COST-BASED FWA NETWORKS OPTIMISATION

Sigit Haryadi
Institute Technology of Bandung-Indonesia
sigit@telecom.ee.itb.ac.id, http://telecom.ee.itb.ac.id/~sigit

Abstract
In this paper I propose a heuristic cost-based topological optimisation method for FWA (Fixed Wireless Access) transport networks. My method minimizes total cost by means of optimizes the number and location of Base Station Controller (BSC) jointly with the transmission network topology. The optimisation method is based on combination of a greedy algorithm and the ChandyRussels algorithm. I test the method on different inputs and analyse the optimality of the resulting network cost. Results show that my proposed algorithm provides optimum solutions. Keywords: FWA (Fixed Wireless Access), greedy algorithm

1. Introduction
The Fixed Wireless Access (FWA) will differ from POTS (Plain Old Telephone Services) network in several ways. The main difference from the users' point of view is that the FWA have a mobility services, with special emphasis on data communications. Thus, the transport network of FWA is going to be a rather complex and high-capacity system. In the development countries, these large and costly systems are going to be deployed in very short timeframes, which emphasizes the importance of a good network design methodology. In fact, the complexity of the system, the high cost factors and the shorter design timeframes call for the use of algorithmic network topology optimisation methods. Optimisation has been used for network topology design for a long time; for example, it is widely applied in the planning process of GSM networks and UMTS networks. However, FWA differs from GSM networking technologies in many ways; the FWA transport technology puts different demands and constraints toward the topology of the network. Therefore, new optimisation methods are needed to consider the new topological constraints. FWA networking technologies differ from UMTS networking technologies because mostly FWA networks have a LDC (Long Distance Centre) as a concentrator in a local area service. The task I address here is twofold: First, I want to find the cost-optimal number and location of BSC (Base Station Controller) nodes in the network. Second, I want to find a cost-optimal interconnection network connecting Base Transceiver Stations to the Base Station Controllers, taking into account specific FWA topological constraints. My primary goal is to perform the two optimisation steps jointly, thus obtaining a solution closer to the global optimum. I formulate the task as a discrete optimisation problem. I am restricted to use some kind of quick heuristics here, because scalability is a primary issue. Typically, greedy algorithm is widely used and provided good results in many different optimisation areas, so I decided to use this method for my network design problem. The FWA transport network is going to be a rather complex and high-capacity system, so I decided to use the Chandy-Russels algorithm combined with a greedy algorithm. The rest of the paper is organised as follows: first, I shortly describe the architecture of FWA, and the network model I use for my investigations. Following the problem definition, I propose a

heuristic optimisation method for solving the problem. I conclude the paper with the analysis of the proposed algorithm simulation. 1.1. The FWA architecture The terrestrial access part of the FWA network consists of two types of network elements called BSC (Base Station Controller) and BTS (Base Transceiver Station). The BSC manages the radio channels of BTS connected to it, concentrates the connections and trunks them to the LDC. The BTS handles the radio channels and forwards the traffic of lower level BTS towards their dedicated BSC, called home BSC. In practice, the FWA network is combined with POTS (Plain Old Telephone Service) network. So the position of some FWA-BSC must be in the LE (Local Exchange). My task is to find where and how many BTSs can be decided as others FWA-BSC.
Figure 1 The hierarchical architecture of the FWA terrestrial access network.

POTS Junction Network

BSC
BTS BTS BTS

BSC
BTS

BSC

BTS

BTS

BTS BTS BTS

BTS

On logical level each BTS is connected directly to its home BSC using ATM PVCs (Permanent Virtual Connections), so the logical topology of the FWA network is a set of star sub networks. On physical level the BTSs are connected to their home BSC directly or through some other BTSs using optical-fibber line or microwave interconnections. It means that the physical level network topology is a set of trees. Due to various technological reasons there is a constraint on how many BTSs can be connected to a BSC [1], on logical level further on this limit will be called BSC degree. On physical level the maximum number of BTS's connected parallel to an upper level BTS is limited, in this case there is a degree constraint for each BTS. Also there is a constraint on the depth of the BTS sub trees, what means how many other BTS's can be placed between a BSC and a BTS on physical level, in other words how many BTS's can be cascaded in the network. This constraint is called cascading constraint further on.

