1: \begin{abstract}

2: % Render unto Caesar, large-scale federated learning (FL) can be achieved in a wireless network without an edge server. 

3: % We demonstrate that simple amplitude modulation and match filters can play similar role as the edge server in federated learning (FL), and hence large-scale FL can be achieved in a wireless network without an edge server. 

4: % We demonstrate that the function of an edge server in federated learning (FL) can be realized by merely analog transmissions and match filtering. 

5: We demonstrate that merely analog transmissions and match filtering can realize the function of an edge server in federated learning (FL).

6: Therefore, a network with massively distributed user equipments (UEs) can achieve large-scale FL without an edge server. 

7: % and hence realize FL in a network with massively distributed user equipments (UEs). 

8: We also develop a training algorithm that allows UEs to continuously perform local computing without being interrupted by the global parameter uploading, which exploits the full potential of UEs' processing power.

9: We derive convergence rates for the proposed schemes to quantify their training efficiency. 

10: The analyses reveal that when the interference obeys a Gaussian distribution, the proposed algorithm retrieves the convergence rate of a server-based FL. 

11: But if the interference distribution is heavy-tailed, then the heavier the tail, the slower the algorithm converges.  

12: Nonetheless, the system run time can be largely reduced by enabling computation in parallel with communication, whereas the gain is particularly pronounced when communication latency is high.

13: These findings are corroborated via excessive simulations.

14: \end{abstract}

15: