Efficient Relay Selection Techniques for D2D Communication under user Mobility and Presence of Obstacles

Abstract

Formalization of device to device (D2D) communication and millimeter wave (mmWave) technology into 5G and beyond cellular networks has given rise to novel challenges of multitude dimensions. Relay selection problem (RSP) is one such fundamental challenge in D2D communication where a user equipment (UE) acts as a relay to divert the communication path between two communicating UEs when they are not in vicinity of each other or when outages occur due to blockages. In this thesis, we developed several relay selection algorithms considering the mobility of the UEs as well as the presence of obstacles. We first developed a network-assisted stochastic integer programming (SIP) model to incorporate uncertainty in network parameters due to UE’s mobility. By utilizing the SIP model, we developed a greedy metric which is computed locally at each node on per-hop basis. This metric predicts the network parameters for upcoming global time instants based on information available at the current global time instant. We have developed relay selection algorithms for both network-assisted and device-controlled scenarios of D2D communication using the developed greedy metric. Here, we have considered the mobility of UEs, but presence of obstacles is not considered. Next, we considered the RSP in the presence of obstacles. Since mmWave suffers from severe penetration losses, a given relay link might get disconnected easily, especially by dynamic obstacles which may change their positions abruptly. We developed a networkassisted probabilistic model which captures the mobility related parameters of UEs and dynamic obstacles by sensing via radars. A detailed analysis for capturing the dynamic obstacles using geometry is presented and an algorithm to select best relay which maximizes average data rate is developed. Here, the orientation in motion of dynamic obstacles is assume to be known at the base station (BS). Orientation of the motion of dynamic obstacles is very difficult to measure accurately at the BS as it may vary rapidly compared to that of the speed and also the obstacles are not usually connected to the BS. We developed a network-assisted model to consider the scenarios where the orientation information is unknown. Using simple and innovative geometric techniques, we derived expressions for probability of blockages and based on them developed a relay selection algorithm. The relay initially provided by the BS at global time instants may get blocked by unknown dynamic obstacles in near future during local time instants, thus leading to huge packet loss and delay. Dynamic obstacles may cause abrupt variations in channel quality and deferring to the BS for an appropriate solution would incur extra delay. Hence, a decision is made locally by source UE: i) to stop communication on the current relay and switch to a new relay by performing directional search in its vicinity, or ii) to continue on the current relay. For the former case, directional search comprises the exploration phase when a new relay is selected. Since the newly selected relay at the exploration phase itself is i vulnerable to blockages, it must be ensured that frequent relay switching is minimized while selecting a relay, as switching has a significant delay overhead. The UE has to locally decide during exploration phase: i) do not select the new relay link as it is likely to be obstructed and go for exploration on another relay link, ii) select the new relay link as it is likely to be free of obstacles and choose it for data transmission, or iii) send more probe packets as decision cannot be made at the current exploration phase and go for further exploration on the same relay link. Both decision problems are modeled using partially observable Markov decision process (POMDP) framework. The channel quality is learned via acknowledgments (ACKs) which are also vulnerable to blockages. Optimal threshold policy is derived for both problems. Later, we gave easy to use stationary policies, which is based on the number of successive ACK successes or ACK failures. Through extensive simulations, we validated our theoretical findings. Our approaches outperform various other classical and state of art approaches.