Joint Mobility Load Balancing and Inter-Cell Interference ... - IEEE Xplore

2 downloads 0 Views 274KB Size Report
Abstract—We consider the joint load imbalance and inter-cell interference (ICI) problems and propose a novel mobility load balancing (MLB) algorithm that is ...
Joint Mobility Load Balancing and Inter-cell Interference Coordination for Self-Organizing OFDMA Networks ¨ u Tuncel and Mutlu Koca Nur Oyk¨ Department of Electrical and Electronics Engineering Bogazic¸i University, Bebek 34342 Istanbul, Turkey Emails: {oyku.tuncel, mutlu.koca}@boun.edu.tr

Abstract—We consider the joint load imbalance and inter-cell interference (ICI) problems and propose a novel mobility load balancing (MLB) algorithm that is compatible with the wellknown ICI coordination (ICIC) approaches relying on fractional frequency reuse (FFR). The algorithm is simulated both under the FFR as well as reuse-1 and -3 frequency planing schemes. In addition, the simulations are performed with commonly used schedulers which show that the proposed MLB+ICIC algorithm provide significant improvements in the degree of load balance, cell-edge spectral efficiency and the number of unsatisfied users. Index Terms—MLB, ICIC, frequency planning

I. I NTRODUCTION Modern multi-cellular systems employing orthogonal frequency division multiple access (OFDMA), such as those envisioned by the 3GPP long term evolution (LTE) [1] standards, feature increasingly dense base station deployments in an effort to provide higher network capacity for large user data traffic.This in turn causes an increase in the inter-cell interference (ICI) especially for the users near the boundary of the cells and thus in their rates. In general “ICI coordination (ICIC)” denotes the set of strategies to improve the network performance by having each cell allocate its resources such that interference experienced in the network is minimized while maximizing spatial reuse. Notice that ICIC has been a key feature of the recently emerged self-organizing networking (SON) concept. In this regard, fractional frequency reuse (FFR) has emerged as an effective ICIC technique in OFDMA based wireless networks. In FFR the overall bandwidth is partitioned within a cell so that the cell-edge users of adjacent cells do not interfere with each other and interference received (and created) by cell-interior users is reduced while using a bigger total spectrum than conventional frequency reuse (FR). The use of FFR in cellular networks leads to natural tradeoffs between performance metrics such as coverage for cell-edge users, network sum-rate and spectral efficiency. As reviewed in [2], [3], two common FFR approaches are strict FFR and soft frequency reuse (SFR). While SFR has higher bandwidth efficiency than strict FFR, it suffers from more interference for This work is supported in part by the Turkish Undersecretariat for Defence Industries within the ULAK project framework and Bogazici University Research Fund under project number 9201.

both cell-interior and cell-edge users. On the other hand, strict FFR is comparatively better in terms of outage probability, network sum-rate and cell-edge user signal-to-interferenceplus-noise-ratio (SINR) measures. Another key reason for inefficient resource utilization and low throughput for the cell-edge users in OFDMA networks is the load imbalance among neighboring cells. In general load balancing (LB) can be done by i) changing the downlink transmit power of evolved Node B (eNB); ii) modifying the handover (HO) regions between adjacent neighbors according to an automatic HO parameter in an operation called mobility load balancing (MLB) [4]; and iii) simply moving a user from a high loaded cell to a relatively less loaded neighboring cell without any HO parameter adjustment. Notice that the LB problem has been considered in a number of works such as [5]–[8]. In [5], [6], resource allocation and mathematical frameworks for MLB are detailed for downlink LTE where the available capacity of only adjacent neighbors of the overloaded cell is considered as the primary design parameter. In [7] and [8], non-adjacent neighbors of overloaded cell are also included in the MLB optimization area. However, with the exception of [9] and [10] which consider a heuristic LB method without any HO parameter adjustment in an FFR network, in the aforementioned works the load imbalance and ICI problems are considered independently. For this reason, we address the load imbalance and ICI problems jointly and propose an MLB algorithm that operates within the FFR frameworks used for ICIC. The proposed joint MLB+ICIC algorithm works in each eNB in a distributed manner considering load conditions of the cells, involves N -th tier neighbors of overloaded cell in MLB optimization area and applies HO parameter adjustment. Both strict FFR and SFR are considered as the frequency planning schemes for ICIC. In both schemes, the inter-cell MLB functionality works in coordination with the intra-cell LB mechanism so as to balance the load distribution between the interior and exterior regions. The proposed algorithm is evaluated with extensive simulations employing commonly used round-robin, proportional fair and best-channel quality indicator (CQI) schedulers, and significant performance improvements are observed in terms of the degrees of load balance, cell-edge spectral efficiency and the number of unsatisfied users in comparison to the systems using conventional frequency reuse schemes.

