On Analog Gradient Descent Learning Over Multiple Access Fading Channels
Open accessThe DOI matches its Crossref record; 63 of the 64 references it cites were found in a registry or an index.
Abstract
We consider a distributed learning problem over multiple access channel (MAC) using a large wireless network. The computation is made by the network edge and is based on received data from a large number of distributed nodes which transmit over a noisy fading MAC. The objective function is a sum of the nodes' local loss functions. This problem has attracted a growing interest in distributed sensing systems, and more recently in federated learning. We develop a novel Gradient-Based Multiple Access (GBMA) algorithm to solve the distributed learning problem over MAC. Specifically, the nodes transmit an analog function of the local gradient using common shaping waveforms and the network edge receives a superposition of the analog transmitted signals used for updating the estimate. GBMA does not require power control or beamforming to cancel the fading effect as in other algorithms, and operates directly with noisy distorted gradients. We analyze the performance of GBMA theoretically, and prove that it can approach the convergence rate of the centralized gradient descent (GD) algorithm in large networks. Specifically, we establish a finite-sample bound of the error for both convex and strongly convex loss functions with Lipschitz gradient. Furthermore, we provide energy scaling laws for approaching the centralized convergence rate as the number of nodes increases. Finally, experimental results support the theoretical findings, and demonstrate strong performance of GBMA using synthetic and real data.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“The convergence behavior demonstrates that the noisy iterates typically introduce non-negligible optimality gap in various FL algorithms, e.g., vanilla gradient method [174], quantized gradient method [175], sparsified gradient method [173], and operator splitting method [87].”
From Edge artificial intelligence for 6g: vision, enabling technologies, and applications (2022) · cites this paper as [174]
Citations
Cites
UCI Machine Learning Repository
Asuncion, A. (2007) · Medical Entomology and Zoology · Cited by 25,370
Goldsmith, A. (2005) · Cambridge University Press eBooks · doi:10.1017/cbo9780511841224 · Cited by 10,658
Mathematical Analysis of Random Noise
Rice, S. O. (1944) · Bell System Technical Journal · doi:10.1002/j.1538-7305.1944.tb00874.x · Cited by 5,480 · cited 1×
“If the channel has a fixed LOS component, the channel distribution can be well approximated by the Rician fading model [47].”
MethodMathematical Analysis of Random Noise
Rice, S. O. (1945) · Bell System Technical Journal · doi:10.1002/j.1538-7305.1945.tb00453.x · Cited by 2,898
Adaptive Federated Learning in Resource Constrained Edge Computing Systems
Wang, S., Tuor, T., Salonidis, T., Leung, K. K., Makaya, C., He, T., & Chan, K. (2019) · IEEE Journal on Selected Areas in Communications · doi:10.1109/jsac.2019.2904348 · Cited by 2,232 · cited 2×
“In recent years, other stochastic gradient-descent (SGD) based methods were developed for federated learning [43], [44].”
BackgroundMethodRobust Stochastic Approximation Approach to Stochastic Programming
Nemirovski, A., Juditsky, A., Lan, G., & Shapiro, A. (2009) · SIAM Journal on Optimization · doi:10.1137/070704277 · Cited by 1,945
Federated Optimization: Distributed Machine Learning for On-Device Intelligence
Konečný, J., McMahan, H. B., Ramage, D., & Richtárik, P. (2016) · arXiv · doi:10.48550/arxiv.1610.02527 · Cited by 1,563
Pegasos: primal estimated sub-gradient solver for SVM
Shalev-Shwartz, S., Singer, Y., Srebro, N., & Cotter, A. (2011) · Mathematical Programming · doi:10.1007/s10107-010-0420-4 · Cited by 1,477
Federated Learning via Over-the-Air Computation
Yang, K., Jiang, T., Shi, Y., & Ding, Z. (2020) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2019.2961673 · Cited by 1,084 · cited 1×
“In [15], [16], multiple antennas were used at the receiver, where beamforming was used to maximize the number of devices scheduled for transmission.”
MethodBertin-Mahieux, T., Ellis, D. P. W., Whitman, B., & Lamere, P. (2011) · Columbia University · doi:10.7916/d8nz8j07 · Cited by 944 · cited 1×
“We used the dataset available by UCI Machine Learning Repository [58], extracted from the Million Song Dataset collaborative project between The Echo Nest and LabROSA [59].”
Method
Research works printed but not linked: 2
- 1998, Introductory lectures on convex programming volume i: Basic course, Lecture Notes Series, 3, 51
- bertin-mahieux, 0, Proc Int Conf Music Inf Retrieval
Cited by
Items 101–200 of 201
Wen, H., Xing, H., & Simeone, O. (2025) · IEEE Transactions on Cognitive Communications and Networking · doi:10.1109/tccn.2025.3640114 · Cited by 9
“The downside of a noisy training process is the convergence time, which may be severely affected by noisy updates [26], [27], [28].”
