Carnegie Mellon University
Harishankar_cmu_0041E_10623.pdf (20.67 MB)

Incentivizing User-centric Resource Allocation in Wireless Networks in Realtime

Download (20.67 MB)
posted on 2021-05-27, 18:35 authored by Madhumitha HarishankarMadhumitha Harishankar
In this thesis, I propose mechanisms for user-centric resource allocation in wireless networks. I consider a series of practical motivating contexts that progressively require lesser trust and reliance on the network provider and allow for more flexible connectivity schemes benefitting end-devices, especially for emerging connectivity use-cases like the IoT. The granularity of typical month-long mobile data plans is such that users must forecast their network usage over a month and assign a single monetary value to its
utility. Finer-grained real-time information about user needs does not play a role in resource allocation, though users determine their needs and launch mobile applications only in realtime. This results in unrealized value for both the end-user and the network operator and further restricts the user
to availing resources that belong only to their subscribed network(s). Inspired by Verizon’s recent PopData offering, I first consider supplementing typical monthly subscription plans with ad-hoc discount offers, wherein users may consume unlimited data for the offered hour for a small
fixed fee. This allows users to realize any additional resource needs for their sessions in realtime by utilizing these simple offers without the risk of incurring a data overage, while also affording the network a predictable
contract revenue. Second, I consider a user-driven approach to acquiring network resources by proposing a model wherein a slice of resources is dynamically created and assigned to a device based on the session needs it
specifies. Devices can then reliably estimate their session performance at the onset. I explore how these models can be made incentive-compatible for the network and the user, show that they can be executed in realtime albeit
at a steep cost to users since they are unable to plan spending optimally in realtime, and that this suboptimality can be alleviated with reinforcement learning techniques. Finally, I remove the inherent device-network trust relationship that exists in these models by allowing devices to seamlessly authenticate with any access point (without subscriptions) and make real-time payments for consumed data, using public and permissionless blockchains, in a scalable and secure manner.




Degree Type

  • Dissertation


  • Electrical and Computer Engineering

Degree Name

  • Doctor of Philosophy (PhD)


Carlee Joe-Wong Patrick Tague

Usage metrics


    Ref. manager