978-1-4799-8088-8/15/$31.00 ©2015 IEEE

(a) FFR

serve users exceeds the number of total available PRBs in cell c, i.e. Nc,tot , or a predefined threshold. The virtual load of sub-band w in cell c is the ratio of total number of required PRBs to satisfy the users connected to sub-band w and total number of PRBs allocated for sub-band w and formulated as  1 w  ρw N (3) c = c,u w Nc,tot w

(b) SFR

u∈Uc

where Uw c is the set of users connected to cell c in sub-band w w and Nc,tot is the total number of available PRBs both in cell c assigned for sub-band w. The virtual load of cell c becomes

Figure 1: FFR and SFR deployments [2].

II. S YSTEM M ODEL In this paper, an MLB algorithm is presented that operates conjointly with strict FFR (which will be referred simply as FFR for the remainder of this paper) and SFR. Contrary to full frequency reuse (reuse-1) or reuse-N where all or a fraction of the whole bandwidth is used in the entire cell, as illustrated in Fig. 1 FFR and SFR partitions the available bands within a cell between the cell center and cell edge. In FFR, interior regions of cells use a common sub-band whereas the cell edges are assigned mutually orthogonal sub-bands to the neighboring cells. In SFR, interior region of a particular cell is allowed to share the cell-edge sub-bands of the neighboring cells. The boundary between cell-center and cell-edge is depicted with an inner circle with a radius Rint in Fig. 1. The number of users assigned to inner and outer bands are dependent on Rint . We consider the downlink traffic where only the path-loss and small scale fading effects are considered. The received SINR for user u in band w from base station c is given as w SIN Rcu =

σ2 +

Pcw hw cu Gcu Pzw hw zu Gzu z∈Qw c ,z=c



(1)

where Pcw and Pzw are the downlink transmit powers of serving w and interfering cells in the sub-band w. hw cu and hzu denote the corresponding channel fading coefficients, which is assumed to be exponentially distributed. The path loss between the base station and the user is calculated as Gcu = c − u−α where c and u are the coordinates of the base station c and user u and α is the path loss coefficient. Qw c is the set of base stations using the same sub-band w as user u. In FFR, a fixed transmit power is applied for all users in all sub-bands, thus, Pcw =P . For SFR, a power control factor β is employed to create two different power classes for interior and exterior users, i.e. Pcint =P and Pcext =βP where β is typically set as 2 or 4. A. Virtual Load The load of a cell is defined as the the ratio of allocated physical resource blocks (PRBs) to the total number of PRBs in a given cell. Under constant bit rate (CBR), the number of PRBs required to serve user u is calculated as w  N cu =

Du w ] (BW R [SIN Rcu P RB )

(2)

where Du is CBR requirement of user u, BWP RB is the transmission bandwidth of a PRB and R [SIN Ruw ] is the achievable spectral efficiency obtained from link-level simulation for a given modulation and coding scheme (MCS). We consider scenarios with different load distributions including overloaded cells where the total number of PRBs required to

ρc =

1 Nc,tot

 

w  N c,u

(4)

w∈W u∈Uw c

where W the set of inner and outer bands in a cell. B. Handover Procedure We consider both inter-cell HO and intra-cell HO. Inter-cell HO is the transfer of users from one cell to an adjacent cell and is modeled based on the A3 event given [1] such that Rj − Rc > Oc,j − Oj,c + H