1.2. Simulations Inputs & Outputs A relevant and currently open design task of the FWA terrestrial access network is to plan a cost optimal physical and logical topology the access network. This task covers the optimisation of: The number and location of BSCs , include LDC and LE as the predefined BSCs The planning of the physical links connecting BTSs to their corresponding BSCs. Minimizes of total cost My proposed algorithm suits for both of the above task, and it is also able to continue the design of an existing network. The inputs to the planning algorithm are: The set of geographical sites of BTSs and cell-traffic forecast The location of some BSCs which predefined in the POTS-LE and during the optimization process they can not be moved Estimation of transmission & switching cost 2. The Proposed Planning Algorithm In this section I will present the formulation of the design problem and the applied mathematical model. The model used is general for these types of optimization problems [1]. I get the following as input : The position of LDC-POTS as a FWA concentrator (as an extra input add to UMTS planning) The cost of transmission link A set S of sites. A site is a geographical point, where a BTS, BSC or both of them can be optionally placed. Let us denote the number of sites by n. A set B of BTS's. This is a subset of S. Two additional subset of S, the predefined and possible locations of BSC devices, denoted by Rpre and Rpos. The cascading constraint, BTS degree constraint and BSC degree constraint denoted by L, DBTS and DBSC. The cost of a BSC device denoted by CostBSC. The capacity independent part of the cost of the link between two sites denoted by cij for all pair i; j of sites. These are typically in direct proportion to the distance of the sites but they can be arbitrary in my model. A function ijcap: R R for all pair of sites which represents the capacity dependent part of the link cost. These are increasing and typically step-like functions. The amount ti of traffic generated by a BTS for all station. A set of predefined links. It must satisfy the constraints on connecting. With these notations my task is to put some BSCs to the possible places and to find a set of links which connect the BTSs to the BSCs and satisfy the constraints. And, naturally, my main goal is to minimize the total cost of the network. In my model, the total cost of the network consists of the cost of BSCs and the cost of the links. The cost of BSCs is in proportion with their number. The cost a link is the sum of its the capacity dependent and independent cost. The cost of the BSC subsystem increases linearly with the number of BSCs, because each BSC has a similar, constant cost value. Meanwhile, the cost of BTS access links decreases with the number of BSCs, because the length of the physical links will be smaller when having more BSCs.

My task is to minimized total cost as follow: (1) Total cost = BSC cos t + link cos t BSCs cost depend on BSC device cost and BSC number, while link cost depend on BTS traffic, link device cost and traffic routing. I formulate the task as a discrete optimisation problem. Considering the complexity of the problem, I am restricted to use some kind of quick heuristics here, because scalability is a primary issue. Typically, greedy algorithm is widely used and provided good results in many different optimisation areas; I decided to use this method for my network design problem. The transport network of FWA is going to be a rather complex and high-capacity system, so I decided to use the Chandy-Russels algorithm combined with greedy algorithm. Step 1 : Considering the FWA problem, LDC must be judged as a centre of star network, so a very big traffic must be added as a dummy traffic to the LDC in the beginning of FWA planning. LE must be judged as a secondary centre of FWA network, so a big traffic must be added as a dummy traffic to the LE in the beginning of FWA planning.[2] LDC Traffic = normal LDC traffic + very big traffic as a dummy traffic (2) LE Traffic = normal LE traffic + big traffic as a dummy traffic (3) Make sure : LDC trafic >> LE trafic >> BTS trafic (4) Step 2 : Run a software program of the combination of greedy algorithm and the Chandy-Russels algorithm. Typically as follow A: Input : a. i. LDC, LE, Rpre , and Rpos1 positions ii. LDC , LE and BTSs Traffic iii. Link by link cost iv. BSC device cost b. Generate: i. M-Matrix of traffic ii. U-Matrix of traffic iii. C-Matrix (link cost) iv. M-Matrix of U/C c. Count: maximum value of each row d. Find: node connection 3. Planning Algorithm Simulation Some running results are given in Table 1. In order to clear my proposed algorithm, figure 2 and figure 3 show the process of first row Table 1.
#Nodes 40 100 200 300 2 Levels network 3 Levels network 4 Level network 175.234 190.112 201.453 240.245 235.398 242.709 305.023 302.267 298.214 367.902 366.765 360.043 Table 1 Total Charge of Simulation Results 5 Level network 203.876 243.980 308.087 378.099

Figure 2 Initial Condition of the Network

LE = 2nd level BSC LE = 2nd level BSC

LE = 2nd level BSC

LDC = highest level BSC

LE = 2nd level BSC

LE = 2nd level BSC

LE = 2nd level BSC

= possible location of BTS and BSC

Figure 3 Final Condition of the Network

LE = 2nd level BSC LE = 2nd level BSC

LE = 2nd level BSC

LDC = highest level BSC

LE = 2nd level BSC

LE = 2nd level BSC

LE = 2nd level BSC

= Low-level BSC (and BTS level) = BTS

4. Conclusion I have proposed a heuristic optimisation algorithm of planning cost-optimal FWA access network which combining a greedy algorithm and the Chandy-Russels algorithm. My algorithm jointly optimizes the number and location of BSC and the link topology connecting BTSs to their corresponding BSCs. Simulation and running of this algorithm efficiently on various inputs.

5. References
[1] HARMATOS, J., JUTNER, A. and SZENTESI, A., Cost-Based UMTS Transport Network Topology Optimisation, Traffic Lab-Ericsson Telecommunication Ltd., Budapest - Hungary, 1 April 1999 [2] HARYADI, S., Mobile Switching Centre Processing Capacity Computations, Seminar Proceeding International Wireless Symposium, 21 September 2005, Aalborg, Denmark.

[3] HARYADI, S., Moble Multmeda Teletraffc Swtchng Processng Time, Seminar Proceeding Mobile Multi Media, 19 September 2005, Kuala Lumpur Malaysia