JOINT RESOURCE ALLOCATION ALGORITHMS FOR UPLINK IN 5G AND LTE NETWORK

Transcription

1 JOINT RESOURCE ALLOCATION ALGORITHMS FOR UPLINK IN 5G AND LTE NETWORK Alaa Ghaith 1, Rima Hatom 2, Abbas Hatom 3 1 HKS Lab, Faclty of Sciences, Lebanese University, Beirt, Lebanon 2 LIP6, UPMC - University of Paris VI, 4 Place Jssie, Paris, France 3 LIP6, UPMC - University of Paris VI, 4 Place Jssie, Paris, France Abstract: Recently, femtocell-based 5G is the hottest radio access technology, that meet the exponentially increased ser-operator demand to enhance: indoor coverage, network capacity and qality of service and decrease latency. Intelligent allocation of the resorces and interference management isses are the sbstantial challenges in sch context. In this jornal, we propose a joint power and channel allocation algorithms with Adaptive Modlation and Coding (AMC) for plink. A linear optimization models for the SC- FDMA-based Uplink transmission are adopted and resolved. Both power and modlation/coding schemes are independently assigned to each ser over each allocated sbchannel. The ser differentiation strategy ensres the QoS garantee with respect to a priority level of each ser: premim High Priority HP sers and best effort BE sers. As trade-off between distribted and centralized architectres, we consider a clstered network architectre. It combines the high performance (centralized) and the low comptational complexity (distribted) advantageos effects. Extensive simlations prove that or proposals otperform different techniqes in the literatre and considerably improve two previosly proposed methods QP-FCRA and Q-FCRA. We consider different metrics for performance evalation sch as spectral efficiency, throghpt satisfaction rate, ser otage and transmission power. Keywords: 5G, LTE, SC-FDMA, AMC, QoS, Resorce Allocation, Femtocell. 1. Introdction 5G radio access technology is adopted as the enhancement of the previos generation technology sch as the 4G that is not entirely deployed yet. 5G will provide better speeds and coverage than the crrent 4G. 5G operates with a 5GHz signal and can offer speeds p to 1Gb/s for tens of connections. Hawei, a major player in the Chinese mobile market, believes 5G will provide speeds 100x faster than 4G LTE offers. Main design objectives for 5G are: 1) realization of massive capacity and connectivity, 2) spport to very diverse set of services, applications, sers and reqirements, and 3) efficient and flexible tilization of all available noncontigos spectrm resorces [2, 6, 8, 7]. Recently, more than 50% of total voice calls and more than 70% of data traffic are established indoors [19]. However, transmitted signal from Macrocell Base Station (MBS) is weakly received by the indoor User Eqipment (UE) de to the penetration loss and barriers absorption effect casing a degradation of the signal qality. Unfortnately, the architectres mentioned above cannot garantee the satisfied indoor coverage. Therefore, basing on the fact that the best way to increase the system capacity is by decreasing cell size or the transmitter-receiver distance, a new architectre appears and consists to deploy a small size home base station randomly by an end ser (femto:10 15 size order). The femtocell base stations also called Femtocell Access Point (FAP) are connected to the MBS throgh a broadband connection. This is the Femtocell architectre that first improve the coverage in indoor environments, and secondly increase system capacity and spatial rese by redcing cell sizes and offloading macrocells [18]. In order to provide a complete framework abot the LTEfemtocell networks, we are motivated to perform the resorce allocation problem for the plink transmission taking into accont many specifications. The 3GPP LTE standard adopts for the plink connection the Single Carrier-Freqency Division Mltiple Access (SC-FDMA). Therefore, in or contribtion, we proposed a SC-FDMA plink optimization resorce allocation problem nder the constraint of contigos allocated physical resorce blocks for each ser. Unlike most previos plink works considering constant power and fixed MCS over all resorce blocks allocated to a ser, we propose to adopt a joint power control and adaptive modlation/coding mechanisms over the resorce blocks depending on the channel stats. Moreover, we introdced a spectrm sensing techniqe, in order to estimate the interference power vales over the occpied resorce blocks and redce the complexity of the optimization problem. Recent works on the SC-FDMA plink resorce allocation and interference management isses are investigated in the literatre [11, 17, 16, 15]. Athors in [16] design an energy efficient model to allocate power and sb-channels nder the main constraints respected by the SC-FDMA adopted for the plink LTE transmission; in particlar, all resorce blocks assigned to a ser mst be contigos, and the power transmitted by a ser over all allocated resorce blocks mst be constant. In addition, each resorce block can be allocated to one ser at most. To resolve the optimization problem of the effective capacity maximization nder the cited constraints as well as the QoS constraints in term of services delay and data rate reqirements, they referred to the Canonical Dality Theory (CDT) for the complexity considerations. In [15], athors proposed a distribted cochannel Radio Resorce Management (RRM) method for the plink transmission in the traditional LTE celllar systems. COPYRIGHT - IJST 14