FL-MAC-RDP: Federated Learning over Multiple Access Channels with Rényi Differential Privacy
Wu, S., Yu, M., Ahmed, M. A. M., Qian, Y., & Tao, Y.-H. (2021) · International Journal of Theoretical Physics · doi:10.1007/s10773-021-04867-0 · Cited by 9
“Some people study joint model devices through wireless channels [9, 11, 13–16], some people consider FL on multiple access channels (MAC) [8, 10, 17, 18], which is denoted as FL-MAC.”
BackgroundSubgradient Descent Learning Over Fading Multiple Access Channels With Over-the-Air Computation
Gez, T. L. S., & Cohen, K. (2023) · IEEE Access · doi:10.1109/access.2023.3291023 · Cited by 9
“Then, the PS broadcasts the updated model back to the nodes via an error-free channel, as commonly assumed in the OTA learning literature [24], [25], [27], [28], [29], [30], [31], [32], [33], [34], as the bandwidth requirement for the downlink transmission does not scale with N , and digital communications can be impl…”
BackgroundMethodInfluentialSelf-citationMhanna, E., & Assaad, M. (2024) · Allerton Conference on Communication, Control, and Computing · doi:10.1109/allerton63246.2024.10735291 · Cited by 8
Fundamental Limits of Over-the-Air Optimization: Are Analog Schemes Optimal?
Jha, S. K., Mayekar, P., & Tyagi, H. (2022) · IEEE Journal on Selected Areas in Information Theory · doi:10.1109/jsait.2022.3177582 · Cited by 8
“In [18], the subgradient estimates are scaled-down appropriately to satisfy the power constraint.”
BackgroundMethodInfluentialOver-The-Air Federated Learning under Byzantine AttacksPreprint
Sifaou, H., & Li, G. Y. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2205.02949 · Cited by 8
“As in [21], [22], [24], [27], [28], we assume that perfect channel state information (CSI) is available at the clients and the CS.”
BackgroundTrust Online Over-the-Air Computation for Wireless Federated Learning
Sun, M., Zheng, J., Du, H., Zhang, H., Niyato, D. T., Kang, J., Wang, J., Ren, J., Gao, L. C., & Wang, Z. (2025) · IEEE Transactions on Mobile Computing · doi:10.1109/tmc.2025.3547148 · Cited by 7
Robust Federated Learning Over the Air: Combating Heavy-Tailed Noise with Median Anchored Clipping
Li, J., Chen, Z., Chong, K. F. E., Das, B., Quek, T. Q. S., & Yang, H. H. (2025) · International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt) · doi:10.23919/wiopt66569.2025.11123216 · Cited by 7
“While such interference enhances privacy protection, it also compromises the reliability of channel transmission, especially when it manifests as impulse interference, rendering the noise exhibiting a heavy-tailed distribution (rather than Gaussian) [16]—this has been consistently demonstrated by…”
MAC Aware Quantization for Distributed Gradient Descent
Chang, W.-T., & Tandon, R. (2020) · Globecom · doi:10.1109/globecom42002.2020.9322254 · Cited by 7
Intelligent reflecting surface-assisted federated learning in multi-platoon collaborative networks
Ma, X., Zhao, J., Liao, J., & Zhang, Z. (2022) · Digital Communications and Networks · doi:10.1016/j.dcan.2022.11.010 · Cited by 7
Over-the-Air Federated Learning Over MIMO Channels: A Sparse-Coded Multiplexing Approach
Zhong, C., & Yuan, X. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3567598 · Cited by 7
“…the following base-lines: (I) SCoM; (II) Algorithm 1 and sparse-coding without compression (top-k ); (III) Algorithm 1 and no sparse-coding; (IV) GBMA [57] and sparse-coding with compression; (V) GBMA [57] and sparse-coding without compression (top-k ); and (VI) GBMA [57] and no sparse-coding.”
Performance Analysis for Channel-Weighted Federated Learning in OMA Wireless Networks
Yan, N., Wang, K., Pan, C., & Chai, K. K. (2022) · IEEE Signal Processing Letters · doi:10.1109/lsp.2022.3154653 · Cited by 7
“Therefore, the broadcast of the global parameter is error-free [6, 10] .”
An Autoencoder-Based Constellation Design for AirComp in Wireless Federated Learning
Mu, Y., Wei, X., & Shen, C. (2024) · International Conference on Communications · doi:10.1109/icc51166.2024.10622489 · Cited by 6
“As a result, it is commonly assumed that the download communication phase is error-free [4], [5], [11]– [13].”
BackgroundPAUSE: Low-Latency and Privacy-Aware Active User Selection for Federated Learning
Peleg, O., Lang, N., Ben-Ami, D., Rini, S., Shlezinger, N., & Cohen, K. (2025) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2025.3627252 · Cited by 6
“Communication latency can be tackled by model compression [7], [8], [9], [10], [11], and via over-the-air aggregation in settings where the users share a common wireless channel [12], [13], [14].”
