You are on page 1of 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224152684

Coordinated Radio Resource Management for the LTE Downlink: The Two-Sector
Case

Conference Paper · June 2010


DOI: 10.1109/ICC.2010.5502785 · Source: IEEE Xplore

CITATIONS READS
7 75

1 author:

Patrick Hosein
University of the West Indies, St. Augustine
120 PUBLICATIONS   644 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Capacity Building and Research on Smart Grid Technology in the Caribbean Region View project

To Determine the Probability and Predictability of Households Redeeming Coupons View project

All content following this page was uploaded by Patrick Hosein on 08 May 2014.

The user has requested enhancement of the downloaded file.


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings

Coordinated Radio Resource Management for the


LTE Downlink: The Two-Sector Case
Patrick Hosein
Huawei Technologies Co., Ltd.
10180 Telesis Court, Suite 365, San Diego, CA 92121, USA
Tel: 858.882.0332, Fax: 858.882.0350, Email: phosein@huawei.com

Abstract—We consider the downlink of an Orthogonal Fre- Work has also been done (see [4], [5], [6], [7]) on inter-
quency Division Multiple Access (OFDMA) network. OFDMA ference management strategies such as Fractional Frequency
transmission technology will be used in the next generation of Reuse (FFR) and Soft FFR whereby certain resources (RBs)
wireless networks such as the LTE (Long Term Evolution) and
WiMAX wireless standards. In such a channel, multiple User are shared with neighbors while others are not (i.e., used by
Equipments (UE) may each be allocated multiple Resource Blocks only one of two neighboring sectors). This resource partition-
(RBs) in each frame. A RB is a collection of subcarriers spanned ing can be static or can be slowly changed. However, even for
over all symbols that are available for data transmission. We are the dynamic case, the solution cannot be changed fast enough
concerned with the optimal allocation of power and bandwidth to continuously maintain optimality. In this paper we consider
(number of subcarriers) with the objective of maximizing the
sector-wide throughput. Note that since we are considering multi- the case whereby the bandwidth partitioning is optimized over
sector performance then the influence of intersector interference both sectors in each frame and refer to this as the Orthogonal
is also taken into account. In general, this problem is intractable Frequency Allocation (or OFA) strategy.
and so we focus on the two sector case to gain intuition that
can then be applied to more general algorithms. We determine Another approach to resource allocation is referred to as
the regions within which (a) a frequency reuse factor of unity is Coordinated Multi-Point Transmission or CoMP ([8]). In this
optimal, (b) orthogonal frequency allocations is optimal and (c) case, channel information is exchanged among sectors and
joint processing of signals from both sectors is optimal. We also one additional form of resource allocation is the simultaneous
show that it is never optimal to combine two or more of these transmission of the same packet to a UE from two or more
solutions for a given scenario.
sectors. The received signal is then combined at the UE
before decoding. We will include this approach as well in
I. I NTRODUCTION our investigation and refer to it as the Joint Processing at
Receiver (or JPR) strategy for resource allocation. Note that
We consider the downlink of an OFDMA network as pro-
this strategy requires the exchange of UE data among sectors
posed, for example in the LTE [1] and WiMAX [2] standards.
in addition to UE channel information and therefore requires
In such a channel, multiple UEs may each be allocated
significant backhaul resources.
multiple RBs in each frame. Different power allocations can
be made to each of these RBs to take into account the We are interested in the global allocation of resources
performance objectives of the sector as well as the interference to maximize system capacity. Unfortunately this problem is
caused on neighboring sectors. The rate achieved by a UE intractable in general. For the case of two sectors with one RB
therefore depends on (a) the downlink channel conditions per sector and with a single UE per sector, it has been shown
of the UE, (b) the number of RBs allocated to each UE, [9] that binary power allocation is optimal (i.e., either both
(c) the power allocated to each RB and (d) the interference UEs serve the UE within its sector with full power or only one
strategy used by each sector. We must therefore optimize is allowed to serve with full power while the other is turned
over these parameters with the objective of satisfying all QoS off). We extend these results to also take into account the
requirements of all UEs in all sectors. bandwidth dimension. In other words, instead of considering
A vast amount of work has been done on the allocation a single RB we consider multiple RBs per sector. Therefore the
of resources within a single sector. For these, it is assumed optimal allocation of RBs to UEs in each sector as well as the
that the Signal to Interference and Noise (SINR) information optimal allocation of power to these RBs must be determined.
for each UE is reported to the basestation which then makes In [10] a more general formulation (more than two sectors)
allocations to satisfy the performance requirements of the UEs of this problem was formulated. For any pair of sectors,
in its sector (e.g., see [3]). These allocations are made over necessary conditions were provided for the solution in which
the entire available bandwidth in each sector and hence we these sectors use orthogonal frequency resources (the OFA
refer to this as the Frequency Reuse Factor of One (or RFO) solution) but these are not sufficient conditions. We focus on
strategy for resource management. The optimal solution for two sectors but provide necessary and sufficient conditions for
each sector can be determined but this solution is not globally each solution type and we also present complete solutions for
optimal. all scenarios. Illustrative numerical results are then provided.