2 They considered different types of services to be taken into accont in the schedling strategy. As the first stage, they classed the served sers with respect to the priority order. In a second stage, by an iterative way, the schedler allocates jointly the Physical Resorce Block (PRB), the transmit power as well as the Modlation/Coding Scheme (MCS). Unfortnately, beginning by the maximm allowed power transmission is not beneficial in sense of energy efficiency and the iterative strategy reslts higher time convergence. These facts are not appropriate for the small-cell-femtocellnetworks where the energy efficiency is more crcial factor comparing with the traditional macrocell network. Previos works, althogh take into accont the selection of the appropriate MCSs according to the channel qality, they don t present enogh description of the Adaptive Modlation and Coding (AMC) techniqe based on the physical layer argmentation. In addition, they jointly allocate sb-channels, power and MCS, bt, effectively, the schedler mst fix power for select the corresponding MCS and/or fix the MCS to calclate the transmit power, which significantly decelerates the convergence. Ths, we propose for the SC- FDMA-based plink femtocell-lte networks a joint resorce allocation problem where the power, the RBs and the MCSs are simltaneosly and optimally allocated while mitigating the interference. Neither the power nor the MCS are enforced to be constant over all RBs allocated by one ser. Doing that we aim to deal with or main scope in this jornal in sense of increasing the data rate and offering an enhanced ser experience while increasing the spectral and power efficiencies. The rest of this paper is organized as follows. In Section II, we describe the system model. In Section III, we present the adaptive modlation and coding techniqe. In Section IV, we present and explain or algorithms on plink resorces allocation approaches. Simlation metrics and reslt evalation are provided in Section V and VI respectively. Finally, the conclsions are drawn in Section VII. 2. System Model and Description We present in this section a global description of or system inclding the network model, the physical propagation model and lastly smmarize all notations sed in the seqel. Two-tier femto-macro network is assmed, where a set F of N Femtocells base stations (FBSs), also known as Femtocell Access Points (FAPs), sbmerge in a macrocell base station coverage area (Figre 1). As femtocell network architectre, we distingish between three different categories known in the literatre: Figre 1: Two-tier Femto-Macro Heterogeneos Network Centralized: in this case, the schedling and the resorce allocation take place niqely at a centralized base station referred as allocator. This central schedling collects information abot available resorces, ser positions and channel qality measrements. All sers send their reqests and the allocator responds considering the available resorces and the interference relative to each ser. The main advantage of this way is the high accracy and service performance since all necessary allocation parameters are known. However, it cannot be applied to dense networks, since the allocator gets heavily loaded and the bottleneck connection increases the network complexity. Distribted: conversely, in the distribted way, each FAP presents a self-schedling to allocate resorces for its own sers. It shold be able to learn abot the srronding environment and efficiently allocate resorce to ensre satisfied ser experience. This approach redces the network complexity especially for high density networks. However, de to the lack of knowledge abot the srronding, this way is less reliable. Clstered: in order to tradeoff between the centralized and the distribted techniqes, a hybrid approach is adopted in many references [14, 13]. It aims to enhance schedling reliability while redcing the network complexity. Several neighboring femtocells constitte one entity basing on the mtal interference criterion. The scenario is explained as follows: o First, each FAP collects the knowledge abot the srronding network by sensing the transmissions of the neighboring cells as well as receiving reports from its own sers. o Second, a set of one-hop neighbors interfering FAPs is formed by each FAP regarding to the received knowledge. The element nmber of this set is the basickey sefl for constitting the clsters, so-called interference degree. o Third, each FAP interference degree is broadcasted to each element of the set. o Final, the FAP that has the highest interference degree is selected as the clster head (CH) and others FAP are the clster members. The clster architectre is a promising approach for rban dense environment, since it redces the bottleneck and complexity while providing relative knowledge abot the FAPs clster members. COPYRIGHT - IJST 15

