WIRELESS COMMUNICATIONS AND MOBILE COMPUTING
Wirel. Commun. Mob. Comput. 2005; 5:77–93
Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/wcm.289
Duplexing, resource allocation and inter-cell
coordination: design recommendations for next
generation wireless systems
A. Alexiou*
,
y
, D. Avidor, P. Bosch, S. Das, P. Gupta, B. Hochwald, T. E. Klein, J. Ling, A. Loza no,
T. L. Marzetta, S. Mukherjee, S. Mullender, C. B. Papadias, R. A. Valenzuela and H. Viswanathan
Bell Laboratories, Lucent Technologies
Summary
Coexistence of different access technologies, hierarchical cellular deployment, a wide variety of data services,
requirements for transparent operation across different technologies, adaptivity to varying network conditions and
mobility and quality of service (QoS) constraints introduce a number of challenges in the design of future
generation systems and the specification of new air interfaces, such as efficiency and flexibility in the utilization of
spectrum, dynamic resource allocation and exploitation of the multiuser diversity and reconfigurable interference
management and inter-cell coordination. In this paper, three critical issues for the design of next generation
systems are addressed: (i) duplexing, (ii) scheduling and reso urce allocation and (iii) interference and inter-cell
coordination. A number of research directions are presented, which constitute promising potential candidates for
next generation systems specification. Copyright # 2005 John Wiley & Sons, Ltd.
KEY WORDS: channel state information; duplexing; radio resource management; scheduling; interference
co-ordination
1. Introduction
Next generation wireless systems are expected to
deliver a wide variety of data services in a hetero-
geneous communication network environment, which
supports transparent operation across a number of
different technologies, hierarchical and ad hoc struc-
tures, adaptivity to varying traffic and propagation
conditions and satisfies certain quality of service
(QoS) constraints. To this end, the design of next
generation systems and the specification of a new air
interface will have to rely on the exploitation of new
resources, such as the channel state information (CSI),
cross layer and contextual information, and the im-
plementation of optimization strategies for the effi-
cient and flexible utilization of the spectrum available,
the dynamic resource allocation exploiting all types of
diversity (time, frequency, code, space, multiuser) and
the reconfigurable interference management and in-
ter-cell coordination.
In this paper, the challenges associated with these
critical issues for the design of next generation sys-
tems are addressed and a number of research direc-
tions are presented, which constitute promising
*Correspondence to: A. Alexiou, Bell Labs, Wireless Research, Lucent Technologies, The Quadrant, Stonehill Green, Swindon,
Wiltshire SN5 7DJ, UK.
y
E-mail: alexiou@lucent.com
Copyright # 2005 John Wiley & Sons, Ltd.
potential candidates for next generation systems
specification.
First, in Section 2, the duplexing scheme selection
is analyzed and the major benefits and drawbacks of
the two traditional candidates, namely time division
duplex and frequency division duplex, are presented
in terms of link reciprocity, link budget and synchro-
nicity and guard requirements. A new duplexing
approach is proposed, which flexibly combines the
features of the two.
Then, in Section 3, the resource allocation issue is
first analyzed in a cellular network setup in the conte xt
of an orthogonal frequency division multiplexing
(OFDM) air-interface and a number of possible options
on how resources can be assigned across adjacent cells
are discussed. Considering resource allocation within a
cell, three promising scheduling techniques are then
presented: a near capacity multiantenna multiuser trans-
mission scheme, the so-called sphere-encoded multiple
messaging, a distributed scheduling scheme supporting
service differentiation and a joint opportunistic beam-
forming and scheduling scheme exploiting multiuser
div ersity.
In Section 4, interference and inter-cell coordina-
tion are addressed. The delay sensitivity of handoff
algorithms and the benefits of fast cell switching, as an
alternative to soft handoff, are discussed along with
the impact on system performance of the structure of
out-of-cell interference, in the absence of inter-cell
coordination. The benefits of superposition coding as
a throughput-optimal encoding technique for com-
pound channels, in which the random state of the
channel, such as the out-of-cell interference, is un-
known to the transmitter, are considered in a realistic
scenario, in which the transmission rates of the super-
position code have to be chosen from a fixed and pre-
determined rate set.
The importance of self-organization in future wire-
less networks is emphasized, as a means of achieving
adaptive and reconfigurable operation, with base sta-
tions probing the environment around them and ad-
justing accordingly a number of parameters, such as
their antenna configuration and transmit power.
