Analytically Characterized Optimal Power Control for Signal-Level-Integrated Sensing, Computing and Communication in Federated Learning

summary

The gist

Signal-level integrated sensing, computing and communication (Sig-ISCC) is envisioned where a single waveform simultaneously supports sensing…

In short

The work addresses designing optimal transmit powers and receive scaling factors for Signal-level Integrated Sensing, Computing, and Communication (Sig-ISCC) in Federated Learning (FL). It proposes a polynomial-time algorithm that transforms a complex non-convex problem into an equivalent convex one. This method ensures that edge devices can simultaneously optimize their signals to minimize communication distortion while satisfying joint target detection requirements.

Key concepts

Signal-level Integrated Sensing, Computing and Communication (Sig-ISCC)
This concept envisions a single radio waveform that performs three tasks at once: sensing (detecting targets), computing (federated learning), and communication. The system uses over-the-air computation where edge devices transmit gradients analogously, and the base station rescales the signal to perform computations.
Joint Power-Control and Receive Scaling Problem (P0)
This is the core optimization problem that seeks to find the best combination of transmit powers for each edge device ($p_k$) and a receive scaling factor ($\alpha$). The goal is to minimize the Mean Squared Error of the aggregated signal while ensuring that a specific joint target detection requirement is met, all under maximum power constraints.
Variable Transformation
The original optimization problem is non-convex, making it very hard to solve. The authors introduce a specific mathematical transformation that maps the original variables (like transmit powers) into new variables. This transformation converts the difficult non-convex problem into an equivalent convex optimization problem, which is much easier and guaranteed to find a global optimum.
Polynomial-time Algorithm
This refers to an efficient computational method that solves a problem in time proportional to a polynomial function of the input size. The proposed algorithm is characterized by iterating over specific intervals and checking conditions, guaranteeing that it finds the globally optimal solution quickly, even when using approximate numerical methods for root-finding.

Terminology used across episodes

This episode discusses

The paper

Analytically Characterized Optimal Power Control for Signal-Level-Integrated Sensing, Computing and Communication in Federated Learning · Read on arXiv

Paul Zheng, Yao Zhu, Xiaopeng Yuan, Yulin Hu, Anke Schmeink

RWTH Aachen University · Wuhan University

In next-generation wireless edge systems, efficient functionality integration is essential to address the growing demands of communication, computation, and sensing. Signal-level integrated sensing, computing, and communication (Sig-ISCC) is envisioned, where a single waveform simultaneously supports sensing, computing and communication via over-the-air computation (AirComp). Meanwhile, federated learning (FL) is widely regarded as a promising distributed machine learning framework that enables network intelligence in a privacy-preserving and secure manner, and exhibits strong synergy with AirComp, which alleviates the communication bottleneck of FL. In this paper, we study uplink Sig-ISCC design for AirComp-FL with joint target detection. We formulate the joint power and receive-scaling control problem, where edge devices' transmitted signals should serve both sensing and AirComp purposes. The goal is to minimize the AirComp aggregation distortion subject to a joint target-detection requirement. Although the resulting problem is non-convex in the original variables, we show that it admits an equivalent convex reformulation after a suitable variable transformation. By exploiting analytical optimality properties, we develop a robust, optimal, and polynomial-time-complexity algorithm that efficiently achieves the optimal transmit powers and receive scaling factor. Simulation results validate the optimality and numerical robustness of the proposed algorithm and show its superior FL performance compared to baseline methods.

DOI: 10.1109/TWC.2026.3747602

Transcript

Introduction to the show: ident: Computer Science. The day in computer science: software, systems and theory on arXiv.

Tom: We're looking at the paper: "Analytically Characterized Optimal Power Control for Signal-Level-Integrated Sensing, Computing and Communication in Federated Learning".

Jane: Here's the gist of the paper. Signal-level integrated sensing, computing and communication (Sig-ISCC) is envisioned where a single waveform simultaneously supports sensing, computing and communication via over-the-air computation (AirComp).

Tom: This work proposes an analytically characterized polynomial-time algorithm that efficiently achieves the optimal transmit powers and receive scaling factor for uplink Sig-ISCC design in AirFEEL.

Jane: System Model Description The system considers a set K of K EDs and a single antenna base station, where each ED performs both local training for federated learning and coordinated joint target detection.

Tom: The objective of the FL training is to minimize the global loss function F(w) by minimizing Fk(w) for each client k in K.

