Fundamental limits on latency in cloud- and cache-aided HetNets

Abstract

A Fog radio access network (F-RAN) is considered as a network architecture candidate to meet the soaring demand in terms of reliability, spectral efficiency, and latency in next generation wireless networks. This architecture combines the benefits associated with centralized cloud processing and wireless edge caching enabling primarily low-latency transmission under moderate fronthaul capacity requirements. The F-RAN we consider in this paper is composed of a centralized cloud server which is connected through fronthaul links to two edge nodes (ENs) serving two mobile users through a Z-shaped partially connected wireless network. We define an information-theoretic metric, the delivery time per bit (DTB), that captures the worst-case per-bit delivery latency for conveying any requested content to the users. For the cases when cloud and wireless transmission occur either sequentially or in parallel, we establish coinciding lower and upper bounds on the DTB as a function of cache size, backhaul capacity and wireless channel parameters. Through optimized rate allocation, our achievability scheme determines the best combination of private, common signalling and interference neutralization that matches the converse. Our converse bounds use subsets of wireless, fronthaul and caching resources of the F-RAN as side information that enable a single receiver to decode either one or both users' requested files. We show the optimality on the DTB for all channel regimes. In case of serial transmission, the functional DTB-behavior changes at fronthaul capacity thresholds. In this context, we combine multiple channel regimes to classes of channel regimes which share the same fronthaul capacity thresholds and as such the same DTB-functional. In total, our analysis identifies four classes; in only three of those edge caching and cloud processing can provide nontrivial synergestic and non-synergestic performance gains. Interestingly, in these three classes, we show that only under parallel fronthaul-edge transmission strategies edge caching becomes obsolete as long as a certain fronthaul capacity is exceeded.

Delivery Time Minimization in Cache-Assisted Broadcast-Relay Wireless Networks with Imperfect CSI

“…3 We shall occasionally avoid indexing the functional dependency for notational simplicity. 4 This assumption is in agreement with the widely used imperfect and perfect CSI setting at transmitting and receiving nodes, respectively. Thus, the UEs have perfect-quality CSI.…”

Fundamental Limits of Cloud and Cache-Aided Interference Management With Multi-Antenna Edge Nodes

“…Reference [7] characterizes the minimum NDT with a multiplicative factor of 2 by considering single-antenna ENs, intra-file coding for caching and general delivery strategies on fronthaul and edge channels. References [20], [23] studied a related scenario with coexisting macro-and small-cell base stations. The work [24] studied an F-RAN model with decentralized placement algorithms.…”

Cache-Assisted Broadcast-Relay Wireless Networks: A Delivery-Time Cache-Memory Tradeoff

“…From [26,Lemma 1], it readily follows that the NDT is a convex function in µ. This means that a cache-aided network shown in Fig.…”

“…Later on for the setting of arbitrary number of edge nodes and receivers a constant factor characterization of 2 has been established in [24]. The effect of channel strength and fading on the delivery time of partially connected F-RANs has been investigated in [25], [26], [27] on the basis of the binary fading model [28] and the linear deterministic model [29].Recently, the effect of Tx-Rx caching on the delivery time of interference networks was studied in two new lines of research. The first being, where distinct nodes, i.e., transmitters and receivers are equipped with caches.…”