1: \begin{abstract}

2: In this work, we propose a communication-efficient hierarchical federated learning algorithm for distributed setups including core servers and multiple edge servers with clusters of devices. Assuming different learning tasks, clusters with a same task collaborate. To implement the algorithm over wireless links, we propose a scalable clustered over-the-air aggregation scheme for the uplink with a bandwidth-limited broadcast scheme for the downlink that requires only a single resource block for each algorithm iteration, independent of the number of edge servers and devices. This setup is faced with interference of devices in the

3: uplink and interference of edge servers in the downlink that are to

4: be modeled rigorously. We first develop a spatial model for the setup

5: by modeling devices as a Poisson cluster process over the edge

6: servers and quantify

7: uplink and downlink error terms due to the interference. Accordingly, we present a comprehensive mathematical approach to derive the convergence bound for the proposed algorithm including any number of collaborating clusters and provide special cases and design remarks. Finally,

8: we show that despite the interference and data heterogeneity, the proposed algorithm

9: not only achieves high learning accuracy for a variety of parameters but also significantly outperforms the conventional hierarchical learning algorithm.

10: 

11: \end{abstract}

12: