Public Safety Priority-Based User Association for Load ... - IEEE Xplore

8 downloads 0 Views 7MB Size Report
Jan 27, 2017 - Corresponding author: K. H. Chang ([email protected]). This research was .... (MA-AC) [9], cell breathing techniques [10], and biased-.
Received July 11, 2016, accepted July 26, 2016, date of publication August 4, 2016, date of current version January 27, 2017. Digital Object Identifier 10.1109/ACCESS.2016.2598198

Public Safety Priority-Based User Association for Load Balancing and Interference Reduction in PS-LTE Systems ZEESHAN KALEEM1,2 AND KYUNGHI CHANG1 , (Senior Member, IEEE) 1 Department 2 COMSATS

of Electronic Engineering, Inha University, Incheon 402-751, South Korea Institute of Information Technology, Wah Campus, Wah Cantonment 45550, Pakistan

Corresponding author: K. H. Chang ([email protected]) This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the ITRC (Information Technology Research Center) support program (IITP-2016-H8501-16-1019) supervised by the IITP (Institute for Information & communications Technology Promotion).

ABSTRACT This paper addresses the issues of user association in multi-tier heterogeneous networks (HetNets) to reduce co-channel interference and provide load balancing for proactively offloading users onto mobile personal cells (mPC). Previously, much of the literature discussed the user’ association problem for HetNets with conventional fixed small cells. The problem discussed in the existing literature is easy to analyze owing to fix nature of the small cells. In this paper, we focus on the mPC instead of fixed small cells, which complicates the user association problem due to its nature of mobility. In this paper, we propose the public safety (PS) users priority-based mPC user association (PS-UA) scheme for load balancing and interference reduction in highly fluctuating PS long-term evolution systems. The proposed scheme improves the user-association problem by minimizing call blocking probability (CBP) according to the network load conditions and PS user priority. Moreover, it further improves user signal-to-interference-and-noise ratio by implementing enhanced intercell interference coordination scheme to further reduce the interference to the offloaded users. System-level simulations confirmed the validity of the proposed PS-UA scheme, because it convincingly reduces the CBP for PS users as compared with the conventional static user association scheme. INDEX TERMS Public safety priority, user association, interference reduction, load balancing, PS-LTE, 5G system. I. INTRODUCTION

Next generation mobile communications (5G) systems are targeting 1000× data rate by deploying an extra layer of low-power small cells in existing macrocell-based homogeneous networks [1]. Moreover, since 5G systems are focusing on a user-centric approach rather than a conventional network-centric approach because of benefits like less power consumption, suitability for high mobility and low-latency applications, reliable for public safety (PS) situations, and can provide the same user experience everywhere. These targets can be achieved by deploying a centrally managed softwaredefined networking (SDN) architecture-based 5G mobile personal cell (mPC) [2], which can fulfill users’ demands by providing ubiquitous connectivity according to users’ traffic loads and situations. The mPC are similar like conventional femotcells but they are connected to an evolved packet core (EPC) of an operator via high-speed wireless backhaul links towards, as shown in Fig. 1. Note that the use of an mPC VOLUME 4, 2016

in itself is not a novel concept [3]; however, the deployment of mPC has become more practical due to the introduction of new carrier type (NCT), such as millimeter waves that can be used as backhaul for mPC. The Mobile and wireless communications Enablers for the Twenty-twenty Information Society (METIS) project has also presented the concept of moving cells or mPC as one of the important candidates for the 5G system. Because its deployment will help to improve the link budget for the end user, and also results in better coverage or higher user throughput, mostly in the cell-edge [4]. Some of the possible deployments suitable for mPC in PS situations and ultra-dense network (UDN) scenario are shown in Fig. 1. Here, SDN controller is connected to the EPC by using application programming interface (API), which will enable the operators to efficiently modify and update network policies. Most of the EPC functionalities like serving gateway (S-GW), packet data network gateway (PDN-GW), and mobility management entity (MME) has been moved to

2169-3536 2016 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

9775

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

FIGURE 1. 5G mPC deployment use cases by targeting high data rate and emergency situations.

the cloud and are running on the top of SDN controller as the applications that can be easily managed from anywhere. Since, the SDN controller has knowledge of the network state, i.e. load associated with base stations in the UDN and the respective interference levels, it can efficiently manage the user association problem that can help to meet the capacity, latency, and quality of service (QoS) requirements of the users. The more details of mPC architecture is already explained in our previous research paper [2]. An mPC deployment has the potential to improve QoS for users by deploying mPC close to the targeted deployment area, such as PS situations and high-traffic zones in commercial areas. The mPC can also offer offloading gain by associating more users with the mPC. The two major deployment cases that is, commercial and PS deployment situations can be clearly seen from the Fig. 1. However, this deployment leads to a serious user association problem [5], creates new challenges for user association because of mPC’s moving nature, and also generates severe co-channel interferences because of unplanned deployment of the mPC. This, in turn, limits the use of conventional static user-association schemes (C-SUA) [6] because those schemes have no capability to dynamically associate users with mPC according to the varying network load conditions, especially in PS situations based on PS service priority. In C-SUA, the term static means that 9776

base-stations are fixed and have no capability to move irrespective of the proposed scheme where the mPC moves. This paper investigates the user-association problem for mPCs deployed in PS situations to reduce the call blocking probability and interference between different base stations (BSs). Furthermore, the major problem of co-channel interference between mPC and the conventional macrocell network is also considered, and major conclusions have been drawn on the suitability of using an enhanced inter-cell interference coordination scheme (eICIC) for interference reduction. A. RELATED WORK

