eess.SPSep 4, 2026

Wireless Foundation Models: State-of-the-Art and Open Challenges

Authors: Alonso M. Pacheco Huachaca, Juan J. Rodriguez Rodriguez, Ahmed Aboulfotouh, Nelson L. S. da Fonseca, Carlos A. Astudillo, Hatem Abou-Zeid

Organizations: Institute of Computing, University of Campinas, SP, Brazil · Department of Electrical and Software Engineering, University of Calgary, Calgary, AB, Canada

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Wireless foundation models (WFMs) have emerged as a promising approach for learning reusable representations from large-scale wireless data and adapting them to downstream tasks. However, the rapidly growing literature remains fragmented across modalities, pretraining objectives, architectures, adaptation strategies, and evaluation protocols, making it difficult to assess progress toward broadly transferable models. This survey provides a systematic analysis of WFMs for physical-layer applications. We first introduce the main WFM design components, including pretraining, backbone architectures, and downstream adaptation. We then organize the literature into five physical-layer task families: signal recognition and demodulation, channel representation learning, RF sensing and localization, beam management, and spectrum sensing and monitoring, while separately examining multi-task PHY models. Across these categories, we analyze how existing models are pretrained, adapted, and evaluated, with particular attention to downstream task diversity and the distinction between in-distribution, partial-shift, and out-of-distribution transfer. Our analysis shows that current WFMs provide increasing evidence of reusable wireless representations, but this evidence varies considerably across task families and evaluation settings. Differences in datasets, modalities, architectures, pretraining objectives, adaptation protocols, and distribution shifts make it difficult to determine which design choices drive transfer and generalization. We conclude by identifying open directions for improving data availability, evaluation rigor, generalization, efficient adaptation, and real-world deployment, providing a unified framework for understanding the current WFM landscape and the requirements for developing more reusable foundation models for future physical-layer wireless systems.

