Edge State Estimation in Industrial IoT: To Offload or Not to Offload?

Edge State Estimation in Industrial IoT: To Offload or Not to Offload?

In industrial IoT systems, efficient state estimation is crucial for real-time monitoring and control. This article delves into two distinct approaches for state estimation: local estimation at each sensor and centralized estimation at the network edge.

The Dilemma: Local or Edge Estimation?

Consider an industrial IoT system where each sensor faces a critical decision: to perform local state estimation or offload the task to edge estimators. Figure 3 illustrates these two scenarios:

Case A: Local Estimation

• Each sensor utilizes its own sensory data {yi(1), yi(2),..., yi(t)} to calculate a local optimal estimate using a recursive algorithm like the Kalman filter [28].* The local estimate is then transmitted to the edge estimator over shared wireless channels.* The edge estimator fuses these local estimates with data from other edge estimators to generate a final estimate.

Case B: Direct Data Delivery to Edge

• Sensors bypass local estimation and directly transmit raw sensory data to edge estimators.* Edge estimators combine all received sensory data to compute an edge estimate.* This edge estimate is then fused with estimates from other edge estimators.

Impact of Offloading Decisions on Delay and Accuracy

The choice between local and edge estimation hinges on a critical trade-off: experienced delay versus estimation accuracy.

• Delay: Local estimation introduces delay when transmitting the local estimate to the edge. Direct data delivery, on the other hand, incurs delay during data transmission to the edge.* Accuracy: Large delays degrade estimation accuracy as the information used becomes outdated.

Choosing the Right Approach

The optimal approach depends on the specific constraints of the IoT system:

• Limited Communication Resources: If bandwidth or power is scarce, local estimation (Case A) might be preferable to reduce communication overhead.* Stringent Accuracy Requirements: When high accuracy is paramount and communication resources are abundant, direct data delivery to edge estimators (Case B) becomes more suitable.

Conclusion

Selecting between local and edge-based state estimation in industrial IoT involves carefully balancing delay and accuracy. By understanding the trade-offs and considering system-specific constraints, we can design efficient and reliable state estimation strategies for next-generation industrial IoT applications.