Self-citationA Survey on Federated Learning in RIS, STAR-RIS, and BD-RIS-Assisted Wireless Networks
Mendes, I. F., Duarte, D. L. F., Guedes, L. G. B., Souza, L. S., Silva, R. S., Souto, V. D. P., & Souza, R. D. (2026) · IEEE Open Journal of the Communications Society · doi:10.1109/ojcoms.2026.3661403 · Cited by 6
“Most relevant to this survey, FL has gained increasing attention in wireless communications, where it has been integrated with RIS, STAR-RIS, and BD-RIS technologies to address key challenges related to convergence speed, latency, and adaptability under dynamic channel conditions [96], [102], [103].”
Model Pruning for Distributed Learning Over the Air
Zhao, Z., Xu, K., Hong, W., Peng, M., Ding, Z., Quek, T. Q. S., & Yang, H. H. (2024) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2024.3486169 · Cited by 6
“Recognizing that (stochastic) gradient descent-based algorithms are often resilient to training noise, [5], [6] suggested directly using OTA computing for distributed learning without estimating CSI, where such an architecture is coined as A-OTA distributed learning.”
A Distributed Stable Strategy Learning Algorithm for Multi-User Dynamic Spectrum Access
Gafni, T., & Cohen, K. (2019) · Allerton Conference on Communication, Control, and Computing · doi:10.1109/allerton.2019.8919920 · Cited by 6
“Model-free learning strategies were developed in [27]–[29].”
Self-citationPruning Analog Over-the-Air Distributed Learning Models with Accuracy Loss Guarantee
Xu, K., Yang, H. H., Zhao, Z., Hong, W., Quek, T. Q. S., & Peng, M. (2022) · International Conference on Communications · doi:10.1109/icc45855.2022.9839266 · Cited by 6
“The error function is defined as the difference between the model parameters at iteration k +1 and the optimal model w ∗ : To facilitate the analysis, we first make a few assumptions which have been used in the proof of [5], [6], [10].”
BackgroundMethodInfluentialSparse Training for Federated Learning With Regularized Error Correction
Greidi, R., & Cohen, K. (2024) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2024.3463505 · Cited by 6
“Additionally, the communication aspects of FL have been a subject of exploration in recent years, spanning both digital [6]– [12] and analog [13]–[16] communications.”
BackgroundSelf-citationOver-the-Air Federated Learning: Rethinking Edge AI Through Signal ProcessingPreprint
Azimi‐Abarghouyi, S. M., Fischione, C., & Huang, K. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.03719 · Cited by 6
“Initially proposed in [41], later enhanced with momentum-based acceleration in [42], and further extended in [43], this scheme demonstrates that AirFL can achieve convergence even without sophisticated signal processing.”
Personalized Federated Learning Over the Air
Li, Z., Chen, Z., Quek, T. Q. S., & Yang, H. H. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3573777 · Cited by 5
NCAirFL: CSI-Free Over-the-Air Federated Learning Based on Non-Coherent Detection
Wen, H., Michelusi, N., Simeone, O., & Xing, H. (2025) · International Conference on Communications · doi:10.1109/icc52391.2025.11161902 · Cited by 5
“While phase-compensated equalization can avoid issue ( ii ), it was only studied for narrow-band transmissions and may not be directly applicable to AirFL over broadband channels [7].”
Lower Bounds for Over-the-Air Statistical Estimation
Lee, C.-Z., Barnes, L. P., & Özgür, A. (2021) · IEEE International Symposium on Information Theory (ISIT) · doi:10.1109/isit45174.2021.9518157 · Cited by 5
Over-the-Air Aggregation-Based Federated Learning in Cache-Enabled Wireless Edge Networks
Hong, J., Park, S., & Choi, W. (2023) · International Conference on Communications · doi:10.1109/icc45041.2023.10279408 · Cited by 5
“…aggregation strategies have explored the channel inversion power control to cancel out the effect of channel fading, and they have shown to achieve a convergence to the optimal model parameters under some constraints on the training model and the simplified channel without pathloss [2]–[4].”
BackgroundOver-the-Air Federated Learning with Phase Noise: Analysis and Countermeasures
Dahl, M., & Larsson, E. G. (2024) · Conference on Information Sciences and Systems · doi:10.1109/ciss59072.2024.10480215 · Cited by 5
“As transmitted symbols carry gradient elements, Wiener phase noise implies that gradient elements transmitted early are received more accurately than those transmitted later.”
BackgroundUnbiased Over-the-Air Computation via Retransmissions
Hellström, H., Fodor, V., & Fischione, C. (2022) · Globecom · doi:10.1109/globecom48099.2022.10001026 · Cited by 5
“of efficient distributed machine learning [11]–[13].”
BackgroundOn Analog Distributed Approximate Newton with Determinantal Averaging
Sharma, G. D., & Dey, S. (2022) · IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC) · doi:10.1109/pimrc54779.2022.9977466 · Cited by 5
“As the rank values increases the hessian approximation moves closer to exact hessian and hence the 5The actual transmission power is the product of PAF and squared-norm of the vector to be transmitted, eg. in GBMA it is P ||gn(θk)|| 2 2 performance of the algorithm improves.”
