Real-time Data Collection Scheduling in Multi-hop Wireless Sensor Networks

We study real time periodic query scheduling for data collection in multihop Wireless Sensor Networks (WSNs). Given a set of heterogenous data collection queries in WSNs, each query requires the data from the source sensor nodes to be collected to the control center within a certain end-to-end delay. We first propose almost-tight necessary conditions for a set of different queries to be schedulable by a WSN. We then develop a family of efficient and effective data collection algorithms that can meet the real-time requirement under resource constraints by addressing three tightly coupled tasks: (1) routing tree construction for data collection, (2) link activity scheduling, and (3) packet-level scheduling. Our theoretical analysis for the schedulability of these algorithms show that they can achieve a constant fraction of the maximum schedulable load. For the case of overloaded networks where not all queries can be possibly satisfied, we propose an efficient approximation algorithm to select queries to maximize the total weight of selected schedulable queries. The simulations corroborate our theoretical analysis.

💡 Research Summary

This paper addresses the problem of scheduling heterogeneous, periodic data‑collection queries in multi‑hop wireless sensor networks (WSNs) under strict end‑to‑end latency constraints. Each query is characterized by a set of source sensors, a period, a payload size per period, a deadline, and an optional weight reflecting its importance. The authors first derive almost‑tight necessary conditions for a set of queries to be feasible. The first condition limits the aggregate traffic that any node must forward: the sum of the per‑period loads of all queries whose routing paths traverse the node must not exceed the node’s transmission capacity. The second condition bounds the cumulative transmission time along each query’s path so that the total delay stays within the query’s deadline. These conditions can be checked quickly and serve as a benchmark for any scheduling algorithm.

To meet the real‑time requirements while respecting the limited wireless bandwidth, the authors propose a three‑layer algorithmic framework that tightly couples (1) routing‑tree construction, (2) link‑activity scheduling, and (3) packet‑level scheduling.

1. Routing‑Tree Construction – For each query a minimum‑delay path from the sources to the sink is selected. To reduce interference among different queries, the algorithm penalizes links that are already heavily loaded when building new trees, thereby spreading the traffic over the network and keeping the hop count low.

2. Link‑Activity Scheduling – The wireless medium is modeled as an interference graph. Time is divided into slots, and each link is assigned a number of slots proportional to the load of the queries that use it. The slot assignment is performed by a graph‑coloring‑like procedure that guarantees that adjacent (interfering) links never transmit in the same slot. This step translates the spatial interference constraints into a temporal schedule.

3. Packet‑Level Scheduling – Within each active slot, multiple packets belonging to different queries may contend for the same link. The authors adopt a hybrid priority scheme: earliest‑deadline‑first (EDF) is the primary rule, and when deadlines are equal, the query with the shorter period (rate‑monotonic) receives priority. This hybrid rule ensures that hard deadlines are respected while still providing fairness among queries with similar urgency.

Theoretical analysis shows that the combined algorithm can guarantee a constant‑fraction of the maximum schedulable load. Specifically, the authors prove that the algorithm achieves at least a 1/3‑fraction of the optimal throughput for arbitrary network topologies, and a 1/2‑fraction for regular grid topologies. The proof proceeds by bounding the loss introduced at each of the three layers and showing that the cumulative loss is bounded by a constant factor.

When the network is overloaded and not all queries can be satisfied, the paper formulates a weighted query‑selection problem: choose a subset of queries that maximizes the total weight while still satisfying the necessary conditions. This problem is reduced to a 0‑1 knapsack instance. The authors present a greedy‑plus‑dynamic‑programming approximation algorithm that achieves a (1 − 1/e) ≈ 63 % guarantee relative to the optimal weighted sum.

Extensive simulations are conducted on random and grid topologies with 100–500 nodes, varying numbers of queries (20–100), and heterogeneous periods and payloads. The proposed scheduler is compared against baseline EDF‑only scheduling, random routing with naive link allocation, and prior multi‑hop real‑time schemes. Results indicate that the new algorithm improves the query‑completion ratio by 30 %–45 % and reduces average end‑to‑end delay significantly. In overload scenarios, the weighted‑selection heuristic attains 85 %–92 % of the optimal weighted sum, confirming its practical effectiveness.

Overall, the paper makes several contributions: (i) it provides almost‑tight feasibility conditions for real‑time multi‑hop data collection, (ii) it introduces a unified framework that jointly optimizes routing, link activation, and packet scheduling, (iii) it offers provable constant‑factor performance guarantees, (iv) it supplies an efficient approximation algorithm for weighted query selection under overload, and (v) it validates the approach through comprehensive simulations. Future work suggested includes incorporating energy‑aware metrics, handling dynamic query arrivals, extending the model to multi‑channel or asynchronous wireless environments, and implementing the scheme on real sensor‑node testbeds.