Beruflich Dokumente
Kultur Dokumente
number of PRBs in the RS. In the utility function, a fairness 8VHU 6,15ĜĜ 6,15ĜĜ 6,15ĜĜ ȡȡȡȡȡȡȡȡȡȡȡ 6,15ĜĜ 1 56
value is multiplied to every average SINR’s to balance the 8VHU 6,15ĜĜ 6,15ĜĜ 6,15ĜĜ ȡȡȡȡȡȡȡȡȡȡȡ 6,15ĜĜ 1 56
system throughput and the cell-edge user throughput. This ȡȡ ȡȡ ȡȡ ȡȡ ȡȡ
ȡȡ ȡȡ ȡȡ ȡȡ
L ȡȡ ȡȡ ȡȡ 6,15ĜĜLM ȡȡ
metric is based on the power control algorithm in [12] and it ȡȡ ȡȡ
ȡ
ȡȡ
ȡ
ȡȡ
ȡ
ȡȡ
ȡ
is originally introduced to compensate the path loss. Fairness 8VHU. 6,15ĜĜN 6,15ĜĜN 6,15ĜĜN ȡȡȡȡȡȡȡȡȡȡȡ 6,15ĜĜN 1 56
value is defined by the product between the fairness factor
that we assume a constant in a system and the user’s total
path loss to the serving BS. The adjusted SINR that is going Fig. 4. Cost Metric for the Hungarian Method. Note that NRS = K
to be maximized, is defined as
Two-dimensional square cost metric is used as an input
SINR(k,m) = SINR(k,m) × Δfairness to the algorithm, and it can be constructed as in Fig. 4
Δfairness (dB) = θ × P L(k,j) (dB), (5) with the size of NRS × NRS . 6-step algorithm in [10] is
applied per-sector basis and it is a modified form from the
where θ is the fairness factor which decides how fairly the original Munkres’ Assignment Algorithm (also known as the
resource allocation is performed and P L(k,j) is the total path Hungarian Algorithm). The algorithm converts the cost metric
loss to BS j from user k whose serving BS is j. It is a simple into a series of equivalent cost metrics by manipulating rows
method for compensating path loss for the cell-edge user, and and columns through the additions and the subtractions. It
it can also be used to give an advantage to the cell-edge continues converting the cost metrics until it reaches a state
users in the allocation algorithm. If a user is located in a cell where an optimal assignment is obvious. When the algorithm
edge and has a large path loss value, the user’s fairness value finishes, the final equivalent cost metric is consisting of non-
becomes large. Then, the allocation is executed with setting a zero or zero elements. Zero elements at the matrix after the
high priority to this cell-edge user. A large value of θ ensures iterations, imply that the assignment should be done for the
more fairness by sacrificing the total system throughput. corresponding zero-element pair. Due to the space limitation,
The optimization problem is defined by the adjusted SINR a specific implementation is not presented in this paper.
IV. A LTERNATING R ESOURCE A LLOCATION F OR of a simulator run, a certain number of users are uniformly
M ULTIPLE C ELLS distributed in 57 sectors. Then, we compute the static gain
This section addresses the problem of the resource alloca- including antenna gain, path loss and shadowing between each
tion in a multi-cell system. The multi-cell resource allocation user and sectors. Each users selects its sector with the largest
problem can be considered as a centralized problem, however, static gain.
not only the global optimality is not known but also a large An adaptive modulation and coding is used in the simulation
amount of overhead information makes the approach imprac- and the type of modulation and coding scheme is selected
tical. In a distributed manner, if we can expect scheduling in based on the SINR estimations over the allocated bandwidth.
neighboring cells, it can be beneficial to perform a distributed We assume a perfect channel quality estimation through the
optimal allocation. However, as it is impossible to correctly sounding reference signal for every TTI. The Exponential
predict neighboring cell’s allocation or interference level, we Effective SINR Mapping (EESM) [2] is used to compute
propose a cell-alternating1 allocation which is to perform the effective SINR and the Block Error Rate (BLER) curve
allocation by rotation among neighboring cells. computed from the link-level-simulation is referred to decide
For example, consider the cell layout as in Fig. 5. Three transmission success and failure for the effective SINR. The
sectors in one cell compose a rotating allocation group. First scheduled user sets its total transmission power using the
sectors in the rotation are numbered as 1 and the allocations following [12]
for those sectors are performed, while the other neighboring P = min{Pmax , 10 · log10 M + P0 + α · PL + Δmcs } (7)
sectors numbered as 2 and 3 maintain the previous allocation.