BackgroundMethodInfluentialOver-the-Air Federated Edge Learning with Error-Feedback One-Bit Quantization and Power Control
Liu, Y., Liu, D., Jin, R., Zhu, G., & Shi, Q. (2024) · IEEE International Conference on Computer and Communications (ICCC) · doi:10.1109/iccc62609.2024.10942260 · Cited by 5
“Recently, implementing AirComp in FEEL has attracted much attention due to the advantages in transmitting an aggregation of high-dimensional updates [17].”
Optimal Power Control for Over-The-Air Federated Edge Learning Using Statistical Channel Knowledge
Yu, X., Xiao, B., Ni, W., & Wang, X. (2022) · 2022 14th International Conference on Wireless Communications and Signal Processing (WCSP) · doi:10.1109/wcsp55476.2022.10039388 · Cited by 4
“When there is no scaling (i.e., ρ n,t = 1 , ∀ n ) and the stepsize , the optimality gap is upper bounded by [9] where the constant c = 1 − 2 βµ h mL m + L + δ is the diameter of the compact convex parameter set, of which the parameter vector w lies in the interior; and r 0 = ∥ w 0 − w ∗ ∥ denotes…”
BackgroundInfluentialAirFL-Mem: Improving Communication-Learning Trade-Off by Long-Term Memory
Wen, H., Xing, H., & Simeone, O. (2024) · IEEE Wireless Communications and Networking Conference · doi:10.1109/wcnc57260.2024.10571203 · Cited by 4
“To provide additional context for this work, other solutions to the problem of deep fading in AirFL in the presence of CSIT include phase-only compensation [8], [9], transmission weight optimization [10], and device selection [11].”
MethodPAUSE: Privacy-Aware Active User Selection for Federated Learning
Peleg, O., Lang, N., Rini, S., Shlezinger, N., & Cohen, K. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10890399 · Cited by 4
Self-citationAccelerated Federated Learning Over Wireless Fading Channels With Adaptive Stochastic Momentum
Su, L., & Lau, V. K. N. (2023) · IEEE Internet of Things Journal · doi:10.1109/jiot.2023.3340014 · Cited by 4
Gez, T. L. S., & Cohen, K. (2026) · IEEE Access · doi:10.1109/access.2026.3744899 · Cited by 4
“This approach is particularly well-suited for mobile applications, such as those in 5G, IoT, and cognitive radio systems, where communication resources are limited with digital [9], [29]–[36] or analog [4], [37]– [40] communications, and due to privacy concerns [2], [41], [42].”
Self-citationSemi-Asynchronous Federated Edge Learning for Over-the-Air Computation
Kou, Z., Ji, Y., Zhong, X., & Zhang, S. (2023) · Globecom · doi:10.1109/globecom54140.2023.10437830 · Cited by 4
“To enhance the performance of wireless FEEL system, several studies related to the AirComp system for FEEL have been done [4]–[6].”
BackgroundMethodSu, L., & Lau, V. K. N. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3177449 · Cited by 4
“…the distributed raw data samples and then training neural networks (NNs) in a centralized way, the local sensors transmit local gradients [4]–[7], model updates [8]–[10], or intermediate computation results [11]–[14] to the central server, where the transmitted signals are further aggregated…”
BackgroundAdaptive Scheduling for Machine Learning Tasks over Networks
Gatsis, K. (2021) · Proceedings of the ... American Control Conference/Proceedings of the American Control Conference · doi:10.23919/acc50511.2021.9482692 · Cited by 4
Distributed Sparse SGD with Majority VotingPreprint
Özfatura, K., Özfatura, E., & Gündüz, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.06495 · Cited by 4
“…estimates over all the workers, instead of receiving each one separately, linearity property can help to employ more efficient communication strategies both in wired (Eghlidi & Jaggi, 2020) and wireless network setups (Mohammadi Amiri & Gündüz, 2020; Sery & Cohen, 2020; Zhu et al., 2019).”
BackgroundByzantine-robust distributed support vector machine
Wang, X., Liu, W., & Mao, X. (2024) · Science China Mathematics · doi:10.1007/s11425-023-2217-2 · Cited by 3
Power-Efficient Over-the-Air Aggregation With Receive Beamforming for Federated Learning
Kalarde, F. M., Dong, M., Liang, B., Ahmed, Y., & Cheng, H. T. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3626720 · Cited by 3
“Various analyses and designs have been proposed for over-the-air analog aggregation [5], [6], [7], [8], [9].”
Mhanna, E., & Assaad, M. (2024) · HAL · Cited by 3
Multi-Antenna Transceiver Design for Federated Learning in Cell-Free Massive MIMO System
Fan, S., Xu, S., Xing, F., Xu, Y., Li, C., & Yang, L. (2025) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2025.3619068 · Cited by 3
“Next, we introduce some assumptions that are classical for the analysis of convergence in FL [29].”
Differentially Private Wireless Federated Learning Using Orthogonal Sequences
Wei, X., Wang, T., Huang, R., Shen, C., Yang, J., & Poor, H. V. (2026) · IEEE Transactions on Information Theory · doi:10.1109/tit.2026.3668977 · Cited by 3
“For a given x ik , client k transmits symbol x i k e − j φ k , where e − j φ k is the phase correction term as suggested in [30].”
