Queueing Theory Models Asynchronous Event-Driven Readout, Matching Prototype Latency at Occupancy

Modern imaging systems increasingly rely on event-driven sensors, but designing efficient readout architectures for these devices presents significant challenges. Dominik S. Górni and Grzegorz W. Deptuch, both from AGH University of Krakow, now present a new analytical framework that models these complex systems using queueing theory. Their work overcomes limitations in existing models by accurately predicting performance across a wide range of operating conditions, from low to high traffic, and importantly, links architectural choices directly to key performance metrics. This achievement enables designers to rapidly optimise critical parameters such as acknowledge period, tiling strategies, and link counts, ultimately paving the way for faster, more efficient imaging systems. The resulting model provides a powerful tool for selecting the best architectural configurations under practical constraints, improving the performance of event-driven sensors.

Researchers acknowledge the need to serialize requests and present an analytical framework that models the system as a queue with deterministic characteristics, implementing losses at the source through one-slot gating. The framework links admitted rates, loss probabilities, utilization, and mean sojourn times with self-consistent relations, achieving a closed-form solution that separates fixed path delay from queueing effects. This allows for rapid sizing during the design phase, while the framework accurately matches post-layout results from a physical prototype across varying traffic loads, reproducing saturation behavior and observed latency growth, unlike traditional abstractions.

Asynchronous Imager Performance via Queueing Analysis

Scientists developed an analytical framework to model event-driven imagers and arrays, focusing on the performance of asynchronous arbiter trees. The study represents the system’s root as a queue with deterministic characteristics and implements losses at the source through one-slot gating, effectively controlling the flow of information. Researchers modeled admitted rates, loss probabilities, utilization, and mean sojourn times using self-consistent relations, achieving a closed-form solution that separates fixed path delay from queueing effects, a crucial step in understanding system performance. The team validated the model against post-layout results from a physical prototype, demonstrating its accuracy across a range of traffic intensities.

To assess performance, the study employed a deterministic approach, modeling the root as a queue with a Poisson arrival process and deterministic service time, allowing for precise calculations of queueing delays. Scientists leveraged the algebraic nature of these relations to rapidly size the system during the design phase, exploring the impact of partitioning into independent tiles. Reducing fan-in lowers arbitration depth and fixed path delay, decreases loss, and improves latency at a fixed acknowledge period, T, with throughput adding across tiles, demonstrating a clear pathway to optimization. This approach allows designers to predict system behavior and avoid saturation while meeting stringent timing resolution requirements, reducing reliance on time-consuming end-to-end simulations.

Queueing Model Predicts Sensor Array Performance Limits

Scientists developed an analytical framework to model event-driven sensor arrays, achieving a detailed understanding of performance limitations and design trade-offs. The work centers on modeling the system as a queue with deterministic elements and implementing losses at the source through one-slot gating. This model successfully links architectural parameters to performance metrics, enabling rapid sizing during the design process. Experiments demonstrate the framework matches post-layout results of a physical prototype across varying traffic loads, reproducing saturation and accurately predicting observed latency growth.

Validation against prototype data confirms the model’s predictive power, even under heavy traffic conditions where classical abstractions diverge. Researchers discovered that reducing fan-in lowers arbitration depth and fixed path delay, demonstrably decreasing loss and improving latency at a fixed acknowledge period, T. Throughput increases linearly across independent tiles, demonstrating scalability. The framework allows for exploration of architectural trade-offs, including the number of arbitration trees, acknowledge period, and buffering policies, revealing how appropriately chosen parameters can achieve sub-microsecond and even tens-of-nanosecond effective timing resolution, a significant improvement over frame-based approaches.

Queueing Model Predicts Detector Performance Rapidly

This research presents a new analytical framework for modelling event-driven systems, commonly used in modern radiation detectors, particle trackers and intelligent sensor arrays. The team developed a model that accurately predicts system performance based on key architectural parameters, offering a valuable tool for design optimisation. By modelling the system as a queue with deterministic elements and incorporating source-level losses, the framework establishes clear relationships between the admitted rate, loss probability, system utilisation and the mean time an event spends in the system. The resulting equations provide a closed-form solution that separates the fixed path delay from queueing effects, allowing designers to rapidly assess trade-offs between factors like the number of arbitration trees, acknowledge period and buffering policies. Validation against post-layout results from a physical prototype demonstrates the model’s accuracy across a range of traffic conditions, accurately reproducing observed saturation points and latency growth where other abstractions fail.