arxiv: 2605.03199 · v1 · submitted 2026-05-04 · 💻 cs.NI

Recognition: unknown

PERFECT: Personalized Federated Learning for CBRS Radar Detection

Shafi Ullah Khan , Madan Baduwal , Vini Chaudhary , Debashri Roy

Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:03 UTC · model grok-4.3

classification 💻 cs.NI

keywords federated learningpersonalized federated learningCBRSradar detectionenvironmental sensing capabilitynon-IID datadynamic spectrum sharingprivacy preservation

0 comments

The pith

Personalized federated learning lets ESC sensors reach 99 percent radar recall while keeping data local.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces PERFECT, a federated learning framework that trains radar detection models locally at each Environmental Sensing Capability sensor and applies personalization to manage data differences. Centralized approaches create privacy risks and fail on non-IID data because sensors observe distinct radar types amid varying interference from LTE and 5G users. PERFECT keeps raw observations at the sensors, aggregates only model updates, and still meets the required detection threshold. A reader would care because this setup supports commercial use of the CBRS band without compromising incumbent naval radar protection or exposing sensor data.

Core claim

PERFECT is a federated learning framework that leverages ESC level personalization for robust and efficient radar detection. It preserves privacy by training models locally on ESC sensors and is the first to effectively handle non-IID scenarios through model personalization where different ESCs observe distinct radar types. Extensive simulations demonstrate that PERFECT achieves the mandated 99 percent recall for radar detection, matching centralized performance while significantly enhancing privacy, efficiency, and scalability for dynamic spectrum sharing.

What carries the argument

The PERFECT framework, which applies personalization to local models within a federated learning process across ESC sensors to accommodate non-IID radar signal distributions.

If this is right

Radar detection proceeds without any raw sensor data leaving the local ESC devices.
The network scales to many geographically dispersed sensors without a central data repository.
Detection accuracy remains high even when sensors encounter different radar signatures.
Overall system efficiency improves because model updates replace large data transfers.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same personalization step could reduce communication costs in other federated sensing networks.
Real deployments would need to test how quickly models adapt when new radar types appear.
The approach might extend to spectrum sharing scenarios beyond CBRS where privacy constraints are similar.

Load-bearing premise

Personalization at each ESC sensor can reliably compensate for the distinct radar types and interference patterns seen at different locations.

What would settle it

A set of simulations in which each ESC observes a unique radar type and the resulting personalized models fall below 99 percent recall would show the claim does not hold.

Figures

Figures reproduced from arXiv: 2605.03199 by Debashri Roy, Madan Baduwal, Shafi Ullah Khan, Vini Chaudhary.

**Figure 1.** Figure 1: An overview of the radar interference detection problem and view at source ↗

**Figure 2.** Figure 2: The proposed PERFECT framework. In a single communi view at source ↗

**Figure 3.** Figure 3: A synthetically generated CBRS spectrum sharing scenario view at source ↗

**Figure 4.** Figure 4: Sample spectrograms featuring Type 1 and Type 2 radars with 5G and LTE signals. It is evident that both radar types exhibit significant sidelobes despite the Type 2 radar having longer pulse width (or lower bandwidth) than Type 1. We construct a variety of signal scenarios corresponding to the hypotheses H1 (overlapping signals) and H0 (nonoverlapping signals). Under hypothesis H1, we define three subcate… view at source ↗

**Figure 6.** Figure 6: Deep residual CNN model architecture with an initial convo view at source ↗

**Figure 7.** Figure 7: Performance of local learning: inference accuracy, recall for view at source ↗

**Figure 8.** Figure 8: (a) Centralized learning performance: achieves view at source ↗

**Figure 10.** Figure 10: PERFECT validation accuracy across each ESC and per-class recall, over communication rounds. view at source ↗

**Figure 11.** Figure 11: Comparison across ESC-1–ESC-5 for two FL setups: view at source ↗

read the original abstract

The Citizens Broadband Radio Service (CBRS) band is pivotal for expanding next-generation wireless services, but its success hinges on robustly protecting incumbent users, such as naval radar systems, from interference. This task is delegated to a network of Environmental Sensing Capability (ESC) sensors, which must detect faint radar signals amidst heavy co-channel interference from commercial LTE and 5G users. Traditional centralized detection models raise significant data privacy concerns and are ill-suited for the Non-Independent and Identically Distributed (non-IID) nature of data from geographically dispersed sensors. To overcome these limitations, we propose a novel Federated Learning (FL) framework PERFECT that leverages ESC level personalization for robust and efficient radar detection. PERFECT preserves privacy by training models locally on ESC sensors. Furthermore, our framework is the first to effectively handle non-IID scenarios through model personalization where different ESCs observe distinct radar types. We demonstrate through extensive simulations that PERFECT achieves the mandated 99% recall for radar detection, matching centralized performance while significantly enhancing privacy, efficiency, and scalability for dynamic spectrum sharing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