Most prior work on load balancing schemes applies only to heterogeneous networks (HetNets) with fixed small cells. The HetNet scenario with fixed small cells is less sensitive to the cell association policy because small cells are fixed with no mobility and emergency service requirements. The mobility of an mPC brings massive inequalities in cell sizes and creates user-association problems. The existing work in the literature on cell association can be generally classified into two main groups [6]: 1) strategies-based on channel borrowing from lightly loaded cells, such as QoS priority-based dynamic fractional frequency reuse (QoS-DFFR) that borrows the band from VOLUME 4, 2016

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

the lightly loaded cells [7], load balancing with selective borrowing (LBSB) [8], etc. 2) strategies-based on traffic transfer to lightly-loaded cells, such as mobility-aware admission control (MA-AC) [9], cell breathing techniques [10], and biasedbased offloading in HetNets [11]. In this paper, we also followed the approach of traffic transfer mentioned under the second group. There has been a lot of efforts in the literature by analyzing the user association problem via optimizing the utility functions for HetNets with fixed small cells, such as maximization of networkwide aggregate utility by partial frequency reuse and load balancing [12], network-wide proportional fairness [13], and α-optimal user association [14]. These schemes have limitations because they have been proposed to optimize system throughput or network efficiency by statically considering user-association problem. Thus, the impact of offloading on UL has not been very well studied and analyzed because of its greater complexity due to varying user’ power and interference situations. Singh et al. [15] proposed a general model to characterize the UL signal-to-interference and noise ratio (SINR) and rate distribution in a mult-tier HetNet as a function of the association rules and power control parameters. However, the proposal has shortcomings because this model can only be applied to specific cases, such as a HetNet with fixed small cells. Moreover, they only modeled the system for biased association by ignoring the further step of applying cell range extension and interference coordination schemes such as eICIC to reduce the UL interference to the offloaded users. Certainly, the existing schemes often ignores the fact that the users and the BS can have different mobility patterns and diverse QoS demands. However, an effective and optimum user-cell association scheme should be able to prioritize users based on urgent needs to provide public protection and disaster relief (PPDR) services during public safety situations and for mission-critical services. For instance, a user who requires services in a disaster situation should be treated differently from the users who are using any commercial application. Thus, the QoS and emergency spectrum resources demand for PS and non-PS users must be fulfilled according to location (i.e., cell-edge or cell center) and the priority of the users in the network. We refer to such additional information about the users as context information. B. CONTRIBUTIONS AND ORGANIZATION OF THE PAPER

In this paper, we propose public safety-users connection priority-based mPC user association (PS-UA) scheme for load balancing and interference reduction in PS-LTE system. This scheme has the capability to associate PS and non-PS users with an mPC according to their connection priority while applying a cell range extension (CRE) offset to the mPC. Finally, to further reduce the interference among the users offloaded to an mPC, and users who are not allowed to offload because of lower priority, we applied an eICIC scheme to reduce the interference among them. In the VOLUME 4, 2016

proposed PS-UA scheme, we considered the user association problem and an eICIC scheme for interference reduction in a HetNet environment. This scheme results in an increase in cell-edge users’ throughput, high gain in the received SINR, more balanced load distribution, and reduction in the call blocking probability (CBP). Moreover, we found that for mPC almost blank subframe (ABS) muting ratio should be dynamically adjusted, irrespective of the fixed small cells scenario. Furthermore, as was analytically proven by Singh et al. [16] for fixed small cells, showing that minimum pathloss association is suitable for the UL scenario, we similarly proved by system-level simulations that minimum pathloss association is suitable for mPC in the PS scenario. Since, in this paper, we are focusing on a PS-LTE system where we have users of different priorities, we also considered a priority indicator, besides only the minimum pathloss factor [16], for user association. II. SYSTEM AND CHANNEL MODELS FOR mPC IN HetNets

