Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Coherence-Aware Distributed Learning under Heterogeneous Downlink Impairments

Mehdi Karbalayghareh, David J. Love, Christopher G. Brinton

TL;DR

This work tackles the challenge of downlink impairments in federated learning over wireless networks with heterogeneous channel coherence times. It develops a coherence-aware framework that uses product superposition to overlay pilots and global model updates, enabling static devices to receive full updates while dynamic devices obtain partial updates via estimated virtual channels. The authors provide a convergence analysis under imperfect CSI and propose PLMF for handling missing parameters, along with optimal pilot-data power allocation. Numerical experiments on MNIST and CIFAR-10 demonstrate substantial gains in exploration efficiency and training accuracy, validating the method's practical impact for FL in heterogeneous wireless environments.

Abstract

The performance of federated learning (FL) over wireless networks critically depends on accurate and timely channel state information (CSI) across distributed devices. This requirement is tightly linked to how rapidly the channel gains vary, i.e., the coherence intervals. In practice, edge devices often exhibit unequal coherence times due to differences in mobility and scattering environments, leading to unequal demands for pilot signaling and channel estimation resources. Conventional FL schemes that overlook this coherence disparity can suffer from severe communication inefficiencies and training overhead. This paper proposes a coherence-aware, communication-efficient framework for joint channel training and model updating in practical wireless FL systems operating under heterogeneous fading dynamics. Focusing on downlink impairments, we introduce a resource-reuse strategy based on product superposition, enabling the parameter server to efficiently schedule both static and dynamic devices by embedding global model updates for static devices within pilot transmissions intended for mobile devices. We theoretically analyze the convergence behavior of the proposed scheme and quantify its gains in expected communication efficiency and training accuracy. Experiments demonstrate the effectiveness of the proposed framework under mobility-induced dynamics and offer useful insights for the practical deployment of FL over wireless channels.

Coherence-Aware Distributed Learning under Heterogeneous Downlink Impairments

TL;DR

This work tackles the challenge of downlink impairments in federated learning over wireless networks with heterogeneous channel coherence times. It develops a coherence-aware framework that uses product superposition to overlay pilots and global model updates, enabling static devices to receive full updates while dynamic devices obtain partial updates via estimated virtual channels. The authors provide a convergence analysis under imperfect CSI and propose PLMF for handling missing parameters, along with optimal pilot-data power allocation. Numerical experiments on MNIST and CIFAR-10 demonstrate substantial gains in exploration efficiency and training accuracy, validating the method's practical impact for FL in heterogeneous wireless environments.

Abstract

The performance of federated learning (FL) over wireless networks critically depends on accurate and timely channel state information (CSI) across distributed devices. This requirement is tightly linked to how rapidly the channel gains vary, i.e., the coherence intervals. In practice, edge devices often exhibit unequal coherence times due to differences in mobility and scattering environments, leading to unequal demands for pilot signaling and channel estimation resources. Conventional FL schemes that overlook this coherence disparity can suffer from severe communication inefficiencies and training overhead. This paper proposes a coherence-aware, communication-efficient framework for joint channel training and model updating in practical wireless FL systems operating under heterogeneous fading dynamics. Focusing on downlink impairments, we introduce a resource-reuse strategy based on product superposition, enabling the parameter server to efficiently schedule both static and dynamic devices by embedding global model updates for static devices within pilot transmissions intended for mobile devices. We theoretically analyze the convergence behavior of the proposed scheme and quantify its gains in expected communication efficiency and training accuracy. Experiments demonstrate the effectiveness of the proposed framework under mobility-induced dynamics and offer useful insights for the practical deployment of FL over wireless channels.

Paper Structure

This paper contains 22 sections, 1 theorem, 44 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Related Works
  3. Contributions
  4. System Model
  5. Channel Model
  6. Baseline Transmission Scheme
  7. Coherence-Aware Device Scheduling and Communication Protocol
  8. Downlink Signaling: Integrated Pilot-Parameter Broadcast
  9. Pilot-Parameter Power Allocation
  10. Uplink Model Aggregation
  11. Convergence Analysis
  12. Preliminaries
  13. Assumptions
  14. Numerical Experiments
  15. Simulation Setup
  16. ...and 7 more sections

Key Result

Theorem 1

Under Assumptions 1–5, for a non-convex L-smooth global loss function, if the learning rate is chosen such that $\eta_\ell \le \frac{1}{2L\tau}$, the product superposition FL scheme using PLMF over $T$ rounds of training satisfies where $\eta_g = \eta_\ell\tau$ is the effective global learning rate (assuming $\eta_\ell = \eta_{k,i}, \forall k \in [K], \forall i \in [\tau]$), and the irreducible e

Figures (5)

  • Figure 1: Wireless FL scenario considered. Our focus is on studying and mitigating the impact of downlink impairments.
  • Figure 2: Heterogeneous link coherence intervals.
  • Figure 3: Training loss versus communication rounds on for the proposed product superposition-based FL scheme (MNIST).
  • Figure 4: Test accuracy comparison between the proposed scheme and conventional baselines, for the MNIST dataset. (a) and (b) show test accuracy versus normalized communication cost at $\lambda = 0.2$ and $\lambda = 0.3$, respectively. (c) presents test accuracy as a function of $\lambda$ with a fixed training duration of $T = 20$.
  • Figure 5: Test accuracy versus normalized communication cost on the CIFAR-10 dataset for the proposed product superposition-based FL, and the conventional FL with ordinary pilots: (a) SNR = 10 dB, and (b) SNR = 30 dB. Shaded regions indicate the standard deviation of the test accuracy.

Theorems & Definitions (4)

  • Remark 1
  • Remark 2
  • Remark 3
  • Theorem 1