Autonomic Downlink Inter-Cell Interference Coordination in LTE Self-Organizing Networks

Transcription

1 Autonoic Downlin Inter-Cell Interference Coordination in LTE elf-organizing etwors Panagiotis Vlacheas, Evangelos Thoatos, Kostas Tsagaris and Panagiotis Deestichas University of Piraeus, Departent of Digital ystes, Piraeus, Greece {panvlah, ethoat, tsag, Abstract This paper provides a atheatical forulation for Downlin Inter-Cell Interference Coordination (ICIC), in the context of a 3GPP Long-Ter Evolution (LTE) elf-organizing etwor (O). O concepts have been introduced in LTE standardization in order to increase the networ perforance and reduce the operational expenditure for operators. Aong the proposed O use cases, ICIC is of utost iportance, since, in OFDMA-based networs, inter-cell interference is the ain factor hindering the achieveent of the high rate requireents, especially in downlin, where broadband services exist. Coordinated usage of resources in the related cells, by eans of a schee that is distributed, autonoic, context-aware, policydriven and nowledge-based, is proposed in this study as an effective ICIC approach. A set of results is provided, for showing the effectiveness of the schee in ters of Quality of ervice and spectral efficiency. Keywords-networ anageent; inter-cell interference coordination; self-organized networs; fuzzy logic; reinforceent learning; policies I. ITRODUCTIO The growing deands on obile networs to support applications with higher rate requireents have driven the need to develop Orthogonal Frequency Division Multiplexing (OFDM) 4G networs, such as WiMAX and 3GPP LTE. LTE provides intra-cell orthogonality between users in both uplin and downlin and therefore, inter-cell interference reains the ain interference source in these systes. Moreover, in order to support the deanding broadband services, a frequency reuse factor of one is used in neighboring cells, that eans these cells use the sae available resources siultaneously. Therefore, the inter-cell interference liits the LTE perforance in ters of throughput and spectral efficiency, especially for users at the cell edge. eedless to say, it is of utost iportance to develop viable interference itigation and/or cancellation techniques. uch techniques have been investigated and deployed with varying degree of success in the previous generations (2G, 3G). They have ainly focused on either ensuring orthogonality between transitted signals in tie, frequency and spatially or by reoving and canceling the interfering signals if orthogonality fails. Power control is also a echanis to control the interference not only in the target cell but also in the cell neighborhood. It is coon practice, even for LTE, to use less transit power near the cell border in order to decrease inter-cell interference (fractional path-loss copensation). However, LTE provides ICIC to liit intercell interference, enabling also inter-eodebs signaling capabilities, including uplin overload indicator (OI) and high interference indicator (HII), and downlin relative narrowband transit power (RTP) indicator. These indicators provide bit aps of interference conditions on per Physical Resource Bloc (PRB) basis. By this way, instead of restricting the resource usage statically with a reuse factor larger than one, which finally leads to less efficiency and throughput due to bandwidth reduction, spectru allocations are applied dynaically and according to traffic and radio conditions. The rest of the paper is organized as follows. ection II presents the state of the art. In section III, we describe the proposed atheatical fraewor for ICIC. ection IV presents in detail the cognitive aspects of the fraewor in ters of self-organization, learning capability and policies. In ection V, siulation results are given. Finally, in section VI, we suarize our wor and we pave the way for future research actions. II. RELATED WORK Inter-cell interference itigation and cancellation has received huge attention as one of the ost iportant echaniss, seeing to optiize networ perforance and satisfy Qo requireents targeted by the operator. The authors in [] give an overview and a coparison (for LTE) on existing interference itigation and cancellation techniques. In suary, in the short ter, a cobination, such as fractional power control and adaptive fractional frequency reuse based on scheduling in high IR regionould for the basis of a robust LTE ICIC strategy. Longer ter gains in ICIC perforance could potentially be achieved through the use of networ MIMO, opportunistic and/or organized beaforing, and distributed power control, as well as coding strategies such as sphere decoding or dirty paper coding. In [2], each eodeb is considered as an agent that perfors both learning, using fuzzy-reinforceent learning, and control in a sporadic context to dynaically adjust the fractional power control to reach optial tradeoffs between cell-edge and neighboring cell perforance in uplin. Results show iportant gain to the networ capacity and the perceived quality for data applications. elf-optiizing schees for interference anageent in downlin of OFDMA networs are investigated in [3], naely power control, fractional frequency reuse and dynaic fractional load. The proposed

