In this article, the problem of training federated learning (FL) algorithms over a realistic wireless network is studied. In the considered model, wireless users execute an FL algorithm while training their local FL models using their own data and transmitting the trained local FL models to a base station (BS) that generates a global FL model and sends the model back to the users. Since all training parameters are transmitted over wireless links, the quality of training is affected by wireless factors such as packet errors and the availability of wireless resources. Meanwhile, due to the limited wireless bandwidth, the BS needs to select an appropriate subset of users to execute the FL algorithm so as to build a global FL model accurately. This joint learning, wireless resource allocation, and user selection problem is formulated as an optimization problem whose goal is to minimize an FL loss function that captures the performance of the FL algorithm. To seek the solution, a closed-form expression for the expected convergence rate of the FL algorithm is first derived to quantify the impact of wireless factors on FL. Then, based on the expected convergence rate of the FL algorithm, the optimal transmit power for each user is derived, under a given user selection and uplink resource block (RB) allocation scheme. Finally, the user selection and uplink RB allocation is optimized so as to minimize the FL loss function. Simulation results show that the proposed joint federated learning and communication framework can improve the identification accuracy by up to 1.4%, 3.5% and 4.1%, respectively, compared to: 1) An optimal user selection algorithm with random resource allocation, 2) a standard FL algorithm with random user selection and resource allocation, and 3) a wireless optimization algorithm that minimizes the sum packet error rates of all users while being agnostic to the FL parameters.

A Joint Learning and Communications Framework for Federated Learning Over Wireless Networks

In this paper, the problem of training federated learning (FL) algorithms over a realistic wireless network is studied. In particular, in the considered model, wireless users execute an FL algorithm while training their local FL models using their own data and transmitting the trained local FL models to a base station (BS) that will generate a global FL model and send it back to the users. Since all training parameters are transmitted over wireless links, the quality of the training will be affected by wireless factors such as packet errors and the availability of wireless resources. Meanwhile, due to the limited wireless bandwidth, the BS must select an appropriate subset of users to execute the FL algorithm so as to build a global FL model accurately. This joint learning, wireless resource allocation, and user selection problem is formulated as an optimization problem whose goal is to minimize an FL loss function that captures the performance of the FL algorithm. To address this problem, a closed-form expression for the expected convergence rate of the FL algorithm is first derived to quantify the impact of wireless factors on FL. Then, based on the expected convergence rate of the FL algorithm, the optimal transmit power for each user is derived, under a given user selection and uplink resource block (RB) allocation scheme. Finally, the user selection and uplink RB allocation is optimized so as to minimize the FL loss function. Simulation results show that the proposed joint federated learning and communication framework can reduce the FL loss function value by up to 10% and 16%, respectively, compared to: 1) An optimal user selection algorithm with random resource allocation and 2) a standard FL algorithm with random user selection and resource allocation.

A Joint Learning and Communications Framework for Federated Learning over Wireless Networks

Problems formulated in terms of logarithmic or exponential equations often use the Lambert W function in their solutions. Expansions, approximations and bounds on W have been derived in an effort to gain a better understanding of the relationship between equation parameters. In this paper, we focus on one of the branches of W, denoted as W-1, we derive tractable upper and lower bounds and we illustrate their usefulness in identifying conditions under which user cooperation can yield a lower outage probability than non-cooperative transmission.

Bounds on the Lambert Function and Their Application to the Outage Analysis of User Cooperation

Multimedia applications are characterized as delay-sensitive and high-bandwidth stipulating traffic sources. Supporting such demanding applications on cognitive radio sensor networks (CRSNs) with energy and spectrum constraints is a highly daunting task. In this paper, we propose a spectrum-aware cluster-based energy-efficient multimedia (SCEEM) routing protocol for CRSNs that jointly overcomes the formidable limitations of energy and spectrum. Clustering is exploited to support the quality of service (QoS) and energy-efficient routing by limiting the participating nodes in route establishment. In SCEEM routing, the number of clusters is optimally determined to minimize the distortion in multimedia quality that occurs due to packet losses and latency. Moreover, the cluster-head selection is based on the energy and relative spectrum awareness such that noncontiguous available spectrum bands are clustered and scheduled to provide continuous transmission opportunity. Routing employs clustering with hybrid medium access by combining carrier-sense multiple access (CSMA) and time-division multiple access (TDMA). TDMA operates for intracluster transmission, whereas CSMA is used for intercluster routing. Thus, a cross-layer design of routing, i.e., of medium access control (MAC) and physical layers, provides efficient multimedia routing in CRSNs, which is revealed through simulation experiments.

