled. This represents the signaling
phase.
Fig. 4. Signaling procedure for the access of a DB burst.
Step 2) RT measures the perceived interference and noti-
fies the relevant result to ; this is the measurement
phase.
Step 3) RT has now acquired the complete set of infor-
mation to check the access rule given by either (13)
or (14) in order to perform the access check phase.
In brief, it checks whether its transmission results
in an increase or a decrease of the overall (possibly
weighted) throughput.
Step 4) The burst transmission is confirmed by means of a
handshake between and .
If the transmission set-up phase fails, a backoff procedure
is begun in accordance with the various possible mechanisms
presented in Section VI.
As regards implementation issues, these are similar to the
ones presented for the RB case.
VI. PERFORMANCE CASE STUDY:THE DYNAMIC BANDWIDTH
CLASS OF TRAFFIC
In this section, we concentrate our analysis on the support
of the DB class of traffic. We present a simulation model and
the detailed performance analysis and results in order to verify
the effectiveness of the optimization algorithm based on (13),
as well as the fairness brought about by adopting a weighted
throughput as an optimization function [see rule (14)].
A. The Simulation Model
We developed an event-based simulator. We simulated an area
of 110 110 m, where 80 RTs are randomly distributed with
a minimum distance of 1 m. During the simulations, the RTs
are assumed immobile [21]. The propagation model consists of
a deterministic geometric attenuation with distance, where the
path loss exponent is four; the set of transmission parameters
used in the simulation is derived from [2]. The target SNR is
common to all the RTs and is equal to 14.7 dB. The ratio be-
tween the maximum power of an RT and the background noise
power is 2 10 .
CUOMO et al.: RADIO RESOURCE SHARING FOR AD HOC NETWORKING WITH UWB 1729
As far as arrivals are concerned, new bursts arrive according
to a Poisson process at rate burst/s and the burst size has a fixed
length of 200 Kb. For each new burst, the source RT is selected
randomly from among all the RTs, whereas the destination RT
is chosen at random within a circular neighborhood, centered
in the source RT with a 55-m radius (single hop transmission).
Each RT handles a finiteDB queue, able to accommodate up to
80 bursts, where upcoming data bursts line up, if enough room
is available. Once a burst becomes head-of-line, the signaling
and measurement phases before burst transmission are carried
out in order to verify the admission criteria stated in Section V.
These two phases are assumed to last approximately 15 ms. If
the admission check fails, the backlogged RT schedules a new
attempt; a number of attempts takes place and, if all of them
are unsuccessful, the burst is discarded. In particular, we con-
sidered five access disciplines:
— without backoff procedure (WOBP), based on (13) and
with ;
— with backoff procedure (WBP) based on (13) and with
;
— with persistent backoff procedure (WPBP) based on
(13) and with ;
— with fairness (WF) based on admission rule (14) and
with ;
— without admission control (WOAC) where a burst is
transmitted at maximum power without any admission
check as soon as it becomes head-of-line.
Since we assume that the transmitter/receiver signaling mes-
sages are always successfully delivered and do not account for
radio resources allocated to signaling, the performance results
obtained represent the upper limit of achievable performance in
a realistic scenario [22].
The results are calculated by averaging over 25 different net-
work topologies with independent traffic patterns.
B. Performance Results
In this section, we report the main performance results de-
rived in the simulation analysis. All the figures report the be-
havior when considering the access disciplines identified above.
In Fig. 5, the overall achievable throughput in the system
is shown, whereas Fig. 6 reports the individual throughput of
an RT: throughput in Fig. 6(a) is averaged during the entire
transmitter lifetime (i.e., by considering the ON and OFF pe-
riods), whereas in Fig. 6(b) the time period considered coin-
cides only with the activity period dedicated to burst emission.
As the offered load increases, access mechanisms with admis-
sion control outperform WOAC: this is coherent with the fact
that data bursts are transmitted only when an increase in the
overall throughput is foreseen. Moreover, for a total offered load
of less than 32 Mb/s with the traffic parameters used, the higher
the number of attempts, the better the performance. When the
total offered load is greater than 32 Mb/s, the WPBP is not the
best choice because a number of RTs spend much of their time
in the backoff state. In fact, in Fig. 6(b), showing the individual
throughput during ON periods, we can observe that low values
of the offered load lead to a decrease in throughput as the load
increases, due to the growing number of active RTs. On the con-
Fig. 5. Overall throughput.
