Optimal Resources Allocation for a Cognitive Network

• Main Authors: Shu-hsien Chu, 朱書賢
• Other Authors: Wen-kuang Kuo
• Format: Others
• Language: en_US
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/55488216217749097579

碩士 === 國立成功大學 === 數學系應用數學碩博士班 === 97 === In this paper, through characteristic analysis of wireless network, we formulate the cognitive radio multi-hop network cross-layer resource management optimization problem into a network maximum throughput problem with 0-1 integer constraint and non-convex constraint. And this structure can be extended to networks considering different objects, e.g., minimum power consumption, minimum worst congestion level, and maximum worst PSNR quality. This problem totally cross three layers, physical layer, data-link layer, and network layer. The link scheduling, channel assignment, flow allocation, routing, and power management issues are involved in a NP-hard mixed-integer-non-linear program (MINLP). Technically, we face two major constraints. For the non-convex constraint, we develop a local linear approach (LLA) algorithm. In this algorithm, we partition the domain of variables in the non-convex function. Then, we find a single variable upper (or lower) approximation function on each partition. By linearizing the approximated function, we find the upper (or lower) linear function and replace the non-convex function with it. For the 0-1 integer constraint, we use branch and bound to find the optimal solution at first. In this network model, there exists many skew subtrees in the binary tree of branch and bound. The skew structure results the speeded up solution. But the time it need for finding solution still grow up exponentially. Thus, we propose a heuristic scheme, Greedy Search (GS). In this scheme, we assume that someone of sorted relax integer variable is 0 or 1 respectively and find the corresponding solutions. Then we compare the solutions and choose the larger one. The process will continue till all relax variables are determined. Moreover, this is a linear time algorithm. The numerical result shows that the approximated solution obtained form our scheme is close to the optimal. And it can solve large networks that SBNB can not.