PERFECT applies personalization to federated learning for CBRS radar detection but leaves the actual mechanism and any supporting experiments unspecified.

read the letter

The paper's main contribution is a federated learning setup called PERFECT for detecting naval radar signals in the CBRS band. ESC sensors train models locally to avoid shipping raw data, and the authors add per-ESC personalization so that sensors seeing different radar types can still reach the required 99% recall while matching what a centralized model would do. This targets a concrete issue in spectrum sharing where data is both private and non-IID because of geography and radar variation. The framing is straightforward and the privacy-plus-scalability angle is a reasonable fit for the domain. What the work does well is name the practical constraints of CBRS sensing and sketch how local adaptation might relax the non-IID problem without a central data repository. That part reads as honest engineering motivation rather than over-claim. The soft spot is that none of the supporting pieces are shown. The abstract states that personalization handles the non-IID case for the first time and that simulations confirm the 99% recall, yet it gives no description of how personalization is implemented, how the non-IID distributions were built, what baselines were used, or any ablation that isolates the personalization step. Without those details the central performance claim stays untestable. There are also no equations or derivations, so the paper is light on formal grounding. This is the sort of applied paper that could interest people working on dynamic spectrum access or FL deployments in wireless systems. A reader already familiar with CBRS policy and basic federated methods might extract the application angle, but anyone wanting to judge the technical advance will need the full methods and results sections. I would bring it to a reading group only after seeing the complete manuscript. It is worth sending for peer review because the problem is real and the high-level direction is plausible, but the authors will have to supply the missing implementation and validation before the claims can be assessed.

Referee Report

2 major / 0 minor

Summary. The manuscript proposes PERFECT, a personalized federated learning framework for radar detection in the Citizens Broadband Radio Service (CBRS) band using a network of Environmental Sensing Capability (ESC) sensors. It claims to overcome privacy issues and non-IID data challenges from geographically dispersed sensors observing distinct radar types by performing local training with ESC-level personalization. Through extensive simulations, the paper asserts that PERFECT achieves the mandated 99% recall for radar detection while matching centralized performance and improving privacy, efficiency, and scalability for dynamic spectrum sharing.

Significance. If the personalization mechanism and simulation results are rigorously validated, the work could advance practical deployment of CBRS by enabling privacy-preserving, scalable incumbent protection in heterogeneous wireless environments where centralized data collection is infeasible.

major comments (2)

[Abstract] Abstract: The central claim that PERFECT 'is the first to effectively handle non-IID scenarios through model personalization where different ESCs observe distinct radar types' lacks any description of the personalization method (e.g., layer adaptation, local initialization from global model, meta-learning step, or clustering). This mechanism is load-bearing for the headline contribution and cannot be assessed from the provided text.
[Abstract] Abstract: The assertion that 'PERFECT achieves the mandated 99% recall for radar detection, matching centralized performance' via 'extensive simulations' provides no experimental setup, non-IID data construction details, baselines, ablation studies isolating personalization, or error analysis. Without these, the soundness of the performance claim cannot be evaluated.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the major comments on the abstract below and will revise it to include concise additional details on the personalization mechanism and simulation context while preserving its brevity.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim that PERFECT 'is the first to effectively handle non-IID scenarios through model personalization where different ESCs observe distinct radar types' lacks any description of the personalization method (e.g., layer adaptation, local initialization from global model, meta-learning step, or clustering). This mechanism is load-bearing for the headline contribution and cannot be assessed from the provided text.

Authors: We agree the abstract should briefly indicate the personalization approach for clarity. The manuscript details that PERFECT initializes a global model via federated averaging and then enables each ESC to perform local adaptation by fine-tuning selected layers on its private non-IID data (distinct radar types per geographic location). We will revise the abstract to add a short clause such as 'via ESC-level local adaptation of the global model on distinct radar observations' so the mechanism can be assessed from the abstract. revision: yes
Referee: [Abstract] Abstract: The assertion that 'PERFECT achieves the mandated 99% recall for radar detection, matching centralized performance' via 'extensive simulations' provides no experimental setup, non-IID data construction details, baselines, ablation studies isolating personalization, or error analysis. Without these, the soundness of the performance claim cannot be evaluated.

