【作者】 彭烈新；

【导师】 朱光喜；

【作者基本信息】 华中科技大学 ， 信息与通信工程， 2007， 博士

【摘要】 近年来,随着无线数据业务(移动Internet,无线多媒体等)需求的日益增长,一般认为全IP架构将成为下一代无线网络的主流架构,即各种业务将在数据包的层面上获得统一。数据包的调度是有线网络中一项较为成熟的技术,但将其用于无线网络却要有挑战得多——无线信道的时变性以及功率的受限性是其中两个最为主要的因素。本论文的主要目标是研究并设计能有效提高系统功率效率的包调度策略,具体而言,重点研究不同通信环境下(单用户与多用户;平坦衰落与频率选择性衰落;理想信道信息与非理想信道信息等)链路时延与功率的折衷关系,进而设计在保障服务质量(Quality of Services, QoS)(时延)的前提下使发射功率最小的次优或简化策略。论文首先讨论了单用户平坦衰落信道下的包调度策略以及功率时延关系。该调度问题可用马尔可夫决策过程(Markov Decision Process, MDP)建模,通过对MDP的分析我们发现:在最优策略下,功率和时延具有凸函数的关系,仿真结果也证实了这一结论。最优策略的求解可通过动态规划得到,但复杂度非常高,因此我们给出了一种简化策略,在该策略中发送速率是当前队列长度与信道状态的一个简单函数,算法复杂度得以大大降低,仿真结果表明其性能较最优策略仅有较少的下降。另外我们还利用Lyapunov稳定性定理证明了该简化策略可以保证队列的稳定性。其次,结合实际环境我们分别分析了在缓冲区为有限长并采用QAM调制情形下的调度策略以及部分信道信息对调度策略性能的影响。在平坦衰落信道中,利用时间分集效应是包调度策略用来降低功率的一个重要手段,而在频率选择性衰落信道中,调度器除了可以利用时间分集效应外,还可以利用频域上的分集以获取进一步的性能增益。我们分析并提出了OFDM系统下的最优和次优包调度策略,对于最优策略,我们给出了两种求优模型:一种是基于无限长时间的平均代价准则MDP模型,另一种是基于稳态概率的线性规划(Linear Programming, LP)模型。在业务到达过程以及队列状态都具有Markov性的条件下,可以证明这两种模型是等价的,因此它们具有相同的最优解,但这两种模型的复杂度都随状态数呈指数增长,随着子载波数量的增加,其求解是异常复杂的。基于此我们提出了两种简化策略:子时隙算法和MLQHR+OFDM算法。子时隙算法将时频二维空间的求优转化为时间一维空间的求优,其复杂度仅随子载波数线性增长。MLQHR+OFDM算法则是借鉴平坦衰落信道下的简化策略,将发送速率限定为子载波信道状态和队列状态的一个简单函数,使复杂度大大降低。仿真结果表明:相比最优策略而言,子时隙算法在功率效率方面只有2%的降低,而MLQHR+OFDM算法虽然功率损耗增加大约10%,但考虑到其低复杂度的优点,我们认为它还是具有一定实用的价值。论文最后讨论了多址信道和广播信道下的多用户调度策略。在这两种信道模型中,系统功率和发射速率都具有凸函数的关系,因而调度策略仍然可以利用时间分集效应降低功率,另外,还可以利用多用户分集进一步提高系统性能。我们具体分析了两用户的情况,证明了系统功率和时延具有凸函数的关系,仿真结果也支持这一结果。与单用户情形类似,我们也分别给出了这两种信道模型下的简化调度策略,仿真结果显示:与最优策略相比,简化算法功率损耗增加大约15%,但复杂度却大大降低。另外我们讨论了多址信道的功率区问题,证明了对于给定的时延,都对应着一个功率区,该功率区的控制面边界上的每个点都对应于一个功率凸组合问题的最优解,最后我们还给出了求功率区的具体算法。更多还原

【Abstract】 With the increasing demand for high-data-rate services, All-IP architecture has been widely regarded as the mainstream one in the next-generation wireless network, i.e., all services are based on packets. The scheduling of packets is a relatively mature technique in the wireline networks, while in the wireless networks it is much more challenging, mainly due to the severity and time-varying nature of the wireless channel as well as the limited power in the mobile terminals. In this dissertation we aim at studying and devising high power-efficiency packet scheduling strategies, specifically, we explore the optimal tradeoff between the queuing delay and power in various communication scenarios (single user and multi-user; flat fading and frequency-selective fading; perfect CSI vs. imperfect one, etc.), and further design the suboptimal strategies to minimize the transmission power under the constraint of QoS (delay) guarantee.We first address the power-delay tradeoff properties and scheduling algorithms in the single-user flat-fading-channel scenario. Under mild conditions, the scheduling problem can be formulated as a MDP (Markov Decision Process), through the analysis of which we can find that the average transmission power is a non-increasing, strictly convex function of the average delay. As the complexity of the optimal solution which can be yielded by the dynamic programming method is very high, we propose a very simple scheme in which the scheduled rate is a logarithm function of the current queue length and channel state. Simulation results show that the performance loss of our suboptimal algorithm is very low. Furthermore, we prove the capability of our algorithm to keep the queue length bounded by utilizing the Lyapunov stability theorem. In addition, we take some practical issues into consideration by analyzing the scheduling algorithm under the condition of finite buffer and QAM modulation mode as well as the impact of the partial CSI over the scheduling performance.In the flat fading channel, temporal diversity is an important means of decreasing transmission power, while in the frequency-selective channels, we can further exploit the spectral diversity in addition to the temporal one. We analyze the optimal scheduling strategy and propose suboptimal ones. In specific, we present two optimization models for the optimum strategy: one is based on the average-cost MDP based on infinite time horizon, the other one is formulated as a Linear Programming (LP) problem based on the“steady-state”probability. It can be proved that under the Markov-chain assumption of both the arrival process and the fading process the two seemingly different models are equivalent in essence. However, the complexity of the two models is both exponential with the states. To reduce the complexity, we design two suboptimal algorithms: the sub-slotting algorithm and the MLQHR+OFDM algorithm. For the former by transforming a two-dimensional optimization problem to one-dimensional one, we can reduce the complexity to just linear with the sub-carrier numbers; as for the latter, it is the extension of the simplified algorithm for the flat-fading scenario and is more useful in the frequency-selective channel due to its very low implementation complexity .Numerical results indicate that the sub-slotting algorithm only incurs as low as 2% performance loss compared with the optimum algorithm, while the performance degradation of ALQHR+OFDM algorithm is somewhat bigger.In the last we discuss the multi-user scheduling strategies in the Multiple-Access Channel and Broadcast Channel, respectively. In specific, we analyze the two-user scenario and prove the same convexity property of power versus delay as in the single-user case. What’s different is that in the multi-user environment an additional form of diversity can be utilized. Similarly as the single-user scenario, we also present the simple algorithms for the above two channels. Compared with the optimum scheduling algorithms, our proposed algorithms exhibit about 15 percent loss in the power efficiency, while their complexities are much lower than the optimum one. In addition we analyze the power region in the MAC channel. We prove that for any given delay, there’s a unique power region in whose dominant boundary every point corresponds to the optimal solution to a convex power combination optimization problem. In addition, we present the detailed algorithm for computing the power region.更多还原