(5)

where Rj and Rc are the reference signal received power (RSRP) values from target cell j and serving source cell c, respectively. Oc,j is the cell individual offset (CIO) of cell c with respect to cell j, Oj,c is the CIO of cell j with respect to cell c and H is the hysteresis. Oj,c and Oc,j are kept symmetric i.e. Oj,c = -Oc,j . This ensures that a user handed over from cell c to cell j is not handed straight back to the cell c to avoid any ping-pong HOs. If the condition in (5) holds in the period of time-to-trigger (TTT), inter-cell HO operation is triggered. Intra-cell HO not being a real HO, the users only change the assigned band and is applied to balance the load between inner and outer bands. Intra-cell LB is achieved by chancing the area of inner and outer regions. III. J OINT MLB AND ICIC A LGORITHM The proposed algorithm presented in Algorithm 1 is implemented in each eNB in a distributed manner. Here the inter-cell MLB directs the cell-edge users of each overloaded cell by adjusting the CIO parameter under the constraint that the neighborhood cells do not get overloaded. Once an overload is detected in a cell, it becomes a source eNB (SeNB) and collects the virtual load information from all adjacent neighboring cells to create a potential target eNB (TeNB) list. The SeNB chooses the first entry from the list, considers it as temporary TeNB (T eN Btemp ) and decreases the temporary CIO value of SeNB with respect to T eN Btemp (OSeN B,T eN Btemp ) in steps of Δ from its current value. For each candidate user satisfying the condition in (5), the load created in T eN Btemp , i.e. ρT eN Btemp , is estimated. If ρT eN Btemp is less than the load threshold that limits HOs of users, i.e. ρth,HO , the current OSeN B,T eN B value is updated to the OSeN B,T eN Btemp and OT eN B,SeN B =-OSeN B,T eN B to keep symmetry. T eN Btemp becomes the actual TeNB and the load status of SeNB and TeNB are changed accordingly. The adjacent neighbors of SeNB are the first tier neighbors and the neighbors of the first tier are the second tier neighbors.

If the first tier neighbors do not have sufficient capacity to serve the users which are candidate for HO operation, first tier neighbors are requested to direct some of their load to the second tier neighbors. This means that the second tier neighbors are also included in LB optimization area. As long as SeNB is overloaded and N -th tier neighbors have insufficient resources, they are requested to direct some of their load to (N + 1)-th tier neighbors of SeNB and (N + 1)th tier neighbors are also included in LB optimization area. In the cases of reuse-1 and reuse-3, MLB implies inter-cell MLB and in the cases of FFR and SFR, the inter-cell MLB is coordinated with the intra-cell LB. For cell c, if the load of the inner band is greater than the outer band the current value of the radius of interior region, Rint,c , is decreased or otherwise increased in steps of Δintra and temporary interior temp temp radius value, Rint,c , is achieved. Rint,c is updated to Rint,c and intra-cell HO operation is triggered for the users who are supposed to change their current band. IV. S IMULATION R ESULTS A. Layout and Parameters The system is simulated with 19 cells each operating with 10 MHz bandwidth which corresponds 48 PRBs. The simulation parameters are listed in Table I. The overloaded cell is the central one denoted as “Cell 1”. Cell overload is created with 70 users with 256 kbps CBR. 4 different cases are considered in order to observe the change with respect to the user distribution in the surrounding cells. These cases are Case 1, Case 2, Case 3 and Case 4 where 20, 25, 30 and 35 users are placed on each first tier neighbors. The rest of the cells are assumed to be very low loaded with 8 users per cell. Initial value of Rint is 0.667 Km. SFR power control factors are 1 and 4 which corresponds to cell-edge user gains of 0 dB and 6 dB, respectively. Δ parameter is chosen as 1. Smaller Δ values cause slower convergence whereas larger ones result in radio link failures. Table I: Simulation parameters. Parameter Cell Layout User Distribution Inter-site Distance Cell Edge SINR System Bandwidth Channel Model Nint Traffic Model Initial CIO configuration CIOmax CIOmin Δ Δintra H ρth ρth,HO SIN Rth,min