Authors: Abstracts conventionally summarize outcomes rather than replicate full experimental protocols. The manuscript provides these elements in Sections 4–5: non-IID data is generated by assigning distinct radar signatures to ESCs based on simulated geographic dispersion; baselines include centralized training, FedAvg, and local-only models; ablations isolate personalization; and error analysis covers recall, precision, and privacy metrics. We will add a brief contextual phrase to the abstract (e.g., 'in extensive simulations with non-IID data from geographically dispersed ESCs observing distinct radar types') to better frame the claim. Full setup, baselines, ablations, and analysis remain in the body due to length constraints. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical framework validated by simulation without derivations

full rationale

The paper introduces the PERFECT personalized FL framework for CBRS radar detection and reports simulation results showing 99% recall matching centralized performance. No equations, closed-form derivations, parameter fittings, or analytical chains are present in the manuscript. Claims rest on empirical simulation outcomes rather than any self-referential mathematical construction, self-citation load-bearing uniqueness theorems, or renaming of known results, rendering the work self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract introduces no explicit free parameters, axioms, or invented entities; relies on standard concepts of federated learning and radar detection without additional postulates.

pith-pipeline@v0.9.0 · 5494 in / 1032 out tokens · 81184 ms · 2026-05-08T17:03:08.125481+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

23 extracted references · 1 canonical work pages · 1 internal anchor

[1]

Amendment of the commis- sion’s rules with regard to commercial operations in the 3550-3650 mhz band,

Federal Communications Commission, “Amendment of the commis- sion’s rules with regard to commercial operations in the 3550-3650 mhz band,” Federal Communications Commission, Report and Order GN Docket No. 12-354, 2015

2015
[2]

Requirements for commercial operations in the u.s. 3.5 ghz citizens broadband radio service band,

Wireless Innovation Forum, “Requirements for commercial operations in the u.s. 3.5 ghz citizens broadband radio service band,” The Wireless Innovation Forum, Technical Specification WINNF-TS-0112, 2018

2018
[3]

Rf dataset of incumbent radar signals in the 3.5 ghz cbrs band,

R. Caromi, M. Souryal, and T. A. Hall, “Rf dataset of incumbent radar signals in the 3.5 ghz cbrs band,”Journal of Research of the National Institute of Standards and Technology, vol. 124, p. 1, 2019

2019
[4]

Dynamic spectrum sharing for 5G NR and LTE coexistence,

MathWorks, “Dynamic spectrum sharing for 5G NR and LTE coexistence,” https://www.mathworks.com/help/5g/ug/ dss-for-5g-nr-and-lte-coexistence.html, 2024

2024
[5]

Detection of incumbent radar in the 3.5 ghz cbrs band,

R. Caromi, M. Souryal, and W.-B. Yang, “Detection of incumbent radar in the 3.5 ghz cbrs band,” in2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP). IEEE, 2018, pp. 241–245

2018
[6]

Deep learning for radar signal detection in the 3.5 ghz cbrs band,

R. Caromi, A. Lackpour, K. Kallas, T. Nguyen, and M. Souryal, “Deep learning for radar signal detection in the 3.5 ghz cbrs band,” in2021 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), 2021, pp. 1–8

2021
[7]

Waves of imagination: Unconditional spectrogram generation using diffusion architectures,

R. Vanukuri, S. U. Khan, T. T. Sarı, G. Secinti, D. Patiño, and D. Roy, “Waves of imagination: Unconditional spectrogram generation using diffusion architectures,” inMILCOM 2025-2025 IEEE Military Communications Conference (MILCOM). IEEE, 2025, pp. 1468–1473

2025
[8]

Pushing the boundaries in cbrs band: Robust radar detection within high 5g interference,

S. U. Khan, M. Kulhandjian, and D. Roy, “Pushing the boundaries in cbrs band: Robust radar detection within high 5g interference,” inMILCOM 2025-2025 IEEE Military Communications Conference (MILCOM), 2025, pp. 1278–1283

2025
[9]

Finding waldo in the cbrs band: Signal detection and localization in the 3.5 ghz spectrum,

N. Soltani, V . Chaudhary, D. Roy, and K. Chowdhury, “Finding waldo in the cbrs band: Signal detection and localization in the 3.5 ghz spectrum,” in2022 IEEE Global Communications Conference (GLOBECOM). IEEE, 2022, pp. 1–6