In this paper, UL of an LTE-Advanced (LTE-A) system is considered. We consider a K -tier (K = 2) HetNet environment with M macrocell (eNB) sites having L hexagonal sectors (L = 3) in each site, and N randomly deployed mPCs in each sector. We denote all BSs (i.e., eNBs and mPCs) by B = {M + N }, and all users, i.e., macro user equipment (MUE) and mPC UE (mPUE) by U . We assumed that B BS and U users in the k th tier are transmitting with power PBk and Puk , respectively. User u can be associated with only one BS at a time. In eICIC, there is a concept of blank subframe and nonblank subframes (where 1subframe = 2 resource blocks), thus each user can be served by either blank (b) or non-blank (n) resource blocks (RBs) out of a total of R RBs. As we are modeling prioritized system, thus there are both the PS and non-PS users in a system. To priorities users, we reserved the blank RBs for high priority PS users to transmit their data while other users (non-PS user) will quit their transmission to reduce interference. The subframe during this time is called as almost blank subframe (ABS). Let the ABS period be denoted as TABS ; hence, υ = n/TABS is the ABS ratio with n is the non-blank RBs time length and the constraint is that n ≤ TABS . In this paper, we introduce two kinds of (n) indicator variables: 1) the association indicator variables βj,u , (b) βj,u ∈ {0, 1} which indicates whether user u is associated with BS j and is using a non-ABS (n) or blank (b) RB, and 2) the PS priority indicator variable δu ∈ {0, 1}, which will be δu = 0 for non-PS user u and will be δu = 1 for PS user u. Moreover, the users can handover to other base stations to balance the system load but for PS users which has high priority in our proposed scenario will be forced to keep attach with its serving base station (that can be either mPC or macrocells). A. LOAD DISTRIBUTION IN HetNets

Since, we need to associate users with BS j based on loading conditions and the PS priority of the users, the normalized 9777

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

load on eNB j for non-blank and blank RBs is as follows [17]: X (n) (n) 1 (n) ρeNBj /mPCj = βj,u rj,u (1) (1 − υ)R u∈U 1 X (b) (b) (b) βj,u rj,u (2) ρeNBj = υR

The shadowing (Shdu,j ) is modeled by a log-normal distribution with a mean of 0 dB and standard deviation X dB, i.e., different channel models have different X values [19]. The fast fading that occurs due to the moving nature of the mPC or user is modeled by using a pedestrian-B (Ped-B) channel model [20].

For mPC j because only PS users have the high priority to access its frequency band for blank RBs (b) during the ABS period, except for special cases where PS users are very less than non-PS users, and enough free RBs are available for nonPS users. The normalized load on mPC j during (b) RBs is represented as follows: 1 X (b) (b) (b) (b) δj,u βj,u rj,u , ρmPCj = υR u∈U  with δu = 1, PS user (3) 0, non-PS user

C. LOAD AND PS PRIORITY-AWARE PER-USER SINR FOR eICIC

u∈U

(n)

(b)

where rj,u and rj,u are the number of required RBs for user u associated with BS j by using (n) or (b) RBs, respectively. Thus, the total load on BS j can be represented as follows: (n)

(b)

ρeNBj /mPCj = (1 − υ)RρeNBj /mPCj + υRρeNBj /mPCj

(4)

Load balancing of the system is measured by using the Jain’s fairness index, described as. P ( j∈B ρj )2 , (5) τ= P B j∈B ρj2 where ρj represents the load of BS j as calculated in Equations (1), (2), and (3). B. CHANNEL MODEL FOR mPC IN HetNets

The signal transmitted by user u in the k-th tier with power Puk is received at BS j with power Puk Hu,j . The channel Hu,j encompasses the effect of pathloss (PLu,j ), shadowing (Shdu,j ), and antenna gains (AGu,j ) as large-scale fading. Moreover, it also has a fast fading (Fu,j ) component that fluctuates rapidly with time as compared to large scale fading. By adopting these notations, we can model the effect of the channel as follows: Hu,j (dB) = AGu,j − PLu,j − Shdu,j − Fu,j

(6)

In this paper, we calculated the pathloss for user u connected to eNB j by using an urban model [18], described as follows: PLu,j (dB) = 15.3 + 37.6 log10 (R)

(7)

Similarly, the pathloss for user u connected to mPC j by using line of sight (LOS) and non-LOS (NLOS) WINNER + B1 [18] is represented as follows:  PLu,j (dB) = 33.0 + 22.7 log10 (R), LOS Scenario 31.1 + 38.3 log10 (R), NLOS Scenario (8) 9778

Based on the above channel model and user transmit power, the UL per-user SINR of user u associated with eNB or mPC j on non-blank (n) RBs can be written as: 2 Pu,j Hu,j (n) , ∀j ∈ B (9) SINRu,j = P 2 (n) ρl Pu,l Hu,l + σ 2 l∈B,l6=j

The UL per-user SINR of user u associated with eNB/mPC j on blank (b) RBs can be written as: (b)

SINRu,j  2  Pu,j Hu,j   , ∀j ∈ B, j 6 = M ,  2 P (b) ρl Pu,l Hu,l + σ 2 =  l∈B,l6=(M ∪j)    (b) 0, ∀j ∈ M , δl,u = 0 (10) (b)

where δl,u = {0, 1} is a priority indicator, and its value is selected depending upon whether the user is located in the CRE to access the blank (b) RBs area, or is a PS user with high priority or a non-PS user with low priority. III. PS PRIORITY-AWARE USER ASSOCIATION AND eICIC ABS RATIO MAXIMIZATION A. MINIMUM PATHLOSS-BASED AND PS PRIORITY-AWARE USER ASSOCIATION