2 algoriths are fully distributed, using inforation available fro neighboring cells and closed for forulas, while bringing noticeable perforance gains in bloc call rate and file transfer tie. In [4], the authors present a cell-specific and realistic HII, in order to overcoe the shortcoings, raised in traditional HII based ICIC schees by the reception of the sae HII by cells suffering fro different inter-cell interference. In this concept, two uplin coordination ethods are siplified and the resulting integration iproves significantly the spectral efficiency and the blocing probability. The authors in [5] study an uplin ICIC echanis, which fully utilizes the flexibility of frequency selective scheduling and rate adaptation. The proposed technique shares resources between the cell-edge and cell-center users, without the need to strictly classify each user equipent (UE) into one of these categories. iulation results prove the flexibility in balancing the perforance of cell edge users and average networ perforance. Based on the careful investigation of the literature, we can conclude to the rear that there is a need for a coon unified fraewor that can support challenging cognitive features, which are currently issing, such as selforganization, learning capability and policies. III. AALYI AD MATHEMATICAL FRAMEWORK In the following analysis, the downlin of an OFDMA syste is considered. We focus on downlin in our study due to the related broadband services, which pose higher rate requireents than those in uplin. The topology consists of one target cell and C neighboring interfering cells. For siplicity reasons, only the first cluster around the target cell is assued to cause significant interference to it, that eans C = 6. Let us assue that denotes the nuber of active users in the target cell and is the nuber of total available subcarriers in the syste. Then, we select n [, ] to represent a user, s [, ] to represent a subcarrier and c [,C ] to represent an interfering cell. Moreover, T s, n is the supported rate of subcarrier s of user n and r n is the requested rate of user n. The paraeter T s, n can be found based on the odulation schee, the coding rate and the IR [6]. We for three arrays in order to represent the subcarrier allocation to users in the target cell with respect to the subcarriers used in each of the interfering cells and the interference power that each interfering cell causes to a specific user at a specific subcarrier. o, denotes if the subcarrier s is used in the interfering cell c, if subcarrier s is assigned to user n in the target cell and I, c denotes the received interference power at the subcarrier s of the user n that is caused by the interfering cell c. These three arrays are the drive wheel of our forulation. The first two arrays have logical values according to the truthfulness of their logical expression as follows:, c, if = subcarrier s is s 0, otherwise used by cell c, if subcarrier s is assigned to user n s, n = (2) 0, otherwise The objective of the ICIC optiization proble, as defined in the context of this paper, is to find the appropriate resource allocation in the target cell, in order to iniize the interference caused at the target cell s users. This eans that the array is the array under investigation. Therefore, the objective function can be fored as follows: subject to in s= n= C n= s= c= n= s= s,c I (4), s [,] s, nts, n rn, n, s,c,n (5) [ ] The constraint (4) satisfies that the resource allocation in the target cell will not exceed the nuber of total available subcarriers. The constraint (5) is used to represent that each subcarrier is allocated to only one user of the target cell, while the constraint (6) guarantees that each user will satisfy his rate requireents. It ust be noted that in our analysis, we use subcarriers as the basic resource units, instead of PRBs. However, this consideration does not affect the analysis, since it ay be easily extended to also assue PRBs. Investigating the optiization proble (3)-(6), it can be easily noticed that the only issing paraeter is the array (3) (6) (), which denotes the subcarriers used in the neighboring cells that cause interference to the target cell. The question is now how to retrieve inforation about this array. This can be easily done using the RTP indicator [7]. RTP is used in downlin LTE syste as the indication of interference level on the PRBs. In our case, PRB is identical to one subcarrier. Besides, a RTP threshold is set to decide whether the interference level is high or not. Therefore, if a subcarrier s is used by a neighbour cell c and is quite interfered, the RTP threshold is exceeded and the paraeter s,c will be equal to logical one. RTP essages are exchanged between eodebs over the so called X2 interface. The introduction of the array in our proble forulation allows a proactive, dynaic, event-triggered ICIC schee for downlin LTE. The ter proactive is used in the sense that harful collisions can be avoided by scheduling resources in the target cell that are either not used by the neighbor cells or less interfered, i.e. with a zero value in the