(a)
(b)
Fig. 6. Individual throughput.
trary, higher values of the offered load result in a higher number
of RTs in the backoff state, depending on the number of at-
tempts: only the RTs in the best locations can maximize their
target function and transmit. Note that an RT is in a “good” po-
sition when it is going to disturb a restricted number of receivers
and contemporarily the achievement of a high bit rate is envis-
aged thanks to the conditions of the current destination.
1730 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 20, NO. 9, DECEMBER 2002
Fig. 7. Burst transmission probability.
Fig. 8. Burst loss probability due to buffer overflow.
As far as WOAC is concerned, both overall and individual
throughput saturate: indeed, when the offered load increases, the
interference level also increases, due to the use of the maximum
power and, as a consequence, the total throughput reaches
its maximum value.
Figs. 7–9 show the results concerning successful burst trans-
mission. We observe that a burst loss occurs due to both the fi-
nite queue length assumption and the backoff mechanism (if it
is not persistent). The procedures with optimization are clearly
better than WOAC: in fact, although WOAC tries to transmit
all the bursts that have arrived, heavy burst losses occur due to
buffer overflow, since queues are heavily backlogged with re-
spect to WPBP, WBP , and WOBP. On the contrary,
in these two latter cases, burst losses due to buffer overflow are
Fig. 9. Burst loss probability due to protocol behavior.
Fig. 10. Mean delay.
negligible, whereas burst losses due to protocol behavior are as
heavy as the number of attempts is low; in the case of WPBP,
the protocol does not discard any burst, but if the offered load
increases, backoff time increases and burst loss, due to buffer
overflow, exceeds the total burst loss probability of WBP.
Nevertheless, although WPBP (for a low offered load)
and WBP (for a high offered load) result in the algorithms
achieving the maximum throughput, there is a tradeoff between
the throughput itself and the mean delay, as we can see in
Fig. 10. In particular, the higher the number of attempts, the
higher the mean delay perceived by a burst. However, the
admission rule in (13) produces a positive impact on burst
CUOMO et al.: RADIO RESOURCE SHARING FOR AD HOC NETWORKING WITH UWB 1731
(a)
(b)
Fig. 11. Performance of the with fairness admission control.
delay: the procedures with admission control actually ensure
reduced delays compared with those measured for the WOAC.
Finally, we analyzed the fairness behavior of the proposed
admission rules (13) and (14). Fig. 11(b) shows the overall
throughput as a function of the offered load for WPBP
(optimization without fairness), WOAC and optimization
with fairness constraints (WF). In this latter case, the overall
throughput saturates to an intermediate level because nonop-
timum transmissions are also accepted, so long as they are
found to be convenient along the line of the fairness constrained
admission rule, even though they cause an overall throughput
reduction. Fig. 11(a) represents the histograms reporting
the number of RTs that have achieved specific throughput
values for a fixed value of the offered load (1.2 burst/s for
each RT). In the case of the optimization rule (13) with no
fairness constraint, the variance of the throughput distribution
is maximum compared with the other algorithms, but the mean
individual throughput is also maximized. When the joint opti-
mization-fairness criterion (14) is used, throughput distribution
is concentrated, whereas the mean individual throughput is
reduced. Performance obtained without any admission control
(WOAC) is the worst. In other words, Fig. 11 confirms that
the aim to guarantee some fairness in resource sharing can be
achieved only by paying a price in terms of a worse perfor-
mance compared with the optimization admission rules.
VII. CONCLUSION
The work presented here focuses on radio resource sharing
issues in ad hoc networking exploiting UWB communications.
The research effort synthesized in this work is twofold.
From a system point of view, the ad hoc concept based on
UWB radio is developed and guidelines for the design of the
MAC protocol are laid out. The basic steps in access procedures
involving both measurements and MAC signaling are defined.
As regards radio resource sharing, key issues are the power
and capacity assignment principles and algorithms. For this
purpose, the power and rate allocation problem has been for-
mulated for both elastic bandwidth data traffic and reserved
bandwidth traffic, the latter being characterized by bandwidth
guaranteed and/or delivery delay thresholds. Assignment has
been posed as an optimization problem that becomes an op-
timum power assignment for guaranteed quality (RB) traffic
and a joint power/rate allocation problem in the case of DB
traffic.
