Federated Learning Meets Random Access: Energy-Efficient Uplink Resource Allocation

TL;DR

This work studies energy-efficient uplink resource allocation for coexisting Federated Learning (FL) and throughput-oriented random access (RA) traffic in a shared wireless uplink. It models FL uplink as FDMA with per-device rate $R_n^{\rm FL}(\rho)$ and RA uplink as ALOHA or S-ALOHA with throughput $Q^{a}(\lambda,\rho)$ and energy $E^{a}(\lambda,\rho)$, and seeks to minimize the total energy subject to an FL latency budget and a minimum RA throughput, comparing ALOHA and S-ALOHA. The authors propose a decomposition-based solution that first characterizes the optimum $\lambda^{\star}(\rho)$ and then performs a grid search over $\rho$, deriving close-to-optimal points for both MAC options. The results reveal that ALOHA yields substantial energy savings when FL energy dominates, while S-ALOHA shines when RA traffic is heavy, offering practical guidance on when to deploy which MAC protocol in mixed FL-and-RA networks.

Abstract

Artificial intelligence-generated traffic is changing the shape of wireless networks. Specifically, as the amount of data generated to train machine learning models is massive, network resources must be carefully allocated to continue supporting standard applications. In this paper, we tackle the problem of allocating radio resources for two sets of concurrent devices communicating in uplink with a gateway over the same bandwidth. A set of devices performs federated learning (FL), and accesses the medium in FDMA, uploading periodically large models. The other set is throughput-oriented and accesses the medium via random access (RA), either with ALOHA or slotted-ALOHA protocols. We derive close-to-optimal solutions to the non-convex problem of minimizing the system energy consumption subject to FL latency and RA throughput constraints. Our solutions show that ALOHA can sustain high FL efficiency, yielding up to 48% lower consumption when the system is dominated by FL traffic. On the other hand, slotted-ALOHA becomes more efficient when RA traffic dominates, yielding 6% lower consumption.

Figures (6)

• Figure 1: System model. A gateway serves two classes of devices sharing the same uplink channel: FL devices transmit model updates in FDMA and throughput-oriented devices acces the channel via RA (ALOHA / S-ALOHA).
• Figure 2: Scheme of the FL round. Instants $t_0, t_1, t_2$ and $t_3$ denote the beginning/end of different phases, while $T_{\rm cpu}^{\rm FL}, T_{\rm idle}^{\rm FL}$, and $T_{\rm tx}^{\rm FL}$ refer to the intervals of computation, idling, and transmission, respectively.
• Figure 3: Top: maximum achievable throughput of the RA protocols as a function of the bandwidth share $\rho$. Bottom: FL and RA average energy consumption as a function of the bandwidth share $\rho$. Results obtained with $N=30$ and optimized $\lambda$.
• Figure 4: Optimized performance varying the number of FL devices $N$. Left: overall system energy consumption. Right: allocated bandwidth share to FL devices $\rho$. ($q=0.178$, $\lambda^\prime=10^4$ pkts/s).
• Figure 5: Optimized performance varying the number of RA fresh transmissions $\lambda^\prime T_{\rm tot}^{\rm FL}$ ($T_{\rm tot}^{\rm FL} = 60$ s). Left: overall system energy consumption. Right: allocated bandwidth share to FL devices $\rho$. ($q=0.178$, $N=30$).