In a HetNet environment with UL transmission, the analytical and simulation results of Singh et al. [16] proved that minimum pathloss-based association (PLA) is the best choice for associating users with BS j, as compared with other existing schemes like maximum SINR-based association and rate biased-based association [6]. In this paper, we proposed the minimum pathloss-based and PS priority-aware user association, in which user u is located at distance D, and a pathloss PLu,j in the k-tier can be associated with BS j if it maximizes the following function: k ψu,j = arg

max

k∈{1,2,....,K }

k k δu,j (PLmin, + biasj ) u,j

(11)

k where PLmin, is the minimum pathloss of user u from BS u,j th j in the k tier, and bias = {0, 5, 10, 15, 20, 25} is the CRE offset to associate more users with mPC j. That is, increasing bias value leads to the CRE for the corresponding BS j, and therefore, the offloading of more users to the corresponding tier. In this paper, we also discuss which bias factor will be suitable in the sense of mPC load balancing and PS prioritybased user association. VOLUME 4, 2016

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

FIGURE 2. System architecture of the proposed PS-UA scheme in PS LTE system.

B. eICIC ABS RATIO MAXIMIZATION

The mPC is moving to provide a high data rate to the users connected to it. In order to offload more users to an mPC, bias is introduced. Although, by increasing bias value more users are associated with the mPC as shown in Fig. 2 by grey circle for which the radius of the grey circle increases as the bias value increases. This will also result in more interference with the offloaded users located in the CRE because of getting high power from the neighbor eNB. Thus, in order to reduce this interference, eICIC was introduced by third-generation partnership project (3GPP) in Rel. 10, by sending an ABS from eNB j during the highly interfering RBs. In this paper, our objective is to find the optimum value υj∗ of ABS ratio υ = n/TABS that can associate a higher number of PS users with the mPC based on their connection priority for load balancing. Thus, the resulting ABS ratio υ optimization can be written as follows: ( υj (i) + 1/TABS , if i = 1 υj (i + 1) = (12) υj (i) + λ(i), if i > 1 where λ is the sign index showing increase or decrease in the number of served users during ABS subframe i. Thus, λ can be represented as follows: ( +1/TABS , if u(i) > u(i − 1) λ (i) = (13) −1/TABS , if u(i) < u(i − 1) From (12), we can deduce that the optimal ABS ratio υj∗ is obtained when no further increase in the number of served VOLUME 4, 2016

users is noticed for BS j. That is, λ detects the decrease in the number of served users in the CRE, or there are no more high-priority users to be served by the mPC. IV. PROPOSED PUBLIC SAFETY PRIORITY BASED USER ASSOCIATION SCHEME FOR mPC

In this section, to associate the users with an mPC based on their public safety priority, we propose a PS-UA scheme. The major steps involved in the proposed scheme described as follows: A. STEP 1 CONTEXT-INFORMATION COLLECTION

In this paper, our goal is to use context-information (CI) to efficiently schedule the users based on their PS priority. We consider three types of CI: 1) location information indicates users’ position, whether located at the cell-edge or in the cell center. In this paper, the users are divided into celllocation . edge or cell center users based on SINR_thresholdu,j location That is, if SINRu,j < SINR_thresholdu,j , then user u will be treated as a cell-edge user for BS j; otherwise, user u will be counted as a cell center user, 2) User connection priority information is gathered based on the number of users sending an emergency connection request to the BS. In this paper, users are in two categories: PS and non-PS users. To prioritize a user based on connection priority, the details are discussed in Table 1, and finally 3) deployment situations can be either PS or commercial deployment, and are decided based on the number of users sending the emergency service request to 9779

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

TABLE 1. Commercial and public safety users priority table in 3GPP ProSe LTE-A.

the BS. In this paper, we only modeled the PS scenario, where the number of PS users will always be higher than the non-PS users.

TABLE 2. Major implementation steps of the proposed PS-UA scheme.

B. STEP 2 USER ASSOCIATION

In this step, the PS-UA scheme will associate the user with BS j by using user’s CI. As we already discussed in Section III-A, minimum pathloss-based association is the optimal approach. Thus, by using (11), the users will be associated with their nearest BS by the ensuing PS priority constraint. To offload more users to an mPC, the bias factor is introduced, which will increase the cell range of the mPC, and in the result will associate more users with mPC. This offloading will help to increase cell-edge user throughput. But, it should be noted that in case the PS user associated with eNB and getting good service would not be offloaded to mPC. C. STEP 3 SDN-BASED eICIC FOR FURTHER INTERFERENCE REDUCTION

In case the interference is not less than a predefined threshold then then SDN-based centrally managed eICIC will be considered to solve the problem that occurred by introducing the bias factor. In eICIC the concept of almost blank subframe (ABS) is defined as a transmission subframe, where no data signal will be transmitted from the eNB, but only the most critical information required for the system will be transmitted to provide support to legacy non-PS users connected to the eNB. The ABS transmission from the eNB is necessary because if no ABS is transmitted, then the users offloaded to the mPC from applying CRE will receive the high power transmit signal from the eNB. Then in turn CRE will result in more interference. Thus, during the subframes where the eNB transmits ABS, the low-power mPC is able to schedule offloaded PS or non-PS users. Thus, the proposed PS-UA will reduce the interference, and hence, will maximize system throughput and cell-edge throughput. The main steps of PS-UA scheme are summarized in Table 2. V. SYSTEM-LEVEL SIMULATIONS FOR PERFORMANCE EVALUATION OF THE PROPOSED PS-UA SCHEME FOR 5G mPC A. SIMULATION ENVIRONMENT AND ASSUMPTIONS