3 corresponding paraeter. In this way, the resource allocation in the target cell taes also into account the context changes, i.e. the load variations in the cells of the syste. eedless to say, such a solution has the additional benefit that it does not need to pre-configure the resource usage in a planned anner, e.g., via Operation and Maintenance (O&M), but it is done via a self-organised and cognitive anner. IV. COGITIVE FEATURE OF THE FRAMEWORK AD CHALLEGE A. elf-organisation The proposed fraewor is consistent with the release 9 of 3GPP guidelines about elf-organized etwors (O) [8]. The fraewor configures the ICIC configuration paraeters, lie reporting thresholds/periods and preferred/prioritized resources, autoatically. Traditionally, this configuration has to be set by the operator for each cell. etting and updating these paraeters autoatically is the tas of a O echanis. Therefore, our atheatical fraewor represents such a distributed O echanis in eodebs for downlin with respect to the available tie/frequency resourceeighborhood relations of the cells (cell topology), context changes (cell traffic load variations) and policies (Qo requireents targeted by the operator). The O functionality of the proposed fraewor is depicted in Figure using a concept ap ethodology [9]. Therefore, ICIC optiizes RTP threshold and the corresponding reporting period, while updating array that deterines resource preferences of the target cell through these RTP essages. By this way, the coordinated usage of resources is et and a higher throughput is achieved by the inter-cell interference reduction. However, the spectral efficiency and the resource utilization ay be reduced (not obligatorily) if the coordination prohibits the use of subcarriers or PRBs that are oentarily in good fading conditions. However, the throughput and IR increase is achieved at the expense of an increased bachaul counication and intra-node processing. This is why a learning capability, which will allow the self-optiization without the use of frequent RTP essages, is required. ICIC Optiizes To iprove Throughput To iprove Paraeters pectral Efficiency Of Increases when Resource Preferences exist eighbor Cells uch as Reporting thresholds/ periods for DL TX Power Decreases when the use of PRBs in good fading conditions is prohibited Figure. Concept Map for the O functionality of the proposed ICIC fraewor. B. Learning Capabilities The introduction of learning capability in the fraewor will allow an autoatic but fast, low processing resource allocation in the target cell without the need of too any signaling essages and the solution of the optiization proble in each step. A well fitted ethod for learning is the fuzzy reinforceent learning (RL) approach. This approach was successfully used for another O use case, i.e. Capacity and Coverage Optiization (CCO), in [0]. The ain proble with RL is when the input state space or output action space is continuous or highly diensional. Therefore, fuzzy logic can be used to provide the required level of abstraction both for state and action spaces. In [0], it is recoended to separately fuzzify each input paraeter and to cobine the fuzzified inputs in the inference stage using the AD fuzzy conjunction operator to for identical states. Within this context, the tas of the RL layer is to find the best consequent for each fuzzy if...then rule. Returning bac in the concept ap of Figure, several states and following actions could be defined using the fuzzy reinforceent learning approach. For instance, ICIC should onitor IR, throughput and spectral efficiency. These etrics are actually the input paraeters, which will be fuzzified through ebership functions (e.g. Low and High ). Then, a nuber of distinct states and actions are produced. A siple rule could be if throughput is low then decrease RTP threshold in order to decide better on if a subcarrier is interfered or if throughput is low then increase RTP reporting period to lead to better ICIC through ore frequent RTP indicator essages. On the other hand, if spectral efficiency is low then increase RTP threshold to ignore occasionally the resource preferences when they prohibit the use of subcarriers in good fading conditions. The previous rules are actually different control loops. Closed control loop anageent is a very crucial process in autonoics. everal states can be activated concurrently and in an autoatic way. If the corresponding actions are conflicting, this will lead to networ instability. Therefore, there is a need for a conflict resolution echanis. This is the reason why operator goals and policies should be taen into account. C. Policies There are two ain cases where operator policies, either high level policies concerning business goals or low level configuration policies, ay apply: Qo requireents targeted by the operator and conflict resolution in the case of O echaniss coordination. In the first case, the operator sets Qo requireents to be satisfied in the networ. o, the networ entities are onitored and different etrics through ey perforance indicators (KPIs) are easured. uch etrics coprise the IR, throughput, spectral efficiency (also depicted in Figure ). When perforance degradation occurs or new networ conditions exist (e.g. due to a new service), specific states are activated, which lead to specific actions. Operator policies ay affect the various thresholds or helping the conflict

