Optimal Pricing and Admission Control for Heterogeneous Secondary Users

Changkun Jiang, Lingjie Duan, Jianwei Huang

TL;DR

This paper explores how a spectrum database operator can maximize its revenue by dynamically pricing and allocating idle spectrum to unlicensed users with random, varying needs. By formulating the problem as a stochastic dynamic program, the authors demonstrate that complex dynamic admission policies often simplify to straightforward stationary rules, and that dynamic pricing can boost revenue by over 30% compared to static pricing when user demand is highly elastic.

What Problem Does This Paper Address?

Database-assisted spectrum sharing (like the FCC's TV white space initiative) allows unlicensed "secondary users" (SUs) to opportunistically access spectrum when primary licensed users are idle. For a database operator to coordinate this effectively, they need an economic incentive—specifically, a way to maximize revenue by charging SUs for access.

Previous research mostly focused on static pricing models where user demand is predictable and uniform. However, real-world SU demands are heterogeneous and stochastic. For example, a "light-traffic" SU might only need the channel for a single time slot to send a text, while a "heavy-traffic" SU might need multiple consecutive slots to download a video. Furthermore, these users arrive randomly. The core challenge is: how should the operator jointly set prices and make admission decisions over time, knowing that admitting a heavy-traffic user right now will block the channel for future, potentially more profitable, users?

Methods

The authors formulate the operator’s revenue maximization problem as a stochastic dynamic programming problem over a finite time horizon. They analyze the system under two distinct schemes:

  1. Static Pricing with Dynamic Admission: Prices are fixed at the start, but the operator can dynamically choose who to admit in each time slot. The authors solve this using standard backward induction.
  2. Joint Dynamic Pricing and Dynamic Admission: Both prices and admission rules can change over time based on demand. Because pricing and admission decisions are tightly coupled (the price affects who requests access, which affects who can be admitted), this is mathematically complex. The authors introduce a novel price-and-admission decomposition approach. In each time slot, they calculate the optimal price for every possible admission strategy (e.g., prioritizing heavy users vs. prioritizing light users), and then select the combination that yields the highest expected future revenue.

The model is initially built for two user types (1-slot vs. 2-slot occupancies) but is later extended to arbitrary slot occupancies and multiple user types.

Datasets

Not discussed in the paper. This is a theoretical and analytical study. The authors use numerical simulations based on synthetic parameters (specifically, varying the demand elasticity/price sensitivity of different user types) to validate their models and compare policies.

Key Insights and Ideas

  • Complex Policies Degenerate to Simple Rules: One of the most surprising mathematical findings is that the highly complex optimal dynamic admission policy often degenerates into a simple, threshold-based stationary (time-independent) policy. For instance, under a wide range of parameters, the optimal move is simply a static rule like "always prioritize heavy-traffic users if available."
  • The "Occupancy Gap" Shrinks Dynamic Pricing Benefits: As the difference in channel occupancy length between heavy and light users increases (e.g., a heavy user needs 5 slots instead of 2), the flexibility of dynamic pricing decreases. Because the channel gets locked up for longer periods, the operator has fewer opportunities to dynamically adjust prices, making simple static pricing nearly as effective.
  • Decomposition is Key to Tractability: The authors proved that you don't need to solve for price and admission simultaneously in a massive joint state space. By proving that admission strategies are exhaustive and mutually exclusive, they successfully decoupled the problem into bite-sized optimizations per time slot.

Key Results

  • Revenue Boost: Dynamic pricing outperforms optimal static pricing by more than 30% when both types of SUs are highly sensitive to price changes (high demand elasticity).
  • When to use Static Pricing: When users are not price-sensitive, dynamic pricing yields less than a 10% improvement. Given the high computational and implementational complexity of changing prices constantly, operators are better off using static pricing in low-elasticity markets.
  • Performance vs. Baselines: The proposed optimal dynamic policy significantly outperforms existing heuristic "switch-over" policies (e.g., a rule of thumb where an operator only admits a heavy user if their price is at least double the light user's price).

Limitations and Future Work

  • Single Channel Assumption: The current model is restricted to a single spectrum channel. In reality, operators manage multiple channels. The authors note that extending this to a multi-channel scenario requires allocating 2D time-frequency chunks, which vastly complicates the Markov Decision Process (MDP).
  • Delay-Intolerant Users: The model assumes that if an SU is not admitted immediately, they leave the system. Future work could incorporate delay-tolerant users who are willing to wait in queues, requiring a queueing-based MDP to analyze pricing and admission.

Summary updated

Abstract

This paper studies how to maximize a spectrum database operator's expected revenue in sharing spectrum to secondary users, through joint pricing and admission control of spectrum resources. A unique feature of our model is the consideration of the stochastic and heterogeneous nature of secondary users' demands. We formulate the problem as a stochastic dynamic programming problem, and present the optimal solutions under both static and dynamic pricing schemes. In the case of static pricing, the prices do not change with time, although the admission control policy can still be time-dependent. In this case, we show that a stationary (time-independent) admission policy is in fact optimal under a wide range of system parameters. In the case of dynamic pricing, we allow both prices and admission control policies to be time-dependent. We show that the optimal dynamic pricing can improve the operator's revenue by more than 30% over the optimal static pricing, when secondary users' demands for spectrum opportunities are highly elastic.
BibTeX
@article{jiang2016optimal,
  author = {Changkun Jiang and Lingjie Duan and Jianwei Huang},
  title = {{Optimal Pricing and Admission Control for Heterogeneous Secondary Users}},
  journal = {IEEE Trans. Wireless Communications},
  year = {2016},
  url = {https://openreview.net/forum?id=JedgbaRdQB},
}
Figure