【作者】 黄新林；

【导师】 王钢；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2011， 博士

【摘要】 由于日益增长的无线接入业务需求以及授权频谱的低利用率现状,使得认知无线电和动态频谱接入技术成为下一代通信网络(即认知无线电网络,Cognitive radio network,CRN)的核心技术。CRN通过机会式的频谱接入来获得巨大的无线带宽,从而提高频谱利用率。然而,CRN没有固定的工作频带以及调制方式,这些传输参数取决于频谱感知、频谱切换、频谱共享以及频谱管理等过程。为了在CRN上实现多媒体传输,本文从上述四个过程出发逐步展开课题的研究。首先,本文提出了一种基于空时数据挖掘和压缩感知的联合频谱感知算法。频谱感知是认知无线电区别于传统无线电的关键技术,感知无线电环境是实现认知无线电传输的基础。由于CRN布置在一定的空间内,不同的簇可能具有不同的稀疏频谱,在联合频谱感知过程中不适合进行信息共享,这一课题目前在联合频谱感知领域没有被很好地解决。本文提出使用基于狄利克雷过程(Dirichlet process, DP)的分层贝叶斯模型来实现感知数据的自动分组,将具有同一稀疏频谱的簇自动分到同一组。在每个组内,可以进行信息共享并实施分层贝叶斯推理得出共享的超参数。因此,基于DP的分层贝叶斯模型充分利用了空间分集信息对感知数据进行非参数分组。然而,顺序采集的压缩感知数据在时间上也是相关的。为了进一步提高频谱判决的精确度,本文采用隐藏马尔科夫模型来描述隐藏的信道状态与顺序采集的压缩感知数据之间的关系,并使用维特比算法来获得最终的超参数以及更高精确度的频谱判决。仿真实验表明空时数据挖掘算法在恢复信号的归一化均方误差、正确检测概率、虚警概率等方面的性能均优于目前的几种频谱感知算法。其次,为了通过频谱切换来保证多媒体应用的服务质量(Quality of Service, QoS),必须对无线电环境进行周期性检测。然而,频谱感知必将引入额外的时延。因此,本文将频谱切换与频谱感知进行联合跨层设计,具体为多媒体传输的自适应频谱感知周期和数据包分配方案。频谱感知周期在本文中定义为相邻频谱感知的时间间隔,并假设次要用户(Secondary user, SU)的频谱感知是周期性的。本文充分考虑主要用户(Primary user, PU)的数据包到达特性,并建立了频谱感知周期与剩余数据包数、频谱感知周期与剩余时间之间的数学模型。如果SU传输多媒体数据包过程中,PU重新占用传输信道,那么SU需要切换信道,并利用新选择的信道在剩余时间内继续传输剩余数据包。因此,较小的剩余数据包数和较大的剩余时间将确保更多的多媒体数据包在要求的时延内到达接收端,从而提高了多媒体应用的QoS性能。基于上述的两个数学模型,为单信道链路推导出了最优的频谱感知周期,为多信道链路给出了信道选择、最优的频谱感知周期、最优的数据包分配策略之间的权衡关系。在多信道链路的情况下,最优的频谱感知周期和最优的数据包分配策略(固定链路选择的信道数)由Hughes-Hartogs和离散粒子群算法(discrete particle swarm optimization, DPSO)的联合算法给出。图像和视频传输仿真实验验证了本文提出的算法给出的信道选择、最优化频谱感知周期、最优的数据包分配策略的正确性和有效性。再次,为了在簇内同其它SU在上行链路中公平地分享可利用频谱资源,本文针对基于非连续正交频分复用技术的认知无线电网络提出了一种分布式、协作式以及填充式的频谱共享机制,并为SU的每个数据包选择最佳的子载波、发射功率以及调制方式。传统的动态分配策略通常假设所有SU通过公共控制信道(Common control channel, CCC)来进行大量重要信息交换,实现协作的频谱共享。为了避免大量信息进行交换,本章提出将机器学习(基于狄利克雷分布的全贝叶斯模型)应用于频谱共享机制(即智能跨层分配策略)。为了研究簇内的目标SU,本文将簇内其它SU等效为一个虚拟SU并使用基于狄利克雷分布的全贝叶斯模型来学习该虚拟SU的子载波分配策略等信息。并且,将学习结果用于估计任一信道的队列时延小于最大忍耐时延的概率。为了计算SU的吞吐量,本文引入了时间窗口来精确定义。时间窗口准确地描述了发射功率一定时,可以同时并行传输的数据包个数。最后,通过最大化基于时延和吞吐量的效用函数,得出智能跨层分配策略,并可以为SU的每一个数据包产生最优的子载波选择、功率以及调制方式分配方案。数据仿真和视频传输的仿真实验验证了智能跨层分配策略的正确性和有效性,且仿真结果与理论值相互吻合以及视频质量优于其它两种典型的动态分配策略。最后,本文研究了频谱管理和路由的跨层设计,并提出了面向链路稳定和信道容量的路由协议。本文针对高速运动的CRN,其快速运动的SU将不断改变网络拓扑结构并使得通信链路不能长时间保持稳定,从而有可能降低端到端的QoS性能以及增加路由协议的复杂度。本文针对联合搜救模型(如进行搜救的无人机),具体研究内容包括:(1)提出使用鸟群觅食模型来描述参与联合搜救的SU的运动特征,并估计出链路稳定概率;(2)根据各个SU的可用信道以及链路稳定概率,提出一种综合考虑节点度、簇内成员到簇头的平均跳数和信道切换次数的分簇算法;(3)根据分簇算法形成的基于簇结构的CRN,提出两种CCC选择方案:节点收缩和DPSO算法。簇头间的CCC以及网关节点的选取,综合考虑了任意两个簇头间的平均传输时延以及CCC的吞吐量;(4)针对快速运动的SU和不断改变的网络结构,提出一种基于链路稳定及信道容量的联合路由选择和信道分配的路由协议。仿真实验结果表明:(1)本文提出的CCC算法具有较高的吞吐量以及较小的传输时延;(2)与其它典型的路由协议相比,本文提出的路由协议具有较好的平均端到端时延和投包率性能。更多还原

【Abstract】 The dramatic increase in wireless access services and low spectrum utilization existing in licensed channels necessitate a new communication paradigm (i.e., cognitive radio network, CRN), whose key techniques are cognitive radio and dynamic spectrum access. Through opportunistic spectrum access, CRN can improve the spectrum utilization. However, CRN do not have a fixed channel and modulation type, and its transmission parameters are determined by spectrum sensing, spectrum mobility, spectrum sharing and spectrum management processes. To realize multimedia transmissions over CRN, this paper will consider the above four processes in our research.First, a cooperative spectrum sensing algorithm is proposed, which integrals compressive sensing (CS) and the spatial-temporal data mining method. The spectrum sensing which tries to aware the radio enviroment is the basic function to realize services