Due to rapid growth of spectrum demand, future wireless networks should be ... With respect to Radio Access Technologies (RATs), this means that there is a ...
2011 Technical Symposium at ITU Telecom World (ITU WT)
An Approach for Automated Spectrum Refarming for Multiple Radio Access Technologies Soroush Ghamari, Malte Schellmann, Markus Dillinger and Egon Schulz Huawei Technologies Duesseldorf, European Research Center, Riesstr. 25D, 80992 Munich, Germany {soroush.ghamari, malte.schellmann, markus.dillinger, egon.schulz}@huawei.com
Abstract—The mobile communication technologies beyond the third generation are characterized by flexibility. Due to rapid growth of spectrum demand, future wireless networks should be able to dynamically allocate resources to maintain the quality of service (QoS) and promote the efficient use of radio spectrum. The dynamic spectrum management is an effective solution for approaching this goal. In this paper, we propose a new heuristic method for sharing spectrum between two different radio access technologies (RAT) in an intra-operator scenario. We designed a new dynamic spectrum manager that aims to prevent overload in RATs and improves spectrum utilization of RATs. Our approach contains two complimentary phases of action, the first phase tries to prevent overload of a RAT in a proactive manner, and the second phase tries to solve an overload situation in case the proactive phase fails to prevent it.
I. I NTRODUCTION Today, available frequency spectra are assigned to different radio services by following a fixed spectrum allocation policy. With respect to Radio Access Technologies (RATs), this means that there is a dedicated band used exclusively by one RAT operated by a single operator. As broadband radio services are expected to continue their recently observed explosive growth also in the future, it can be foreseen that the demand in additional frequency bandwidth cannot be satisfied any longer by following the conventional spectrum allocation scheme. Hence, new measures to unlock frequency spectrum need to be found. A suitable measure for that is known under the term of ”spectrum sharing”, where different radio systems are granted a coordinated access to the same spectrum band [1]. Based on the assumption that different radio systems will not have the same demands for bandwidth at the same time, a heavily loaded RAT may ”borrow” frequency resources from another RAT that are temporarily unused. This approach not only relaxes the problem of spectrum shortage, but also significantly improves the efficiency of overall spectrum utilization. Spectrum sharing between heterogeneous radio services is considered in the context of the utilization of the so-called White Spaces in the frequency spectrum of TV broadcast services [2]. Today, these White Space frequencies are used by secondary access of devices for project making and special events (PMSE); however, their operation is also of temporary nature and hence there is a high potential for the use of this spectrum by other radio devices. Opening this spectrum for an opportunistic access of new additional radio devices requires those devices to get aware of the current spectrum usage,
© Copyright 2011 ITU - All rights reserved
allowing them to avoid interferences and potential distortions of the other systems operating in the same band. This can be enabled by the cognitive radio technology. Currently, a lot of research is carried out in this context, and there exist many international collaboration research projects seeking for practical solutions, like the FP7 projects FARAMIR [3], CogEU [4], QoSMOS [5] and Saphyre [6]. The FP7 SACRA project [7] builds on similar ideas, but extends the scope to multiple frequency bands that are available for opportunistic spectrum access. If multiple different radio services are demanding a limited spectrum resource, decision methods for a suitable resource allocation become of interest. For this use case, spectrum auctioning methods have been developed that try to satisfy a maximum of service requests while maximizing the spectrum utilization [8]–[10]. These techniques are also in the focus of the FP7 project Saphyre [6], which elaborates on various solutions for efficient sharing of physical resources, in particular spectrum. A special scenario for spectrum sharing is an inter-operator scenario where multiple operators are using the same RAT, which allows for a flexible inter-operator resource management. Here, spectrum borrowed to another operator does not necessarily need to be freed, as long as the resources used for the simultaneous access of different operators are orthogonal. Major research in this field has been conducted in the scope of the FP7 E3 project [11] and its predecessors FP6 E2R phase I and II. Finally, another scenario is of high relevance today: As many operators operate multiple RATs simultaneously, contiguous blocks of spectrum may be reallocated directly from one RAT to another. This reallocation is done manually today based on the spectrum demand of each RAT measured on the long-term perspective; the corresponding process is termed ”spectrum refarming”. However, with the increasing adaptivity and flexibility of future systems, a dynamic spectrum refarming of spectrum may be enabled, where spectrum may flexibly be reallocated according to current and dynamically changing needs of the different operated RATs. In this paper, we present an automated refarming methodology realizing a self-planning approach, which enables an automatic reallocation of spectrum between different RATs. II. S CENARIO FOR SPECTRUM SHARING We assume an operator operating two RATs, namely LTE and UMTS/HSDPA, in its sites. Two bands of 5 Mhz spectrum
187
Fig. 1.
