Optimal Information Updating based on Value of Information

P. R. Kumar, Rahul Singh, Gopal Krishna Kamath, P. R. Kumar 0001

Abstract

We address the problem of how to optimally schedule data packets over an unreliable channel in order to minimize the estimation error of a simple-to-implement remote linear estimator using a constant "Kalman'' gain to track the state of a Gauss Markov process. The remote estimator receives time-stamped data packets which contain noisy observations of the process. Additionally, they also contain the information about the "quality'' of the sensor\ source, i.e., the variance of the observation noise that was used to generate the packet. In order to minimize the estimation error, the scheduler needs to use both while prioritizing packet transmissions. It is shown that a simple index rule that calculates the value of information (VoI) of each packet, and then schedules the packet with the largest current value of VoI, is optimal. The VoI of a packet decreases with its age, and increases with the precision of the source. Thus, we conclude that, for constant filter gains, a policy which minimizes the age of information does not necessarily maximize the estimator performance.

Introduction: Beyond Age of Information

In modern networked control systems—ranging from smart grids to autonomous vehicle fleets—remote estimators must track physical processes using data transmitted over unreliable channels. Traditionally, the Age of Information (AoI) has been the primary metric for optimization. The core assumption of AoI is that "fresher is better": minimizing the time elapsed since the last update was generated should, in theory, minimize estimation error.

However, this paper argues that AoI is an insufficient metric when data sources have varying levels of precision. A high-quality sensor might produce a slightly older packet that remains more informative than a brand-new packet from a noisy, low-quality sensor. This research introduces the Value of Information (VoI) as a superior metric for scheduling, balancing both data freshness and measurement quality.

System model for a single process with multiple sources and a remote estimator.

The Value of Information Index

The authors model the physical state as a Gauss-Markov process. The remote estimator uses a linear filter with a constant gain to track this state. To optimize scheduling, the paper proposes a VoI Index for each packet $\psi$ in the queue at time $t$. This index, $W^2_\psi(t)$, quantifies the expected estimation error and is composed of two primary factors:

  1. Sensing Noise: The inherent variance or "quality" of the source that generated the packet.
  2. Process Noise: The uncertainty that accumulates as the packet ages, reflecting how much the physical process has changed since the measurement was taken.

The proposed Index Policy is a "greedy" approach: at each transmission opportunity, the scheduler selects the packet with the highest Value of Information (which corresponds to the lowest expected estimation error index).

Proving Optimality

A significant contribution of this work is the mathematical proof that this simple index-based rule is globally optimal. Using a Markov Decision Process (MDP) framework and backward induction, the authors demonstrate that prioritizing the packet with the lowest current error index minimizes the cumulative squared estimation error over an infinite time horizon.

Crucially, the index is shown to be time-consistent. If Packet A is superior to Packet B at arrival, it remains superior as they both age in the queue. This property reduces the computational complexity of the scheduler to $O(\log n)$, as it only needs to maintain an ordered queue rather than re-calculating indices for every packet at every time step.

Scaling to Multi-Process Systems

The paper extends this logic to more complex scenarios where multiple independent processes compete for the same communication resources.

Extension to multi-process systems sharing a single communication link.

By applying the index rule across different process queues, the authors show that the "curse of dimensionality" typically associated with multi-process scheduling can be mitigated. The system can be decomposed into manageable sub-problems where the scheduler simply compares the top VoI indices from each process's queue to decide which packet to transmit next.

Conclusion and Future Directions

This research provides a rigorous theoretical bridge between information theory and control. By formalizing Value of Information, it demonstrates that minimizing data age is not always equivalent to maximizing estimator performance.

While the paper provides a strong theoretical foundation, future work remains to extend these proofs to systems with time-varying Kalman gains, vector-valued processes, and more complex multi-hop network topologies. Additionally, practical implementations may need to account for scenarios where source noise variances are not known a priori and must be estimated dynamically.

Summary updated

BibTeX
@inproceedings{kumar2019optimal,
  author = {P. R. Kumar and Rahul Singh and Gopal Krishna Kamath and P. R. Kumar},
  title = {{Optimal Information Updating based on Value of Information}},
  booktitle = {Allerton},
  year = {2019},
  doi = {10.1109/allerton.2019.8919810},
}
Figure