After one transmission time interval (TTI), sectors numbered where Pmax is the maximum transmit power from the user, P0
as 2 do the allocation in same way, and so do sectors numbered is the power to be contained in one PRB, α is the path loss
as 3 in next TTI. By doing so, we can avoid a sudden change compensation factor, PL is the path loss, M is the number
of the neighboring sectors’ interferences. On the other hand, of PRBs assigned to the user, and Δmcs is modulation and
since the same numbered sectors change allocation at the same coding dependent value signaled from the base station. The
time, they cannot predict the interference level from each closed loop power control is not considered. Details of the
other. However, their distances are far enough to ignore their simulation parameter are described in Table I.
influence. It can be considered to group with more sectors than
three in order to have farther distance among simultaneous TABLE I
allocation sectors. However, if the rotation group becomes S IMULATION PARAMETER
bigger, the time for sustaining allocation become longer. This Simulation Time 500ms/drop, 10 drops
Layout 19 cells- 3sectors/cell (Wrap-around)
results in being vulnerable to fast fading channels. Scenario (Inter-Sector Distance) Urban Macro(500m)
Urban Micro(200m)
Rural Macro(1732m)
Suburban Macro(1299m)
Traffic Model Full buffer
Thermal Noise Density -174dBm
Bandwidth 10Mhz - 50 PRBs
BLER Target 0.1
HARQ Synchronous
Available MCS QPSK[1-10]
16QAM[12-21]
64QAM[22-29]
Max user power 250mW
Standard deviation of slow fading 8 dB
α, P0 (Power Control) 0.6 , -57 dBm
Propagation model [11]
Fast fading model [11]
60
globally optimal solution.
ACKNOWLEDGEMENT
40
This work was supported by the Korea Science and En-
gineering Foundation (KOSEF) grant funded by the Korea
20 government (MEST). (No. R01-2008-000-20042-0)
R EFERENCES
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
[1] IEEE Std 802.16, “IEEE Standard for Local and metropolitan area
Fairness factor (theta) networks. Part 16: Air Interface for Fixed Broadband Wireless Access
Systems”, October, 2004
Fig. 7. 5% user throughput for θ [2] 3GPP TR 25.814 V7.1.0. Physical layer aspect for evolved Universal
Terrestrial Radio Access (UTRA).
[3] J. Lim, H. G. Myung, K. Oh and D. J. Goodman, “Proportional
Fair Scheduling of Uplink Single-Carrier FDMA Systems”, in IEEE
International Symposium on Personal, Indoor and Mobile Radio Com-
simultaneous allocation, the throughput hardly varies for θ munications (PIMRC), Sep. 2006.
while the response of cell-alternating allocation to θ is quite [4] F. D. Calabrese et al. “Search-Tree Based Uplink Channel Aware
Packet Scheduling for UTRAN LTE,” in IEEE Vehicular Technology
clear. It is because the cell-alternating allocation is better Conference (VTC) 2008 Spring, Marina Bay, Singapore, May 2008.
to track the neighboring sector’s interference and to have [5] O. Nwamadi, X. Zhu and A. Nandi, “Dynamic Subcarrier Allocation
more stable interference. At the same time, it shows how the For Single Carrier - FDMA System,” in Proceedings of 16th European
Signal Processing Conference, EUSIPCO 2008, Lausanne, Switzerland,
Aug. 2008.
TABLE II [6] M. Moretti and A. Todini, “A resource allocator for the uplink of multi-
THROUGHPUT OF THE URBAN MACRO SCENARIO cell OFDMA systems,” in IEEE Transactions on Wireless Communica-
tions, vol. 6, no. 8, pp.2807-2812, Aug. 2007.
α Simul- Cell- Increase [7] F. D. Calabrese et al. “Performance of a Radio Resource Allocation
taneous Alternating Rate Algorithm for UTRAN LTE Uplink,” in IEEE Vehicular Technology
0 Sector Average 8.07(Mbps) 10.39(Mbps) 28.8% Conference (VTC) 2007 Spring, Dublin, Ireland, Apr. 2007
5% User 47.87(Kbps) 27.22(Kbps) -43.1% [8] H.W. Kuhn, “The Hungarian method for the assignment problem,” in
0.3 Sector Average 8.13 10.28 26.5% Naval Research Logistic Quarterly, vol. 2, no. 1, pp. 83-97, 1955.
5% User 47.18 31.42 -27% [9] 3GPP TR 25.892, “Feasibility Study for Orthogonal Frequency Division
0.6 Sector Average 8.14 9.82 20.6% Multiplexing (OFDM) for UTRAN enhancements”, V 6.0.0, June 2004.
5% User 44.42 45.99 3.5% [10] R. A. Pilgrim, Munkres’ Assignment Algorithm. Modified for Rectan-
0.9 Sector Average 8.15 9.16 12.8% gular Matrices, Course notes, Murray State University.
5% User 43.88 55.43 26.3% [11] “Guidelines for evaluation of radio interface technologies for IMT-
1.1 Sector Average 8.17 8.34 2.1% Advanced”’ Doc. 5D//TEMP/90-E, 2nd Meeting of Working Party 5D
5% User 45.30 67.16 48.3% Dubai, 24 June - 1 July 2008.
1.5 Sector Average 8.15 7.70 -5.6% [12] 3GPP TS 36.213 Evolved Universal Terrestrial Radio Access (E-UTRA);
5% User 47.50 69.25 45.8% Physical layer procedures (Release 8) V8.2.0