Finally, the paper is concluded in Section 5.
2. Duplexing
Traditionally, the decision on how to partition the
resources available for communication between up-
link and downlink has boiled down to two clear
possibilities: either time or frequency separation.
Each contending option has accepted advantages as
well as clear drawbacks. Time division duplex (TDD)
scheme uses the same frequency band but alternates
the transmission direction in time. Frequency division
duplex (FDD) scheme requires separate frequency
bands for uplink and downlink transmission. An
implicit understanding developed over the years is
that TDD was attractive in microcell systems while
FDD was preferred in wide-area systems.
In this section, the duplexing issue is considered in
the context of a new air interface design, in an effort to
go beyond the traditional paradigm where only pure
FDD and TDD are evaluated, bringing other options
into consideration.
2.1. Paired Versus Unpaired Spectrum
The allocation of spectrum is a highly political issue,
performed by certain national and international agen-
cies and mostly beyond the control of equipment
manufacturers. However, the allocation of spectrum
strongly conditions the duplexing choice because, if
unpaired spectrum is allocated, FDD cannot be used.
Although unpaired spectrum is indeed easier to find,
the historical tendency is nonetheless to assign paired
spectrum for wide-area systems, in which case both
options are possible.
2.2. Link Asymmetry
An immediate advantage of TDD is that it enables an
asymmetric allocation of degrees of freedom between
uplink and downlink although in general not dynami-
cally on a cell-by-cell basis, but rather on a syst em-
wide basis. A similar asymmetry in FDD would
require uneven spectrum blocks, highly unlikely and
very rigid. The relevance of link asymmetry, none-
theless, is still unclear. The UMTS forum forecasts a
2.3 ratio of downlink over uplink for 2010 but this will
depend largely on emerging and yet-to-be-envisioned
applications. In that sense, TDD may be considered as
a favored option but not decisively so because of the
difficulty in predicting whether the asymmetry ratio
will be significantly different from unity.
2.3. Link Reciprocity
Link reciprocity is usually regarded as the most
attractive feature of TDD, which naturally enables it,
at least for low-to-moderate normalized Doppler
spreads. As a result of reciprocity, sophisticated trans-
mit processing schemes that necessitate instantaneous
78 A. ALEXIOU
ET AL.
Copyright # 2005 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2005; 5:77–93
channel information become feasible [1]. The lack of
reciprocity in FDD, in turn, makes these schemes
dependant on the relay of CSI through feedback,
which tends to incur unacceptable delays if conven-
tional transmission techniques are employed.
Two questions arise concerning the link reciprocity
issue:
Whether the majority of users fall within the
Doppler range where reciprocity holds with sufficient
accuracy;
What is, quantitatively, the value of channel infor-
mation availability at the transmitter;
In a cell operating individually, the advantage in terms
of downlink throughput is sizeable if the number of
base station antennas is sufficient [22].
2.4. Link Budget
Let us consider an FDD system that is radiating a steady
power level P. Consider now TDD. If the power level
during the active part of the duplex is kept at P (same
amplifiers used), there is a 3 dB increase in the thermal
noise floor because of the doubling in bandwidth. W ith
a path loss exponent of 3.8, for instance, this results in a
17% reduction in range. In exchange, the ‘average’
transmit po wer over time is halved, which has no impact
on the base station but would extend the battery life in a
mobile terminal. If, in contrast, it is preferred to keep
the range unchanged with respect to the FDD case, the
power radiated during the active part must be doubled.
This would require bigger amplifiers and it would also
result in the same ‘average’ transmit power and thus the
same battery life.
2.5. Synchronicity and Guards
In FDD, uplink and downlink are orthogonal in
frequency, provided there is sufficient separation be-
tween the corresponding blocks. Each side of the link
requires guard bands to accommodate filter roll-offs
(Figure 1). In TDD, temporal orthogonality is only
possible if cells have synchronized uplink and down-
link switch patterns plus gu ard times to account for
propagation delays. This is in addition to guard bands
roughly equal to those in FDD (Figure 2).
Orthogonality is essential in wide-area systems,
otherwise catastrophic interference may take place.
z
The overhead represented by the guard times can
be made as small as desired by extending the duplex
time (time that either link is active). Guard times are
therefore an issue only because the duplex time should
be kept short in order to minimize physical-layer
contribution to latency and also to ensure channel
reciprocity over the widest possible range of Doppler
spreads.
