Optimisation of Radio Access Network Operation ... - CiteSeerX

9 downloads 728 Views 319KB Size Report
functional architecture the following functional blocks are introduced: (i) Dynamic Spectrum Management (DSM), (ii). Dynamic Self-Organising Network Planning ...
Optimisation of Radio Access Network Operation introducing self-x functions use cases, algorithms, expected efficiency gains J. Belschner, P. Arnold

H. Eckhardt, E. Kühn

Deutsche Telekom Laboratories Darmstadt, Germany [email protected] [email protected]

Alcatel-Lucent Bell Labs Stuttgart, Germany {edgar.kuehn, harald.eckhardt}@alcatel-lucent.de

E. Patouni, A. Kousaridas, N. Alonisioti

A. Saatsakis, K. Tsagkaris, P. Demestichas

Department of Informatics and Telecommunications, University of Athens, Athens, Greece {elenip, akousar, nancy}@di.uoa.gr

University of Piraeus, Department of Digital Systems, Piraeus, Greece {asaatsak, ktsagk, pdemest }@unipi.gr

Abstract— With the deployment of next generation (4G) mobile radio systems an additional radio access network is established. A variety of different Radio Access Technologies (RATs) will be operated in parallel. In this framework, the Long Term Evolution (LTE) system, specified by 3GPP, will have to co-exist with WiMAX, mobile 2G/3G networks and Wireless Local Area Networks (WLANs). To cope with this increasing diversity and complexity mechanisms for self-optimisation, self-organisation, self-healing, self-configuration (self-x) are essential to guarantee cost efficient and high quality network operation. Within the project E³ [1] self-x functionalities for different use cases and different elements of a mobile radio access network are developed. Aim of this paper is to give and overview about the interworking of different self-x functionalities and to present three exemplary use cases.

introduced: (i) Dynamic Spectrum Management (DSM), (ii) Dynamic Self-Organising Network Planning and Management (DSNPM), (iii) Reconfiguration Control Module (RCM) and (iv) Joint Radio Resource Management (JRRM), see Figure 1. A. Aspects of Functional Architecture In the following, the functionality and scope of these functional blocks are described.

Multi / Meta Operator Single Operator

DSM/DSA

I.

INTRODUCTION

Evolution of mobile radio networks is driven by the demand for new, high bit rate consuming applications and services. The technical answer is the development of new and more powerful radio technologies and integration into existing mobile radio networks. This leads to significantly higher complexity and heterogeneity while pressure for maintaining manageability and cost efficiency of the networks is continuously increasing. E³ answer is introduction of solutions for obtaining higher flexibility and efficiency in usage of radio, hardware and computational resources by cognition, selforganisation and self-optimisation. This paper presents the interworking of cognitive radio self-x functions in E³ environments and hereafter three exemplary self-x use cases namely Handover Parameter Optimisation, Protocol Stack SelfConfiguration & Topology Self-Organisation and Knowledgebased Proactive Context Handling. II.

INTERWORKING OF COGNITIVE RADIO SELF-X FUNCTIONS IN E3 ENVIRONMENTS

For self-organisation and self-optimisation in the E³ functional architecture the following functional blocks are

The work reported here has been partly funded by EU FP7 research programme ICT

DSNPM SON decision taking SON self-x functions

RCM

JRRM

Multi Radio

SON SON

Radio Technology Specific

RRM

Figure 1

Vendor Specific SW/HW

Cut-out of E3 functional architecture pillars

1) The Dynamic Spectrum Management provides the framework for spectrum assignment decisions by DSNPM. It is based on regulatory guidelines, operator policies and preferences as well as negotiations with other DSM instances. 2) The Radio Resource Management (RRM) acts on a given set of radio, hardware and computational resources. Its goal is to optimise resource assignment to users and services. There is a radio technology specific scope (RRM) and a multi radio scope (JRRM) providing optimised access selection and resource assignment among all available radio technologies.