To evaluate the performance of the proposed PS-UA scheme, system-level simulations are performed based on the 3GPP 9780

specifications and assumptions under the public safety scenario. The main simulation parameters are summarized in Table 3. We focus on the urban macro layout scenario because it is mandatory for both general and public safety scenarios. VOLUME 4, 2016

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

TABLE 3. System-level simulation parameters.

The deployment layout consists of 7 eNB sites following a hexagonal layout with an inter-site distance (ISD) of 500m. In the deployed scenario, the center site with 3 sectors is considered the region of interest (ROI), with the UE in other cells only providing interference. The ROI consists of eNBs, mPCs, non-PS UE, and PS UE with all UE are deployed outdoor, with 60% of deployed UE being PS-UE while 40% is non-PS UE in a cell. We deployed 5mPC per sector, and all the simulations are done under the channel model discussed in Section II-B. B. SIMULATION RESULTS AND DISCUSSIONS

In this section, we compare the performance of the proposed PS-UA scheme for mPCs. We compared the proposed adaptive PS-UA scheme performance with the conventional static user allocation (C-SUA) scheme. In order to prove the benefits achieved from the moving nature of the mPC, we also compared the performance with the fixed small cell deployed in the HetNet environment. 1) RANDOM WALK MOBILITY MODEL FOR MOVING mPC

There are two types of mobility model in the literature that can be used in the simulation of networks: traces and synthetic models. Different synthetic entity mobility models for safety networks are available in the literature, but we considered the random walk mobility model for moving mPC in our simulation. We opted for this model because it is closer to the practical environment, since most of the phenomenon in nature are random. VOLUME 4, 2016

In this mobility model, an mPC moves from its current location to a new location by randomly choosing a direction and speed for travel. The new speed and direction are both chosen from pre-defined ranges, that is, from [Speed_min, Speed_max] and [0, 2π ], respectively. Each movement in the random walk mobility model occurs in constant time interval t at the end of which a new direction and speed are calculated. Fig. 3 shows an example traveling pattern of an mPC using the random walk mobility model starting from a randomly chosen position, while the speed of the mPC is set at 3Km/h.

FIGURE 3. Traveling pattern of an mPC using the random walk mobility model.

2) PER-USER AVERAGE RECEIVED SINR UNDER THE PS-UA SCHEME

In order to test the performance of the proposed PS-UA scheme, we simulated it for the public safety scenario under two different situations: 1) without giving priority to PS users when associating them with an mPC and 2) by giving priority to PS users when associating them with an mPC. Furthermore, we also compared the performance of the C-SUA scheme which did not give any priority to PS users while associating with an mPC, and also neglected the location and deployment situation for associating the users. From the simulation results shown in Fig. 4, we can notice that the conventional C-SUA scheme results in severe SINR loss because of not having the capability to dynamically associate the users based on their connection priority. The other main reason for SINR degradation in the C-SUA scheme is CRE area because in C-SUA the user even will be offloaded to a small cell by giving bias, but the offloaded users will still receive high interference power from the eNB. Thus, in order to cope with this situation the proposed PS-UA with PS priority based user association scheme outperforms the C-SUA scheme because it cares about users’ priority and mPC load conditions when associating the users. Furthermore, the proposed scheme also applies eICIC scheme for interference reduction; hence it results in better SINR. 9781

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

FIGURE 4. Per-user received SINR under the proposed PS-UA scheme compared to the conventional SRA scheme for a PS scenario.

3) EFFECT OF VARYING CELL RANGE EXTENSION OFFSET ON USER ASSOCIATION

In this section, we checked the effect of varying bias values within the range 0-25 dB with steps of 5 dB to provide cell range extension for fixed and mobile personal cells. We simulated this to check the effect of varying CRE on user association percentage variations for fixed and mPCs. From the simulation results in Fig. 5, we can see that there is fluctuation in the number of user associations with an mPC. That is, even at some high CRE values, the number of associated users is quite less. The reason for this uneven association is the PS priority-based users association for mPC irrespective of only minimum pathloss-based user association in a fixed small cell. If there are more PS users in the vicinity of an mPC, then first of all that PS users will be allowed to associate with the mPC, and non-PS users will be allowed only to associate with the mPC when enough free spectrum is available. Hence, from these results we conclude that for mPC high bias value selection is not always a good option because of the dynamically fluctuating environment and prioritization of users and some optimization schemes are needed to find a solution. Hence, in this paper, the proposed PS-UA scheme adopts the appropriate bias factor according to mPC load situations and connection priority.

