Periodic Reporting for period 2 - FOGHORN (FOG-aided wireless networks for communication, cacHing and cOmputing: theoRetical and algorithmic fouNdations)
Reporting period: 2018-12-01 to 2020-05-31
The subject matter of the project has only grown in importance from the time of submission of this proposal. The fog architecture described in the project has been included by 3GPP in the 5G standard network architecture, and the integration of computing and communication has become the subject of a novel line of research in the information-theoretic literature.
The FOGHORN project is organized into three main WPs.
WP1: Fog-aided wireless networks for communication. A key feature of a fog-aided network is that control and data planes can be split across edge and cloud. Accounting for the complementarity and synergy of cloud and edge processing, the main question addressed by WP1 is: As a function of key system parameters such as traffic type, number of antennas, ranging from regular to massive antenna arrays, density of deployment, fronthaul architecture, including single- and multi-hop, what is the optimal functional split at control and data planes? What are corresponding optimal algorithmic solutions and performance trade-offs? A key aspect of this WP is the modelling of heterogeneous traffic requirements from the three generic services of 5G, namely enhanced Mobile BroadBand (eMBB), massive Machine-Type Communication (mMTC).
WP2: Fog-aided wireless networks with edge caching. The second work package targets the optimal operation of a fog-aided wireless network with edge caching capabilities. A key tenet of the proposed approach is that edge caching, fronthaul, and wireless transmission should be jointly designed so as to leverage the synergistic and complementary features of cloud and edge processing. The WP is based on a novel network information-theoretic asymptotic framework that enables a quantitative, tractable and insightful analysis of latency in a fog-aided wireless network accounting for fronthaul and wireless transmissions in the presence of caching.
WP3: Fog-aided wireless networks for computing. Edge and mobile cloud computing enables the offloading energy-consuming tasks from resource-constrained mobile devices, which are typically mobile broadband users or machine-type devices, to nearby servers at the edge or at the cloud. The goal of this project is the optimization of network resources for the execution of computational tasks at edge and cloud.
WP1: Fog-aided wireless networks for communication
The main contributors to WP1, beside the PI, have been Dr Jingjing Zhang (Post-Doc) and Rahif Kassib (PhD student), both working at KCL and supported through this project. There have also been a number of external collaborators, namely Dr Jinkyu Kang and Prof Joonhyuk Kang (KAIST); Dr Jihong Park (Oulu); Prof Petar Popovski and Dr Kasper Trillingsgaard (Aalborg Unviersity); Malihe Aliasgari and Prof Joerg Kliewer (NJIT); Jaein Kim and Prof Inkyu Lee (Korea University); Prof S.-H. Park (Chonbuk University); Prof Shlomo Shamai and Prof Yonina Eldar (Technion); Ghizi Mountaser and Prof Toktam Mahmoodi (KCL); Morteza Varasteh, Boozoo Rassouli, and Prof Deniz Gunduz (Imperial College); Seongah Jeong (Samsung). The main achievements are briefly described as follows.
• Reference  tackles a problem of central importance for Task 1.1 (Data link control) by focusing on uplink communications and investigating the cloud-edge allocation of two important network functions: the control functionality of rate selection and the data-plane function of decoding. Three functional splits are considered: 1) distributed radio access network, in which both functions are implemented in a decentralized way at the RRSs; 2) cloud RAN, in which instead both functions are carried out centrally at the RCC; and 3) a new functional split, referred to as Fog-RAN (F-RAN), with separate decentralized edge control and centralized cloud data processing. Using the adaptive sum-rate as the performance criterion, this work is the first to prove that the considered F-RAN functional split can provide significant gains in the presence of user mobility.
• When transmission is strongly constrained by energy and delay, as in IoT applications, analog transmission to edge processors in fog architectures may be preferable over conventional digital transmission. In the context of Task 1.1 this is investigated in references  from an information-theoretic viewpoint. These works study the zero-delay transmission of a Gaussian source over an additive white Gaussian noise with a one-bit analog-to-digital converter (ADC) front end and a possibly correlated side information at the receiver. The design of the optimal encoder and decoder is derived analytically, providing useful design insights into optimal
• The problem of scheduling and radio resource allocation in F-RAN models, which is the subject of Task 1.3 is addressed in references . In particular, in reference  a Virtual Reality (VR) mobile social network is considered, whereby the states of all interacting users should be updated synchronously and with low latency via two-way communications with edge computing servers. The resulting end-to-end latency depends on the relationship between the virtual and physical locations of the wireless VR users and of the edge servers. In this work, the problem of analyzing and optimizing the end-to-end latency is investigated for a simple network topology, yielding important insights into the interplay between physical and virtual geometries.
• In references  and  fronthaul and wireless transmission resources are optimized for an F-RAN architecture assuming massive MIMO. Specifically in  hybrid downlink beamforming techniques are optimized in the presence of centralized cloud processing, while  considers uplink channel estimation and tackles the problem of jointly optimizing the pilot sequences and the pre-RF chains analog combiners.