4 resolution in different control loops, as denoted in the previous subsection. Recently, O entities coordination has been suggested [3] as a ean to enforce operator policies. In this context, the proposed ICIC fraewor ay interact with other O echaniss. As a draft exaple, CCO algorith increases the downlin transission power assigned to a PRB, in order to increase the capacity and/or the coverage, which leads to ore inter-cell interference. Then, ICIC will be activated to reduce the inter-cell interference. O coordination will be the result of our future wor and is currently under investigation. V. EVALUATIO In order to evaluate the proposed fraewor, indicative siulations have been ade. A genetic algorith (GA) has been used for the solution of the inter-cell interference iniization proble, since it can be successfully applied for non-convex probles where the search space is large. Here, we present a snapshot of our siulation, naely the resource allocation in one slot. The siulation paraeters are denoted in Table I. We consider 45 users with different Qo requireents (24 with 28 Kbps, 5 with 256 and 6 with 52) in the target cell. Table I. iulation paraeters uber of ites 7 Frequency Reuse Factor ite-to-ite Distance 500 Cell Radius 250 ectors per ite Carrier Frequency 2GHz yste Bandwidth 5MHz Data ubcarriers 300 ubcarrier pacing 5KHz lot Duration 0.5sec Path-Loss Model *log0 (D), D: distance in hadowing Long-noral with std. dev. 8dB Theral oise Density -74dB/Hz Max Tx Power at eb 43dB Tx Antenna Gain 8dB Rx Antenna Gain 0dB User Distribution Unifor User Qo Levels 28Kbps, 256Kbps, 52Kbps pectral Efficiency (bps/hz) 8,00 7,50 7,00 6,50 6,00 5,50 5,00 4,50 4,00 3,50 3,00 pectral Efficiency per Generation Generations With RTP Without RTP Figure 2. pectral Efficiency vs generations with and without RTP essages As expected, the usage of the RTP indicator gives a solution with 4.6dBW less total received interference power at the target cell. In Figure 2, the spectral efficiency at each step of the algorith (00 generations) is been illustrated with and without RTP indicator essages. In both cases, the achieved spectral efficiency is larger than the 3GPP target requireent [], i.e..53 bps/hz, for the LTE downlin perforance. In this case, the use of RTP indicator does not prohibit the usage of subcarriers with good IR, thus leading to an increase on spectral efficiency. Finally, in Figure 3, the total throughput of the target cell at each step of the algorith is depicted, showing that the use of RTP indicator leads to a throughput increase. All the previous results prove the effectiveness of the proposed fraewor and the iproveent achieved by inter-ebs counication through RTP essages. Throughput (Kbps) 8500, , , , , , , , , , , ,00 Total provided throughput With RTP Without RTP Generations Figure 3. Total throughput in target cell vs generations with and without RTP essages VI. COCLUIO In this paper, we present a atheatical forulation for downlin ICIC in LTE O. The proposed forulation allows the coordination between ebs through RTP indicator and in this concept, the consideration of context changes, i.e. the load variations in the syste cells. Moreover, a fraewor, explaining how self-organization, learning capability and policies could be introduced into the proposed schee, is analysed in detail. Indicative results show the effectiveness of the proposed ICIC schee and the iproveent achieved by the use of the RTP indicator. Future plans include ore elaboration on cognitive aspects and policies, as well as the interaction between ICIC and other O use cases. ACKOWLEDGMET The research leading to these results has been perfored within the Univerelf project ( and received funding fro the European Counity's eventh Fraewor Prograe (FP7/ ) under grant agreeent n REFERECE

5 [] Gary Boudreau, John Panicer, ing Guo, Rui Chang, eng Wang; ophie Vrzic, ortel, Interference coordination and cancellation for 4G networs, IEEE Counications Magazine, Vol. 47, Issue 4, pp. 74-8, [2] Mariana Dirani and Zwi Altan, elf-organizing networs in next generation radio access networs: Application to fractional power control, Elsevier Coputer etwors, Vol. 55, Issue 2, pp , 20. [3] R. Cobes, Z. Altan, M. Haddad and E. Altan, elf-optiizing strategies for interference coordination in OFDMA networs, accepted at the IEEE ICC 20 Worshop on Planning and Optiization of Wireless Counication etwors, Kyoto, 20. [4] Guangrong Zhang; Chao Zhang, Jun Zhang; Guo Wei, A ovel Uplin Interference Coordination chee Using High Interference Indicator, IEEE 72nd Vehicular Technology Conference Fall (VTC 200-Fall), 200. [5] Mazin Al-halash, Farid Khafizov, Zhijun Chao, Interference constrained soft frequency reuse for uplin ICIC in LTE networs, IEEE 2st International yposiu on Personal Indoor and Mobile Radio Counications (PIMRC), 200. [6] Fa-tang Chen, Gen-lin Tao, A ovel MC election Criterion for upporting AMC in LTE yste, International Conference on Coputer Application and yste Modeling (ICCAM 200), 200. [7] David Astely, Eri Dahlan, Anders Furusar, Ylva Jading, Magnus Lindstro, tefan Parvall, LTE: the evolution of obile broadband, Vol. 47, Issue 4, pp. 44-5, [8] 3GPP TR , Evolved Universal Terrestrial Radio Access etwor (EUTRA); elf-configuring and self-optiizing networ (O) use cases and solutions, June 200. [9] Jennifer Turns, Cynthia J. Atan, Robin Adas, Concept Maps for Engineering Education: A Cognitively Motivated Tool upporting Varied Assessent Functions, IEEE Transactions on Education, vol. 43, no. 2, pp , May [0] R. Razavi,. Klein and H. Claussen, elf-optiization of Capacity and Coverage in LTE etwors Using a Fuzzy Reinforceent Learning Approach, IEEE 2st International yposiu on Personal Indoor and Mobile Radio Counications (PIMRC), 200. [] 3GPP TR 25.93, Requireents for Evolved UTRA (E-UTRA) and Evolved UTRA (E-UTRA), v