FIGURE 5. Effect of varying the bias factor to provide cell-range extension in fixed and mobile personal cells. (a) Fixed small cell with constantly increasing percentage of user association, and (b) mPC with dynamically fluctuating users’ association percentages.

4) ABS MUTING RATIO MAXIMIZATION AND APPROPRIATE CRE BIAS OFFSET OPERATING POINT SELECTION

The users’ system edge throughput for a scenario with 5 mPC per eNBs is shown in Fig. 6. The results are presented for different CRE bias offset and different ABS muting ratios υ ranging from 1/8 to 7/8. For each ABS muting ratio, CRE bias offset values from 0 to 25 dB are simulated. The reason for simulating different ABS muting ratios υ is to find the optimum ABS muting point υ ∗ for eICIC in case of fixed small cell and mobile personal cell. Based on the simulation results in Fig. 6, we see that deployment of an mPC with an eNB almost doubles the edge throughput over the fixed small cell with eNB scenario at 9782

FIGURE 6. Variation in user edge throughput in accordance with CRE bias for fixed and mobile personal cell under the PS scenario.

3dB CRE bias offset and no eICIC. However, it should be noted that only by increasing CRE bias offset, the users’ edge throughput will not increase because the offloaded users on fixed or mobile cells will get the interference from the neighbor cells. This trend can be clearly noticed from Fig. 6 where user throughput decreases for both the fixed VOLUME 4, 2016

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

and mobile cases after reaching a peak at some CRE bias offset values. Thus, higher CRE bias offsets can only be beneficial when eICIC is enabled to negate the interference effect on the offloaded users. We considered the proposed PS-UA scheme when applying CRE or eICIC for an mPC that can dynamically allocate the resources according to the situation, as compared to the C-SUA scheme. Thus, from Fig. 6 it can be clearly noticed that PS-UA outperforms the C-SUA scheme in users’ edge throughput when compared at the same ABS muting ratio and offset values. However, there are some rare cases where C-SUA applied to fixed small cell perform better than the proposed scheme such as for ABS muting ratio of 4/8 with CRE bias offset of 14.5 dB. This happens because the proposed scheme did not allow offloading of non-PS users that existed in the mPC CRE area to care the quality of service of PS users. Moreover, we can also see that the optimum ABS muting ratio υ ∗ is 2/8 with a CRE bias factor of 8.5 dB for an mPC, and after that the peak of the user edge throughput for different ABS muting ratio υ and CRE bias offset also decreases. The reason for this decrease is quite often because, as bias offset increases more users try to offload to an mPC or fixed small cells, but due to high load conditions, throughput decreases. Thus, in order to compensate this throughput degradation loss, the density of mPC or fixed small cells deployed should be increased. This scenario will be considered in our future works.

FIGURE 7. Call blocking probability for PS and non-PS users under the proposed PS-UA and C-SUA schemes in different load conditions.

5) CALL BLOCKING PROBABILITY REDUCTION UNDER THE PS-UA SCHEME

Call blocking probability is minimized under the proposed PS-UA scheme as compared with the C-SUA scheme. From Fig. 7, it can be clearly noticed that when the proposed PS-UA scheme and the C-SUA schemes are compared for PS users for different load situations, the proposed scheme clearly outperforms C-SUA for all load conditions, because a PS user will always be allowed to offload to an mPC in a high load situation, which in turn will reduce the CBP. On the other hand, because of prioritizing PS users in the VOLUME 4, 2016

FIGURE 8. Load balancing under PS-UA and C-SUA schemes.

proposed scheme, the non-PS users suffers little high compared with the C-SUA scheme. The reason for this increment is quite obvious because non-PS users will get fewer chance to offload to an mPC which increases CBP. But this degradation is compensated for because overall system CBP decreases in the proposed scheme. Thus, the proposed scheme is suitable for high loading conditions because it provides a convincing decrease in the overall system CBP. 6) LOAD BALANCING UNDER THE PROPOSED PS-UA SCHEME

Load balancing of the system is measured by using the Jain’s fairness index, as described in (5). The larger the value of τ , the more balanced the load distribution among the given cells, and vice versa, where the value of τ always lies between the interval [1/B, 1]. Fig. 8 plots the load balancing index (LBI) for PS UE, non-PS UE, and the whole system under the proposed PS-UA and C-SUA schemes. The simulation results clearly show that by increasing the number of users, the LBI decreases because of the availability of limited resources, and hence the system becomes unbalanced. It is clearly seen that the PS-UA scheme can achieve high LBI in terms of the PS UE and for the overall system as compared to the C-SUA scheme. For example, when the PS-UA scheme is compared with C-SUA for a number of users at around 60 per cell there is 4.3% and 1.4% gain in LBI for PS users and the overall system, respectively. The reason for more balanced load distribution in the PS-UA scheme is provisioning of priority to PS users when offloading to the mPC. For C-SUA, however, we can easily notice that it can achieve a lower LBI than PS-UA scheme because it has an equal probability of offloading users which disturbs the balance of small cells. Moreover, the quality of service will also degrade. Thus, the proposed PS-UA scheme outperforms the existing schemes in terms of load balancing even under high loading conditions. 7) USERS AVERAGE THROUGHPUT PERFORMANCE UNDER THE PS-UA SCHEME