InfluentialTowards Federated Learning Over the Air: Why Scaling Up Helps?
Zhu, J., Das, B., Παππάς, Ν., & Yang, H. H. (2025) · International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt) · doi:10.23919/wiopt66569.2025.11123201 · Cited by 3
“In response, a line of recent studies [3]–[5] suggested integrating over-the-air computations into the FL model training procedure, leveraging the superposition properties of radio waveforms for fast and scalable parameter aggregation.”
Fundamental limits of over-the-air optimization: Are analog schemes optimal?
Jha, S. K., Mayekar, P., & Tyagi, H. (2021) · Globecom · doi:10.1109/globecom46510.2021.9685770 · Cited by 3
“We remark that the convergence analysis in [37] required additional smoothness assumptions and is not valid for our setting.”
BackgroundMethodInfluentialOn the Necessity of Aligning Gradients for Wireless Federated Learning
Chang, W.-T., Seif, M., & Tandon, R. (2021) · IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc51858.2021.9593177 · Cited by 3
Over-the-Air Computation for Distributed Machine Learning and Consensus in Large Wireless Networks
Frey, M., Bjelaković, I., & Stańczak, S. (2022) · Applied and numerical harmonic analysis · doi:10.1007/978-3-031-09745-4_13 · Cited by 3
Over The Air Federated Learning in the Presence of Impulsive Noise
Shaban, Z., Prasad, R., & Kumari, P. (2024) · IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) · doi:10.1109/icasspw62465.2024.10625955 · Cited by 2
“OTA-FL allows clients to simultaneously transmit updates using analog signaling over the uplink channel in a non-orthogonal manner, hence optimizing available temporal and spectral resources [2, 3].”
MethodSubgradient Descent Learning with Over-the-Air Computation
Gez, T. L. S., & Cohen, K. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10095134 · Cited by 2
“Each node n ∈ 1 , ..., N experiences at time t k a block-fading channel ˜ h n,k with gain The channel fading is assumed to be i.i.d. across time and nodes, with mean µ h and variance σ 2 h as in [4], [6], [8], [11].”
BackgroundMethodInfluentialSelf-citationFederated Linear Bandit Learning via Over-the-air Computation
Wang, J., Jiang, Y., Liu, X., Wang, T., & Shi, Y. (2023) · Globecom · doi:10.1109/globecom54140.2023.10437441 · Cited by 2
“In addition, we assume that perfect channel state information (CSI) is available at all edge devices and the server as in most of the existing works on AirComp [14], [16], [18].”
MethodTang, B., Hu, B., Qu, Z., & Ye, B. (2024) · IEEE Internet of Things Journal · doi:10.1109/jiot.2024.3406918 · Cited by 2
“…that the edge server in FEEL only needs to get the aggregated data transmitted by the edge devices, a new framework called Air-FEEL was introduced [37], [38], which is characterized by allowing all the devices to transmit signals simultaneously so that the edge server can directly access the…”
BackgroundOver-the-air Federated Policy Gradient
Yang, H., Huang, L., Dey, S., & Shi, L. (2024) · International Conference on Communications · doi:10.1109/icc51166.2024.10622210 · Cited by 2
“With transmitter synchronization and phase compensation [26], the central controller at time step k can receive the following signal where n k ∼ N (0 , σ 2 I d ) is the additive white Gaussian noise, and h i,k is the channel gain suffered by agent i at time step k , which is a random variable with…”
BackgroundCommunication-Efficient Federated Learning via Sparse Training with Regularized Error Correction
Greidi, R., & Cohen, K. (2024) · Allerton Conference on Communication, Control, and Computing · doi:10.1109/allerton63246.2024.10735294 · Cited by 2
“Addressing these issues is a dynamic and evolving field with numerous prior works spanning various areas, including weight pruning [17]–[19], over-the-air methods [13]–[16], [20], model or gradient compression [21]–[30], quantization and sparsification.”
BackgroundSelf-citationAge-Aware Partial Gradient Update Strategy for Federated Learning Over the Air
Du, R., Zhu, J., Li, Z., & Yang, H. H. (2026) · International Conference on Communications · doi:10.1109/icc59461.2026.11587476 · Cited by 2
“However, the conventional Top-k method updates only the largest-magnitude gradients, and superimposed sparse updates lose sparsity, causing significant reconstruction errors in large-scale client scenarios [12].”
CoBAAF: Controlled Bayesian Air Aggregation Federated Learning from Heterogeneous Data
Gafni, T., Cohen, K., & Eldar, Y. C. (2022) · Allerton Conference on Communication, Control, and Computing · doi:10.1109/allerton49937.2022.9929426 · Cited by 2
“Other related works on OTA FL methods can be found in [18]–[22] and references therein.”