Value /Assumption Hexagonal - Single Sector Uniform 1 Km 10 dB 10 MHz (48 PRBs) Rayleigh flat fading 24 PRBs 256kbps CBR 0 dB 6 dB -6 dB 1 dB 0.01 km 3 dB 95% of total PRBs 85% of total PRBs -7.04 dB

For convenience, we denote the systems with no intra-cell LB and no inter-cell MLB as no MLB and the systems where N -th tier neighbors of overloaded cell are involved in MLB operation as MLB-N .

Algorithm 1 Joint MLB and ICIC Algorithm 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23: 24: 25: 26: 27: 28: 29: 30: 31: 32: 33: 34: 35:

for t = c to C do N ←1  Begin from 1-st tier neighbors while ρc ≥= ρth do  Overload condition SeN B ← c L ← sort (create a list of potential TeNBs and sort according to the load status) while ρc ≥= ρth ∧ i < size(L) do i←1 T eN Btemp ← L(i) CIOtemp ← OSeN B,T eN Btemp while ρSeN B ≥= ρth ∧ i ≤ size(L) do CIOtemp ← CIOtemp − Δ u ← group (estimate and group the users that satisfy A3 event HO for a given CIOtemp ) if ρT eN Btemp ≤ ρth,HO then T eN B ← T eN Btemp OSeN B,T eN B ← CIOtemp OT eN B,SeN B ← −OSeN B,T eN B ρSeN B ← ρSeN B − ρSeN B,u ρT eN B ← ρT eN Btemp end if end while i←i+1 end while if i > size(L) ∧ ρc ≥= ρth then N ← N + 1 (add (N + 1)-th tier neighbors to the optimization area) end if end while if reuse method is F F R ∨ SF R then if ρouter > ρinner then c c temp Rint,c ← Rint,c + Δintra else temp Rint,c ← Rint,c − Δintra end if temp Rint,c ← Rint,c end if end for

B. Performance Evaluation Metrics The performance is evaluated using the following metrics: 1) Inter-cell MLB index (MLBI): Inter-cell MLBI measures the degree of LB between cells and given as C ( ρc )2 Jinter (x) = c=1 (6) C C c=1 ρc 2 where C is the number of cells.  The inter-cell MLBI takes value in the interval of C1 , 1 and a larger index means a more balanced load distribution between cells. 2) Intra-cell LB index (LBI): Intra-cell LBI measures the degree of LB between sub-bands of cells and given as C 2 (ρw − ρw c ) Jintra (x) = c=1 c . (7) C A smaller index means a more balanced load distribution between bands.

1

0.95 0.8 0.9

0.85

0.8

0.75 0

C. Results The results show the comparative performances of the joint ICIC+MLB algorithm and the conventional schemes. The simulations are performed with round-robin, proportional fair and best-CQI schedulers to better observe the impact of scheduling on the performance. Fig. 2 shows the number of unsatisfied users in Cell 1 for different frequency planning schemes. As seen in the figure even though the number of unsatisfied users are reduced by MLB for all planning schemes, FFR and SFR 6 dB outperforms the reuse-1, reuse-3. For FFR and SFR 6 dB, no unsatisfied users remain after MLB-2. Since the number of unsatisfied users is calculated based on the virtual load statel, this implies that involving second tier neighbors is sufficient to solve the overload condition in Cell 1 for FFR and SFR 6 dB. That is why we compare no MLB with first and second tier MLB systems, i.e. MLB-1 and MLB-2.