For the proposed PS-UA scheme, we find that the proposed PS-UA outperforms in terms of edge, average, and peak 9783

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

users around 4.3%. The reason for this improvement is quite obvious as both high-priority PS UE and low priority non-PS UE have the opportunity to access the spectrum of the mPC according to the load situation. Furthermore, we also find that for an mPC, the optimum ABS muting ratio is 2/8 with a CRE bias factor of 8.5 dB. The reason for this decrease is quite often because, as bias offset increases, more users try to offload to an mPC or fixed small cells, but due to high load conditions edge throughput decreases. Thus, in order to compensate this throughput degradation loss, the density of the mPC deployed should be increased. This scenario will be considered in our future works. FIGURE 9. User throughput under PS-UA and C-SUA schemes.

throughout. Specifically average throughput of UE shows 95.5% and 52.91% improvement by using PS-UA compared to C-SUA and PS-UA w.o. PS priority schemes, respectively, as shown in Fig. 9. This throughput improvement is because both the high-priority PS UE and low priority non-PS UE has the opportunity to access the spectrum of mPC according to the load situation, and RBs are also dynamically allocated based on their service priority to all users in a system by considering their location and priority. Furthermore, the application of fractional power control (FPC) scheme also helps to reduce the interference between the users. In LTE, FPC is used to calculate the power of the users on physical uplink shared channel (PUSCH) based on their distance from the base-station by using:    10 log10 (MPUSCH ) PPUSCH = min Pmax , (14) +PO_PUSCH + α · PL where α is the cell-specific pathloss compensation factor, PL is the downlink pathloss estimate of the serving cell, MPUSCH are the numbers of allocated resource blocks to each UE, Pmax is the UE maximum allowed allocated power, and PO_PUSCH is the parameter to control SINR target. Therefore, all these factors brings the opportunistic throughput gain for both PS and non-PS UE, and improved system throughput. Hence, the proposed scheme outperforms the C-SUA scheme due to its capability to dynamically adapt according to the system load conditions. VI. CONCLUSIONS

This paper proposes a new public-safety priority-aware user association scheme for interference reduction and load balancing in a PS-LTE system. To the best of the author’s knowledge, the proposed PS-UA scheme is the first to take care of context-information by considering the moving nature of an mPC and the PS priority of users when associating users with an mPC. This results in a significant gain in user edge throughput and load balancing, and a reduction in call blocking probability as compared to the existing static user association schemes in the literature. For instance, by using the proposed PS-UA scheme, there is load balancing gain for 9784

REFERENCES [1] J. G. Andrews et al., ‘‘What will 5G be?’’ IEEE J. Sel. Areas Commun., vol. 32, no. 6, pp. 1065–1082, Jun. 2014. [2] Z. Kaleem, Y. Li, and K. Chang, ‘‘Architecture and features for 5G mobile personal cell,’’ in Proc. Int. Conf. ICT Converg. (ICTC), JeJu, South Korea, Oct. 2015, pp. 164–166. [3] C.-H. Lee, S.-H. Lee, K.-C. Go, S.-M. Oh, J. S. Shin, and J.-H. Kim, ‘‘Mobile small cells for further enhanced 5G heterogeneous networks,’’ ETRI J., vol. 37, no. 5, pp. 856–866, Oct. 2015. [4] J. F. Monserrat, G. Mange, V. Braun, H. Tullberg, G. Zimmermann, and Ö. Bulakci, ‘‘METIS research advances towards the 5G mobile and wireless system definition,’’ EURASIP J. Wireless Commun. Netw., vol. 2015:53, pp. 1–16, Mar. 2015. [5] H. S. Dhillon, R. K. Ganti, F. Baccelli, and J. G. Andrews, ‘‘Modeling and analysis of K-tier downlink heterogeneous cellular networks,’’ IEEE J. Sel. Areas Commun., vol. 30, no. 3, pp. 550–560, Apr. 2012. [6] Q. Ye, B. Rong, Y. Chen, M. Al-Shalash, C. Caramanis, and J. G. Andrews, ‘‘User association for load balancing in heterogeneous cellular networks,’’ IEEE Trans. Wireless Commun., vol. 12, no. 6, pp. 2706–2716, Jun. 2013. [7] Z. Kaleem, B. Hui, and K. Chang, ‘‘QoS priority-based dynamic frequency band allocation algorithm for load balancing and interference avoidance in 3GPP LTE HetNet,’’ EURASIP J. Wireless Commun. Netw., vol. 2014:185, pp. 1–18, Nov. 2014. [8] S. K. Das, S. K. Sen, and R. Jayaram, ‘‘A dynamic load balancing strategy for channel assignment using selective borrowing in cellular mobile environment,’’ Wireless Netw., vol. 3, no. 5, pp. 333–347, Oct. 1997. [9] L. B. Le, E. Hossain, D. Niyato, and D. I. Kim, ‘‘Mobility-aware admission control with QoS guarantees in OFDMA femtocell networks,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Budapest, Hungary, Jun. 2013, pp. 2217–2222. [10] Y. Bejerano and S.-J. Han, ‘‘Cell breathing techniques for load balancing in wireless LANs,’’ IEEE Trans. Mobile Comput., vol. 8, no. 6, pp. 735–749, Jun. 2009. [11] S. Singh and J. G. Andrews, ‘‘Rate distribution in heterogeneous cellular networks with resource partitioning and offloading,’’ in Proc. IEEE Global Commun. Conf. (GLOBECOM), Atlanta, GA, USA, Dec. 2013, pp. 3796–3801. [12] K. Son, S. Chong, and G. Veciana, ‘‘Dynamic association for load balancing and interference avoidance in multi-cell networks,’’ IEEE Trans. Wireless Commun., vol. 8, no. 7, pp. 3566–3576, Jul. 2009. [13] T. Bu, L. Li, and R. Ramjee, ‘‘Generalized proportional fair scheduling in third generation wireless data networks,’’ in Proc. IEEE INFOCOM, Apr. 2006, pp. 1–12. [14] H. Kim, G. de Veciana, X. Yang, and M. Venkatachalam, ‘‘Distributed α-optimal user association and cell load balancing in wireless networks,’’ IEEE/ACM Trans. Netw., vol. 20, no. 1, pp. 177–190, Feb. 2012. [15] S. Singh, X. Zhang, and J. G. Andrews, ‘‘Uplink rate distribution in heterogeneous cellular networks with power control and load balancing,’’ in Proc. IEEE ICC, Jun. 2015, pp. 1275–1280. [16] S. Singh, X. Zhang, and J. G. Andrews, ‘‘Joint rate and SINR coverage analysis for decoupled uplink-downlink biased cell associations in HetNets,’’ IEEE Trans. Wireless Commun., vol. 14, no. 10, pp. 5360–5373, Oct. 2015. [17] J. Ben Abderrazak, A. Zemzem, and H. Besbes, ‘‘QoS-driven user association for load balancing and interference management in HetNets,’’ in Proc. 6th Int. Conf. Netw. Future (NOF), Sep./Oct. 2015, pp. 1–3. VOLUME 4, 2016

