⚡ electrical engineering

Multi-Sensor Scheduling for Remote State Estimation over Wireless MIMO Fading Channels with Semantic Over-the-Air Aggregation

This paper proposes a novel multi-sensor scheduling framework for remote state estimation over MIMO fading channels that leverages semantic over-the-air aggregation and a tractable dynamic programming approach to achieve superior estimation accuracy and power efficiency compared to existing methods.

Original authors: Minjie Tang, Photios A. Stavrou, Marios Kountouris

Published 2026-02-06

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Minjie Tang, Photios A. Stavrou, Marios Kountouris

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to guess the exact location of a runaway balloon (the "state") that is drifting through a windy, chaotic city. You have a team of M sensors scattered around the city, each holding a camera to take pictures of the balloon. Your goal is to figure out where the balloon is as accurately as possible, but you have two big problems:

The Wind (Fading Channels): The air is turbulent. Sometimes the signal from a camera gets blocked or distorted before it reaches your computer.
The Battery (Power Limits): Every time a camera takes a picture and sends it, it uses up battery. If all 100 cameras shout their pictures at you at the exact same time, they might crash into each other (causing confusion), or they might drain their batteries so fast they die before the job is done.

This paper proposes a clever new way to manage this team of sensors, called SemOTA (Semantic Over-the-Air Aggregation). Here is how it works, broken down into simple concepts:

1. The Old Way: Everyone Shouts at Once

In traditional methods, all sensors try to send their data simultaneously to save time.

The Problem: It's like a crowded room where everyone is shouting at once. While you get a lot of information, the noise is overwhelming, and everyone burns through their batteries quickly. If you have too many sensors, the system becomes inefficient and expensive to run.

2. The New Way: The "Smart Editor" (SemOTA)

The authors suggest a system where the sensors don't just shout randomly. Instead, they act like a team of editors with a Smart Editor (the remote estimator) in charge.

Semantic Meaning: The system doesn't care about every picture. It only cares about the pictures that actually help fix the guess about the balloon's location. If a sensor sees something that doesn't change your guess much (like a picture of a tree when you already know the balloon is in the sky), it stays silent.
The "Over-the-Air" Magic: When the sensors do decide to speak, they don't wait in line (which takes time). They all speak at the exact same moment, but because they are "semantic," their voices blend together in the air to form a single, clearer message, rather than a messy shout.

3. The Decision Process: The "Game Plan"

How does the system decide who speaks and who stays quiet?

The Dynamic Game: The authors modeled this as a complex game played over a set period (like a 100-round match). The goal is to win the game (get the best guess) while spending the least amount of battery points.
The "Q-Function" (The Scorecard): They created a mathematical "scorecard" that predicts: "If I let Sensor A talk right now, will it help me more than it costs in battery?"
The Shortcut: Calculating the perfect scorecard for every single second is too hard for a computer to do in real-time. So, the authors invented a shortcut (using a mathematical trick called "PSD cone decomposition"). This shortcut gives them a "good enough" rule that is fast to calculate but still very smart.

4. The Results: Smarter and Cheaper

The paper tested this new system against three other common methods:

Random Shouting: Sensors shout if their picture is "loud" enough (but they often crash into each other).
Waiting in Line: Sensors take turns (but this takes too long and misses the balloon).
Everyone Shouts: All sensors talk at once (accurate but drains batteries).

The Findings:

Accuracy: The new SemOTA system was almost as accurate as "Everyone Shouts," but much better than the random or waiting methods. It reduced the error in guessing the balloon's location by a huge margin (over 100 times better in some cases).
Battery Life: Because it only lets the most helpful sensors speak, it used 10 times less power than the other methods.

Summary Analogy

Think of the old system as a chaotic town hall meeting where 100 people try to give advice at once. It's loud, confusing, and exhausting.

The new SemOTA system is like a smart moderator. The moderator listens to the room, figures out who has the most important news right now, and lets only those few people speak simultaneously. Their voices blend together perfectly to give a clear answer, while the other 95 people save their energy for when they are truly needed.

The Bottom Line: This paper proves that by being "smart" about who talks and when, you can get a much clearer picture of the world without burning out your batteries.

Technical Summary: Multi-Sensor Scheduling for Remote State Estimation over Wireless MIMO Fading Channels with Semantic Over-the-Air Aggregation

Problem Statement

The paper addresses the challenge of remote state estimation for unstable dynamic plants monitored by multiple geographically distributed sensors over wireless multiple-input multiple-output (MIMO) fading channels. The core problem lies in balancing two conflicting objectives: maximizing estimation accuracy and minimizing transmit power consumption.

Existing approaches face significant limitations:

Dedicated Resource Allocation: Traditional schemes (e.g., ALOHA, TDMA) assign dedicated resources to sensors, leading to low spectral efficiency and performance degradation due to packet collisions or access latency.
Conventional Over-the-Air (OTA) Aggregation: While OTA allows simultaneous analog transmission to improve spectral efficiency and reduce latency, it requires all active sensors to transmit simultaneously. This imposes a prohibitive transmit power burden that scales with the number of sensors, making it energy-inefficient for large-scale networks.
Existing Scheduling Strategies: Current sensor scheduling methods often rely on measurement thresholds or error covariance thresholds. These approaches can lead to redundant transmissions or fail to account for measurement heterogeneity and channel variations, resulting in suboptimal power usage and estimation performance.