System states.
(dedicated bands) are assigned to LTE and UMTS. In addition, a 5 Mhz spectrum band (flex band) can be assigned to any of the RATs according to their specific traffic loads and demands. Flexible band (FB) can be dynamically exchanged between LTE and UMTS. This scenario is of particular importance in emerging phase of LTE: Currently many UMTS smart phones are operating without support for LTE. With population growth of LTE smart phones, in near future it is likely to observe an uneven traffic demand between LTE and UMTS devices, especially in hot spot areas. With applying dynamic spectrum management, operators can provide the desired service in peak traffic hours or places. For transferring the flex band from one RAT to the other we define four states. Each state represents a system situation. Figure 1 illustrates the four states of the system. State one and two are the main states of the system. State one represents the RAT which only operates on 5 Mhz dedicated band (DB). State two characterizes the RAT which can balance its traffic in both DB and FB. States three and states four are the transition states for moving from state one to state two and vice versa, respectively. In fact, states three and four prepare the FB for reallocation from one RAT to the other. The transition of flex band should be done in a way that increases the operator’s utility. For this purpose, we designed a module, called dynamic spectrum manager (DSM). III. S OLUTION SPECIFICATION For reallocating flex band it is necessary to monitor information from both RATs on their current system conditions. Therefore, we integrate DSM as a new component in the network management system (NMS) architecture. The NMS monitors the status of the RATs and sets off alerts according to conditions that may impact the system performance. We define the amount of free resources in the dedicated band, as one of the status or key performance indicator (KPI). The conditions for alerting the DSM module is defined as reduction of free resources to a certain amount of value. The developed DSM contains two phases of action. The first phase of DSM has proactive characteristics and tries to avoid crisis before it happens. The second phase has reactive characteristics, it tries to manage the occurred crisis in case the proactive phase fails to prevent it. We define the crisis as overloading of a RAT. Our methodology for the proactive phase is to identify cues of an
Fig. 2.
Proactive section.
impending crisis and apply activities that will be carried out to avert it. The proactive phase of the crisis plan should identify the triggers that typically set off a crisis. Successful proactive action can eliminate the need for reactive crisis responses. The reactive phase of the DSM represents how to manage the occurred actual crisis and return the system to a desired situation, which is here solving the overload situation. A. Proactive section Figure 2 illustrates the proactive part of the designed DSM in form of a flowchart diagram. Ellipse is used for showing states, rectangle is used for actions, rhombus as conditional branches. We assumed that at initial stage RAT 1 (LTE) is operating in 5 Mhz dedicated band (state 1) and RAT 2 (HSDPA) owns the flex band (state 2). NMS continuously monitors the amount of free resources as the key performance indicator in RAT 1 and provides alerts for DSM if the KPI falls below a certain threshold. We applied a dynamic threshold for identifying cues of an impending crisis. The threshold is defined as: Γ = ∆t · λ1 · R1avg , (1) where ∆t is a time interval and λ1 is the effective arrival rate of users in RAT 1. The effective arrival rate is defined as: λ1 = ρ1 − ν1 ,
(2)
where ρ1 is the measured user arrival of LTE in a cell and ν1 is the measured rate of users that leave LTE resources in the cell. R1avg is the average amount of the necessary resources required for serving a single user with minimum quality of
188
service. Therefore, Γ is the amount of necessary resources that RAT 1 needs in order to serve the users requesting service during the next ∆t minutes. By violation of the KPI, it is indicated that RAT 1 is running out of resources. Hence, NMS alerts the DSM module, the DSM oversees the situation in both RATs for reallocating the FB by evaluating an utility function. We develop an utility function in order to measure the change in the total relative load in the system caused by band reallocation. The relative load is defined as the amount of consumed resources normalized to the amount of available resources in a RAT; and the total relative load is the summation of LTE and HSDPA’s relative load. With evaluating the total relative load we aim to maximize the total amount of available resources as sum over both RATs. Therefore, we have to minimize the sum of relative loads over both RATs. We defined an utility function that serves the mentioned aim. By evaluating utility function, DSM decides about reallocation of the FB. The utility function is defined as: Near future
Present z }| { z }| { exch + L ], U = [L1 + L2 ] − [Lexch 1 2
(3)