Synchronicity is essential in TDD. This is an
inconvenience with respect to FDD but not insur-
mountable given the availability of low-cost GPS
technology. In the following, we present a guard
time evaluation in order to provide a quantitative
insight on the issue.
Consider the possibility of base-to-base interference
due to lack of synchronicity or else due to propagation
delays. To first-order, propagation between elevated
(over the clutter) base stations is dominated by a
direct path and a reflected path. The resulting path-
loss versus distance is plotted in Figure 3. Remarkably,
z
If the system has a microcell component with street-level
or indoor bases, these may not need to be synchronous.
Fig. 1. Guard bands in frequency division duplex (FDD).
Fig. 2. Guard bands in time division duplex (TDD).
Fig. 3. Received power (dBm) versus distance for a two-ray
propagation model (reflected path) and for the classic COST
231-Hata [23] (direct path) with 1 W transmit power.
DESIGN RECOMMENDATIONS FOR NEXT GENERATION WIRELESS SYSTEMS 79
Copyright # 2005 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2005; 5:77–93
the path-loss exponent for base-to-base propagation
(beyond 1–2 km) is roughly the same as for mobile-
to-base, with such mobile in the clutter. The intercept,
however, is about 47 dB higher. Coupled with higher
transmit power and antenna gain at bases with respect to
mobiles, this confirms that base-to-base interference
would be catastrophic and must be avoided. Specifi-
cally, it takes about 32 rings of cells for the base-to-base
path-loss to have grown to the level of mobile-to-base
path-loss at the second ring. Therefore, guard times
should protect against interference originating up to
several tens of rings, which might be tens of km
depending on cell size and base height. Fortunately,
though, earth curvature poses a limit to the detrimental
consequences of base-to-base propagation. For typical
base heights, this effect sets in at about 30–35 km and
hence it suf fices to guard up to roughly that distance. At
3.3 ms/km (speed of light), this distance corresponds to
about 100 ms guard time. (For added margin, this could
be increased to 150–200 ms.) The amount of overhead
can now be estimated depending on the time between
uplink–downlink transitions, which we refer to as
duplex time. For 1 ms duplex time, for instance, the
overhead would amount to 10–20%. For 2 ms duplex
time, it would constitute 5–10%. The above analysis
provides reasonable assumptions when design choices
are made.
2.6. Link Continuity
A drawback of TDD comes from the periodic inter-
ruptions in the links, which are active only for a
part of the time (usually but not necessarily 50%).
Interestingly, this issue did not exist in circuit-
switched voice systems and thus has not been part
of the traditional discussions on FDD versus TDD
duplexing. This is a new issue that is caused by, and
central to, packet-switched data traffic.
§
Besides higher bandwidth efficiencies, one of the
central goals in the design of future generation sys-
tems is to achieve an order-of-magnitude reduction in
latency. This is being recognized as a necessary
condition for the support of certain envisioned appli-
cations (such as gaming). With discontinuous links,
no message—not even a 1-bit acknowledgement—
can be relayed back with a delay inferior to the duplex
time. This implies that the time taken by a basic
roundtrip at the physical layer level cannot go below
a few ms and thus the aggregate delay experienced by
a packet running through a scheduler and subject to
ARQ (automatic repeat request) can easily be on the
order of 10 ms. This latency propa gates through the
protocol stack posing serious problems to the upper
layers and causing bottlenecks. As a result, some of
the throughput improvements enabled at the physical
layer may not be realized, an issue that becomes
increasingly important as data rates grow.
In summary, the primary issues on which the choice
of a duplexing scheme rests appear to be link recipro-
city and link continuity, each of which favors a
different choice. Similarly, the remaining issues (syn-
chronicity, link budget, symmetry etc.) do not point to
a clear preferred choice either. In light of these facts,
the question that naturally arises is whether it is
possible to combine FDD and TDD in such a way
that the best of each is preserved.