978-1-4244-6404-3/10/$26.00 ©2010 IEEE


This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings

II. T HE S INGLE R ESOURCE B LOCK C ASE III. T HE M ULTIPLE R ESOURCE B LOCK C ASE
We consider the case of two sectors indexed by i ∈ {1, 2} In this section we consider the more interesting case of
and two UEs indexed by j ∈ {1, 2} such that each sector multiple RBs per sector. Our objective is to maximize the
serves the UE with the same index value. In this section we total throughput over both sectors. Therefore in each sector
focus on the case of M = 1 RBs. Each sector can allocate a we must determine which RBs are allocated to each UE
total of P power with power pk allocated by sector k to its and how much power should be allocated to each of the
UE. We use ri to denote the rate of the UE in sector i and gij scheduled RBs given the total power constraint. Therefore we
to denote the path gain from sector i to UE j. The objective are maximizing over both power and bandwidth while taking
is the sum rate maximization over both UEs. The path gains into account interference between sectors. We show that three
are normalized by the thermal noise so that the thermal noise distinct solutions are possible and the optimal one depends
is unity. We also normalize the rate by the bandwidth of a on the signal and interference channel conditions of each UE.
resource block. The rates achieved by each sector is then given In Figure 1 we depict these strategies. The color of the arrow
by denotes the RB being served, the solid arrows denote the signal
    and the dotted arrows denote interference.
p1 g11 p2 g22
r1 = ln 1 + , r2 = ln 1 + . (1) Given that binary power allocation is optimal for each
1 + p2 g21 1 + p1 g12
RB (the single RB solution) then there are three possible
Note that, for simplicity, we use Natural Logarithms rather combinations for each RB, (a) both sectors each serve their UE
than Base 2 Logarithms for the Shannon capacity since the over the RB, (b) exactly one sector uses the RB, or (c) both
constant term can be taken into account in the bandwidth sectors use the RB to serve a single UE. One other possibility
normalization. Suppose that we wish to maximize the total is that neither sector uses a particular RB. However one can
throughput R = r1 + r2 . First suppose that we fix the power easily show that by using this resource the throughput can be
allocation in sector 2 at p∗2 and maximize over the power further increased and hence this case can be eliminated. We
allocation in sector 1. We have next show that if one RB is used by both sectors then all RBs
    