Z. Kaleem, K. Chang: Public Safety Priority-Based User Association for Load Balancing and Interference Reduction

[18] Radio Frequency (RF) System Scenarios Version 12.0.0, document 3GPP TR 36.942, 2014. [19] Study on LTE Device to Device Proximity Services; Radio Aspects Version 12.0.1, document 3GPP TR 36.843, 2014. [20] Guidelines for Evaluation of Radio Transmission Technologies for IMT-2000, document ITU-R M.1225, 1997.

ZEESHAN KALEEM received the B.S. and M.S. degree in electronics engineering from the University of Engineering and Technology, Peshawar, and Hanyang University, Korea, in 2007 and 2010, respectively, and the Ph.D. degree from the Electronics Engineering Department, Inha University in 2016. From 2010 to 2012, he was a Lecturer with Namal College, Pakistan (an associate college of the University of Bradford, UK). Since 2016, he is currently an Assistant Professor with the Electrical Engineering Department, COMSATS CIIT Wah Campus, Pakistan. He is the author of IEEE articles and conference papers. He also holds US/PCT and Korean patents. His research interests include interference management in 3GPP LTE-A and 5G systems. He was a recipient of the Higher Education Commission Scholarship, Pakistan and Jungseok Scholarship to pursue the M.S. and Ph.D. degree from Hanyang University and Inha University, Korea, respectively, due to his excellent academic career.

VOLUME 4, 2016

KYUNGHI CHANG (SM’98) received the B.S. and M.S. in electronics engineering from Yonsei University, Seoul, South Korea, in 1985 and 1987, respectively, and the Ph.D. degree in electrical engineering from Texas A&M University, College Station, Texas, in 1992. From 1989 to 1990, he was with the Samsung Advanced Institute of Technology as a member of the research staff and was involved in digital signal processing system design. From 1992 to 2003, he was with the Electronics and Telecommunications Research Institute as a Principal Member of the technical staff. During this period, he led the design teams working on the WCDMA UE modem and IMT-Advanced radio transmission technology (RTT). He is currently with the Electronic Engineering Department, Inha University, where he has been a Professor since 2003. His current research interests include RTT design for beyond 3GPP LTE-A and 5G systems, cross-layer design, and public safety and mobile ad hoc networks. Dr. Chang has served as an Editor-In-Chief and an Executive Director from 2010 to 2012 and in 2013, respectively, for the Journal of Korean Institute of Communications and Information Sciences. He is currently an Executive Director for business affairs for education, KICS. He has also served as an Editor of ITU-R TG8/1 IMT.MOD. He is currently a Chair of expert committee in SafeNet Forum. He is a recipient of the LG Academic Awards (2006), Haedong Best Paper Awards (2007), IEEE ComSoc Best Paper Awards (2008), and Haedong Academic Awards (2010).

9785