Goal-oriented Communications for the IoT: System Design and Adaptive Resource Optimization

Paolo Di Lorenzo, Mattia Merluzzi, Francesco Binucci, Claudio Battiloro, Paolo Banelli, Emilio Calvanese Strinati, Sergio Barbarossa

TL;DR

This paper proposes a fundamental shift in how we design Internet of Things (IoT) networks: moving from traditional "bit-perfect" communication to Goal-Oriented (GO) communication. By optimizing the entire pipeline—from sensing to computation—to fulfill a specific task (like image classification or field reconstruction) rather than just reconstructing raw data, the authors demonstrate that IoT systems can achieve massive gains in energy efficiency and resource management.

What Problem Does This Paper Address?

Traditional wireless communication is built on the Shannon paradigm: the goal is to deliver every bit of data from point A to point B with zero errors. However, in the modern IoT era, data is often collected for a specific purpose, such as an AI model making a decision at the "edge."

Reconstructing every pixel of a high-resolution image is incredibly wasteful if the goal is simply to detect if a specific object is present. This "bit-level" obsession creates a bottleneck in resource-constrained environments where spectrum, energy, and computing power are scarce. The authors tackle this by asking: How can we optimize the network to fulfill the goal with the minimum possible cost?

Methods

The authors introduce a framework consisting of two main components:

  1. GO-COM (Goal-Oriented Communication Module): Located at the IoT device, this module "distills" data. Instead of sending raw samples, it extracts only the features necessary for the task. This might involve task-specific compression (like a learned convolutional encoder) or "effective sensing," where the device only samples data when it is truly informative.
  2. GO-OPT (Goal-Oriented Optimization Engine): This is the "brain" of the network management. It uses Lyapunov stochastic optimization and Deep Reinforcement Learning (DRL) to dynamically allocate resources. It decides on CPU clock frequencies, transmit power, and compression factors in real-time to balance "Goal Effectiveness" (e.g., accuracy, latency) against "Goal Cost" (e.g., energy, bandwidth).

The beauty of this approach is its cross-layer design: it doesn't just look at the radio link; it integrates sensing, communication, computation, and learning into a single optimization problem.

Datasets

The paper evaluates its framework across three distinct use cases:

  • Edge Inference: Uses simulated image data processed through a Convolutional Encoder.
  • Cooperative Sensing: Uses a synthetic signal field synthesized from a Fourier basis to simulate spatially correlated physical phenomena.
  • Federated Learning: Uses the standard MNIST dataset (handwritten digits) to demonstrate how the system handles distributed model training under strict latency and accuracy constraints.

Key Insights and Ideas

  • Effectiveness over Fidelity: The paper formalizes "Goal Effectiveness" as a metric. A communication is successful if the task is completed (e.g., the car brakes in time), even if the underlying data transmission was "lossy" by traditional standards.
  • The "Aha" of Feature Extraction: In the edge inference case, the authors show that a task-specific Convolutional Encoder (CE) vastly outperforms traditional compression like JPEG or simple downsampling because it only preserves features that matter to the AI classifier.
  • Dynamic Adaptation: The system is "AI-native." It doesn't just follow a fixed rule; it learns to adapt. For example, in the Federated Learning test, when the accuracy requirement was bumped from 0.7 to 0.8 mid-run, the GO-OPT engine automatically increased power consumption to meet the new goal.

Key Results

  • 2x Efficiency Gain: In edge inference scenarios, the proposed GO strategies achieved a roughly 2x gain in cost-effectiveness compared to conventional downsampling schemes.
  • Resource Savings in Sensing: For cooperative sensing, the framework significantly reduced the number of active sensors needed to reconstruct a field, especially when the data was highly correlated.
  • Latency Guarantees: In the Federated Learning use case, the system successfully maintained a strict 0.2-second latency per iteration while still meeting evolving accuracy targets, proving it can handle time-critical applications.

Limitations and Future Work

While the results are promising, the authors acknowledge several open frontiers:

  • Theoretical Foundations: We still need a robust information-theoretic framework that fully encapsulates "semantic" and "goal-oriented" characteristics beyond simple rate-distortion theory.
  • Time-Varying Goals: The current work assumes the goal is relatively stable. Future research needs to address scenarios where the communication goal changes dynamically over time (e.g., a drone switching from "search" mode to "rescue" mode).
  • Causality and Value of Information: Exploring how the timing of data affects its value for decision-making remains a complex, unsolved piece of the puzzle.

Summary updated

Abstract

Internet of Things (IoT) applications combine sensing, wireless communication, intelligence, and actuation, enabling the interaction among heterogeneous devices that collect and process considerable amounts of data. However, the effectiveness of IoT applications needs to face the limitation of available resources, including spectrum, energy, computing, learning and inference capabilities. This paper challenges the prevailing approach to IoT communication, which prioritizes the usage of resources in order to guarantee perfect recovery, at the bit level, of the data transmitted by the sensors to the central unit. We propose a novel approach, called goal-oriented (GO) IoT system design, that transcends traditional bit-related metrics and focuses directly on the fulfillment of the goal motivating the exchange of data. The improvement is then achieved through a comprehensive system optimization, integrating sensing, communication, computation, learning, and control. We provide numerical results demonstrating the practical applications of our methodology in compelling use cases such as edge inference, cooperative sensing, and federated learning. These examples highlight the effectiveness and real-world implications of our proposed approach, with the potential to revolutionize IoT systems.
BibTeX
@article{lorenzo2023goal,
  author = {Paolo Di Lorenzo and Mattia Merluzzi and Francesco Binucci and Claudio Battiloro and Paolo Banelli and Emilio Calvanese Strinati and Sergio Barbarossa},
  title = {{Goal-oriented Communications for the IoT: System Design and Adaptive Resource Optimization}},
  journal = {IEEE Internet Things Mag.},
  year = {2023},
  doi = {10.1109/iotm.001.2300163},
}
Figure