p1 g11 p∗2 g22 are also used by both sectors. Similarly if one RB is used by
R∗ = max ln 1 + + ln 1 +
p1 1 + p∗2 g21 1 + p1 g12 exactly one sector then all RBs are used by exactly one sector.
Assume that this is not the case and that (for example) both
We can combine the two logarithms to obtain the logarithm of
sectors transmit over RB 1 and that only sector 1 transmits over
the product of their arguments. Since the logarithm function
RB 2. Sector i uses power pik over RB k ∈ {1, 2}. A necessary
increases monotonically then we can find the optimal solution
condition for optimality is the Kuhn-Tucker conditions (i.e.
by maximizing over the product term. If we ignore all constant
water filling). We denote the rate achieved in sector i over RB
terms this is equivalent to maximizing over
j by ri (j) and so for sector 1 we have
p1 g11 p∗ g22 (1 + p∗2 g21 + p1 g11 )  
G(p1 ) = ∗ + 2 . p11 g11
1 + p2 g21 (1 + p∗2 g21 )(1 + p1 g12 ) r1 (1) = ln 1 +
1 + p21 g21
We can simplify this as
and
b + cp1 r1 (2) = ln {1 + p12 g11 } .
G(p1 ) = ap1 +
d + ep1
From the Kuhn-Tucker conditions the rate derivatives with
where a = g11 /(1 + p∗2 g21 ), b = p∗2 g22 (1 + p∗2 g21 ), c =
respect to the allocated power must be equal and so we must
p∗2 g22 g11 , d = 1 + p∗2 g21 and e = (1 + p∗2 g21 )g12 are all
have
positive. Taking the derivative of G with respect to p1 we g11 g11
=
obtain 1 + p11 g11 + p21 g21 1 + p12 g11
∂G cd − eb
=a+ (2) and hence
∂p1 (d + ep1 )2
p12 g11 = p11 g11 + p21 g21 . (4)
and if we take the second derivative we obtain
∂2G −2e(cd − eb)(d + ep1 ) In sector 2 since the power allocated to RB 2 is zero then the
= . (3) gradient with respect to power for the second RB evaluated at
∂p21 (d + ep1 )4
zero must be greater than or equal to the gradient for the first
Therefore if cd ≥ eb then the first derivative (2) is always non- RB and hence we obtain the following,
negative and hence the sum rate is maximized at p1 = P . If g22 g22
cd < eb then the second derivative (3) is always positive and ≤
1 + p12 g12 1 + p21 g22 + p11 g12
hence G is a convex function and so it attains a maximum at
either 0 or P . Hence in either case the optimum is achieved at resulting in
an extreme point no matter the value of p∗2 . The same argument p12 g12 ≥ p21 g22 + p11 g12 .
holds for the second sector with the conclusion that the optimal
Substituting from Equation 4 we get
power allocations are binary, either both sectors transmit with
full power or only one of them transmits with full power. g12 (p11 g11 + p21 g21 ) ≥ g11 (p21 g22 + p11 g12 )
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings

which can be simplified to These equations can be solved to obtain g11 /m1 = g22 /m2 .
Using the fact that m1 + m2 = M we obtain
g12 g21 ≥ g11 g22 .
g11 g22
m1 = M, m2 = M.
Now since we know that g11 > g12 and g22 > g21 then this g11 + g22 g11 + g22
statement cannot be true and hence both types of resource al- Note that these may not be integer values. However, because
locations cannot co-exist. This means that the optimal strategy the objective function is convex then the nearest integer
is that either both sectors use the RFO approach or both use solution will be optimum. We can use these values to obtain
the OFA approach. Let us now determine the optimal solution the power allocations for each RB and then determine the sum
for each of these strategies. rate to obtain
We first consider the RFO strategy. For this case we show  
that uniform power allocation for each sector is optimal. P
ROF A = M ln 1 + (g11 + g22 ) (6)
Again consider two RBs in each sector (RB 1 and RB 2). M
As above we can write the Kuhn-Tucker Conditions for the
power allocation in each sector to obtain, Note that one possible solution is that all RBs are used by
only one of the sectors. For example, it is optimal for sector
p12 g11 + p22 g21 = p11 g11 + p21 g21 , 1 to use all RBs if
   