3 2.1. Propagation Model The indoor femtocell base stations (FAPs) are assmed to be in a corridor or in rooms. Ths, two link types are considered: corridor-to-corridor link referred to the line of sight (LOS) case and corridor-to-room link referred to the non-line of sight (NLOS) case. In addition, for rooms farther away from the corridor, wall-absorptions mst be applied to the walls parallel to the corridors. Frther, Floor Loss (FL) of the vertical radiations for propagation from floor to floor is modeled, and the Floor Loss mst be added to the path-loss calclated for each floor. Therefore, the A1-type generalized path loss model is considered as the propagation model for the freqency range 2 6 GHz developed in WINNER [6]. The path loss models are ths smmarized in the following form: fc GHz PL Alog10 d m B C log10 X 5.0 (1) In Eqation 1, d is the distance transmitter to receiver in [m], f c is the carrier freqency in [GHz], the fitting parameter A incldes the path-loss exponent, parameter B is the intercept, C describes the path loss freqency dependence, the shadow fading distribtion is log-normal. X is an optional, environment-specific term Notations A list of notations is presented in what follows smmarizing essential parameters sefl in this paper: F = {F 1,..., F N } is the set of FAPs, where N is the total nmber of femtocells. H = { 1,..., nhp } is the set of HP sers. B = {v 1,..., v nbe } is the set of BE sers. S = {S(1),..., S(L v )} is the set of SINR thresholds for different MCSs. Where L v is the total nmber of MCS levels. In or work, we consider 6 MCS levels L v as shown in Table 1. C = {C(1),..., C(L v )} is the set of the MCS efficiencies. R denotes the demand of the ser H B. K = {1,..., K} is the set of available physical resorce blocks (RBs). Δ (k) is the binary resorce allocation vector for ser H B, with 1 or 0 in position k according to whether the RB k is sed or not. α,k (l) is the binary resorce and MCS allocation vector for ser, with 1 or 0 in position k according to whether the MCS l is selected by the ser on RB k or not. P n k is the power transmitted from FAP F n to its ser on the RB k, where P min < the ser or P n k P n k = 0 otherwise. < P max if RB k is sed by P max is the maximm power transmission fixed by the operator. P min is the antenna power sensitivity. γ,k is the reqired SINR for ser on RB k. n, k m n P P m ' n n k. g m 2 k. g g is the gain of the physical link between the FAPs F n and the ser. It is the inverse vale of the path-loss vale jst modeled above. σ 2 is the Additive White Gassian Noise (AWGN) variance. For each resorce block, if the SINR vale at a mobile receiver is less than its minimm reqired threshold, it sffers from interference. Ths, the schedler discards the transmission link on sch RB and tries other resorces. Therefore, fixing a niqe SINR threshold vale is a hard condition that threats to redce the spectral rese and the global system throghpt while increasing the rejected rate. Conseqently, the ser satisfaction rate and the overall system capacity radically decrease. Motivating by this fact, we propose to switch between several SINR thresholds depending on the channel stats on the considered resorce block. The transmission rate is harmonically selected accordingly to the adopted SINR threshold, AMC. 3. Adaptive Modlation and Coding Techniqe Adaptive Modlation and Coding (AMC) is a smart techniqe that optimizes sage of the network resorces and efficiently exploits the channel capacity depending on its qality [4]. Therefore, schedling with AMC atomatically reacts to the channel flctations and dynamically matches the modlation and coding rate with the radio link qality for each RB. This is a rate control mechanism as shown in Table 1 and Figre 2. Each system fixes its proper TBLER that mst not be exceeded. With respect to this BLER threshold, the AMC approach provides a set of discrete SINR thresholds picked corresponding to the predefined set of MCSs [7, 14, 5]. Table 1: Modlation and Coding Schemes (MCSs) Modlation Type Coding Rate Efficiency (bits/sym) SINR level (db) QPSK 1/ QPSK 3/ QAM 1/ QAM 3/ QAM 2/ QAM 3/ In fixed MCS, the base station transmits signal with the same MCS that is the same for all allocated resorce blocks. So, if the MCS selected has a high-order, it is exclsively sitable for good condition RBs; else, nfortnately, the performance degrades and the power efficiency decreases since a considerable amont of transmitted information bits is erroneos received by the ser eqipment. Otherwise, if the low-order MCSs has applied to avoid losing the transmission on the corresponding RBs. For better channel qality it leads dissipating available spectrm and profitless decreasing rating amont; as the reslts, throghpt, spectral efficiency and the system capacity significantly decrease. (2) COPYRIGHT - IJST 16

4 Figre 2: SINR thresholds selection for different MCSs Alternatively, adaptive MCS allocation is a promising way to efficiently allocate RBs in accordance to their relative statses. Ths, the overall operator-ser goals are thoghtflly achieved with respect to two major considerations: 1) Throghpt and spectral efficiency, and 2) Availability and maintenance link qality. With regard to the description above, the benefits of the AMC techniqe clearly shine to tradeoff between desirable throghpt and satisfied system performance. 