where L1 and L2 are the current relative loads on RAT 1 and RAT 2. For statistical purposes, we average the load over its history. Lexch and Lexch are the relative loads on dedicated 1 2 band of RAT 1 and RAT 2 after exchanging the FB. In our scenario, RAT 2 is already using the FB. After reassigning the FB to RAT 1, the load on the DB of RAT 2 doubles, i.e. Lexch = 2 · L1 . 1
(4)
This is because user served on the FB will be moved to the DB. Correspondingly, the load on DB of RAT 1 reduces to half the value, Lexch = L2 /2. (5) 2 A positive utility function represents: By exchanging the flex band the summation of the relative load of HSDPA and LTE on the system decreases. In other words, if the present total relative load (L1 + L2 ) is larger than the estimated future total load (Lexch + Lexch ), reallocating of the flex band helps 1 2 to balance the relative load on the system and prevents the impending overload in RAT 1. Following figure 2, if the utility function is a positive number, FB will be closed to new users in RAT 2 and a timer is set. With this action we provide the necessary time for RAT 2 to prepare the FB for releasing and give the possibility to the users in the FB to end their session without the need for a handover 1 . After elapsing the indicated time, we set RAT 2 to state 4 if and only if U > 0. At this state RAT 2 hands over the remaining users in FB to DB in order to free the FB. Once the FB is freed, we change the state of RAT 2 to state 1. In response RAT 1 is switched to state 3 to configure the FB for its use. After finalizing the configuration, RAT 1 will change to state 2 and balance its users on both the DB and the FB. Accordingly, RAT 1 and RAT 2 have swapped their states.
Fig. 3.
Reactive section.
On the other hand, if the utility function is negative, the reallocation of the flex band causes the total relative load on the system to increase. Therefore the DSM does not reallocate the FB. Thereupon, with updating the KPI’s threshold to a lower amount we give the system time to develop further. Reducing the threshold can be done by decreasing ∆t in equation (1). The successive violation of the updated threshold represents RAT 1 is approaching to the overload level. If the amount of free resources in RAT 1 falls below the lowest threshold, Γmin = ∆tmin · λ1 · R1avg , (6) it indicates that RAT 1 completely ran out of resources and the DSM triggers the reactive section. With applying reactive section we try to handle and solve the overload. B. Reactive section Figure 3 illustrates the reactive section in DSM. Triggering the reactive section means RAT 1 is overloaded and is in need of additional resources. But before granting the FB to RAT 1, we have to make sure that the reallocation of the flex band improves the overall situation. Before band reallocation it is necessary to check the amount of free resources in dedicated band of RAT 2. The amount of free resources should be large enough in order to accommodate the current users in the FB plus the future users arriving during a predefined time interval ∆tf ut . This value is calculated as follows:
1 For example radio resource manger could be configured so that users with nearly empty queue may get more resources to end their sessions.
189
ψ = N f lex · R2avg + ∆tf ut · λ2 · R2avg ,
(7)
Fig. 4. The numerical decision space of the proactive and the reactive utility functions in equations (3) and (8).
where N f lex is the number of current users in the flex band. If the amount of free resources in DB of RAT 2 is larger than ψ we close the FB to new users and set the timer. Meanwhile RAT 2 prepares the FB for releasing. Thereafter we check the blocking rate of RAT 1. If it is larger than the indicated threshold, we evaluate the total relative load and overload of the system after band reallocation. We extended the utility function in equation (3) in order to measure the amount of change in total relative load and overload in the system caused by band reallocation. The utility function can be described as: Future
Present z }| { z }| { exch exch exch + θ + L + θ ], U = [L1 + θ1 + L2 + θ2 ] − [Lexch 1 1 2 2 (8) where θ1 and θ2 are the overloads in RAT 1 and RAT 2. θ1exch is the predicted overload during next minutes after RAT 1 obtains the flex band, α
θ1exch
}| { z (∆tf ut · λ1 · R1avg ) −R1f ree , = R1total
(9)
where α is the resources necessary for serving future users during next ∆tf ut minutes. The future time interval ∆tf ut must be chosen in a way that the assumed future effective user arrival rate λ1 does not change significantly over ∆tf ut minutes. R1f ree is the amount of free resource in both the DB and the FB; and R1total is the total amount of resources in RAT 1. θ2exch is the predicted overload during next minutes after RAT 2 release the flex band, β
z }| { [(∆tf ut · λ2 · R2avg ) + (N f lex · R2avg )] −R2f ree exch θ2 = , R2total (10) where β is the resources necessary for serving future users during next ∆tf ut minutes plus the resources necessary for serving the current users in the flex band. R2f ree is the amount of free resource in the DB of RAT 2 and R2total is the total amount of resources in RAT 2. A negative utility function represents: By exchanging the FB, the total relative load and probable overload on the system
Fig. 5.