Number of unsatisfied users in Cell 1

45

40

35

30

25

20

15

10

5

0

no MLB MLB−1 MLB−2

Reuse−1

no MLB MLB−1 MLB−2

Reuse−3

no MLB MLB−1 MLB−2

FFR

no MLB MLB−1 MLB−2

SFR 0 dB

no MLB MLB−1 MLB−2

SFR 6 dB

Figure 2: Number of unsatisfied users for Case 2 Fig. 3 depicts the inter-cell MLBI among Cell 1 and its first tier neighbors and the intra-cell LBI among 19 macrocells varying with inter-cell and intra-cell LB iterations. In Fig. 3(a), we observe that after MLB-1 the most balanced load distribution is achieved with FFR. The index achieved by SFR 6 is close to FFR whereas reuse-1 provides relatively less balanced distributions and reuse-3 has the worst performance. After MLB-2, inter-cell MLBI increases for all network types. Because of the power factor in SFR 6 dB, the load is more

SFR 6 dB no MLB SFR 6 dB MLB−1 SFR 6 dB MLB−2 FFR no MLB FFR MLB−1 FFR MLB−2

1

Reuse−1 no MLB Reuse−1 MLB−1 Reuse−1 MLB−2 Reuse−3 no MLB Reuse−3 MLB−1 Reuse−3 MLB−2 SFR 6 dB no MLB SFR 6 dB MLB−1 SFR 6 dB MLB−2 FFR no MLB FFR MLB−1 FFR MLB−2

20 40 60 80 Number of inter−cell MLB iterations

(a) Inter-cell MLBI

100

Intra−cell LBI

where |Uc | is the number of users in cell c. 4) Spectral efficiency: Spectral efficiency is defined as the number of bits transmitted per second per Hz. The empirical cumulative distribution function (ECDF) of celledge spectral efficiency gives a clear indication about how MLB enhances the performance. The other tools used to observe MLB performance are the mean cell-edge and the total system spectral efficiency. All of the performance metrics are evaluated for Case 2. The different cases are compared only for the mean cell-edge spectral efficiency.

balanced between inner and outer bands compared to FFR as illustrated in Fig. 3(b). Among 19 cells the load difference between inner and outer bands is the greatest for Cell 1 and we observe that intra-cell LBI is decreasing very fast while the Cell 1 is directing cell-edge users to its neighboring cells. After the cell-edge users of Cell 1 are directed, they create a load imbalance between inner and outer bands of first tier neighbors resulting in an increment in intra-cell LBI.

Inter−cell MLBI

3) Number of unsatisfied users: Number of unsatisfied users is detailed in [5] and calculated as   1 Z = max |Uc | ∗ 1 − ,0 . (8) ρc

0.6

0.4

0.2

0 0

20

40 60 80 Number of intra−cell LB iterations

100

(b) Intra-cell LBI

Figure 3: LBI varying with LB iterations for Case 2 Fig. 4 shows the performance of the proposed algorithm for three different schedulers considering ECDF plots of celledge spectral efficiency of Cell 1 Case 2. For round-robin scheduler, without MLB, the cell-edge spectral efficiency takes the values between 0.25 - 0.5 b/s/Hz. After MLB-1 and MLB2, FFR outperforms the other schemes and the minimum spectral efficiency that %50 of the cell-edge users achieve becomes 1.3 and 1.35 b/s/Hz, respectively. For proportional fair scheduler, the cell-edge spectral efficiency is between 0.9 - 1.4 b/s/Hz without MLB. When MLB-1 is applied, FFR and SFR 6 dB achieve the highest with 2.5 b/s/Hz. Proportional fair scheduler is a channel aware scheduler and that is why it provides a better performance than round-robin scheduler. Because reuse-1 utilizes the whole bandwidth, when it is used with proportional fair scheduling a performance improvement is observed over reuse-3 after MLB. Finally, in the best-CQI scheme, cell-edge users can not be scheduled for reuse-1 and reuse-3. However, for FFR and SFR, the users in the inner and outer bands are scheduled separately. Therefore some celledge users, located in the outer band and with relatively better channel conditions, are more likely to be scheduled. That is why we compare best-CQI scheduler for only FFR and SFR6 dB. For all schedulers, involving second tier neighbors and employing joint MLB+ICIC improves the performance. Table II and III depict the mean cell-edge spectral efficiency of Cell 1 for the four different cases. Notice that, from Case 1 to Case 4 the gain of MLB-1 over no MLB decreases, whereas the gain of MLB-2 over MLB-1 increases. It shows the relation between MLB performance and the number of users located in the first tier neighbors of overloaded cell. In these tables, besides SFR-6, performance for SFR-0 dB is also given. Notice that FFR with round robin scheduler and SFR 6 dB with proportion fair scheduler perform better than other schemes. In best-CQI scheduling only a small number of user can be scheduled and there is a big gap between users in terms of spectral efficiency for which we do not consider the mean

