Research Of Packets Scheduling On Power And Delay Tradeoff In Cellular Networks

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.