Loads of LTE and HSDPA.
will increase. So the system is not triggered to change the FB. A positive utility function on the other hand represents the total relative load and probable overload will be relaxed by band reallocation. After obtaining all the necessary conditions for band reallocation, RAT 2 is set to state 4 and handover all the current users in the FB to the DB and subsequently we set RAT 2 to state 1. Afterwards RAT 1 shifts to state 3 and prepares the FB for the new users. After finalizing the configuration, RAT 1 will change to state 2 and balance its users on both the DB and the FB. Accordingly, RAT 1 and RAT 2 have swapped their states. Figure 4 demonstrates the numerical decision space of the proactive and the reactive utility functions. The numerical decision space represents the amount of utility value for different amount of loads and overloads in RAT 1 and 2. The red area represents positive utility values (U > 0) and the blue area expresses negative utility values (U < 0). The green line (U = 0) is the border of positive and negative utility values. The dashed line separates the numerical decision space of the proactive and the reactive utility functions. The DSM relocates the FB if the calculated utility value is a positive number. Figure 4 demonstrates that the utility function is positive if and only if the current load and overload on RAT 1 is at least two times larger than the load and overload on RAT 2. IV. S IMULATION R ESULTS We use simulation results for demonstration purpose. We demonstrate the advantage of the DSM by depicting the performance of the system both with and without applying the DSM, for a typical hotspot scenario, and compare them respectively. Furthermore, for illustrating the advantage of the DSM, we count the number of served users with and without applying the DSM and compare them. The scenario is simulated in a system-level. Environment model consists of 48 macro cells and all cells are of equal size. LTE and UMTS are operated at each site simultaneously. FTP traffic is modelled, according to the frequently used FTP model introduced in [12]. A constant number of uniformly distributed users (back ground users) with infinite data queue are moving through the operator’s playground, constituting
190
Fig. 6.
Averaged overall relative load.
Fig. 7.
Total number of served users by RAT 1 and RAT 2.
the basic system load. An average number of 20 users per RAT in each cell is defined. Hot spot covers a single cell, and the velocity of the users in the hot spot is set to zero. New users enter to the playground with constant arrival rate λ. The user arrival rate is modelled by a Poisson distribution. LTE and HSDPA user arrival rate in hotspot is set to 8 and 4 user per minute, respectively. FTP traffic supporting a scalable data rate is assumed. The data rate depends on the number of users simultaneously served in the cell. The average number of packets per session is set to 5, their distribution is modelled as a Poisson distribution. The size of the packets is modelled as truncated log-normal distribution with a mean of 2 Mbytes. As the total number of available resources need to be shared between all the users, the allocated user data rate decreases with increasing number of users. However, to ensure that the data transmission of a user does not fall below a limit, a minimum data rate is guaranteed for each FTP user. The minimum rate for each user is 50 kbit/s in the LTE system and 30 kbit/s in the UMTS system. Furthermore, in order to avoid that a single user occupies too many resources, its data rate is also bound to a maximum. For LTE, maximum data rate is limited to 1 Mbit/s and 500 kbit/s for UMTS. Figure 5 shows the loads on LTE and HSDPA system separately for the hot spot cell,
section of DSM is triggered and reallocates the FB to the LTE. Accordingly, HSDPA hands over all users from FB to the DB. The jump in load of HSDPA is the result of this handover. On the other hand, LTE, after obtaining the FB, balances its load on both bands by handing over the users from DB to FB. Thus the overload in LTE is effectively prevented by applying DSM and no user is blocked. Figure 6 shows the operator’s averaged overall relative load in the hot spot cell according to
LTE’s load = L1 + θ1 ,
(11)
V. C ONCLUSION
HSDPA’s load = L2 + θ2 .