BackgroundSelf-citationAdaptive Gradient Methods For Over-the-Air Federated Learning
Wang, C., Chen, Z., Yang, H. H., & Παππάς, Ν. (2023) · IEEE International Workshop on Signal Processing Advances in Wireless Communications (SPAWC) · doi:10.1109/spawc53906.2023.10304465 · Cited by 2
“To harness such uncertainty while maintaining the benefits of the A-OTA computations, numerous works [14], [16] have been proposed to improve the system performance, among which the majority of these approaches are built upon FedAVG [1] or FedSGD [15] for gradient descent-based global model updates.”
BackgroundMethodInfluentialGP-FL: Model-Based Hessian Estimation for Second-Order Over-the-Air Federated LearningPreprint
Hamidi, S. M., Bereyhi, A., Asaad, S., & Poor, H. V. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2412.03867 · Cited by 2
“To address this, we use AirComp to perform gradient aggregation directly over the air [12], [14]–[16], [18], [19].”
Timely Parameter Updating in Over-the-Air Federated LearningPreprint
Zhu, J., Zhao, Z., Li, X., Du, R., Jin, S., & Yang, H. H. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.19103 · Cited by 2
“Recognizing the noise resilience of (stochastic) gradient descent-based algorithms, several studies [6], [10] proposed bypassing explicit CSI estimation and directly utilizing OAC for distributed learning.”
A Masked Pruning-Based Algorithm for Dimensionality Reduction in Federated Learning Systems
Gez, T. L. S., & Cohen, K. (2024) · Allerton Conference on Communication, Control, and Computing · doi:10.1109/allerton63246.2024.10735315 · Cited by 1
“This approach is particularly well-suited for mobile applications, such as those in 5G, IoT, and cognitive radio systems, where communication resources are limited with digital [8], [28]–[35] or analog [3], [36]–[39] communications, and due to privacy concerns [1], [40], [41].”
BackgroundSelf-citationWeighted Over-the-Air Federated Learning
Azimi‐Abarghouyi, S. M., Tassiulas, L., & Fischione, C. (2025) · IEEE International Conference on Machine Learning for Communication and Networking (ICMLCN) · doi:10.1109/icmlcn64995.2025.11140426 · Cited by 1
“The studies in [16], [17] delve into blind over-the-air FL without any compensation for the detrimental effects of wireless channels.”
InfluentialRethinking Federated Learning Over the Air: The Blessing of Scaling Up
Zhu, J., Das, B., Xie, Y. C., Παππάς, Ν., & Yang, H. H. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3631503 · Cited by 1
“Moreover, increasing the number of clients can improve energy efficiency, thereby reducing the transmit power [7].”
ECPRESSO: Efficient Gradient Compression for Over-the-Air Federated Learning
Zhu, J., Du, R., Zhao, Z., Quek, T. Q. S., & Yang, H. H. (2025) · International Workshop on Radio Frequency (RF) Computing · doi:10.1145/3737906.3767101 · Cited by 1
OFLight: Lightweight Gradient Compression for Over-the-Air Federated Learning
Zhu, J. F., Yang, H. Q., Παππάς, Ν., & Poor, H. V. (2026) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon68281.2026.11579004 · Cited by 1
“Recognizing the inherent noise resilience of (stochastic) gradient descent-based algorithms, [5], [9] proposed bypassing CSI estimation directly using OTA computation for distributed learning.”
Age-Aware Edge-Blind Federated Learning via Over-the-Air Aggregation
Elshazly, A. E., & Arafa, A. M. (2026) · International Conference on Communications · doi:10.1109/icc59461.2026.11587676 · Cited by 1
On the Convergence of Federated Learning with Stochastic Quantization
Li, W., Li, Y., & Du, J. (2021) · IEEE International Conference on Cloud Computing and Intelligence Systems (CCIS) · doi:10.1109/ccis53392.2021.9754534 · Cited by 1
Fundamental Limits of Distributed Optimization over Multiple Access Channel
Jha, S. K., & Mayekar, P. (2023) · IEEE Information Theory Workshop (ITW) · doi:10.1109/itw55543.2023.10161622 · Cited by 1
“For our upper bound, we use the well-known scaled transmission scheme from [7].”
A Momentum-Based Wireless Federated Learning Acceleration With Distributed Principle Decomposition
Dong, Y., Wang, L., Chi, Y., Hu, X., Zhang, H., Yu, F. R., & Leung, V. C. M. (2023) · IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) · doi:10.1109/icasspw59220.2023.10193196 · Cited by 1
“…can drastically reduce the demand for bandwidth via the OAC technology, the unbiased estimators for the global model and the global gradient were obtained based on strong assumptions on the independent and identically distributed (i.i.d.) local gradients [10] or i.i.d. locations of workers [11–17].”
MethodFederated Learning with Partial Gradients Over-the-Air
Wang, W., Chen, Z., Παππάς, Ν., & Yang, H. H. (2023) · IEEE International Conference on Sensing, Communication, and Networking (SECON) · doi:10.1109/secon58729.2023.10287423 · Cited by 1
“While numerous works have been conducted to investigate over-the-air FL with gradient descent for model update, among which the convergence [9], [14], generalization [5], privacy-preservation [15] and personalization performance [6] are explored, they ignored the fact that the limited number of…”
BackgroundCombining Scheme Design for the User Multi-Antenna Cell-Free Massive MIMO System
Fan, S., Xu, S., & Zhang, S. (2024) · IEEE International Conference on Communication Technology (ICCT) · doi:10.1109/icct62411.2024.10946278 · Cited by 1
“Next, we introduce some assumptions that are classical for the analysis of convergence in federated learning [1], [12].”
