As one of the most popular and attractive frameworks for model training, federated edge learning (FEEL) presents a new paradigm, which avoids direct data transmission by collaboratively training a global learning model across multiple distributed edge devices, thus overcoming the disadvantage of centralized machine learning in resource limitations, delay constraints, and privacy issues. However, due to the heavy cost of communicating gradient among edge devices, sharing the parameters of a large-scale neural network can still be time-intensive. To alleviate this bottleneck, an efficient scheme, called SignSGD has been recently proposed, where the one-bit gradient quantization with majority vote is featured at edge devices. Nevertheless, the performance of one-bit aggregation will inevitably deteriorate due to the undesirable propagation error introduced by wireless channels. To address this issue, we propose in this work a novel reconfigurable intelligent surface (RIS)-aided one-bit communication optimization scheme under orthogonal frequency division multiple access (OFDMA) to relieve the negative influence of communication error on the SignSGD-based FEEL. Specifically, a learning convergence analysis is firstly presented to quantitatively characterize the impact of wireless communication error measured by the union bound on pairwise bit error rate (BER) on the performance of SignSGD-based FEEL. Immediately, a unified communication-learning optimization problem is further formulated to jointly optimize the sub-band assignment strategy, the power allocation vector, and the RIS configuration matrix. Numerical experiments show that the proposed design achieves substantial performance improvement compared with the state-of-the-art approaches.

Device-free handwriting recognition, which can be widely applied to Human–Computer Interaction (HCI), is expected to monitor the handwriting change of entities without actively involving any physical devices. The key intuition is that different handwriting movements induce different multi-path distortions in WiFi signals and generate distinct patterns in the time-series of Channel State Information (CSI). In this paper, a proposed WiFi-based system namely Wima, by means of which, we are in pursuit of an advanced solution to recognize letters written by hand in the air. Unlike existing WiFi-based handwriting recognition systems, Wima extracts features directly from the original CSI and then transmit them to subsequent classifiers immediately, a deep-learning-based method is firstly introduced in our system to automatically extract high-level features. More specifically, Wima transforms CSI characters like amplitude and phase into images, which are afterwards delivered to extract features by adopting Convolutional Neural Networks (CNN) respectively. The final simulation with commercial IEEE 802.11n NICs shows that Wima can achieve air-writing recognition with a better performance when compared with some existing routine approaches.

As an emerging distributed learning framework, federated edge learning (FEEL) can efficaciously resolve the delay requirements and privacy concerns by enabling collaborative modeling among the edge devices under the premise of data localization. However, the communication bottlenecks, e.g., model damage and signal deviation, will critically diminish the convergence performance due to the restricted resources and the undesirable wireless fading. To overcome this challenge, one feasible way is to integrate the reconfigurable intelligent surface (RIS) into the FEEL system, to reinforce the communication quality by adaptively reconfiguring the signal propagation environment. However, the significant premise to effectively exploit the RIS in most of the prior works is the estimation of exact instantaneous channel state information (CSI), which is extremely thorny and potentially incurs additional communication overhead. To tackle this issue, we investigate in this paper the RIS-aided FEEL system under the realistic supposition where only the statistical CSI is available among devices. Specifically, considering the wireless outage caused by the uncertainty of non-line-of-sight components, we rigorously derive an explicit convergence upper bound of the RIS enabled FEEL framework with outage. Accordingly, a resource configuration problem with the goal of minimizing the sum of outage-probability is further formulated by jointly configuring the RIS configuration matrix and the bandwidth allocation. To seek the solutions, we carefully design a general Bernstein-Type inequality in this paper, and thus the probabilistic outage objective function can be effectively handled in an equivalent manner. Simulation experiments verify that our design can accomplish a significant promotion compared against state-of-the-art baselines.

As a prominent and appealing paradigm, federated edge learning (FEEL) aims to orchestrate collaborative training across a multitude of distributed edge devices without the need for sensitive information transfer. However, device heterogeneity and wireless transmission distortion, which compromise training robustness and efficiency, severely hinder FEEL deployment in real-world applications. To this end, we propose in this article a novel FEEL framework empowered by simultaneous transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) with flexible aggregation. This framework harmonizes devices to train models with heterogeneous intensity in each communication round, while the deployment of STAR-RIS significantly boosts the quality of wireless aggregation. Here, we emphasize that there may exist a crucial tradeoff between the heterogeneous local training intensity and the transmission quality, particularly when edge devices operate under a limited energy budget. To illuminate this perspective, we rigorously derive a novel explicit upper bound that captures the joint impact of local training accuracy and the mean square error of wireless aggregation on FEEL convergence performance. Our theoretical results indicate that a naive focus on improving the local training accuracy within a constrained energy budget may ultimately detract from overall training performance. This finding sharply contrasts with existing research, which typically aims to accelerate convergence by increasing local training intensity while neglecting the impact of wireless aggregation distortion. To strike the ideal balance, we formulate a mixed-integer nonlinear programming problem to guide the joint design of the beamforming at devices and the base station, the configuration mode of STAR-RIS, and the local training intensity. Comprehensive experiments on representative datasets demonstrate that our proposed framework achieves significant performance improvements compared with existing baselines.

In this paper, the problem of model aggregation for the federated edge learning (FEEL) over a realistic wireless network is investigated, where multiple distributed edge devices collaboratively train a global learning model using local data. In the considered model, the one-bit gradient quantization with majority vote is adopted at edge devices, which sends the local gradient sign to the edge server instead of transmitting high-dimensional stochastic gradients directly. After aggregating the quantified signs, the edge server sends back only the majority decision to significantly minimize the transmission overhead since all communications are compressed to one bit. Nevertheless, it turns out that the quality of training will be inevitably deteriorated by the undesirable propagation error introduced by wireless channels. To address this issue, we propose in this work a reconfigurable intelligent surface (RIS) assisted one-bit communication scheme under orthogonal frequency division multiplexing (OFDM) system to reduce the signal distortion during iterative model exchange of FEEL. Specifically, a learning convergence analysis with respect to wireless communication error is first established. After that, we further formulate a unified communication-learning design problem to jointly optimize the power allocation vector and the RIS configuration matrix. Numerical results demonstrate that the proposed design achieves substantial performance improvement compared with the state-of-the-art solutions.