无线推送与缓存通过将热门内容提前推送至网络边缘或用户终端，具有分流忙时数据流量、降低用户下载时延、提升网络吞吐量等功能，是未来无线通信网络的关键性技术之一。然而，由于用户请求的不确定性，尚未充分明确无线推送与缓存是否带来能效增益。当推送内容被请求时，无线推送与缓存通过增加数据传输时间、挖掘高信道增益机会来提升能效；反之，则会带来推送与缓存资源的浪费。面对该问题，本文考虑了不同类型的用户请求预测信息，研究了基于跨层跨域协同的高能效信息推送与缓存调度机制，衡量了信息推送带来的能效增益。 在确定性用户请求时间条件下，本文首先研究了能效最优的联合推送与缓存机制，刻画了信息推送的能效上界。对于正交多址接入下的单用户信息推送，以及非正交多址接入下的多用户组播推送场景，本文基于网络微积分的方法将能耗最小化问题建模为变分问题，并考虑了用户发出请求后的不同读取过程以及缓存器大小约束。通过对变分问题结构的分析，本文将其等效转化为凸优化问题并进行有效求解。基于对最优策略性能的分析与比较，研究显示组播推送通过挖掘组播增益，可以显著提高网络能效，促使本文更深入的探究组播推送技术。 考虑随机用户请求，本文探究了组播推送网络中的高能效联合推送与推荐机制。通过内容推荐引导和聚合用户兴趣，并基于物理层组播波束成形灵活选择被推送的用户，本文采用跨域和跨层协同以提升网络推送能效。基于随机规划理论，能效最大化问题被建模为多阶段随机规划问题。为了使问题可解，本文通过将变量参数分离优化，并基于分支定界等方法提出了次优联合推送与推荐算法。为衡量分离优化带来的性能损失，本文基于半正定松弛提出了能效上界的生成算法。 将用户请求的时间维度引入缓存网络，本文基于随机用户请求延时信息探究了高能效的信息推送方法。为了提高推送能效，本文引入机会传输机制，同时考虑用户服务质量需求约束。在给定机会推送策略时，推导得到了平均时延、平均能耗和时延中断概率，并根据推导结果建立最大化能效的非凸优化问题。为了降低推送策略设计的复杂度，本文基于梯度下降和罚函数方法提出了次优算法，并分析得到了机会推送系统的放缩性质，探究了基于用户请求概率和传输时延的推送门限。此外，在考虑用户缓存代价时，本文基于动态规划和强化学习方法，在用户请求延时概率分布已知或未知的情况下，提出了最大化基站收益的联合推送与缓存策略。
Joint Pushing and caching holds the promise of offloading the peak-hour traffic, reducing the downloading delay, and improving the network throughput by proactively caching popular content items at the network edge or users, thereby being considered as one of the key technologies in future communication networks. However, due to the uncertainty of users' requests, it is not clear whether joint pushing and caching brings energy efficiency (EE) gains compared to the traditional on-demand transmissions. If the pushed content items are requested, joint pushing and caching is capable of improving EE by increasing the transmission duration and exploiting opportunistic transmissions. Otherwise, the energy and storage resources consumed by content pushing are wasted. To address this issue, this paper investigates the impact of different types of user request information, presents corresponding energy-efficient pushing and caching policies based on cross-layer and cross-domain collaborations, and evaluates EE gains brought by content pushing. Considering deterministic users' requests, this paper studies the optimal joint pushing and caching policy that characterizes the upper bound on EE of content pushing. In the scenarios with orthogonal multiple access or nonorthogonal multiple access, we formulate the energy minimization problem as the variational problem based on the network calculus approach, while considering the receiver buffer size and content consumption constraints. Based on the analysis of problem structure, we equivalently convert the formulated problem into a convex optimization problem that can be solved efficiently. According to the comparison of presented policies, the multicast-based content pushing can substantially improve the EE by exploiting multicasting opportunities, thereby motivating the further study of multicast networks in this paper. When only the user request probability is available, this paper studies the energy-efficient joint pushing and recommendation policy in multicast networks. By guiding users' interests via content recommendation and flexibly clustering uses via multicast beamforming, we improve the EE based on cross-layer and cross-domain collaborations. The EE maximization problem is formulated as a multi-stage stochastic programming problem. Since the problem is intractable, suboptimal joint pushing and recommendation policies are presented based on branch-and-bound methods. To evaluate the performance of presented policies, we provide an algorithm to generate an upper bound on EE by using the semi-definite relaxation. When taking the random request delay into account, this paper investigates the energy-efficient content pushing methods. To improve EE while incorporating Quality-of-Service requirements, the opportunistic transmission is introduced and the delay constraints are considered. For a given opportunistic pushing policy, we derive the average delay, average energy consumption, and delay-outage probability, based on which the EE maximization problems are formulated as non-convex problems. In order to reduce the computation overhead of pushing design, we present suboptimal algorithms based on the gradient descent and penalty function methods. Moreover, we derive the scaling property and obtain the request probability and delivery delay thresholds for content pushing. In addition, when taking storage costs into consideration, this paper presents joint pushing and caching policies to improve the profile of base stations. Dynamic programming and reinforcement learning are respectively used in the cases with or without statistical request delay information.