4. AMC-based Joint Resorce Allocation approaches In this section, we present or optimization schedling and resorce allocation problem for SCFDMA plink femtocell networks. Or objective is to perform an optimal resorce allocation problem that jointly assigns RBs, transmit power vales and MCSs for each active FUE in order to flly satisfy HP sers and then serve BE sers as mch as possible Problem Formlation Or allocation and interference mitigation problem are formlated while careflly taking into accont the Uplink transmission specifications. First, we cite the Uplink interference scenarios that differ from the downlink case. Then, we introdce the spectrm sensing techniqe as a beneficial way before formlating the resorce allocation problem. Uplink Interference Scenarios We classify three interference scenarios as smmarized in Table 2. Manage these interference scenarios is the main challenge in order to provide an efficient plink transmission. Table 2: Uplink interference scenarios in two-tier network Scenario Aggressor Victim Interference Type 1 MUE FAP Cross-tier 2 FUE MBS Cross-tier 3 FUE FAP Co-tier MUE to FAP interference When the MUE commnicates with the far-away Macro base Station (MBS), this latter gives the order for the MUE to increase its transmission power to overcome the path-loss and other propagation attenation effects along the MUE-MBS trajectory. The goal is to achieve a minimm SINR at the MBS receiver side allowing srpassing the interference effect and reaching the MCS considered. If the MUE location becomes close to a femtocell region, severe cross-tier interference occrs on the RBs commonly sed by the MUE and a FUE of the corresponding FAP. This interference type is considered very freqently. In or schedling strategy, we benefit from the HP/BE ser's differentiation to resolve the cross-tier interference between the MUEs and the FAPs. In fact, when the MUE passes nearby a FAP, it atomatically switches its commnication from the MBS to the corresponding FAP. FUE to MBS interference The cell-edge FUE aims to increase its plink transmitted power trying to attain the target SINR that garantees reliable transmission. When the FUE is in the vicinity of the MBS, the interference occrs only if the FUE ses the same freqency as the MUE in transmission mode. However, in or plink schedling strategy, we neglected the inflence of the FUE transmitted power on the MBS. Indeed, by definition, the femtocell is a short range region. FUE to neighboring FAP interference The third scenario represents the co-tier plink interference, where FUE is the case of the interference and neighboring FAP is the victim. In fact, the cell-edge FUE radiates with an enogh power to its FAP respecting the reqired SINR vale. This reslts a leakage of power to the adjacent FAP throgh the walls. De to the freqency rese policy, the cell-edge FUE signal and a neighboring femtocell-ser signal interfere at the corresponding neighboring FAP over the common RBs. Conseqently, a severe interference degrades the signal reception. In or plink schedling approach, we focs on the third interference scenario and we benefit from the potential spectrm sensing techniqe next presented to be aware of the srronding interference transmissions Spectrm Sensing Phase The spectrm sensing techniqe is performed to detect and analyze the wireless spectrm. It is originally sed to identify free spaces considered as available resorces in the whole spectrm for dynamic cognitive radio (CR) transmissions. For the LTE Heterogeneos Network, by performing the spectrm sensing techniqe, each FAP listens to its srronding environment and detects over all available RBs, the free and occpied ones; frther, it can measre the interference level on the occpied RBs by measring the received signal energy. Indeed, the FAP can more reliably detect the closed transmissions and consider them as aggressor interference. The most applications introdcing the spectrm sensing techniqe are mainly based on the occpied/free decision. Whereas, or plink allocation problem is based on the spectrm-sharing approach, so the occpied/free decision is not the niqe critical factor, bt also the energy vale of the detected signal on each occpied RB. To give a considerable knowledge abot the interference on the corresponded RB. In the downlink, the interference power vale is nknown and is introdced in the downlink model as variable to be fond. In the plink, the interference powers are assmed known de to the spectrm sensing; this significantly simplifies or COPYRIGHT - IJST 17

5 schedling model and redces its complexity as well as the convergence time needed after the first allocation process. Resorce Allocation Phase We take into accont the specific constraints of the SC-FDMA Uplink transmission. Particlarly, the 3GPP LTE standard adopts the localized SC-FDMA for the plink transmission where all RBs allocated to each ser are forced to be contigos to each other. By this way, each ser reserves a continos portion of the overall licensed spectrm. The previos plink literatre works consider that each ser allocates a constant power over all assigned sbcarriers, so the total power transmitted by each ser eqipment allocated K RBs is given by the following Eqation [15]: n P K. P n k Frthermore, some of the previos works enforce each ser to perform a fixed MCS over all assigned RBs, hence, the spectral efficiency of each ser becomes constant for these RBs. These two assmptions are closely correlated. In fact, the selection of the MCS depends on the channel stats over each RB, which is represented by the SINR level on sch RB. So, the constant power over all allocated RBs imposes a constant channel stats and conseqently constant MCS over those RBs. In contrast, or plink resorce allocation