The need for a distributed algorithm means that suboptimum
algorithms are defined, that assume a step-by-step approach.
The major point here is that, even though (sub) optimization
also requires the application of an admission control rule for DB
traffic, good performance is obtained in terms of throughput and
delay. The fairness issue has also been considered within this
framework.
Further directions for the development of this research are:
1) the assessment of performance analysis when both classes of
traffic are considered; 2) the impact of routing: data bursts can
travel from source to destination by means of a multihop route;
there is a strict relationship between optimum route choice
and optimum radio resource assignment on each given link;
3) energy consumption requirements should be introduced, that
would contribute a reduction in power/rate intrinsic unfairness;
and 4) the effect of a moderate mobility should be considered;
this will also reduce unfairness, but could imply a higher
signaling load and resource allocation stability.
REFERENCES
[1] M. Z. Win and R. Scholtz, “Impulse radio: How it works,” IEEE
Commun. Lett., vol. 2, pp. 10–12, Jan. 1998.
[2] M. Z. Win and R. A. Scholtz, “Ultra-wide bandwidth time-hopping
spread-spectrum impulse radio for wireless multiple-access communi-
cation,” IEEE Trans. Commun., vol. 48, pp. 679–690, Apr. 2000.
[3] [Online]. Available: http://www.UWB.org
[4] F. Ramirez-Mireles and R. A. Scholtz, “N-Orthogonal time-shift-mod-
ulated signals for ultra-wide bandwidth impulse radio modulation,” in
Proc. IEEE Communication Theory Mini Conf., Nov. 1997.
[5] S. S. Kolenchery, J. K. Townsend, and G. Bilbro, “Report on perfor-
mance of local power control in peer-to-peer impulse radio network
with bursty traffic,” in Proc. IEEE GLOBECOM ’97, Nov. 1997, pp.
910–916.
[6] F. Ramirez-Mireles and R. A. Scholtz,“Multiple access performance of
ultra-wideband spread spectrum impulse radio using block waveform
modulation,” in Proc. IEEE ICC, vol. 2, June 1988, pp. 775–779.
[7] , “Wireless multiple-access using SS time-hopping and block wave-
form pulse position modulation, part 1: Signal design and part 2: Mul-
tiple-access performance,” in Proc. ISITA Symp., Oct. 1998.
[8] M. Frodigh, P. Johansson, and P. Larsson, “Wireless Ad Hoc net-
working—The art of networking without a network,” Ericsson Review,
vol. 4, 2000.
[9] J.-P. Hubaux, T. Gross, J.-Y. Le Boudec, and M. Vetterli, “Toward
self-organized mobile Ad-Hoc networks: The terminodes project,”
IEEE Commun. Mag., vol. 39, pp. 118–124, Jan. 2001.
1732 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 20, NO. 9, DECEMBER 2002
[10] J. C. Haartsen, “The bluetooth radio system,” IEEE Pers. Commun., vol.
7, pp. 28–36, Feb. 2000.
[11] B. P. Crow, I. Widjaja, L. G. Kim, and P. T. Sakai, “IEEE 802.11 wireless
local area networks,” IEEE Commun. Mag., vol. 35, pp. 116–126, Sept.
1997.
[12] F. Ramirez-Mireles, M. Z. Win, and R. A. Scholtz, “Signal selection for
the indoor impulse radio channel,” in Proc. IEEE VTC Conf., vol. 3, May
1997, pp. 2243–2247.
[13] M. Z. Win, R. A. Scholtz, and M. A. Barnes, “Ultra-Wide bandwidth
signal propagation for indoor wireless communications,” in Proc. IEEE
ICC, June 1997, pp. 56–60.
[14] F. Cuomo and C. Martello, “MAC principles for an ultra wideband wire-
less access,” in Proc. IEEE GLOBECOM, San Antonio , TX, Nov. 2001,
pp. 3548–3552.
[15] S. Lal and E. S. Sousa, “Distributed resource allocation for
DS-CDMA-based multimedia Ad-Hoc wireless LAN’s,” IEEE J.
Select. Areas Commun., vol. 17, pp. 947–967, May 1999.
[16] C. Martello and D. Bocchetta, “Power controlled MAC protocols for
wireless Ad-Hoc networks,” Proc. European Wireless, pp. 319–326,
Feb. 2002.