The authors propose a framework to optimize sensor scheduling that leverages Semantic Over-the-Air (SemOTA) aggregation. The goal is to selectively activate only the most informative sensors at any given time to maintain high estimation accuracy while drastically reducing power consumption, all within the context of random MIMO fading channels.

Methodology

The proposed methodology integrates semantic communication principles with dynamic programming to solve a stochastic optimization problem.

System Model:
- Plant Dynamics: Modeled as a first-order linear time-invariant system with Gaussian process noise.
- Communication: Sensors transmit measurements via a shared MIMO fading channel using OTA aggregation. The received signal is a superposition of channel-scaled measurements plus additive white Gaussian noise (AWGN).
- Estimation: A remote estimator employs a Kalman filter to compute the minimum mean-squared error (MMSE) estimate of the plant state based on the aggregated noisy measurements.
Problem Formulation:
The sensor scheduling task is formulated as a finite-horizon stochastic dynamic programming (DP) problem. The objective is to minimize a cost function comprising:
- The trace of the estimation error covariance (representing estimation accuracy).
- The weighted sum of transmit powers (representing energy consumption).
  The decision variables are binary scheduling indicators ( $\delta_{m,k}$ ) determining whether sensor $m$ transmits at time slot $k$ .
Optimal Solution Analysis:
- The authors derive the optimal solution using backward induction and Q-functions.
- Semantic Structure: Analysis of the optimal Q-function reveals that the optimal scheduling policy possesses a "semantic structure." This means the decision to activate a sensor depends dynamically on the real-time estimation error covariance and the instantaneous channel state information (CSI). Sensors are activated only when their marginal contribution to estimation accuracy justifies the power cost.
Approximate Solution (Tractable Algorithm):
Since the exact DP solution involves computationally expensive iterative expectations, the authors derive a tractable upper bound on the Q-function.
- PSD Cone Decomposition: By applying a positive semidefinite (PSD) cone decomposition to the aggregated channel gain, they obtain a closed-form upper bound.
- Algorithm 1: Based on this bound, an efficient approximate scheduling algorithm is developed. This algorithm computes scheduling decisions at the remote estimator and broadcasts them to the sensors, who then transmit only if selected.

Key Contributions

The paper outlines four primary contributions:

SemOTA Scheme for MIMO: Development of a SemOTA framework tailored for MIMO fading channels. Unlike conventional OTA, it improves energy efficiency by prioritizing informative measurements without sacrificing the zero-access-latency and high-spectral-efficiency benefits of OTA.
DP Formulation and Structural Analysis: Formulation of the scheduling problem as a finite-horizon stochastic DP problem. The authors characterize the structural properties of the optimal Q-function, demonstrating that the optimal policy adapts to real-time variations in both error covariance and channel states.
Tractable Approximation: Derivation of a closed-form upper bound on the Q-function using PSD cone decomposition. This enables the development of a low-complexity, approximate scheduling policy and a corresponding remote estimation algorithm.
Performance Validation: Simulation results confirming that the proposed scheme outperforms existing baselines in both estimation accuracy and power efficiency.

Numerical Results

The proposed SemOTA scheme was evaluated against three baselines:

Baseline 1: Measurement threshold with ALOHA.
Baseline 2: Error covariance threshold with random TDMA.
Baseline 3: Conventional OTA (all sensors active).

Findings:

Estimation Accuracy: As the number of sensors increases, Baseline 1 and Baseline 2 suffer severe performance degradation due to collisions and latency. The proposed SemOTA scheme maintains robust performance, achieving more than a two-order-of-magnitude reduction in Normalized Mean Square Error (NMSE) compared to the threshold-based baselines. While it exhibits slightly higher NMSE than Baseline 3 (full OTA), it remains within the same order of magnitude.
Power Efficiency: The proposed scheme operates at least one order of magnitude lower in average transmit power compared to all baselines. This is attributed to the selective transmission of only critical sensor measurements.

Significance and Claims

The paper claims that the proposed SemOTA-based scheduling framework effectively bridges the gap between the high spectral efficiency of OTA aggregation and the energy constraints of wireless sensor networks. By introducing a "semantic" layer that prioritizes information content over raw transmission, the method achieves a superior trade-off between estimation accuracy and power consumption.

The authors emphasize that their approach preserves the key advantages of conventional OTA (zero access latency, high spectral efficiency) while mitigating its primary drawback (high power demand). The derivation of a tractable upper bound allows for practical implementation, making the solution viable for real-time remote state estimation in dynamic MIMO environments. The results suggest that semantic-aware scheduling is a critical enabler for scalable, energy-efficient industrial monitoring and autonomous systems.