P g11 P g11
for sector 1 and M ln 1 + > (M −1) ln 1 + +ln(1+P g22 )
M M −1
p21 g22 + p11 g12 = p22 g22 + p12 g12 ,
and a similar expression holds for sector 2.
for sector 2. We can solve these two equations to obtain Next we consider the JPR strategy. In other words, a UE can
p21 = p22 and also p12 = p11 . This means that for these be served by simultaneous transmissions from both sectors.
two RBs equal power must be allocated. We can repeat for We assume that the received signals from both sectors are
any pair of RBs to finally conclude that, if all RBs are used combined to obtain the achievable rate. We ignore possible
by both sectors then uniform power allocation is optimal for beamforming gains which will require additional information
each sector. Therefore the maximum sum rate for the RFO exchange. We also initially ignore any (open loop) transmit
strategy is given by diversity gains but will later take this into account.
     We first investigate the case in which the JPR strategy is
P g11 P g22 used for all RBs. We first check if it is possible to use some
RRF O = M ln 1 + + ln 1 +
M + P g21 M + P g12 RBs to serve UE 1 and others to serve UE 2. Suppose that
(5) RB 1 is used by both sectors to serve UE 1 and RB 2 is used
Now consider the case in which each RB is used by exactly to serve UE 2. The corresponding achieved rates are given by
one sector. First note that the transmissions from each sector
r1 = ln(1 + p11 g11 + p21 g21 ), r2 = ln(1 + p22 g22 + p12 g12 ).
experience no interference from its neighbor. This means that
the optimal strategy for each sector is the uniform allocation For sector 1 we can compute the derivative of the rate with
of power over all RBs that are used for transmission. Let mi respect to power for each RB and set them equal (Kuhn-Tucker
denote the number of RBs allocated to sector i with m1 +m2 = conditions) to obtain
M . The power allocated to each of the RBs used on sector i g11 g12
is therefore P/mi where P is the total power. The total rate = .
1 + p11 g11 + p21 g21 1 + p12 g12 + p22 g22
achieved in sector i is therefore
  We can repeat for sector 2 to obtain
P
ri = mi ln 1 + gii . g21 g22
mi = .
1 + p11 g11 + p21 g21 1 + p12 g12 + p22 g22
If we consider mi as a continuous variable (i.e., very large
We can divide these two equations to obtain g11 g22 = g12 g21
M ), we can determine the second derivative of the rate with
which is false since g11 > g12 and g22 > g21 and so we
respect to mi to obtain
conclude that only 1 UE can be served for the JPR strategy.
∂ 2 ri P gii P gii Given this fact we can now determine the optimal power
2 =− 2 − <0
∂mi mi + mi P gii (mi + P gii )2 allocations.
Assume that UE 1 is served in all RBs. The derivative of
and hence the rate is a convex function of the number of allo- the rate with respect to power in sector 1 must be the same
cated RBs. We can therefore use the Kuhn-Tucker conditions for all RBs and hence we must have
to obtain the optimal solution. Equating the gradients of the
g11
sector throughputs with respect to the number of allocated RBs =c
1 + p1i g11 + p2i g21
we get
    for some constant c or
P g11 P g11 P g22 P g22
ln 1 + − = ln 1 + − .
m1 m1 + P g11 m2 m2 + P g22 p1i g11 + p2i g21 = κ (7)
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings

for some constant κ. Summing over all RBs i we obtain P/2

P/2
P (g11 + g21 ) = M κ. (8)
Therefore, using 7 and 8, the rate for RB i is given by
  Frequency Reuse One
P
ri = ln(1 + p1i g11 + p2i g21 ) = ln 1 + (g11 + g21 )
M
P
and so the sum rate is given by
 