strategy is based on the power control mechanism where the transmit power may be different over the allocated RBs with respect to the minimm reqired SINR level on each one. In addition, it is not mandatory that each ser is assigned the same MCS over all allocated RBs since the interference stats becomes different for each RB, which allows increasing the ser spectral efficiency and the overall system throghpt Problem Resoltion Or plink resorce allocation and interference management model handles the link adaptation isse in terms of both: 1) power control and 2) Adaptive Modlation/Coding mechanisms. The overall transmitted power is minimized nder ser demands constraints; at the same time, the overall data rate is maximized nder performance constraints. In this paper, each FAP exploits the spectrm sensing for getting on each occpied RB the interference level and allows the control for each ser of the power transmission with additional srronding knowledge. This information is provided to the clster head to which the FAP is attached. So, the optimization model becomes simpler since the power constraint is ths relaxed to match the linear form which will be discssed next. On each assigned RB, the power is controlled independently. Ths, for this prpose the overall transmit power of the whole system is ths minimized nder the constraint to be as enogh as to satisfy the minimm SINR reqired for a reliable signal reception and decoding. On the other hand, the AMC techniqe is adopted in order to maximize the overall system data rate, increase the spectrm rese and redce the otage probability. Several consective SINR thresholds are taken into accont corresponding to several MCSs with different efficiencies are adopted. We allow that each ser performs on each RB a specific MCS nlike some previos approaches [15]. Therefore, independently for each RB, or optimization problem starts by allocating the highest order MCS respecting the corresponding SINR threshold. If it fails to reach this SINR level, it switches to the lower order MCS corresponding to lower SINR threshold while minimizing the transmitted power. The SC-FDMA transmission mode is considered as an additional constraint for or optimization problem. The RBs assigned to one ser are independently distribted respecting the channel stats only withot taking into accont any arrangement rle. To better nderstand the optimization model, let s describe the problem constraints as follows: Constraint (a) ensres that the SINR threshold S(l) corresponding to the MCS l vale mst be obtained if this scheme is sed on the allocated RB k, α,k (l) = 1. De to the spectrm sensing techniqe introdced before the transmission, the overall power interference level sensed by each ser on each RB k becomes a known vale eqal to E (k). (b) takes into accont the antenna sensitivity. (c) means that the total sm of powers transmitted by a FAP cannot exceed a maximm vale P max. (d) ensres the contigos condition of the allocated RBs for each ser. (e) ensres that HP ser reqirements mst be flly satisfied over the sm of the allocated RB. in (f), s v represents a slack vector, it is defined as the difference between the reqired and the obtained throghpt for the BE sers. s v (k) needs to be minimized. (g) ensres the exclsivity, at least one MCS shold be sed on the RB k if it is allocated. (h) denotes that α,k (l) is a binary variable Δ (k) is a binary variable and that it is impossible to se more than MCS level l on the same RB k if it is allocated. (i) ensres the orthogonality assignment between sers in the same femtocell. This model is solved as a linear optimization problem sing the IBM ILOG CPLEX optimization solver [1]. In award, the plink schedling task can be smmarized as follows: throgh the spectrm sensing techniqe, the FAP detects the energy transmitted, exploits, and sends it to the Clster Head. The Clster Head collects information from all FAPs attached to its clster (FAPs members) and at each schedling period, it associates the set of RBs/powers/MCSs that mst be sed by each FAP. All FAPs will obtain the RBs/powers/MCSs vectors to be sed and will not change ntil the next schedling period. As the Clster Head associates for each FAP its vector, it will have the visibility of the interference dring this schedling period. Conseqently, the intra-clster problem does not arise. For inter-clster case, as the Clster Heads do not coordinate with each other, it is possible that two or more neighboring and interfering FAPs are attached to two different clsters and they are assigned the same resorces. In this case, the resoltion happens at the FAPs in the third phase of the algorithm sing a Bernolli distribtion to keep or discard COPYRIGHT - IJST 18