[17] E. M. Royer and C. K. Toh, “A review of current routing protocols for
Ad Hoc mobile wireless networks,” IEEE Pers. Commun., vol. 6, pp.
46–55, Apr. 1999.
[18] N. Bambos, “Toward power-sensitive network architectures in wireless
communications: Concepts, issues, and design aspects,” IEEE Pers.
Commun., vol. 5, pp. 50–59, June 1998.
[19] S. Lu, V. Bharghavan, and R. Srikant, “Fair scheduling in wireless packet
networks,” IEEE/ACM Trans. Networking, vol. 7, pp. 473–489, Aug.
1999.
[20] A. Baiocchi, F. Capriotti, F. Cuomo, and C. Martello, “Distributed radio
resource sharing with UWB radio,” in Proc. IST Mobile Communica-
tions Summit, Barcelona, Spain, Sept. 2001, pp. 753–758.
[21] D. Kim, “Rate-regulated power control for supporting flexible trans-
mission in future CDMA mobile networks,” IEEE J. Select. Areas
Commun., vol. 17, pp. 968–977, May 1999.
[22] I. F. Akyildiz, D. A. Levine, and I. Joe, “A slotted CDMA protocol with
BER scheduling for wireless multimedia networks,” IEEE/ACM Trans.
Networking, vol. 7, no. 2, pp. 146–158, Apr. 1999.
Francesca Cuomo received the Laurea degree in
electrical and electronic engineering in 1993 (magna
cum laude), and the Ph.D. degree in information
and communications engineering in 1998 from the
University of Rome “La Sapienza,” Italy.
In 1995, she joined Coritel, Rome, Italy, a research
institute on telecommunications, and was responsible
for two consecutive years of the SWAP project in the
radio access area. Since 1996, she has been a Re-
searcher at the INFOCOM Department at the Univer-
sity of Rome. She participated in the European ACTS
INSIGNIA project dedicated to the definition of an Integrated IN and B-ISDN
network. She is now participating in the European IST Whyless.com project
focusing on adoption of the ultra-wideband radio technology for the definition
of an open mobile access network. In this project, she is responsible for the
WP4–Network Resource Manager. She participated also in several national re-
search projects. Her main research interests focus on modeling and control of
broadband integrated networks, intelligent networks, architectures and protocol
for wireless networks, mobile and personal communications, QoS guarantees,
and real-time service support in the Internet.
Cristina Martello received the Laurea degree in
electronic engineering (magna cum laude) from the
University of Rome “La Sapienza,” Italy, in 2000.
Currently, she is working toward the Ph.D. degree in
information and communications engineering at the
same university.
She is participating in the European Whyless.com
project on a mobile access network based on
ultra-wideband radio technology. She also collabo-
rates with Co.Ri.Tel. in the INFOCOM Department,
University of Rome “La Sapienza,” (a research
consortium on telecommunications) for the development of dynamic radio
resource control in the third-generation of mobile wireless systems. Her
main research interests include radio resource control techniques adaptive to
traffic conditions and radio channel quality for cellular radio systems and the
synthesis of MAC protocols for ad hoc oriented networks.
Andrea Baiocchi (M’91) received the Laurea degree
in electronics engineering in 1987 and the Dottorato
di Ricerca (Ph.D. degree) in information and commu-
nications engineering in 1992, both from the Univer-
sity of Roma “La Sapienza,” Italy.
Currently, he is an Associate Professor in commu-
nications at the University of Roma “La Sapienza.”
His main scientific contributions are on traffic control
in ATM networks, queuing theory, and radio resource
management in cellular mobile networks. His current
research interests are focused on traffic models and
dimensioning algorithms for IP networks and on mobile computing, specifi-
cally TCP adaptation and packet scheduling over the wireless access interface.
These activities have been carried out also in the frameworkof national and in-
ternational (European Union, ESA) projects, and have resulted in about sixty
publications in international journals and conference proceedings.
Fabrizio Capriotti received the Laurea degree in
telecommunication engineering (magna cum laude)
from the University of Rome “La Sapienza,” Italy,
in 2001.
Currently, he works with Co.Ri.Tel., in the
INFOCOM Department, University of Rome “La
Sapienza,” (a research consortium on telecom-
munications). He is participating in the European
Whyless.com project on a mobile access network
based on the ultra-wideband technology; his work
regards the definition of MAC protocols for ad hoc
oriented networks and the interworking with IP networks supporting QoS.