An Research On The Problem Of Passive Optical Access Network Optimization

Triggered by the explosion in the Internet population and traffic, broadband access has become one of the central issues in telecommunication systems. The requirements for new access network technology are not only the provision of broadband services but also reliability as a social information infrastructure. Taking account of these requirements, in this paper we propose Passive Optical Network technology as the promising access network technology, which can provide broadband access and can smoothly evolve to the age of FTTH because of its numerous advantages such as less cost, more customers and wide rangesupport. To minimize the installation costs it\is necessary to optimize network toplogy, Therefore, it has a great significant to study Passive Optical Access Network (POAN) optimization.The optimization of POAN contains two problems: One is how to assign the Optical Branching Devices (OBDs) to Optical Network Units(ONUs), and the other is how to determine the locationas of these OBDs. Until now, this type of combinatorial problems are usually solved by linear programming, interger programming, analytical methods, heuristic methods and so on. But these methods have many disadvantages such as .low searching efficiency, converging slowly, converging to local optimization solution easily and not converging. The GA that this paper designed can deal with these problems well because of its simplicity, robustness and high searching efficiency.This paper first describes the conception of access network and Passive Optical Network. Then we introduce the architectur of POAN. After we introduced the Genetic Algorithm (GA), we focus on how to use GA to implement the toplogical optimization of POAN. In this paper, how to design the fitness function, encoding method, selection method, crossover method and mutaion method are discussed in detail. The simulation experiments are carried out and the simulation results verified the e'ffectivness of the GA.Also, we discussed the performance of GA and how its performance is affected by its main parameters. By comparing GA with other algorithms such as Simulated Annealing Algorith and Hopfield Net method we lead a conclusion that the effectivness of the GA is better than other algorithms.Then a new improving scheme of GA which we call it HGSAA (Hybrid Gentic-Simulatd Annealing Algorithm) is put forward based on the existing condition, and the new algorithm works more efficiently than the old one because it has not only GA's global searching capability but also SAA's local searching capability. The simulatio results show the new algorithm is more effective than GA. The minimum cost is reduced and the clustered network structures have been produced with the ONUs are evenly distributed or clustered around the OBDs and the clustered network structures are well ordered with a satisfactory degree of symmetry after the HGSAA has runed 100 generation.At the end of this paper, we summarize the whole work have been done and point out the direction we should strive to.