P
M ln 1 + (g11 + g21 ) .
M
Orthogonal Frequency Allocation
Note that if we instead had served UE 2 then we would obtain
a similar expression. Therefore the maximum sum rate for the P/2
JPR strategy is given by
  P/2
P
RJP R = M ln 1 + (max{g11 + g21 , g22 + g12 })
M
Joint Processing at Receiver
(9)
We now investigate if we can mix RBs using the JPR Fig. 1. Possible RRM Solution Strategies
strategy with RBs using non-JPR strategies and improve
capacity. We first show that we cannot have a mixture of RFO
and JPR RBs. Assume that in RB 1 both sectors serve their to be used. Again using the same approach that was used for
respective UEs (RFO) while in RB 2 both sectors serve (for solving the case in which all RBs were assigned using the
example) the UE in sector 1 (JPR). Let pi1 denote the sector OFA approach then we can compute the sum rate in this case
powers for RB 1 and pi2 those for RB 2. Again using the to be
Kuhn-Tucker Conditions we can equate the power gradients  
  P1 g11 + P2 g22
for sector 1 to obtain RRF O = M ln 1 + .
g11 g11 M
=
1 + p11 g11 + p21 g21 1 + p12 g11 + p22 g21 Note that RRF  
O > RJP R and hence the sum rate, for this
and for sector 2 we have subset of RBs, can be increased by using a OFA strategy for
g22 g21 the subset rather than the JPR strategy that was used. Hence
= .
1 + p21 g22 + p11 g12 1 + p12 g11 + p22 g21 we get a contradiction. Therefore RBs using OFA and JPR
We can combine these to obtain strategies cannot be combined. Therefore RBs must either be
allocated completely using either the JPR or the OFA strategy.
g22 (1 + p11 g11 + p21 g21 ) = g21 (1 + p21 g22 + p11 g12 ). Note that we find that that ROF A > RJP R when all RBs
Solving we obtain were used strictly for one or the other strategy. However, for
the joint processing case there may also be beamforming and
g21 − g22
p11 = <0 diversity gains. The transmit diversity gain depends on several
g11 g22 − g12 g21 factors. One important factor is how tightly the transmissions
where the inequality is due to the fact that g22 > g21 and from both antennas (which are at different basestations) can
g11 > g12 . This violates our assumption that p11 > 0 and be synchronized. If this synchronization is not tight then the
hence RFO and JPR RBs cannot be combined. transmit diversity gain may small. To take into account the
Finally we check if OFA and JPR RBs can co-exist. Suppose transmit diversity gain we can include a gain factor that is
that the optimal solution contains both OFA and JPR RBs. dependent on the channel gains for the served UE. In this
Consider the set of JPR RBs. Denote the total power allocated case we obtain
by sector i to these RBs by Pi and denote the number of   
RBs in this set my M  . Note that this must be the optimal R P
JP R = M ln 1 + max {g1k + g2k + α(g1k , g2k )}
allocation for JPR transmissions otherwise the sum rate could M k∈{1,2}
have been improved by using the optimal one. Using the same (10)
approach as was used for the case in which all RBs used the The transmit diversity gain will then determine the region in
JPR approach then the sum rate for this subset of RBs is which JPR outperforms OFA.
  We have therefore shown that there are distinct regions in
  max{P1 g11 + P2 g21 , P2 g22 + P1 g12 }
RJP R = M ln 1 + . which RFO, OFA and JPR are optimal. One can therefore
M investigate the size of such regions for various parameter com-
Now using the total power in each sector for this set of RBs we binations to determine, under what conditions, each strategy
can determine the optimal solution if the OFA strategy were should be deployed.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE ICC 2010 proceedings

1
IV. N UMERICAL R ESULTS
Next we provide an illustrative numerical example. We
assume M = 100 RBs (although the results are independent
0.95 JPR
of this factor if we compare spectral efficiencies). If power is

Distance of UE 2 from Sector 2


uniformly allocated over all RBs then the power per RB is
denoted by p. We normalize all distances by the sector radius
so that the sector radius is 1 unit. We assume that the UE in 0.9 OFA
each sector is placed on a line joining the antennas of each
sectors. We then vary the distance of each UE from its serving
sector along this line. We assume that the spectral efficiency 0.85 RFO
of a UE at the edge of the cell (including interference from the
neighboring sector) is 0.2 bps/Hz. We use a path loss exponent
of 3.5 so that the path loss for a UE at a distance of x from
0.8
a sector is x−3.5 . Therefore, for a UE at the edge we have
 