over CRN, which is the main difference from traditional wireless communications. Since the CRN is deployed in a certain area, different clusters may have different sparseness spectrum states and is not appropriate for information sharing in data mining, which has not been addressed very well in the spectrum sensing field. Hence, in our cooperative spectrum sensing, the Dirichlet process (DP) prior is employed to make an automatically grouping among different clusters. In each group, the Bayesian inference is used for information sharing and one common sparseness hyper-parameters is discovered in each group. Hence, the DP prior is very suitable to our heterogenous CRN and collects the spatial information of the CS data. Moreover, the sequential CS data are not independent to each other. To exploit the time-domain relevance among sequential CS observations, the hidden markov model is employed to describe the relationship between hidden subcarrier state and sequential CS data, and the Viterbi algorithm is used to find out the final hyper-parameters and make a high resolution spectrum decision for each secondary user (SU). Simulation results show that our proposed algorithm successfully exploits the spatial-temporal relationship to obtain higher spectrum sensing performance in terms of normalized mean square error, probability of correct detection, and probability of false alarming compared with some recent research works.Second, the spectrum mobility is used to guarantee the quality of service of multimedia application, and the spectrum sensing is necessary to detect the radio enviroment. However, spectrum sensing will cause extra delay. Hence, the optimal spectrum sensing frequency and packet loading schemes are discussed in this paper, which integrals spectrum sensing in spectrum mobility management. Here the sensing frequency means how frequently a CR user detects the radio enviroment, and the spectrum sensing is assumed to operated periodly. This paper well considers the packet arrive rate of PU and derives the math model between the sensing frequency and the number of remaining packets that need to be sent, as well as the relationship between sensing frequency and the new channel availability time during which the CRN user is allowed to use a new channel (after the current channel is re-occupied by primary users) to continue packets transmission. Hence, a smaller number of remaining packets and a larger value of new channel availability time will help to transmit more multimedia packets within a required delay and thus a higher QoS. Then by using the above relationship models, we select appropriate sensing frequency under single channel case, and study the trade-offs among the number of selected channels, optimal sensing frequency and packet-loading scheme under multi-channel case. The optimal sensing frequency and packet-loading solutions for multi-channel case are obtained by using the combination of Hughes-Hartogs and discrete particle swarm optimization (DPSO) algorithms. Our experiments of image and video packets transmission demonstrate the validity of our sensing frequency selection and packet loading schemes.Thrid, to fairly share the available spectrum resource of the uplink in each cluster, a distributed, cooperative and overlay spectrum sensing mechanism called cognitive cross-layer scheduling scheme is proposed