• Multi-tenant for cellular architectures are instead considered in , where the spectrum available for downlink transmission is partitioned into private and shared subbands, and the participating operators cooperate to serve the user equipments on the shared subband. In order to enable interoperator cooperation, the cloud processors of the participating operators are also connected by finite-capacity backhaul links. Inter-operator cooperation may hence result in loss of privacy. A novel approach based on the idea of correlating quantization noise signals across RUs of different operators is proposed to control the trade-off between distortion at UEs and inter-operator privacy.
• Centralized cloud is expected to mostly take place via network function virtualization on commercial off-the-shelf servers. In order to mitigate the impact of straggling decoders in this platform, a novel coding strategy is proposed in references , whereby the cloud re-encodes the received frames via a linear code before distributing them to the decoding processors. This work is the first to demonstrate that coding is instrumental in obtaining a desirable compromise between decoding latency and reliability.
• Beside communication, a key task to be carried out by edge nodes and wireless devices is edge positioning. Within Task 1.3 reference  studies for the first time the use of Full-Dimensional multiple-input multiple-output (FD-MIMO) base stations (BSs) to provide simultaneously high-rate data communication and mobile 3D positioning in the downlink. Analysis and comprehensive numerical results demonstrate that the proposed schemes can effectively cater to both data communication and positioning services, providing only minor performance degradations as compared to the more conventional cases in which either function is implemented.
• The grand objective of 5G wireless technology is to support three generic services with vastly heterogeneous requirements: enhanced mobile broadband (eMBB), massive machine-type communications (mMTC), and ultra-reliable low-latency communications (URLLC). The problem of engineering the coexistence of different traffic types in F-RAN architectures is studied in references . Reference  is the first to study the impact of coding of MAC frames for transmission on the fronthaul links from edge nodes to the cloud. Significant reliability-latency trade-off gains are demonstrated.
• The coexistence of all three generic traffic services is studied in  for a single-cell scenario. This work studies for the first time the potential advantages of allowing for non-orthogonal sharing of RAN resources in uplink communications from a set of eMBB, mMTC and URLLC devices to a common base station. The approach is referred to as Heterogeneous Non-Orthogonal Multiple Access (H-NOMA), in contrast to the conventional NOMA techniques that involve users with homogeneous requirements and hence can be investigated through a standard multiple access channel. The concept of reliability diversity is introduced as a design principle that leverages the different reliability requirements across the services in order to ensure performance guarantees with non-orthogonal RAN slicing. This study reveals that H-NOMA can lead, in some regimes, to significant gains in terms of performance trade-offs among the three generic services as compared to orthogonal slicing.
• The coexistence URLLC and eMBB traffic is studied in  in multi-cell systems with inter-cell interference. A novel functional split is proposed, whereby URLLC traffic is served by the Edge Nodes (ENs) of a cellular network, while eMBB communications are handled centrally at a cloud processor as in a Cloud-RAN system. This solution guarantees the low-latency requirements of the URLLC service by means of edge processing, e.g. for vehicle-to-cellular use cases, as well as the high spectral efficiency for eMBB traffic via centralized baseband processing. In addition to the conventional orthogonal multiplexing of the two services, these works evaluate the performance of H-NOMA. Both uplink and downlink are analyzed by considering practical aspects such as fading, lack of channel state information for URLLC transmitters, rate adaptation for eMBB transmitters, analog and digital fronthaul, and different coexistence strategies, such as puncturing.
WP2: Fog-aided wireless networks with edge caching
The main contributors to WP2, beside the PI, have been Dr Jingjing Zhang (Post-Doc) and Rahif Kassab (PhD student), both working at KCL and supported through this project. There have also been a number of external contributors, namely: Tim Chang Dr Avik Sengupta and Prof Ravi Tandon (University of Arizona); Dr Mohammadreza Azimi (NJIT); Prof Igor Stanojev (University of Wisconsin-Platteville); Prof Giuseppe Durisi, R. Devassy, Guido Ferrante (Chalmers University); Prof Elif Uysal-Biyikoglu (METU); Jasper Goseling (University of Twente); Andrea Matera and Prof Umberto Spagnolini (Politecnico di Milano); Seok-Hwan Park (Chonbuk University); Roy Karasik, Michael Zeide, and Prof Shlomo Shamai (Technion); Petar Popovski (Aalborg University). The main accomplishments in this WP are briefly described as follows.
