Efficient Transmission in the Presence and Absence of Data Privacy Using Digital Twins and Federated Optimization

Abstract

Mobile uplink transmission scheduling arises in many applications where devices periodically upload measurements, features, or model updates. Effective online schedulers typically rely on information that is considered private to its associated devices. For a mobile device, this may include its mobility profile, recent location(s), and experienced channel conditions. Schedulers use this type of information to perform tasks such as shared bandwidth and channel time slot assignments. In practice, however, this information may or may not be available to the scheduler due to data-privacy requirements. This thesis studies efficient transmission in both regimes while respecting practical limits on bandwidth and energy.

In the first part, we study vehicles participating in FL over a shared wireless channel and seek to minimize the duration of each update period so that aggregation can proceed as quickly as possible. We formulate the joint transmission-time scheduling and bandwidth-assignment problem for each round as a mixed-integer nonlinear program and establish its NP-completeness. To obtain practical solutions, we develop approximation methods that perform a binary search on the round duration using fractional relaxations, followed by dependent rounding to produce feasible schedules. Simulation results show near-optimal update times when compared with an exact solver on the same instances, demonstrating both efficiency and scalability in the centralized, non-private setting.

In the second part, we address scheduling when data privacy must be preserved. A Digital Twin (DT) may protect information that is considered private to its associated physical system, motivating the use of DTs to interact with the scheduler without exposing raw state information. We consider three energy-constrained transmission-scheduling problems: minimizing total transmission time under (i) fixed-power or (ii) fixed-rate time slotting with power control, and (iii) maximizing uploaded data within a fixed time horizon. We propose a real-time federated optimization framework in which the scheduler iteratively exchanges only aggregate signals with DTs to compute global fractional solutions without revealing private information; dependent rounding then yields implementable channel schedules for the physical systems. Experiments demonstrate consistent makespan reductions, near-zero bandwidth and energy violations, and millisecond-order end-to-end runtime on typical edge servers. To the best of our knowledge, this is the first framework that enables privacy-preserving channel sharing across DTs while maintaining strong performance.

Together, these results provide a unified treatment of mobile uplink scheduling under bandwidth, energy, and data privacy considerations. The work shows how joint scheduling and fractional bandwidth assignment can accelerate learning and synchronization when full state is available and how DT assisted federated optimization can retain much of this performance when privacy requires that state remain local.