3) The DSNPM comprises a decision taking part which decides on the optimal configuration of the radio access networks including spectrum and radio technology assignment per cell. It has an overview about the network at least in a certain geographical area. It decides on node specific resource assignment on an abstracted level as input for the RCM. The DSNPM comprises a second functional part focused on single network nodes and closest neighbours. It provides self-x functions e.g. for self-configuration, self-optimisation and selfhealing. There is for both DSNPM parts a radio technology specific scope e.g. for optimisation of configuration parameters and of RRM resource usage. Also, there is a radio technology overarching scope for optimising the interactions between radio technologies as e.g. for so-called vertical handover or inter-RAT load balancing. Self-Organisation functionality is here regarded as operator domain specific. 4) The Reconfiguration Control Module acts on computational and hardware resources. RCM optimisation takes place by reassigning these node resources locally between cells from the same or different radio technologies. B. Functional interaction The interaction of the introduced functional optimisation blocks with each other and with the legacy functions RRM and HW specific platform functions is analysed below. 1) In case of a DSNPM reconfiguration decision, there is an impact on Radio Resource Management (RRM and JRRM) and on DSNPM internal self-x functions. Reconfiguration of resources assigned to a distinct cell/radio technology requires an adaptation of configuration parameter settings for RRM and DSNPM self-x functions. Examples are: i) Admission Control in RRM needs to be adapted to changed resources either to prevent overload situations or to allow a higher traffic volume. ii) The scheduler is impacted by the given bounds of the resource pool. If the cognitive part changes these bounds, the scheduler has to be adapted accordingly. iii) For load reporting the modified resource situation has to be configured. However, load balancing automatically acts on the new load situation, decoupled from resource change. iv) In case of instantiation of a new cell or deletion of an already existing one, DSNPM self-x functions are involved in Self-Configuration of the new cell’s parameters, also resolving the impact from or on neighbour’s configurations as e.g. cell ID, neighbour list and interference coordination schemes. 2) The DSNPM self-x functionality interacts with RRM and RCM in cases of e.g. self-configuration, self-optimisation and self-healing: i) Natural task of the DSNPM self-x functionality is to act on RRM and JRRM resource usage configuration parameters in an automated, continuous process e.g. for handover optimisation, load balancing, interference coordination and scheduler optimisation. HO- and congestion-control are based on the cell load, neighbour cell relations and especially in the inter-RAT handover case on the availability of resources for radio bearers.

ii) Interaction from DSNPM self-x on RCM exists in case of a RAT internal change of cell resources, especially when capacity cells are switched on or off due to power saving reasons. 3) If a whole cell shall be shut down due to a cognitive decision from DSNPM decision taking or self-x functions part, handovers for all served terminals have to be initiated to empty the cell. It may take some time until the resources of the cell can be completely released. So far, synchronisation between DSNPM, RCM and RRM/JRRM is required. C. Support Functions For supporting the different cognitive self-x functions for decision taking and self-organisation, there are two categories of functional extensions required, namely for monitoring and reconfiguration control. The monitoring functions require partly extensions to the legacy measurements defined for resource management to support DSNPM as e.g. i) existence (status) and properties of operational cells, ii) load status per QoS class and number of terminals and iii) multistandard capabilities of attached terminal. The mutual interworking of the different cognitive self-x functions for decision taking and self-organisation shown above implies the need for reconfiguration control in order to coordinate reconfiguration and optimisation processes. This comprises e.g. resource reservation, sequence control and synchronisation of reconfiguration processes. III.

HANDOVER PARAMETER OPTIMISATION

A. Introduction Handover (HO) is the basic radio resource management functionality for transferring a mobile stations (MS) connection from one base station to another and is especially needed to support MS mobility. Handover parameters decide on the point in time when a HO is carried out. Since the correct execution time is crucial for a successful HO, incorrect parameters settings lead to failures (handovers executed too early/late) or to unnecessary handover executions. Tuning of handover parameters is a task with high manual effort in current wireless networks. Therefore self-optimisation algorithms shall be able to find best HO parameters automatically. B. Algorithms The principle mode of operation of a HO parameter selfoptimisation algorithm is shown in Figure 2. Information about the network status and the current network’s settings form the input parameters of the optimisation algorithm which then derives improved parameters. To avoid a real network’s operation with incorrect settings, correct settings have to be found by the optimisation algorithm as fast as possible. This brings up the need for heuristic algorithms. A first approach for optimisation algorithms can be a rule based algorithms that represent expert’s knowledge. Rules can be defined like: “If more than x% of all handovers fail, reduce configuration parameter y by z”. Besides that, other heuristic algorithms suitable for large-scale optimisation problems like genetic algorithms are applicable.

Network status: Number of unnecessary handovers during last x seconds Number of handover failures during last x seconds Network handover parameter settings: Hysteresis settings Time to trigger settings Network status and Network settings form the algorithm input parameters

Algorithm output: Optimised handover parameter settings

Optimisation Algorithm (Rule based, genetic, …)

Figure 2

Principle mode of operation of HO self-optimisation

C. Expected Efficiency Gains Handover parameter self-optimisation is supposed to bring improvements in two domains: • •

Reduction of manual effort. In a network with implemented handover parameter optimisation no manual optimisation effort for this task has to be spent. Improved handover quality compared to networks with suboptimal configuration. If in a network without selfoptimisation HO parameters are not or only sometimes optimised manually (e.g. for cost saving reasons), HO parameter self-optimisation will be able to improve handover quality.

D. Simulation Results LTE system level simulations were carried out. Figure 3 shows rates of HO failures and unnecessary HOs with static hysteresis settings. As low hysteresis settings lead to a high number of unnecessary HOs and high settings to many failures, reasonable settings for hysteresis are within the area of 2-4 dB. Figure 4 shows the settings generated by a rule based self-x algorithm. Two rules were defined: a) If a failure occurs, increase hysteresis setting by 0.5 dB, b) If an unnecessary HO occurs, decrease hysteresis setting by 0.1 dB.

Figure 4

IV.

Hysteresis settings generated by rule based algorithm

PROTOCOL STACK SELF-CONFIGURATION & TOPOLOGY SELF-ORGANISATION

A. Introduction In a complex multi-RAT environment, mobile devices and several network elements have intrinsically embedded cognitive capabilities. These interconnected elements have the ability to dynamically change their configuration and compose clusters or form collaborations. The goal of this use case is twofold, to describe: a) the dynamic change of the protocol stack configuration, including either the dynamic switching from one technology to another one or the dynamic adaptation of a specific protocol through parameter tuning or using component-based approaches, b) the dynamic re-organisation of topology structures for a specific optimisation goal (energy consumption, delay or throughput) that will optimise the local and global behaviour of the corresponding network area. B. Algorithms In Figure 5 a composite scenario is considered, where a mobile device initially attempts to solve a performance degradation problem in a self-organising manner by negotiating with neighbouring devices for a topological change.

The generated values are in the expected range from 2 to 4dB. Using the settings generated by this rule base algorithm lead to 8.4% unnecessary HOs and 1.7% failures (see Figure 3 for comparison).

Figure 3

HO failures and unnecessary HOs with static HO parameters Figure 5

Steps of the algprithm for protocol stack self-configuration and topology self-organisation

However, the involved elements do not agree on the topology re-organisation and the mobile device decides on another alternative in cooperation with the network. Such alternative is the protocol re-configuration procedure. The main stages for the self-organisation are: a) the discovery of selforganisation opportunities and the identification of available resources, b) the negotiation between the involved elements to specify their participation and the rules of their collaboration, c) the decision on the goal-driven optimal formation, satisfying global and local performance metrics, and d) the control of the new formation after its establishment [3]. In Figure 5 only the first two phases are analysed. As regards the self-configuration procedure it includes two phases: a) the distributed decision making for protocol configuration, where the new configuration of the protocol stack is also specified and b) the dynamic configuration of RAT protocol components, which concerns operations for the appropriate component instantiation and dynamic binding, i.e. state management operations [2]. C. Expected Efficiency Gains The introduction of self-organisation capabilities will facilitate the management of the increased complexity of network parameters and structures, using decentralised techniques that do not observe single point of failure or bottleneck problems as well as show high scalability features. The introduction of protocol self-configuration features enables the administration of different versions of protocol implementations, remote management procedures and helps the smooth integration of emerging standards. In addition, the distributed decision making approach can be also used to dynamically check whether the network computational resources reach their full utilisation and therefore schedule proactive operations to avoid respective bottlenecks, as it is discussed in section IV.D. D. Simulation Results In this section we examine how the introduced protocol reconfiguration functionality affects the network side, considering the decision making requests produced by mobile devices. Specifically, we model a reconfigurable system as a distributed transactional system and consider two classes of mobile devices: reconfigurable and semi-autonomous; the difference between them lies on the degree they support decision making functionality.

Figure 6

Lower and Upper Global Bounds of the Asymptotic Response time for Reconfigurable Terminals

Our work uses multiclass queuing networks for the system model based on the findings in [4]. The outcome of this model is the derivation of the global bounds for the asymptotic network response time for reconfigurable mobile devices versus the number and frequency (think time) of reconfiguration decision requests as illustrated in Figure 6 [2]. It should be noted that in this model the network computational resources are considered (the CPU and disk resources of the network node and context server). V.

KNOWLEDGE-BASED PROACTIVE CONTEXT HANDLING

A. Introduction DSNPM is enhanced with learning attributes that yields knowledge and experience for predicting problems or new environment conditions and act proactively for solving them. The target of knowledge-based proactive context handling in a wireless network segment is to predict short term problems or changes and proactively proceed to self-optimisation based on the experience of the system gained in the past. DSNPM is constantly checking the context and trying to predict environment conditions that occurred in the past with the same or similar parameter values. As soon as DSNPM identifies that the same or similar context will appear again in the service area the adaptation mechanisms are triggered and the past solution is retrieved in order to be sent to the service area and be implemented. B. Algorithms The objective of the DSNPM is to provide the medium and long term decision upon the reconfiguration actions a network segment should take, by considering certain input information, and by applying optimisation functionality, enhanced with learning attributes [5]. Figure 7 depicts its overall description. “Context” collects the status of the elements of the network segment, and the status of their environment in terms of traffic requirements, mobility conditions, configuration used, and the QoS levels offered. Discovery mechanisms provide information on the QoS that can be achieved by alternate configurations. Context information will be used from the system to update network Key Performance Indicators (KPIs) and to address possible problematic situations.

Figure 7

DSNPM for B3G wireless network segments

C. Simulation Results and Efficiency Gains One of the main functionalities, as mentioned above, is the ability of DSNPM to identify past contexts and retrieve their corresponding solutions. Figure 8 depicts the successful matching probability evolution as an indicative example in LTE RAT simulations [8].

The probability for successful matching is increasing since each time the system captures an unknown context has the ability to store its characteristics and solution. Thus, after a short time the system has “learnt” all the different contexts and their solutions and beyond that point it is capable to provide the appropriate solutions directly. Figure 9 depicts also the evolution of the mean response time for several context IDs in X axis. It is clear that the time needed in order to provide the solution for the contexts is decreasing by exploiting its learning capabilities. 6,00 5,00 4,00 Time

“Profiles” provides information on the capabilities of the elements and terminals of the segment, as well as preferences, requirements and constraints of users and applications. This information is necessary during the optimisation process in order to decide on the most appropriate reconfiguration. “Policies” Policies designate rules that should be followed in context handling. Sample rules can specify allowed QoS levels per application, allocations of applications to RATs and assignments of configurations to transceivers. Regarding the DSNPM decisions, they are targeted at producing a feasible network reconfiguration that can be categorised at the following levels: • Application layer: guaranteed QoS levels assignment to applications; • Network layer: traffic distribution to specific transceivers and corresponding RATs as well as network entities interconnection; • Lower/PHY layer: Number of network element transceivers involved in decisions, RATs to be activated, spectrum selection and radio parameters configuration per RAT. The optimisation process exploits RATs capabilities so as to provide users with the maximum possible QoS level. To do so, several approaches are envisaged. One approach would be to find the best configurations that maximise an objective function, which takes into account the user satisfaction that derives from the QoS levels offered, the cost at which they are offered, and the cost of the reconfigurations [6].In addition, DSNPM exhibits self-x capabilities by mobilising specialised learning attributes that will yield knowledge and experience. Such attributes may regard context (capability to identify previously tackled situations and their suitable solutions [7], [8]), profiles (certain user classes may be better served via a specific RAT), as well as policies (storage of data on NO policies and exploitation of them in future situations).

2,00 1,00 0,00 5 4 2 4 3 2 2 3 1 3 4 5 3 3 2 5 3 4 2 2 1 5 2 3 3 2 1 3 4 3 Conte xt

Figure 9

VII. REFERENCES [1] [2]

[3]

[4]

Success Probability

0,60 0,50 0,40

[6]

0,30 0,20 0,10

[7]

0,00 Tim e

[8] Figure 8

Successful matching probability evolution

CONCLUSION

This paper described the interworking of different self-x functionalities in E³ context and presented three exemplary use cases. These three use cases stand for the variety of application areas that self-x will cover in mobile radio networks. The expected efficiency gains differ like the area of application and will in sum help to provide high quality of service and cost efficiency in a dynamic complex multi RAT environment.

[5]

0,70

System response time evolution

VI.

0,90 0,80

3,00

E³ web site https://ict-e3.eu/ E. Patouni, N. Alonistioti and L. Merakos , “Protocol Reconfiguration: Analysis of Network Response Time", submitted to ICC 2009, June 1418, Dresden, Germany A. Kousaridas and N. Alonistioti, “On a synergetic architecture for cognitive adaptive behavior of future communication systems”, International Symposium on World of Wireless, Mobile and Multimedia Networks (WoWMoM), pp. 1 - 7, 2008. Marin Litoiu, “A performance analysis method for autonomic computing systems”, ACM Transactions on Autonomous and Adaptive Systems (TAAS), v.2 n.1, p.3-es, March 2007 A. Saatsakis et al, ”Functional Architecture for the Management and Control of Reconfigurable Radio Segments in the Wireless B3G Era”, in Proc. 21st Wireless World Research Forum (WWRF) meeting, Stockholm, October 2008 K. Tsagkaris et al “Distributed radio access technology selection for adaptive networks in high-Speed, B3G infrastructures”, Int. J. of Commun. Syst., Wiley, Vol. 20, Issue 8, pp 969-992, August 2007 A. Saatsakis et al, “Enhanced Context Acquisition Mechanisms for Achieving Self-Managed Congnitive Wireless Network Segments”, in Proc. ICT-Mobile Summit Conference, 2008. A. Saatsakis et al, “Cognitive Radio Resource Management for Improving the Efficiency of LTE Network Segments in the Wireless B3G World”, in Proc. IEEE DySPAN 2008, Chicago, 14th– 17th October, 2008

Suggest Documents