Title and authors: Tom: So, we're looking at this paper titled "Analytically Characterized Optimal Power Control for Signal-Level-Integrated Sensing, Computing and Communication in Federated Learning". Jane, what's your initial take on the title? It sounds like a lot to unpack.

Jane: It’s pretty dense, Tom. "Signal-level integrated sensing, computing and communication" is the core idea here—basically one signal doing three things at once: sensing data, running computations, and communicating with the base station.

Tom: Right. And they tie that into federated learning—that's distributed machine learning where devices learn from each other without sharing their raw data. So, it’s about making sure this single waveform works for both sensing and that learning process simultaneously.

Lu: The authors are Paul Zheng, Yao Zhu, Xiaopeng Yuan, Yulin Hu, and Anke Schmeink. They seem to be coming from a background where they tackle complex system designs at the intersection of wireless communication and distributed machine learning.

Meng: I don't see any immediate production implication from the authors' names right now. What matters is whether this theoretical framework actually translates into something we can deploy on a phone or a sensor array without it needing super powerful hardware.

Lalam: Based on the abstract, they are specifically studying uplink Sig-ISCC design for AirComp federated learning with joint target detection. That means the core problem they're tackling is how to set the transmit powers and receive scaling factor so that both goals—sensing and communication—are met at the same time.

The paper's summary: Tom: Okay, so what does this paper actually propose? It seems to be solving a joint power control and receive-scaling problem. They’re trying to minimize the "optimality gap," which is essentially how much performance we lose by not having the perfect power settings.

Jane: That sounds like they are optimizing two things at once: the transmit powers for each edge device, and a single receive scaling factor that the base station uses to adjust what it hears. This whole setup has to satisfy a specific sensing requirement alongside minimizing distortion in the AirComp part of federated learning.

Lu: The paper reformulates this original problem, which starts out non-convex, into an equivalent convex optimization problem after they use a specific variable transformation involving the transmit powers and the scaling factor. That’s a key mathematical step to make it solvable efficiently.

Meng: I worry about that transformation part. If you can't solve it easily in practice, then the math doesn't matter when you have real traffic hitting your system at scale. Is this transformation just a trick for their proof?

Lalam: The paper claims they developed an analytically characterized polynomial-time algorithm to solve this joint power and receive-scaling problem. They state it has provable global optimality, which is pretty strong if the math holds up.

The paper's improvements: Tom: So, the improvement here isn't just that they solved a hard math problem; it’s that they found an algorithm that works fast. They claim their proposed method has a runtime of about zero point zero zero one seconds compared to about zero point zero one seconds for the solver they compared it to, which is a ten times speedup.

Jane: That speedup is significant when you're dealing with real-time systems where you need fast decisions on how much power to send or how much the receiver should scale its signal down. It’s about making sure the system can adapt quickly as things change.

Lu: The paper also focuses on robustness. They proved that their algorithm is robust to root-finding inexactness, meaning even if their numerical solver doesn't find the absolute perfect answer instantly, it still lands on a result that’s close to optimal and stays good.

Meng: Robustness is crucial for deployment because real hardware and real-world measurements are never perfectly clean. If the algorithm breaks because of tiny numerical errors in finding a crossing point, then it’s useless when we try to run it on actual edge devices. How robust is this error handling specifically?

Lalam: The paper shows that the optimal power allocation for this Sig-ISCC federated learning preserves the federated learning performance even when you add that sensing constraint. That's a big claim because it means the sensing task doesn't ruin the learning task, unlike other methods they tested.

Conclusion: Tom: So, to wrap up, this paper presents an analytically characterized polynomial-time algorithm for solving the joint transmit power and receive scaling design problem in uplink Sig-ISCC federated learning. The main thing is that it turns a non-convex problem into a convex one using a specific variable transformation.

Jane: What this means for us is that we can dynamically determine optimal transmit powers and receive scaling factors simultaneously, allowing edge devices to optimize their signals for both AirComp distortion and the target detection requirement.

Lu: And the algorithm they developed is provably fast, polynomial-time, and robust to numerical errors during its calculation. That's a solid foundation for building up more complex integrated systems like this.

Meng: For deployment at scale, the paper’s claim that this optimal power allocation preserves FL performance despite the sensing constraint is what makes it interesting from an engineering standpoint—it means we can actually use it when we impose those sensing tasks.

Lalam: Yeah, the authors show that they can systematically search through all possible sets of participating edge devices by iterating over index pairs. That adds another layer of control for designing which devices transmit at maximum power based on those sensing constraints.

More episodes

← Home