6 the resorces sed. 5. Performance Metrics The performance evalation of or proposal is based on the following QoS metrics: 5.1 Throghpt Satisfaction Rate (TSR) TSR denotes the satisfaction degree of a ser with respect to the reqested resorces. For each ser attached to a FAP F n F, TSR() is defined as the ratio of the allocated nmber of RBs to the reqested ones and can be expressed as follows: K, TSR k k / R 1 For a network with U sers, the TSR metric can be ths given by: TSR TSR / U 5.2 Spectrm Spatial Rese (SSR) SSR denotes the average portion of FAPs sing the same RB within the network. Therefore, it is defined as the mean vale of RBs spatial rese. The SSR metric can be ths expressed K as follows: 1 SSR k K F 5.3 Rate of rejected sers k 1 H B 6. Performance Evalation We compare performance of or method with three different techniqes: 1) UPLINK FMC-QRAP (same problem bt with fixed modlation and coding) 2) Downlink AMC-QRAP presented above to show the difference between the OFDMA and the SC-FDMA schemes where we compare the downlink and plink ser experience. 3) QP-FCRA [12] which considers power control bt with fixed modlation and coding in the downlink. We consider a typical UPLINK LTE SC-FDMA frame with a system bandwidth of 10 MHz and a total nmber of K = 100 RBs. Different network densities are stdied. It ranges from 50 FAPs for low density networks and reaches 200 FAPs in higher density deployments. The FAPs are randomly distribted in a 2-D 400m 400m area, within 10m 10m residences. Users are distribted niformly in the residence with a maximm nmber of 10 sers per FAP. The nmber of sers, their bandwidth reqirements as well as their locations is varied at each simlation. In each FAP we allow p to 4 HP sers and 6 BE sers. The simlation reslts for each of the described metrics are presented below: 6.1 Throghpt Satisfaction Rate It represents the percentage of HP and BE sers not admitted in the network dring the schedling period. Recall that, once accepted, HP sers are completely satisfied, whereas for BE sers, their satisfaction degree will be maximized. 5.4 Average channel efficiency It represents for each resorce block (RB) the mean vale of achieved efficiency over the whole network expressed in bits/symbol. It is ths expressed as follows: L 1 v Eff averagek Cl, k l F 5.5 Transmission power l 1 H B We compte the transmission power allocation on each tile. This parameter is compted regarding: i) The variation in demand or sers reqirements, to show the impact of the nmber of allocated RBs for each ser on the total received power. ii) the average distribtion of power over the different available resorce blocks in the network sed for transmission. 5.6 Fairness The Jain s fairness index [3], which determines how fairly the resorces are distribted among N existing sers, is considered to highlight the performance of the algorithm with regards to ser satisfaction rate. It expressed as follows: N TSR 1 (3) N 2 N. TSR 1 2 Figre 3: Throghpt vs. Network Density As mentioned earlier, the throghpt difference between downlink and plink is noticeable in both FDD and TDD schemes. It is clearly shown, in Figre 3, how the difference is very small in downlink and plink when AMC is sed. In fact, UPLINK AMC-QRAP performs very closely to downlink. When it comes to sing AMC, we see the efficiency and the gain compared to plink with fixed modlation and coding scheme. When comparing UPLINK AMC-QRAP and UPLINK FMC-QRAP, in low density scenario an average throghpt of 99% and 98% is respectively achieved. Where in high density scenarios, the throghpt for UPLINK AMC- QRAP remains high with abot 98% satisfaction whereas the other scheme decreases to below 87%. A very important point is noticed also in this figre: when considering high density scenario, the plink scheme with AMC can perform even better than QP-FCRA for downlink sing fixed modlation and coding, reaching 98% and 96% respectively. This is also de to the spectrm sensing phase sed in UPLINK AMC-QRAP. 6.2 Spectrm Spatial Rese COPYRIGHT - IJST 19

7 Figres 4a and 4b plot respectively the mean SSR for the network for all densities and the average SSR per tile for the high density network. In the first figre we notice how the AMC scheme, for both downlink and plink scenarios, gives a better spectral efficiency and rese compared to the FMC scheme. Figre 5: Otage Probability Figre 4a: Mean SSR vs. Network density The AMC clearly increases the performance of the algorithm for both plink and downlink compared to the scheme with fixed modlation and coding. When comparing both plink and downlink in the AMC mode, we can observe that the performance of the plink reaches better reslts. This is de mainly to the spectrm sensing approach. Moreover, with a better view of the srronding in the plink, the spectrm sensing phase redces the convergence time needed after the first allocation, ths enhancing the ser experience and admission ratio. 6.4 Fairness Figre 4b: Mean SSR vs. Network density When comparing the above reslts, we can clearly see an improvement in different metrics. However, it is very important to verify the fairness of the algorithm and show if the resorces are eqitably distribted among sers. However, in the plink mode we observe a lower SSR compared to the downlink. This is mainly de to the fact that in the plink mode the allocated resorces need to be contigos, where this constraint does not exist in the downlink, as mentioned earlier. Ths, reslting a spread distribtion of some of the resorces allowing more flexibility for the downlink mode. 6.3 Rate of rejected sers When looking at Figre 5 we can make several observations on the otage probability or the nmber of rejected sers. In fact, for low interference scenarios, UPLINK AMC-QRAP is better performing than the other schemes with less than 0.5%of rejection is observed compared to arond 1% and 1.2% for the downlink schemes. If we look at the high density scenario, the benefit is evidenced with a rejection of abot 2% compared to more than 13%, 7% and 10% respectively for UPLINK FMC-QRAP, Downlink AMC-QRAP and QP- FCRA. Figre 6: Fairness Figre 7: CDF power transmission per femtocell ser The Jain s Fairness Index, calclated as the average for entire the network and inclding both types of sers, is sed to stdy this metric. We note that in the best case, it reaches 1 and it is attained when all sers receive the same allocation. Figre 6 shows the fairness for the different methods. It is observed that, for the low density networks, UPLINK AMC-QRAP and COPYRIGHT - IJST 20