(12)
In this paper, we have presented a new heuristic method for sharing the spectrum between two different RATs in an intra-operator scenario. The presented methodology consists of two complimentary phases, the proactive phase and reactive phase. The proactive phase of this methodology represents how to avoid potential overload in RATs; and the reactive phase expresses how to manage an actual overload in case the proactive phase fails to avoid it. The simulation results illustrate that with applying the proposed dynamic spectrum management we can effectively prevent overload in RATs and balance the operator’s overall load. Furthermore, a higher number of total users can be served by applying the DSM. The simulation results confirm that the designed dynamic spectrum manager is a useful reference for intra-operator spectrum management plans.
Lines with circle marker demonstrate the system load without applying the DSM. In this scenario FB is constantly allocated to the HSDPA system. At t = 11 min the DB of LTE is fully loaded (100%) and starts to block new users. As a result, load on the LTE passes 100% and reaches to 150% (50 percent overload). At the same time just 20% of HSDPA’s resources is used, furthermore it is using FB. The DSM scenario is shown with squared markers lines in figure 5. FB initially is allocated to UMTS, the RAT equally balance the users in both DB and FB. The load on both RATs is continuously increasing. At time equal to 6 minutes, LTE uses about 60 percent of available resources in the dedicated band. In order to balance the total relative load and preventing probable overload, the proactive
2
Averaged overall relative load =
1X Li + θi . 2 i=1
(13)
The blue curve is the load history of the operator during one hour without applying dynamic spectrum management. The red curve shows the load history of the operator with applying dynamic spectrum management. The averaged total relative load on the operator’s RATs is significantly decreased. Figure 7 shows the total number of users served by RAT 1 and RAT 2 in the hot spot cell. Served users represent the number of users that successfully terminated their session during the operation time plus the number of users currently served by the system. It is clearly seen that by applying the DSM approach, a significantly higher number of total users can be served.
191
R EFERENCES [1] J.M. Peha, ”Sharing spectrum through spectrum policy reform and cognitive radio,” Proceedings of the IEEE, vol. 97, no. 4, pp. 708-719, Apr. 2009. [2] S. Shellhammer, A. Sadek and W. Zhang, ”Technical challenges for cognitive radio in the TV white space spectrum,” Information Theory and Applications Workshop, Feb. 2009. [3] FP7 project, ”FARAMIR, Flexible And spectrum-aware Radio Access through Measurement and modelling In cognitive Radio system,” Internet: http://www.ict-faramir.eu, ongoing. [4] FP7 project, ”COGEU, COgnitive radio systems for efficient Sharing of TV white Spaces in EU context,” Internet: http://www.ictcogeu.eu/deliverables.html, ongoing. [5] FP7 project, ”QoSMOS – Quality of Service and MObility driven cognitive radio Systems,” Internet: http://www.ict-qosmos.eu, ongoing. [6] FP7 project, ”SAPHYRE, ShAring PHYsical REsources mechanisms and implementations for wireless networks,” http://www.saphyre.eu/publications/deliverables, ongoing. [7] FP7 project, ”SACRA – Spectrum And energy efficiency through multiband Cognitive RAdio,” Internet: http://www.ict-sacra.eu, ongoing. [8] C. Kloeck, D. Grandblaise, J. Luo, G. Dimitrakopoulos, ”Multi-level spectrum auction through radio access,” IEEE Mediterranean Electrotechnical Conference (MELECON), pp. 591-594, May. 2006. [9] M. Li, X. Li, H. Ji, G. Dimitrakopoulos, ”Virtual bidder group auction mechanism for dynamic spectrum access,” 20th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, pp. 2797-2801, Apr. 2010. [10] X. Zeng, Z. Feng, V. Le, Y. Xue, Y. Lin, ”An auction based joint radio resource management scheme and architecture in a multi-operator scenario,” IEEE Vehicular Technology Conference (VTC), pp. 2797-2801, May. 2008. [11] FP7 project, ”E3, End-to-End Efficiency,” Internet: https://icte3.eu/project/deliverables/deliverables.html. [12] 3GPP, ”Feasibility study for orthogonal frequency division multiplexing for UTRAN enhancement,” TR 25.892, V 6.0.0, 2004-06.
192
