基于快速MCVI算法实现CR信道接入*

在认知无线电(CR)技术中,无线网接入是一个极为重要的问题。针对这一问题,当前主流的解决思路包括博弈算法和基于部分可观测马尔科夫决策过程模型(POMDP)的算法。相比于博弈算法,基于POMDP模型的算法具有更好的环境适应性。在此背景下,本文提出了基于POMDP模型的快速蒙特卡罗值迭代算法(Fast MCVI)解决无线网接入问题。与其它解决POMDP模型的算法不同,该算法可解决值连续状态空间下的POMDP模型,具有更好的可靠性和稳定性。另外,相比于传统MCVI算法,快速MCVI算法使用非可支配排序遗传算法(NSGA2)进行优化,加快了算法收敛速度,使其在相同运行时间内能获得更好的决策结果。实验证明,通过值连续状态空间的POMDP模型对CR接入问题建模并使用快速MCVI算法进行决策,网络吞吐率比传统MCVI算法提高了1~1.7个百分点,比贪心算法提高了2.8~5个百分点。

CR Channel Access Based on the Fast MCVI Algorithm

In the technology of the Cognitive Radio (CR), the channel access is a very important problem. To solve this problem, the main ideas include the game algorithm and the algorithms based on the partially observable Markov decision process (POMDP). Compared with the game algorithm, the algorithms which are used to solve the POMDP have the batter adaptability of environmental. In this context, this paper proposes the Fast Monte Carlo Value Iteration Algorithm (MCVI), which is a useful algorithm for POMDP, to solve the CR channel access. Different from the other algorithms, the Fast MCVI can solve the POMDP model with the continuous state space. By this way, the result will be more reliable and stable. In addition, compared with the original MCVI, the Fast MCVI is improved by the Non-Dominated Sorting Genetic Algorithm (NSGA2). After the modification, the convergent rate of Fast MCVI is improved and it can get the batter decisions in shorter period of time. The experimental results indicate that the Fast MCVI achieves 1% ~ 1.7% increases in the network throughput over original MCVI and achieves 2.5% ~ 5% increases over the game algorithm.