2022
[10]

Detection of co- existing rf signals in cbrs using ml: Dataset and api-based collection testbed,

C. Tassie, A. Gaber, V . Chaudhary, N. Soltani, M. Belgiovine, M. Loehn- ing, V . Kotzsch, C. Schroeder, and K. R. Chowdhury, “Detection of co- existing rf signals in cbrs using ml: Dataset and api-based collection testbed,”IEEE Communications Magazine, vol. 61, no. 9, pp. 82–88, 2023

2023
[11]

Spec-scan: Spectrum learning in shared channel using neural networks,

R. Hazari, G. Singh, D. Renjith, D. Krishnan, P. B, H. Olufowobi, and D. Roy, “Spec-scan: Spectrum learning in shared channel using neural networks,” in2025 IEEE 22nd Consumer Communications & Networking Conference (CCNC), 2025, pp. 1–6

2025
[12]

Senseoran: O-ran-based radar detection in the cbrs band,

G. Reus-Muns, P. S. Upadhyaya, U. Demir, N. Stephenson, N. Soltani, V . K. Shah, and K. R. Chowdhury, “Senseoran: O-ran-based radar detection in the cbrs band,”IEEE Journal on Selected Areas in Com- munications, vol. 42, no. 2, pp. 326–338, 2024

2024
[13]

In-Network Fusion for High Interference Signal Detection within CBRS Band,

S. U. Khan, M. Kulhandjian, and D. Roy, “In-Network Fusion for High Interference Signal Detection within CBRS Band,” inIEEE International Conference on Computer Communications (INFOCOM), May 2025

2025
[14]

Fair: Federated incumbent detection in cbrs band,

M. Troglia, J. Melcher, Y . Zheng, D. Anthony, A. Yang, and T. Yang, “Fair: Federated incumbent detection in cbrs band,” in2019 IEEE Inter- national Symposium on Dynamic Spectrum Access Networks (DySPAN), 2019, pp. 1–6

2019
[15]

Communication-efficient learning of deep networks from decentralized data,

B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” inProceedings of the 20th International Conference on Artificial Intelligence and Statistics (AISTATS). PMLR, 2017, pp. 1273–1282

2017
[16]

Federated learning: A survey of core challenges, current methods, and opportunities,

M. Baduwal, P. Paudel, and V . Chaudhary, “Federated learning: A survey of core challenges, current methods, and opportunities,” Computers, vol. 15, no. 3, 2026. [Online]. Available: https://www.mdpi. com/2073-431X/15/3/155

2026
[17]

Advances and open problems in federated learning,

P. Kairouz, H. B. McMahan, B. Avent, A. Bellet, M. Bennis, A. N. Bhagoji, K. Bonawitz, Z. Charles, G. Cormode, H. Eichneret al., “Advances and open problems in federated learning,”Foundations and Trends® in Machine Learning, vol. 14, no. 1–2, pp. 1–210, 2021

2021
[19]

Federated Learning with Personalization Layers

[Online]. Available: http://arxiv.org/abs/1912.00818

work page internal anchor Pith review arXiv 1912
[20]

Deep Learning for Radar Signal Detection in the 3.5 GHz CBRS Band,

R. Caromi, A. Lackpour, K. Kallas, T. Nguyen, and M. Souryal, “Deep Learning for Radar Signal Detection in the 3.5 GHz CBRS Band,” in2021 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN). IEEE, 2021, pp. 1–8

2021
[21]

Effects of radar interference on lte base station receiver performance,

F. H. Sanders, “Effects of radar interference on lte base station receiver performance,” Institute for Telecommunication Sciences, Tech. Rep., 2013

2013
[22]

Procedures for laboratory testing of environmental sensing capability sensor devices,

F. H. Sanders, J. E. Carroll, G. A. Sanders, R. L. Sole, J. S. Devereux, and E. F. Drocella, “Procedures for laboratory testing of environmental sensing capability sensor devices,”National Telecommunications and Information Administration, Technical Memorandum TM, pp. 18–527, 2017

2017
[23]

Deep residual learning for image recognition,

K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV , USA, June 2016, pp. 770–778

2016
[24]

Deepradar: A deep-learning-based environmental sensing capability sensor design for cbrs,

S. Sarkar, M. Buddhikot, A. Baset, and S. K. Kasera, “Deepradar: A deep-learning-based environmental sensing capability sensor design for cbrs,” inProceedings of the 27th annual international conference on mobile computing and networking, 2021, pp. 56–68

2021