• Task 1.1 focuses on the asymptotic information-theoretic analysis of edge caching in F-RAN systems. The key reference  considers a fog-aided wireless network architecture The interplay between cloud processing and edge caching is addressed for the first time from an information-theoretic viewpoint by investigating the fundamental limits of a high Signal-to-Noise-Ratio (SNR) metric, termed normalized delivery time (NDT), which captures the worst-case coding latency for delivering any requested content to the users. The NDT is defined under the assumptions of either serial or pipelined fronthaul-edge transmission, and is studied as a function of fronthaul and cache capacity constraints. Placement and delivery strategies across both fronthaul and wireless, or edge, segments are proposed with the aim of minimizing the NDT. Information-theoretic lower bounds on the NDT are also derived. Achievability arguments and lower bounds are leveraged to characterize the minimal NDT in a number of important special cases, including systems with no caching capabilities, as well as to prove that the proposed schemes achieve optimality within a constant multiplicative factor of 2 for all values of the problem parameters
• The analysis in the key reference  has been extended to account for a number of practical aspects including heterogeneous content popularity , partial wireless connectivity , Device-to-Device (D2D) transmissions , linear precoding , imperfect CSI , and secrecy constraints . All these works present information-theoretic analyses that provide engineering insights into the role of each of the mentioned aspects. For instance, in , assuming out-of-band D2D communication, an optimal caching-D2D cooperation strategy is proposed based on a novel scheme for an X-channel with receiver cooperation that leverages tools from real interference alignment. Insights are provided on the regimes in which D2D communication is beneficial.
• While the works above focus on offline caching, references - study the scenario in which the set of popular content is time-varying, hence necessitating the online replenishment of the ENs' caches along with the delivery of the requested files. The analysis is centered on the characterization of the long-term Normalized Delivery Time (NDT), which captures the temporal dependence of the coding latencies accrued across multiple time slots in the high signal-to-noise ratio regime. Analytical results demonstrate that, in the presence of a time-varying content popularity, the rate of fronthaul links sets a fundamental limit to the long-term NDT of F- RAN system.
• Task 2.2 concerns the non-asymptotic regime in terms of SNR and packet duration. This is studied in reference , where edge caching is combined with cloud-aided transmission in order to compensate for the limited hit probability of the caches at the base stations (BSs). This paper investigates the benefits of fractional caching in a scenario with a cloud processor connected via a wireless fronthaul link to a BS, which serves a number of mobile users on a wireless downlink channel using orthogonal spectral resources. Numerical results demonstrate meaningful advantages for fractional caching due to the interplay between caching and HARQ transmission. The role of short-packet codes on the average delay and on the peak age of information is instead studied in reference .
• Practical online caching strategies are studied within Task 2.2 in . Standard Time-to-Live (TTL) cache management prescribes the storage of entire files, or possibly fractions thereof, for a given amount of time after a request. As a generalization of this approach, this work proposes the storage of a time-varying, diminishing, fraction of a requested file. Accordingly, the cache progressively evicts parts of the file over an interval of time following a request. The strategy, which is referred to as soft-TTL, is justified by the fact that traffic traces are often characterized by arrival processes that display a decreasing, but non-negligible, probability of observing a request as the time elapsed since the last request increases.
WP3. Fog-aided wireless networks for computing
The main contributors to WP3, beside the PI, have been Dr Hyeryung Jang (Post-Doc) and Arsham Mostaani (PhD student), both working at KCL and supported through this project. There have also been a number of external contributors, namely: Dr Seongah Jeong (Samsung); Joonhyuk Kang (KAIST); Alireza Bagheri and Prof Bipin Rajendran (NJIT). While, according to the original plan, WP3 should not have started until after the end of the first reporting period, the emergence of two technologies that were not considered during the writing of the proposal have motivated the consideration of research topics within WP3 at an earlier stage. The two technologies at hand are Unmanned Aerial Vehicle (UAV) communications and neuromorphic computing. The main accomplishments in this WP are briefly described as follows.
• UAVs have been recently considered as means to provide enhanced coverage or relaying services to mobile users (MUs) in wireless systems with limited or no infrastructure. In paper , a UAV-based mobile cloud computing system is studied in which a moving UAV is endowed with computing capabilities to offer computation offloading opportunities to MUs with limited local processing capabilities. The problem of jointly optimizing the bit allocation for uplink and downlink communication as well as for computing at the UAV, along with the cloudlet's trajectory under latency and UAV's energy budget constraints is formulated and addressed by leveraging successive convex approximation (SCA) strategies.
• Neural Networks have become the de-facto standard tool to carry out supervised, unsupervised, and reinforcement learning. Their emergence has built upon the unprecedented availability of computing power in data centers and cloud computing platforms. This reliance on a massive amount of computation and energy has, however, prevented the deployment of powerful Artificial Intelligence (AI) algorithms based on neural networks on low-power mobile or embedded devices. In contrast to artificial neural networks, the human brain is capable of performing more general and complex tasks at a minute fraction of the power required by state-of-the-art supercomputers. Insights from neuroscience reveal that the main reasons for the energy efficiency of the brain may lie in the space-time computing capabilities of spiking neurons that leverages sparse time-based information encoding, event-triggered plasticity, and low power inter-neuron signalling. Based on this observation, neuromorphic signal processing and learning machines have recently emerged as a low-power alternative to energy-hungry artificial neural netw