A Spectrum-Aware Clustering for Efficient Multimedia Routing in Cognitive Radio Sensor Networks

Explosive growth in Information Technology has enabled many innovative application areas such as large-scale outdoor vehicular networks for vehicle-to-vehicle communications. By providing time-sensitive and location-aware information, vehicular networks can contribute to a safer and more efficient driving experience. However, the performance of vehicular networks requires robust and real-time data communications and is impacted by high mobility, intermittent connectivity, and unreliability of the wireless channel. In this paper, a novel adaptive distributed cooperative medium access control (ADC-MAC) protocol is proposed in order to address the inherent problems in the IEEE 802.11 standard when employed in vehicular networks. ADC-MAC exploits spatial diversities to maximize the system throughput as well as the service range of vehicular networks. This is accomplished through adaptively selecting the most suitable helper and transmission mode for transmit/receive pairs among direct transmission (DT), cooperative relay (CR) transmission and two-hop relay (TR) transmission, in accordance with the channel quality and the positioning of relay nodes. Both our Markov Chain modeling based theoretical analysis and ns-2 simulation experiments show that our ADC-MAC protocol outperforms existing schemes under the same network scenarios and maximizes the achieved system throughput and service distance.

/pdf/a-novel-adaptive-distributed-cooperative-relaying-mac-5861e80ffz.pdf

A Novel Adaptive Distributed Cooperative Relaying MAC Protocol for Vehicular Networks

We propose a new analytical approach to evaluate the average packet error rate (PER) of a conventional packet transmission system over a quasi static fading channel, by presenting an integral inequality lemma. The basic idea of the approach is that, given the PER for the AWGN channel as a function of signal-to-noise ratio (SNR), the average PER over Rayleigh fading channel can be generally upper bounded by a quite simple inequality, i.e.,1 - exp(-wo/γ), for both coded and uncoded schemes, where wo, defined by an integral expression, corresponds exactly to the inversion of coding gain; and this bound is tight in the high SNR region or for long packet systems. We further apply the integral inequality to extend our research to more general Nakagami-m fading channel.

A General Upper Bound to Evaluate Packet Error Rate over Quasi-Static Fading Channels

Accurate available bandwidth estimation in IEEE 802.11-based ad hoc networks

An Adaptive Multirate Auto Rate Fallback (AMARF) protocol is proposed for IEEE 802.11 WLANs. The key idea is to assign each data rate a unique success threshold, which is a criterion to switch one rate to the next higher rate, and the success thresholds can be changed dynamically in an adaptive manner according to the running conditions, such as packet length and channel parameters. Moreover, the proposed protocol can be implemented without any change to the current IEEE 802.11 standards. Our in-depth simulation shows that AMARF yields significantly higher throughput than other existing schemes including Auto Rate Fallback (ARF) scheme and its variants, in various running conditions.

Adaptive Multirate Auto Rate Fallback Protocol for IEEE 802.11 WLANS

This letter presents a simplified expression for the average packet error rate (PER) for an uncoded packet transmission system with large packet length over slow Rayleigh fading channels, by assuming that the bit error rate (BER) has an exponential form. Then the throughput performance of an uncoded packet transmission system with automatic repeat request (ARQ) is investigated, and the optimal packet length is derived to maximize the throughput. The theoretical analysis is validated by simulations.

On the Throughput and Optimal Packet Length of an Uncoded ARQ System over Slow Rayleigh Fading Channels

This letter proposes a new approach to model the intra-flow contention problem in multi-hop networks. The model takes into consideration of neighboring interference, hidden-node collision and multi-rate scenario. It can be easily used to do admission control or to calculate the end-to-end capacity of a given multi-hop route. Simulation results validate its accuracy.

Yong Xi

Papers

A General Upper Bound to Evaluate Packet Error Rate over Quasi-Static Fading Channels

Accurate available bandwidth estimation in IEEE 802.11-based ad hoc networks

Adaptive Multirate Auto Rate Fallback Protocol for IEEE 802.11 WLANS

On the Throughput and Optimal Packet Length of an Uncoded ARQ System over Slow Rayleigh Fading Channels

Modeling intra-flow contention problem in IEEE 802.11 wireless multi-hop networks