Hassani, M., & Gholizadeh, R. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.16917 · Cited by 1
“This method has two major benefits, namely (i) accessing to larger data provided by different clients, and (ii) utilizing the computing power of all the clients to train a general model [1], [2], [3], [4], [5], [6], [7], [6], [8].”
MethodDai, Y., Wang, J., & Zhang, L. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2403.20205 · Cited by 1
“This method stimulates the development of stochastic approximation algorithms in machine learning [7, 11, 19, 26, 27, 39, 41] and now is widely used in distributed deep learning [4], multiple access channel [42] and low-rank matrix factorization [10].”
MethodTowards Scalable and Secure Federated Learning: Key Issues and Future ResearchPreprint
Daniels, M., Gallus, S., & Wood, R. (2025) · doi:10.36227/techrxiv.173894842.25024919/v1 · Cited by 1
Can Inherent Communication Noise Guarantee Privacy in Distributed Cooperative Control ?Preprint
Ma, Y., Spurgeon, S. K., Li, T., & Chen, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2601.07997 · Cited by 1
“The CSI assumption is commonly adopted to simplify the analysis of communication noise (Sery and Cohen, 2020; Cao et al., 2020), and the SISO large-scale fading channel model is widely considered in robotic networks and vehicular systems (Khuwaja et al., 2018).”
Non-Coherent Over-the-Air Federated Learning: Protocol, Convergence, and Device SchedulingPreprint
Wen, H., Michelusi, N., Simeone, O., Yang, Y., & Xing, H. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.08312 · Cited by 1
“[13] proposed a gradient-based multiple access protocol for AirFL, in which devices compensate for the channel phase only.”
2020 Index IEEE Transactions on Signal Processing Vol. 68
(2020) · IEEE Transactions on Signal Processing · doi:10.1109/tsp.2021.3055469
Over-the-Air Computation Aided Federated Learning with the Aggregation of Normalized GradientPreprint
Fan, R., An, X., Zuo, S., & Han, H. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2308.09082
“In [24], rather than minimizing the MSE of gradient aggregation, the gap to minimal training loss is characterized for the first time when the loss function is smooth only or with strong convexity as well, without adjusting any signal amplification factor.”
BackgroundA Genetic Algorithm-Based Approach to Power Allocation in Rate-Splitting Multiple Access SystemsPreprint
Fajemilehin, T. O., & Cohen, K. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2309.04196
“The channel state information (CSI) is available at the BS via pilot signals, as commonly assumed in the literature [11], [12], [20]–[22], [25].”
MethodSelf-citationAnalog Over‐the‐Air Federated Learning: Design and Analysis
Yang, H. H., Chen, Z., & Quek, T. Q. S. (2023) · Wiley eBooks · doi:10.1002/9781119913924.ch10
Fundamental Limits of Distributed Optimization Over Multiple Access Channel
Jha, S. K. (2024) · IEEE Transactions on Communications · doi:10.1109/tcomm.2024.3395291
“Due to its applications in federated learning [23], many interesting schemes have been recently proposed for this problem [11], [30], [32], [33], [37], [38], [39], [40].”
Personalized and Robust Federated Learning in Heterogeneous EnvironmentsPreprint
Johnson, A. E., Lee, B., Smith, C., Kim, D., Martínez, E., Noor, F., Chen, G., Park, H., & Gallus, S. (2025) · doi:10.36227/techrxiv.175355119.90288089/v1
Differentially Private Federated Edge Learning
Zhou, Y., Fang, W., Shi, Y., & Letaief, K. B. (2025) · Wireless networks · doi:10.1007/978-3-031-96649-1_8
Decentralized Federated Large Model Tuning over Edge Networks with Zeroth-Order Optimization
Chen, Z., Park, J., He, B., Yuan, Y., & Yang, H. H. (2025) · IEEE/CIC International Conference on Communications in China (ICCC) · doi:10.1109/iccc65529.2025.11148609
“…each UE synchronously performs q - point RGE-based gradient estimate in decentralized networks, whereby the process of computing local gradient approximation, transmitting model updates, and aggregating the model updates are fundamentally redesigned compared with FO approaches [16], [17].”
Majority Vote Compressed Sensing for Over-the-Air Histogram Estimation
Jeong, J., Hellström, H., Özgür, A., Fodor, V., & Fischione, C. (2025) · International Conference on Communications · doi:10.1109/icc52391.2025.11160930
“Since it is significantly more challenging to compensate for those in AirComp, non-coherent AirComp schemes have been developed that eliminate the need for phase correction [9], [10], [20]–[22].”