Table III: Mean cell-edge spectral efficiency of Cell 1 in b/s/Hz for proportional fair scheduler

Cell−edge users spectral efficiency ECDF

1 0.9 0.8 0.7 Reuse−1 no MLB Reuse−1 MLB−1 Reuse−1 MLB−2 Reuse−3 no MLB Reuse−3 MLB−1 Reuse−3 MLB−2 SFR 6 dB no MLB SFR 6 dB MLB−1 SFR 6 dB MLB−2 FFR no MLB FFR MLB−1 FFR MLB−2

0.6 0.5 0.4 0.3 0.2 0.1 0 0

0.2

0.4

0.6 0.8 1 1.2 1.4 Cell−edge users spectral efficiency for Cell 1(b/s/hz)

1.6

1.8

Case 1 Case 2 Case 3 2

Case 4

no MLB MLB-1 MLB-2 MLB-1 MLB-2 MLB-1 MLB-2 MLB-1 MLB-2

Reuse-1 1.40 2.25 2.40 1.81 2.30 1.76 2.33 1.44 1.91

Reuse-3 0.96 1.99 2.01 1.30 1.97 1.23 1.98 1.06 1.60

FFR 0.97 2.80 2.80 2.48 2.58 2.40 2.55 1.58 2.35

SFR 0 dB 0.88 2.61 2.41 2.17 2.42 2.08 2.45 1.38 2.23

SFR 6 dB 1.35 3.14 3.14 2.68 2.91 2.51 2.85 1.81 2.86

(a) Round-robin scheduler

Cell−edge users spectral efficiency ECDF

1

it uses the total bandwidth of the system without partitioning. In best-CQI scheduling, after HO, the bandwidth efficiency is not affected by MLB for reuse-1 and reuse-3, so the total spectral efficiency remains unchanged. Additionally, the bestCQI scheduler gives the highest total spectral efficiency.

0.9 0.8 0.7 Reuse−1 no MLB Reuse−1 MLB−1 Reuse−1 MLB−2 Reuse−3 no MLB Reuse−3 MLB−1 Reuse−3 MLB−2 SFR 6 dB no MLB SFR 6 dB MLB−1 SFR 6 dB MLB−2 FFR no MLB FFR MLB−1 FFR MLB−2

0.6 0.5 0.4 0.3 0.2 0.1 0 0.5

1

1.5 2 2.5 3 Cell−edge users spectral efficiency of Cell 1 (b/s/Hz)

3.5

Table IV: System spectral efficiency in kb/s/Hz for Case 2 4

Roundrobin scheduler Proportional fair scheduler BestCQI scheduler

(b) Proportional fair scheduler 1

Cell−edge spectral efficiency ECDF

0.9 0.8 0.7 0.6 0.5

no MLB MLB-1 MLB-2 no MLB MLB-1 MLB-2 no MLB MLB-1 MLB-2

Reuse-1 1.85 1.84 1.68 3.25 3.23 2.95 4.69 4.69 4.69

Reuse-3 1.03 1.02 0.95 1.52 1.50 1.38 1.66 1.66 1.66

FFR 1.68 1.41 1.37 2.52 2.26 2.19 2.90 2.80 2.79

SFR 0 dB 2.06 1.59 1.50 3.17 2.64 2.54 3.69 3.45 3.36

SFR 6 dB 1.81 1.52 1.41 2.93 2.61 2.48 3.69 3.54 3.43

0.4 SFR 6 dB no MLB SFR 6 dB MLB−1 SFR 6 dB MLB−2 FFR no MLB FFR MLB−1 FFR MLB−2

0.3 0.2 0.1 0 0

0.5

1

1.5

2

2.5

V. C ONCLUSION 3

Cell−edge spectral efficieny Cell 1 (b/s/Hz)