2.7. Band Switching Duplexing
Band switching duplexing has been proposed in [2]
and can be described as follows. Given paired spec-
trum blocks, instead of reserving a block for up link
and the other for downlink, alternate their use every
T s, as depicted in Figure 4. With this scheme reci-
procity is achieved and the channel can be estimated
in each band when it is used for uplink and then
exploited when it is used for downlink. Synchronicity
and guard times are still needed, as in TDD. Both links
are always active (except on guard times). The switch-
ing time T must be selected, so that the channel can be
measured in one time period and then the measure-
ment can be used in the following time period, under
the assumption that the channel has not changed
considerably during that time. In terms of channel
measurement accuracy the switching time needs to be
selected as small as possible. The minimum switching
time is determined by the down time necessary
between reception and transmission and inter-base-
station synchronization time requirements.
Note that band switching is both TDD and FDD. It
is TDD because every unit of bandwidth is used,
alternatively, half of the time for uplink and half of
the time for downlink; it is FDD because, at every
point in time, half the spectrum is used for uplink and
half for downlink.
§
This also includes packet-switched voice, which may be a
replacement for circuit-switched voice.
Fig. 4. Band switching duplexing.
80 A. ALEXIOU
ET AL.
Copyright # 2005 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2005; 5:77–93
Implementation of this duplexing scheme may
introduce some challenges that need to be assessed.
Nevertheless, it provides the best set of tradeoffs and
at the same time the TDD alternative remains an
option in case unpaired spectrum is allocated as
many parameters, such as guard times, synchronicity
etc. are reusable. If link reciprocity fails to provide the
expected gains band switching can be easily reduced
to conventional FDD.
3. Scheduling and Resource Allocation
The efficient utilization of spectrum requires dynamic
resource allocation strategies with the flexibility to
adapt to varying wireless network conditions, user
requirements and QoS constraints, will be one of the
major criteria for the design of a new air interface. In
this section, resou rce allocation strategies are first
analyzed in a cellular network setup in the context
of an OFDM air-interface. Not restricted to a specific
air-interface, three promising resource allocation
within the cell and multiuser scheduling techniques
are then considered.
3.1. Resource Allocation Across Adjacent Cells
within the Context of an OFDM Air-Interface
OFDM is a promising modulation scheme for future
wireless communications, as it can provide large data
rates with sufficient robustness to radio channel im-
pairments. In OFDM, a large number of orthogonal,
overlapping, narrow band sub-channels or sub-
carriers, transmitted in parallel, divide the available
transmission bandwidth. The separ ation of the sub-
carriers is theoretically minimal, such that there is a
very compact spectral utilization. The attraction of
OFDM is mainly due to how the system handles the
multipath interference at the receiver. Multipath gen-
erates two effects: frequency selective fading and
intersymbol interference (ISI). The ‘flatness’ per-
ceiv ed by a narro w-band channel overcomes the former
and modulating at a very lo w symbol rate, which makes
the symbols much longer than the channel impulse
response, diminishes the latter. Using powerful error
correcting codes together with time and frequency
interleaving yields even more robustness against fre-
quency selective fading, and the insertion of an extra
guard interval between consecutive OFDM symbols can
reduce the effects of ISI even more.
When Orthogonal Frequency Division Multiple
Access (OFDMA) schemes are employed, user-
specific subcarriers (tones) are assigned. In a multicell
environment the challenge is to manage resources—
in this case different tones—in an efficient manner, so
that inter-cell interference effects are minimized. In
this section, different resource allocation strategies
and their relative merits are discussed.
A taxonomy of the various options on how re-
sources can be assigned across adjacent cells in an
OFDM system is depicted in Figure 5. Specifically,
the issue of utilizing the sub-carriers or tones of a
given carrier across adjacent cells is addressed. Each
option has a distinct out-of-cell interference charac-
teristic that determines performance. In the following
paragraphs, these options are outlined and their rela-
tive merits discussed. The broad classification at the
Fig. 5. Taxonomy of resource allocation options in multicell orthogonal frequency division multiplexing (OFDM).
DESIGN RECOMMENDATIONS FOR NEXT GENERATION WIRELESS SYSTEMS 81
Copyright # 2005 John Wiley & Sons, Ltd. Wirel. Commun. Mob. Comput. 2005; 5:77–93
![Fig. 3. Received power (dBm) versus distance for a two-ray propagation model (reflected path) and for the classic COST 231-Hata [23] (direct path) with 1W transmit power.](/figures/fig-3-received-power-dbm-versus-distance-for-a-two-ray-1knqcm1z.webp)
