[Paper Review] Greedy Multi-Channel Neighbor Discovery

Niels Karowski, Konstantin Miller+1 more|arXiv (Cornell University)|Jul 13, 2018

Anomaly Detection Techniques and Applications20 references4 citations

TL;DR

This paper proposes a family of low-complexity greedy algorithms—GREEDY and CHAN TRAIN—for passive, asynchronous neighbor discovery in multi-channel wireless networks using periodic beaconing. The algorithms generate recursive listening schedules that guarantee complete discovery while simultaneously minimizing worst-case discovery time, mean discovery time, and maximizing the number of discoveries over time, outperforming IEEE 802.15.4's Passive Scan by up to 4x in mean discovery time and 300% in neighbor discovery rate under realistic conditions.

ABSTRACT

The accelerating penetration of physical environments by objects with information processing and wireless communication capabilities requires approaches to find potential communication partners and discover services. In the present work, we focus on passive discovery approaches in multi-channel wireless networks based on overhearing periodic beacon transmissions of neighboring devices which are otherwise agnostic to the discovery process. We propose a family of low-complexity algorithms that generate listening schedules guaranteed to discover all neighbors. The presented approaches simultaneously depending on the beacon periods optimize the worst case discovery time, the mean discovery time, and the mean number of neighbors discovered until any arbitrary in time. The presented algorithms are fully compatible with technologies such as IEEE 802.11 and IEEE 802.15.4. Complementing the proposed low-complexity algorithms, we formulate the problem of computing discovery schedules that minimize the mean discovery time for arbitrary beacon periods as an integer linear problem. We study the performance of the proposed approaches analytically, by means of numerical experiments, and by extensively simulating them under realistic conditions. We observe that the generated listening schedules significantly - by up to factor 4 for the mean discovery time, and by up to 300% for the mean number of neighbors discovered until each point in time - outperform the Passive Scan, a discovery approach defined in the IEEE 802.15.4 standard. Based on the gained insights, we discuss how the selection of the beacon periods influences the efficiency of the discovery process, and provide recommendations for the design of systems and protocols.

Motivation & Objective

• To design low-complexity, energy-efficient neighbor discovery algorithms compatible with IEEE 802.11 and IEEE 802.15.4.
• To guarantee complete discovery of all neighbors in asynchronous, multi-channel environments without time synchronization.
• To simultaneously optimize worst-case discovery time (WDT), mean discovery time (MDT), and number of discoveries over time (NDoT).
• To characterize the class of listening schedules that are complete and optimal for all three metrics.
• To provide design recommendations for beacon period selection to balance flexibility and discovery efficiency.

Proposed method

• Introduces recursive listening schedules that are complete and pointwise optimal for the cumulative distribution function of discovery times.
• Develops the GREEDY family of algorithms that construct recursive schedules under assumptions on beacon period sets (F3), ensuring optimality for WDT, MDT, and NDoT.
• Proposes CHAN TRAIN as an extension of GREEDY to reduce channel switching overhead.
• Formulates the minimum MDT discovery problem as an integer linear program (ILP) for offline computation in general cases.
• Analyzes performance across three beacon period families: F1 (all possible sets), F2 (IEEE 802.15.4 compatible), and F3 (extended, IEEE 802.11 compatible).
• Employs analytical evaluation, numerical experiments, and extensive simulations under realistic conditions to validate performance.

Experimental results

Research questions

• RQ1 Can listening schedules be constructed that simultaneously minimize worst-case discovery time, mean discovery time, and maximize the number of discoveries over time?
• RQ2 What is the class of complete listening schedules that are optimal for all three discovery metrics?
• RQ3 How do different beacon period sets (F1, F2, F3) affect the performance and feasibility of optimal discovery strategies?
• RQ4 Can low-complexity algorithms achieve near-optimal performance while remaining practical for resource-constrained IoT devices?
• RQ5 What design principles should guide beacon period selection to balance flexibility and discovery efficiency?

Key findings

• The GREEDY and CHAN TRAIN algorithms achieve up to a 4-fold reduction in mean discovery time compared to IEEE 802.15.4's Passive Scan.
• The algorithms improve the number of neighbors discovered per unit time by up to 300% compared to Passive Scan.
• For beacon period sets in family F3, the algorithms generate recursive schedules that are optimal for WDT, MDT, and NDoT simultaneously.
• The ILP-based approach minimizes MDT in the general case but suffers from high computational complexity, limiting it to offline use in moderate-sized networks.
• Family F3 provides the best balance of flexibility and efficiency, supporting optimal discovery with low complexity.
• Family F2 allows for completeness and WDT-optimality with near-optimal MDT and NDoT performance, while F1 does not guarantee optimality for MDT or NDoT.