Durgesh Singh

Millimeter wave (mmWave) device to device (D2D) communication is highly susceptible to obstacles due to severe penetration losses. Dynamic obstacles may cause unpredictable fluctuations to D2D channel quality and hence a D2D relay initially chosen by base station (BS) might undergo failed transmissions resulting in severe packet loss and delay. This local information regarding link quality deterioration needs to be informed to the BS by the user equipments (UEs) which may result in some delay. Also, exploring a new relay on mmWave channel results in significant delay due to directional search. Hence the following optimal sequential decision must be made when packet loss occurs: whether to explore for a new relay link considering exploration cost, or to continue communication via the existing relay. We model this sequential decision problem locally at each UE as partially observable Markov decision process to capture uncertainty in D2D links while minimizing delay. We derive an optimal threshold policy for both positively and negatively correlated links. We also provide a simplified and easy to implement policy based on successive acknowledgment failures for positively correlated links. Through simulation, we validate theoretical findings and demonstrate that our approach outperforms existing state of the art algorithms.

Blockage due to obstacles in millimeter wave (mmWave) device to device (D2D) communication is a prominent problem due to their severe penetration losses. Potential user equipments (UEs) in vicinity of the source UE must be explored in order to select a new relay when the current link gets blocked. However, dynamic obstacles are not known in advance and thus may cause unpredictable fluctuations to D2D channel quality causing newly selected relay link also to be susceptible to blockage. This might cause frequent relay switching leading to call drops and high energy consumption. We have proposed the idea of reducing frequency in relay exploration and switching and thus average end-to-end delay (in seconds) at the expense of additional exploration time units (few milliseconds) during beam alignment. We seek to learn the uncertainty in D2D link qualities by modeling the problem as finite horizon partially observable Markov decision process (POMDP) framework locally at each UE. We have derived an optimal threshold policy which maps the state to set of actions. We then give a simplified and easy to implement stationary threshold policy which counts the number of successive acknowledgment successes/failures for making decisions of selecting or not selecting a given relay locally. Through extensive simulation, we validate our theoretical findings and demonstrate that our approach captures the trade-off between average exploration time and average end-to-end (E2E) delay in presence of dynamic obstacles.

Millimeter wave (mmWave) device to device (D2D) communication is highly susceptible to obstacles due to severe penetration losses and requires almost a line of sight (LOS) communication path. D2D channel condition is local to devices/user equipments (UEs) and hence is \textit{not} directly visible to the base station (BS). Thus quality of the D2D channel needs to be propagated to BS by UEs which may incur some delay. Hence the solution provided by BS to UEs using this gathered channel information might become less useful to establish communication due to moving obstacles. These types of obstacles might not be known in advance and hence may cause unpredictable fluctuations to the D2D channel quality. Hence we seek to learn the D2D channels using the finite horizon partially observable Markov decision process (POMDP) framework to model the uncertainty in such kind of network environments with dynamic obstacles. The objective is to minimize delay when channel quality deteriorates, by making UEs choose locally the best possible decision between i) to continue on the current relay link on which communication is taking place or ii) to switch to another good relay by exploring other possible UEs in its locality. We derive an optimal threshold policy which tells the UE to take appropriate decision locally. Later, we give a simplified and easy to implement stationary threshold policy which counts the number of successive acknowledgement failures, based on which UE make appropriate decision locally. Through extensive simulation, we demonstrate that our approach outperforms recent algorithms.

Challenges in relay selection for D2D communication arise due to the mobility of nodes which brings uncertainty in various network parameters. We first developed a network assisted stochastic integer programming (SIP) model to incorporate uncertainty which predicts the network parameters for upcoming time instance based on information available at current time instance. We converted SIP model to an equivalent deterministic mixed integer non-linear program (MINLP) model and proved its hardness result. By exploiting the constraints of MINLP, we developed a distributed greedy metric termed as connectivity factor (CF) which is calculated locally at each node on per-hop basis. It captures the nodes mobility and hence takes care of link reliability which in turn controls packet loss and delay. It can be computed in O(n) time, where n is the number of transmitters interfering with the given link. Our approach is applicable to any mobility model with relevant distributions of mobility parameters known. We constructed perceived graph based on CF values to devise network assisted and device controlled relay selection algorithms for given source-destination pairs. Extensive simulation results show significant improvements in packet loss and average end to end delay by our approach over a recent implementation of AODV based algorithm.