8 DOWNLINK QP-FCRA are better performing that the others. While all the methods degrade with the network load, UPLINK AMC-QRAP keeps a high degree of fairness and is less affected by the density, where it varies from 99.5% to 98.5% compared to 96% to 93% for DOWNLINK AMC- QRAP and 98% to 87% for UPLINK FMC-QRAP, respectively for both low and high density networks. their reqests are spread over a larger nmber of resorces if failed to provide the reqired SINR for the considered MCS. 6.5 Power allocation In this part we investigate the power transmission and the energy efficiency of or algorithm. We se several metrics to stdy the power control mechanism and compare the different methods. We first start by showing the CDF of the power transmission per ser over the entire network plotted in Figre 7. In this plot we can clearly observe that for UPLINK AMC- QRAP, 90% of sers have a total transmission power less than 15 mw, compared to 18 mw for Downlink AMC-QRAP. This rate grows to approximately 40 and 50 mw for UPLINK FMC-QRAP and Downlink QP-FCRA respectively. This difference is obviosly de to the AMC techniqes which allowed the resorces to be sed with less transmission power when a lower MCS is needed, reslting in a lower interference over the RB and hence less transmission power reqired. Figre 9: Transmit Power vs. Demand Figre 10 represents the average distribtion of transmission power per tile over the entire spectrm. As it can be seen from the figre, when sing the AMC, the power per tile needed has an average lower than 0.3 mw independently of the network load. Whereas this ratio raises to more than 0.6 mw per tile for the schemes with fixed modlation and coding. Figre 10: Mean Power per RB vs. Network Density Figre 8: Transmit Power vs. Distance The power transmitted vs. distance is plotted in Figre 8 where we can see the evoltion in the total power sed based on the distance from the center of the FAP. A power minimization model is sed, as the algorithm tries to allocate the best channel to edge sers far from their FAP. They will se the channel with less interference so they can consme a lower power and redce energy consmption, especially when ser eqipment's are sensitive to battery and energy savings. The difference for both plink and downlink schemes sing AMC compared to the same scenarios withot AMC is clearly highlighted. With AMC, when failed to provide the reqired SINR on some of the resorce blocks, sers in the plink scheme will shift to a lower MCS which in most of cases will redce their power transmission and the interference on neighboring FAPs, and in trn will redce the transmission power of the other sers. Figre 9 displays the transmitted power and its variation over demand. Depending on sers reqests they will be allocated a nmber of resorces to achieve their reqirements. The nmber of resorces depends on the MCS on each resorce. We can see from the figre how the power per demand increases almost linearly. Indeed, when sers ask for a few resorces we tend to allocate them with a high order MCS. Demanding sers might be more difficlt to satisfy and hence Let s have a more detailed view on the average power transmission on each tile of the spectrm by looking at the Figre 11. As shown for the for different network densities, we can observe that the distribtion of power for the UPLINK is mch smoother that the downlink case. As clearly shown in Fig. 11(d) for the high density network, we observe how the power flctates for the Downlink AMC-QRAP compared to the UPLINK AMC-QRAP. In fact, since the PAPR in the OFDMA is higher compared to the SC-FDMA, we can see the power distribtion accordingly. (a) Low density - 50 FAPs (b) Medim density FAPs COPYRIGHT - IJST 21

9 (c) Large density FAPs (d) High density FAPs mitigate the reslted interference, where the power transmission of each ser mst respect a minimm reqired SINR level at the receiver side. To avoid rejecting ser who cannot reach this fixed SINR threshold on its allocated resorce, we have proposed to adopt a set of several SINR threshold and respecting them adaptively to the effective channel qality on sch resorce block. Each SINR threshold corresponds to a specific modlation and coding scheme (MCS) selected nder a block error rate performance constraint. We began by the highest SINR threshold vale corresponding to the highest order MCS in order to maximize the data rate. This is what we refer to the rate control or AMC mechanism. The proposed problems are modeled in accordance to the linear programming tools. As reslts, the system capacity, the overall throghpt, the spatial spectrm rese and the spectral efficiency are all enhanced as proved by extensive simlations sing the IBM ILOG CPLEX solver. Finally, we consider that delivering the high-speed services and application of the next generation