for non-contiguous orthogonal frequency-division multiplexing based CRN, which can generate the optimal subcarrier selection, power and modulation allocation for each multimedia packet from SU. Conventional dynamic scheduling schemes assume significant information exchange among all SUs through a common control channel (CCC), to realize cooperative spectrum sharing. To avoid such a heavy traffic information exchange, a cognitive method to learn the traffic profile is proposed in our spectrum sharing mechanism. From the viewpoint of the target SU, the other SUs in the cluster can be grouped as a virtual SU, and the target SU uses the Dirichlet-prior based fully Bayesian model to update the statistical distribution of subcarrier selection strategy profile of the virtual SU. Moreover, such a statistical distribution is used to estimate the probability of queue waiting time less than a threshold. To learn the throughput performance of SU, the time window is introduced to accurately define the throughput of SUs. The time window can clearly demonstrate how many packets can be transmitted simultaneously over multiple subcarriers, compared with the required packets transmission rate. Finally, maximizing the delay and throughput-based utility function, the cognitive cross-layer scheduling scheme generates the optimal subcarrier selection, power and modulation allocation for each multimedia packet. The data and real video transmission are simulated to validate the correctness of our cognitive cross-layer scheduling schemes. The simulation results match with theoretical analysis very well, and the reconstructed video quality using our proposed scheduling scheme is superior to the other two recently proposed schemes.Last, the cross-layer design of spectrum management and routing design is considered, and a stability-capability oriented routing protocol is proposed. In high-mobility CRNs, the fast topology changes decrease the stability of transmission link and the end-to-end QoS performance and thus the complexity of routing scheme. In this paper, the cooperative searching (i.e., unmanned aerial vehicle, UAV) scenario is considered, and (1) a realistic mobility model is proposed to describe the movement of highly mobile airborne nodes (i.e., UAVs), and estimate the link stability performance based on node movement patterns; (2) a CRN topology management scheme based on a clustering model is also proposed which considers radio link availability, and the cluster-heads (CHs) are selected based on the node degree level, average number of hops and channel switching from member nodes to the CH; (3) two new CCC selection schemes are proposed, which are based on the node contraction concept and the DPSO algorithm. The inter-cluster control channels and gateways are selected from the CHs, considering the average delay of control information transmission between two CHs as well as the total throughput of control channels; (4) a novel routing scheme is proposed that tightly integrates with channel assignment scheme based on the node capacity for the high mobility scenario. Our simulation results show that our proposed CCC selection scheme has high throughput and small transmission time. Compared to other popular CRN routing approaches, our proposed routing scheme achieves lower average end-to-end delay and higher packet delivery ratio.更多还原