Majority Vote Compressed SensingPreprint
Hellström, H., Jeong, J., Özgür, A., Fodor, V., & Fischione, C. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2510.18008
“Others operate with even more stringent assumptions, where neither magnitude nor phase can be corrected at the devices [10], [26].”
Li, W., Lv, T., Zhao, X., Yu, X., Yuan, X., Ni, W., & Peng, M. (2025) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2025.3639518
“OTA-FL allows devices to send their models using the same resource block [24], [25] and aggregate the models in the analog domain.”
Estimation Bias in Over-the-Air Federated Learning
Wang, Y. (2024) · KTH Publication Database DiVA (KTH Royal Institute of Technology)
Online Learning and Optimization in Communication Networks
Wang, J. (2023) · TSpace (University of Toronto)
Beyond gradients : zero-order approaches to optimization and learning in multi-agent environments
Mhanna, E. (2024) · theses.fr (ABES) · doi:10.70675/7eb7acfcz07cbz4f12z9e08z1b5ebe87ca4c
Cai, Z., Zheng, Y., Chen, H., Liu, D., Wang, B., & Zhu, G. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2604.12401
“In OTA, all client transmits their local gradients synchronously with uncoded transmission [8], [21]–[24].”
Channel-Adaptive Robust Aggregation for Over-the-Air Federated Learning in Heterogeneous Networks
Fatima, Z., Shaban, Z., Jamal, Y., Shah, N., Prasad, R., & Bharath, B. N. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11464541
“However, existing OTA-FL frameworks [11–13] typically enforce fixed local update counts before aggregation, reducing adaptability to heterogeneous clients.”
Closing Price Prediction of the Stocks Using Gradient Descent Based Linear Regression Method
Das, A. K., Chakraborty, S., Sengupta, P., Ghosh, A., Mallick, A., & De, P. (2026) · Lecture notes in networks and systems · doi:10.1007/978-3-032-20253-6_43
Optimizing Utility–Privacy Tradeoff in Federated Learning Over Fading Channels
Mu, Y., Wei, X., Eldar, Y. C., & Shen, C. (2026) · IEEE Transactions on Machine Learning in Communications and Networking · doi:10.1109/tmlcn.2026.3706257
“As a result, it is often assumed that the download communication phase is error-free [13], [41], [67], [68], [69].”
Communication-Efficient Over-the-Air Federated Learning via Lightweight Gradient Compression
Zhu, J., Yang, H. H., Παππάς, Ν., & Poor, H. V. (2026) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2026.3707448
“In response, a line of recent studies [4], [5], [6], [7] have suggested integrating over-the-air (OTA) computations into the model aggregation process of FL; whereby using analog transmissions, instead of digital ones, to upload parameters, the clients can exploit the superposition property of the…”
Device Assignment and Model Splitting Optimization for Resilient and Secure Multi-Hop Split Learning
Wei, D., Zhou, D., Liu, Y., & Chen, M. (2026) · IEEE Transactions on Wireless Communications · doi:10.1109/twc.2026.3713859
“However, all of these works [14], [15], [16], [17], [18], [19] considered a single-hop distributed learning training (e.g., FL) where all devices directly upload their models to the server.”
Ma, Y., Spurgeon, S. K., Li, T., & Chen, B. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2607.25564
“Assuming perfect CSI is a standard convention adopted to facilitate the analysis of communication noise [33], [34].”
Polar Code Based Federated Learning: Convergence Analysis and Resource AllocationPreprint
Xiao, H., Kang, W., & Liu, N. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2608.13961
“8 and take N for 16 , 32 , 64 , 128 in the optimization problem.”
InfluentialCommunication Design for Dynamic Distributed Learning Systems
Kalarde, F. M. (2026) · TSpace (University of Toronto)
Approximate Newton Methods for Distributed Learning over Communication-Constrained Wireless Networks
Sharma, G. (2025) · Maynooth University ePrints and eTheses Archive (Maynooth University)
Jang, J., Lyu, H., & Yang, H. J. (2026) · arXiv (Cornell University) · doi:10.48550/arxiv.2609.23084
Metadata
- Authors
- Sery, T., & Cohen, K.
- Affiliations
- Ben-Gurion University of the Negev (IL)
- Venue
- IEEE Transactions on Signal Processing
- Published
- 2020
- Volume
- 68
- Pages
- 2897-2911
- ISSN
- 1053-587X (ISSN-L)
- 1941-0476
- OpenAlex
- W2969343398
- Cited by
- 172
- References
- 64
- Citations by year
- 2026: 8 · 2025: 24 · 2024: 29 · 2023: 29 · 2022: 28 · 2021: 37 · 2020: 15
- FWCI
- 12.81
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Transactions on Signal Processing (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- United States-Israel Binational Science Foundation (2017723)
- Topics
- Physical Sciences › Computer Science › Computer Networks and Communications › Distributed Sensor Networks and Detection Algorithms
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Artificial Intelligence › Stochastic Gradient Optimization Techniques
- Keywords
- multiple access channel, fading channels, distributed learning, federated learning, noisy gradients, convergence rate, energy scaling law, finite-sample error bounds, large-scale wireless networks
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.