wireless systems reqire complementary improvement on both physical and link layer levels. Intelligently and jointly allocating power, sb-channels and power resorces while enhance the transmission mode, provides a competent isse for the ftre generation standards. References Figre 11: Average power transmission per tile 7. Conclsion The essential qestions we aimed to answer here are: How we can provide higher data rates, higher energy efficiency, better spectrm tilization, enhanced performance and increased capacity for the advanced LTE-based Femtocell networks; and what are the degrees of freedom that can be exploited yet? In this jornal, we are oriented according to both: mlti-ser schedling on the data link layer and basic transmission enhancements on the physical layer. We have first exhibited the main featres offered by the high-speed SmallCell-LTE network while citing its main challenges as a network sbmerging in the macrocell area. We have got an overview of the literatre works on the plink schedling and resorce allocation algorithms. After that, we began to explain or VIII. contribtion on the management of the radio resorce allocation of the SC-FDMA-based plink connection basing on the adaptive modlation and coding(amc) techniqe and the power control where an additional hard constraint imposes that the resorce blocks assigned to each ser mst be IX. adjacent. This is referred as: Uplink AMC QoS-based resorce allocation with power control or plink AMC- QRAP. the clster architectre is considered as a hybrid between the centralized and the distribted architectres. Therefore, the power control mechanism is introdced to I. IBM, 01.ibm.com/software/integration/o-ptimization/cplexoptimizer II. Gpta, A., Kmar Jha, R.: A Srvey of 5G Network: Architectre and Emerging Technologies. Recent Advances in Software Defined Networking for 5G Networks III. Hahne, E.L.: Rond-robin schedling for max-min fairness in data networks. IEEE J. Sel. Areas Commn., vol. 9, no. 7, 1991 IV. Zhang, G.: Sbcarrier and bit allocation for real-time services in mltiser OFDM systems. IEEE International Conference on Commnications, 2004, vol. 5, pp V. Zalonis, A., Milio, N., Dagres, I., et al.: Trends in Adaptive Modlation and Coding. Advances in Electronics and Telecommnications, vol. 1, April 2010 VI. D. WINNER II: Winner II channel models part I- channel models. Tech. Rep., Sept VII. Tarhini, C., Chahed, T.: On Capacity of OFDMAbased IEEE WiMAX inclding Adaptive Modlation and Coding (AMC) and Inter-Cell Interference. 2007, pp Boras, C., Diles, G., Kokkinos, V., Papazois, A.: Optimizing hybrid access femtocell clsters in 5G networks. 10th International Conference on Broadband and Wireless Compting, Commnication and Applications, 2015 Boras, C., Diles, G.: Resorce management in 5G femtocell networks. 10th International Conference on Broadband and Wireless Compting, Commnication and Applications, 2015 X. Park, H., Hwang, T.: Energy-Efficient Power Control of Cognitive Femto Users for 5G Commnications. COPYRIGHT - IJST 22

10 XI. XII. XIII. XIV. XV. XVI. XVII. XVIII. XIX. IEEE Jornal on Selected Areas in Commnications, 2015 Hatom, R., Hatom, A., Ghaith, A., Pjolle, G.: QoSbased Resorce Allocation with Link Adaptation for SC-FDMA Uplink in Heterogeneos Networks. 12th ACM International Symposim on Mobility Management and Wireless Access, (MOBIWAC), September 2014 Hatom, A., Langar, R., Aitsaadi, N., Botaba, R., Pjolle, G.: QoS-based Power Control and Resorce Allocation in OFDMA Femtocell Networks. IEEE Global Commnications Conference (GLOBECOM), 2012, pp Ladanyi, A., Lopez-Perez, D., Jttner, A., Chy, X., Zhang, J.: Distribted Resorce Allocation for Femtocell Interference Coordination via power minimization. IEEE Global Commnications Conference, (GLOBECOM), 2011 Lee, J., Amott, R., Hamabe, K., Takano, N.: Adaptive Modlation Switching Level Control in High Speed Downlink Packet Access Transmission. 3G Mobile Commnication Technologies - IEE, May 2002, pp Afifi, A., Elsayed, K. M.F., Khattab, A.: Interferenceaware radio resorce management framework for the 3GPP LTE plink with QoS constraints. IEEE Symposim on Compters and Commnications (ISCC), 2013, pp Elayobi, S-E., Foresti, B.: On Inter-Cell Interference and Adaptive Modlation in OFDMA WiMAX Systems. IEEE Global Telecommnications Conference, (GLOBECOM), 2006, pp. 1 5 Eichinger, J. M., Gao, Y., Jose, S., et al.: Adaptive Modlation and Coding in a SC-FDMA System. US Patent B2, December 2013 Robson, J., Bevan, D., Boe-Lahorge, M.: Radio Resorce Allocation for Celllar Wireless Networks. US Patent B2, Febrary nd Intl. Conf. Home Access Points and Femtocells, Presentations by ABI Research, Picochip, Airvana, IP.access, Gartner, Telefonica Espana. Athor Profile Alaa Ghaith PhD in Telecommnications and Electronics. DEA and Engineering diploma in Telecommnications and Electronics. Worked in the R&D department of FranceTelecom in France for three years on signal processing and mobile commnications (3G and 4G). Active research in some Eropean projects. Worked in the development department of a sb-contractor with Ericsson in Lebanon for three years. Active research on Telecommnications and signal processing. Special attention to the blind analysis and recognition of signals. He is now with HKS research laboratory in the faclty of Sciences I, at Lebanese University. COPYRIGHT - IJST 23