(c) Best-CQI scheduler

Figure 4: Performance of MLB jointly with ICIC approaches for different schedulers for Case 2

A novel joint ICIC+MLB algorithm is proposed which improves the system performance by providing more balanced load distributions among cells and between inner and outer regions. The system performance is measured in terms of celledge spectral efficiency and number of unsatisfied users and with round-robin, proportional fair and best CQI schedulers. R EFERENCES

spectral efficiency for this case. Table II: Mean cell-edge spectral efficiency of Cell 1 in b/s/Hz for round-robin scheduler

Case 1 Case 2 Case 3 Case 4

no MLB MLB-1 MLB-2 MLB-1 MLB-2 MLB-1 MLB-2 MLB-1 MLB-2

Reuse-1 0.40 0.68 0.76 0.53 0.66 0.42 0.58 0.42 0.52

Reuse-3 0.42 0.86 0.95 0.60 0.85 0.45 0.74 0.45 0.61

FFR 0.44 1.50 1.51 1.29 1.35 1.03 1.27 0.77 1.24

SFR 0 dB 0.26 1.32 1.33 0.99 1.25 0.57 1.10 0.57 1.01

SFR 6 dB 0.52 1.36 1.40 1.16 1.32 0.94 1.27 0.75 1.24

Table IV shows the change in the total system spectral efficiency in kb/s/Hz for different frequency planning schemes and schedulers. When a user executes inter-cell HO, the resource remaining for existing users of the newly handed-over cell reduces. As a result, a small degradation is observed in the total spectral efficiency of the network. For all three schedulers, overall spectral efficiency is the highest for reuse-1 since

[1] 3GPP, Radio Resource Control (RRC): Protocol Specification (Release 11), TR 36.331, 2012. [2] T. Novlan, J. G. Andrews, I. Sohn, R. K. Ganti and A. Ghosh, “Comparison of Fractional Frequency Reuse Approaches in OFDMA Cellular Downlink,” IEEE Globecom, Miami, USA, pp. 1-5, Nov. 2010. [3] A. G¨ok and M. Koca, “Performance evaluation of frequency planning and scheduling schemes in OFDMA Networks,” IEEE BlackSeaComm 2014, Chisinau, Moldova, pp. 149-153, 2014. [4] S. H¨am¨al¨ainen, H. Sanneck and C. Sartori, LTE Self-Organising Networks (SON):Network Management Automation for Operational Efficiency, Wiley Publishing, 2011. [5] I. Viering, M. D¨ottling, A. Lobinger, “A mathematical perspective of self-optimizing wireless networks,” IEEE ICC’09, Dresden, 2009. [6] A. Lobinger, S. Stefanski, T. Jansen and I. Balan, “Load Balancing in Downlink LTE Self-Optimizing Networks, in Proc. IEEE VTC 2010Spring, Taipei, Taiwan, pp. 1-5, 2010. [7] N. Zia and A. Mitschele-Thiel, “Self-organized neighborhood mobility load balancing for LTE networks,” IFIP Wireless Days, pp. 1-6, 2013. [8] L. Zhang, Y. Liu, M. Zhang, S. Jia and X. Duan, “A Two-Layer Mobility Load Balancing in LTE self-organization networks,” IEEE 13th ICCT, Jinan, China, pp. 925-929, 2011. [9] Z. Li, H. Wang, Z. Pan, N. Liu and Xiaohu You, “Dynamic Load Balancing in 3GPP LTE Multi-Cell Fractional Frequency Reuse Networks,” IEEE VTC 2012-Fall, Qu˜ebec City, pp. 1-5, 2012. [10] K. Son, S. Chong and G. Veciana, “Dynamic association for load balancing and interference avoidance in multi-cell networks, IEEE Trans. on Wireless Com., vol. 8, no. 7, pp. 3566-3576, 2009.