p
0.2 = log2 1 +
1+p 0.75
0.75 0.8 0.85 0.9 0.95 1
and so p = 0.175. Distance of UE 1 from Sector 1
For the JPR case we model the transmit diversity gain as
follows. We know that, for the case of 2 transmit antennas with Fig. 2. Regions within which each strategy is optimal
equal path losses, the transmit diversity gain is approximately
3dB. In our case the path losses from the two antennas are
different so we instead assume that the gain is with respect to is never optimal) then such a strategy can be eliminated from
the lowest gain channel. Furthermore, since the transmissions consideration when designing radio resource algorithms.
do not originate from the same location then there is a loss In future work we plan to address the issue of fairness.
due to synchronization and so we assume a transmit diversity Given multiple UEs per sector, which strategy is best for
gain of 1.8 dB. This corresponds to allocating resources while maintaining intra-sector UE fairness
and inter-sector sector fairness. We also plan to investigate the
α(g1k , g2k ) = min{g1k , g2k }. case of three sectors facing each other which is a more realistic
scenario.
In Figure 2 we plot the regions in which each of the three
strategies is optimal. The horizontal axis depicts the distance of R EFERENCES
UE 2 from sector 2 while the vertical axis depicts the distance [1] E. Dahlman et. al., “The 3G Long-Term Evolution – Radio interface
of UE 1 from sector 1. We only plot the results near the edge of concepts and performance evaluation”, Proceedings of the IEEE VTC
the sectors since RFO is optimal elsewhere. For this particular Spring, May 2006.
[2] ”IEEE standard for local and metropolitan area networks part 16: Air
scenario we find that as we approach the cell edge the OFA interface for fixed and mobile broadband wireless access systems,” in
approach becomes optimal and then this is followed by the IEEE Std 802.16e-2005, 2005.
JPR approach. Note that when both UEs are at the edge, the [3] P. Hosein, “On the Optimal Allocation of Downlink Resources in
OFDM-Based Wireless Networks”, Lecture Notes in Computer Science,
average sector spectral efficiencies of RFO, OFA and JPR are vol. 3970/2006 , Ch. 15.
0.200, 0.216 and 0.304 respectively and so significant gains [4] C. Koutsimanis, “Inter-Cell Interference Coordination Techniques for
can be obtained for the edge UEs. However this comes at the Multi-Cell OFDMA Networks Supporting Narrow Band and Elastic
Services”, Master of Science Thesis, Royal Institute of Technology, May
cost of additional backhaul communications. RFO requires no 2007.
such communication while OFA requires a small amount of [5] R. Bosisio and U. Spagnolini, “Interference Coordination vs. Interfer-
communication to indicate which RBs each sector is allowed ence Randomization in Multicell 3GPP LTE System”, Proc. WCNC
2008, Las Vegas, USA, April 2008.
to use. However, for JPR the data itself must be transported [6] N. Reider, “Inter-Cell Interference Coordination Techniques in Mobile
to both sectors. Networks”, Master of Science Thesis, Budapest University of Technol-
ogy and Economics, Feb. 2007.
V. C ONCLUSIONS AND F UTURE W ORK [7] J. Heyman, “Intercell Interference Management in an OFDM-based
Downlink”, Master of Science Thesis, Linkopings University, June,
We considered the optimal radio resource management 2006.
strategies for the downlink of a two sector OFDMA network. [8] 3GPP TR 36.814 V1.0.0, Further Advancements for E-UTRA: Physical
We showed that three distinct strategies were possible and that Layer Aspects
[9] A. Gjendemsjo, D. Gesbert, G. E. Oien and S. G. Kiani, “Optimal power
no two can co-exist for a specific scenario. Therefore one can allocation and scheduling for two-cell capacity maximization,” in Proc.
compute the regions within which each of these strategies is RAWNET (WiOpt), Boston, Apr. 2006.
optimal. The size of the regions will of course depend on the [10] R. Etkin, A. Parekh and D. Tse, “Spectrum Sharing for Unlicensed
Bands,” IEEE JSAC Special Issue on Adaptive, Spectrum Agile and
parameters of the network and, for the case of JPR, the specific Cognitive Wireless Networks, Vol. 25, No. 3, pp. 517-528, April 2007.
transmit diversity model used. If one finds that for a particular
network a specific strategy is optimal within a small region (or

View publication stats

You might also like