Figures & tables

Abbrev.MeaningAbbrev.MeaningAbbrev.Meaning
AMCAutomatic modulation classificationCERChannel estimation and recoveryCIRChannel impulse response
CSIChannel state informationEnvREnvironment reconstructionFDDFrequency-division duplexing
FMCWFrequency-modulated continuous waveI/QIn-phase/quadrature signalISACIntegrated sensing and communications
LoS/ NLoSLine-of-sight / non-line-of-sightLSILocalization and spatial inferenceMCModulation classification
MIMOMultiple-input multiple-outputOFDMOrthogonal frequency- division multiplexingOODOut-of-distribution setting
PEFTParameter-efficient fine-tuningRxReceiver / received signalSFMSpectrum forecasting and monitoring
TABLE I: List of abbreviations and shorthand terms used in the survey.
SurveyYearComparison DimensionsTutorialPHY Task Coverage
 FM spec.W. nativeModel levelPTAdapt.Eval. shiftSID/DemChannel R.RF Sens. /Loc.BeamSpectrum
Telecom Foundation Models Survey [ 18 ]2024✓∼\sim✗∼\sim∼\sim✗∼\sim✗✗✗✗✗
Channel Foundation Models Survey [ 15 ]2025✓✓✓✓∼\sim∼\sim✓✗✓✗✗✗
Large AI Models Survey [ 13 ]2025∼\sim✗∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim
Wireless Large AI Models Survey [ 8 ]2025∼\sim✗∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim∼\sim
Network Traffic FM Review [ 16 ]2026✓✓✓✓✓∼\sim✗✗✗✗✗✗
TABLE II: Comparison of surveys related to wireless foundation models.
Fig. 1: An input observation xx is mapped to a transferable latent representation z\mathbf{z} via a shared, pretrained backbone fθf_{\theta} . Task-specific heads ϕt\phi_{t} provide lightweight adaptation, producing predictions y^t\hat{y}_{t} that are optimized against true labels yty_{t} via the multi-task objective defined in Equation 1 .
Fig. 2: Representative WFM pretraining objective families.
Fig. 3: Representative WFM pretraining objective families.
Fig. 4: Representative WFM pretraining objective families.
Fig. 5: Relationship map of backbone families used in WFMs. Solid arrows indicate variants derived from Transformer-based backbones. Dashed arrows pointing into dashed nodes indicate bridge architectures that combine backbone families, while dashed lines without arrowheads indicate conceptual, lightweight, or alternative design relations. Edge labels summarize the main reason for each connection.
Fig. 6: Overview of foundation-model adaptation strategies, ordered from lower update cost to higher adaptation flexibility.
Fig. 7: Hierarchical taxonomy of physical-layer WFM tasks. Major categories group functionally related physical-layer problems, and the individual task-level labels provide compact categories for organizing the surveyed foundation model and self-supervised learning methods.
ModelYearSignal Recogn.Channel Rep. LearningRF Sens. & Local.Beam Mgmt.SpectrumTotal
SIDDemCERCPCDiagFBLSIActRFFEnvRBMPrcSSSFM
RadioFM [ 88 ]2024⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot✓⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot1
ChannelGPT [ 39 ]2024⋅\cdot⋅\cdot✓✓⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot2
SSRadio [ 28 ]2024⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot✓✓2
6G-RadioFM [ 89 ]2024⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot✓⋅\cdot⋅\cdot⋅\cdot⋅\cdot✓⋅\cdot2
WiFo [ 29 ]2024⋅\cdot⋅\cdot⋅\cdot✓⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot⋅\cdot1
TABLE III: Taxonomic coverage matrix for physical-layer wireless foundation models.
DatasetTask(s)StageModalityR/SAccessOriginLabelsUsed in
RF-SSIDPTSpectrogramRN/SCUnlabeled[ 9 , 103 , 116 ]
RF Signal Class. [ 118 ]SIDFT/EvalSpectrogramRPub.ULabeled[ 9 , 103 , 116 ]
WiFi-CSI [ 119 ]SIDPTCSIRN/SUUnlabeled[ 9 ]
RADIOML2016.10A [ 120 ]SIDFT/EvalRaw IQSPub.ULabeled[ 30 , 66 , 31 , 48 ]
RADIOML2016.10B [ 120 ]SIDFT/EvalRaw IQSPub.ULabeled[ 48 ]
RADIOML2016.04C [ 120 ]SIDFT/EvalRaw IQSPub.ULabeled[ 48 ]
TABLE IV: Unified summary of datasets used by physical-layer wireless foundation models.
ModelPT Modality/DataPT ObjectiveBackboneFT Task/DataAdaptationEval. Shift
6G WavesFM [ 9 ]Spectrogram ∗ CSI [ 119 , 127 ]Masked Rec.ViTRF cls. – Spectrogram [ 118 ]Frozen FT Partial FT PEFTPartial
IQFM [ 30 ]IQ ∗ContrastiveCNNModulation cls. – IQ ∗ ; IQ [ 120 ]Frozen FT PEFTID + OOD
Multimodal WFM [ 103 ]Spectrogram ∗ IQ ∗Masked Rec.Multimodal ViTRF cls. – Spectrogram [ 118 ]Frozen FT Partial FT PEFTPartial
Unified FM [ 92 ]IQ [ 122 , 123 , 124 ] CIR [ 138 ]Masked Rec.TransformerTechnology rec. – IQ [ 122 , 123 , 124 ]Partial FT Frozen FTID
SSRadioRep [ 99 ]IQ ∗ContrastiveCNNModulation cls. – IQ ∗Frozen FT Full FTID
Lightweight Edge FM [ 66 ]IQ [ 122 , 123 , 124 , 120 ] CIR [ 138 ]HybridMLPTechnology rec. – IQ [ 122 , 123 , 124 ] Modulation cls. – IQ [ 120 ]Frozen FT Full FTID
TABLE V: Overview of WFMs for signal recognition and demodulation tasks.
ModelPT Modality/DataPT ObjectiveBackboneFT Task/DataAdaptationEval. Shift
ChannelGPT [ 39 ]CSI ∗ Env. map ∗GenerativeTransformerChnl. gen. – CSI ∗ + Env. map ∗ Chnl. pred. – CSI ∗PEFTID
WiFo [ 29 ]CSI [ 130 ]Masked Rec.TransformerChnl. pred. – CSI [ 130 ]Zero-shotPartial
LWM [ 90 ]CSI [ 129 ]Masked Rec.TransformerChnl. diag. – CSI [ 129 ]Frozen FTPartial
6G WavesFM [ 9 ]Spectrogram ∗ CSI [ 119 , 127 ]Masked Rec.ViTChnl. est. – OFDM [ 128 ]Frozen FT Partial FT PEFTPartial
WirelessGPT [ 91 ]CSI ∗ CSI [ 128 ]Masked Rec.TransformerChnl. est. – CSI ∗ Chnl. pred. – CSI [ 130 ]PEFT Frozen FTPartial
Unified FM [ 92 ]IQ [ 122 , 123 , 124 ] CIR [ 138 ]Masked Rec.TransformerChnl. diag. – CIR [ 138 ]Partial FT Frozen FTID
TABLE VI: Overview of WFMs for channel representation learning tasks.
ModelPT Modality/DataPT ObjectiveBackboneFT Task/DataAdaptationEval. Shift
RadioFM [ 88 ]CIR ∗Masked Rec.TransformerIndoor loc. – CIR ∗Partial FT Frozen FTID
6G-RadioFM [ 89 ]Spectrogram ∗Masked Rec.ViTActivity rec. – CSI [ 141 ]Frozen FTOOD
6G WavesFM [ 9 ]Spectrogram ∗ CSI [ 119 , 127 ]Masked Rec.ViTActivity rec. – CSI [ 141 ] Localization – CSI [ 137 ]Frozen FT Partial FT PEFTPartial
Multimodal WFM [ 103 ]Spectrogram ∗ IQ ∗Masked Rec.Multimodal ViTActivity rec. – CSI [ 141 ] Positioning – CSI [ 137 ] RF fingerprinting – IQ [ 126 ]Frozen FT Partial FT PEFTPartial
WirelessGPT [ 91 ]CSI ∗ CSI [ 128 ]Masked Rec.TransformerActivity rec. – CSI ∗ Env. recon. – CSI ∗Frozen FT PEFTPartial
IQFM [ 30 ]IQ ∗ContrastiveCNNAoA – IQ ∗ RF fingerprinting – IQ [ 126 ]Frozen FT PEFTID + OOD
TABLE VII: Overview of WFMs for RF sensing and localization tasks.
ModelPT Modality/DataPT ObjectiveBackboneFT Task/DataAdaptationEval. Shift
LWM [ 90 ]CSI [ 129 ]Masked Rec.TransformerBeam pred. – CSI [ 129 ]Frozen FTPartial
IQFM [ 30 ]IQ ∗ContrastiveCNNBeam pred. – IQ/beam [ 143 ]Frozen FT PEFTOOD
MUSE-FM [ 52 ]Scene ∗ W. signal ∗Task-driven SupervisionTransformerMU prec. – Scene ∗ + W. signal ∗MT-FT Zero-shotID †
CSI-CLIP [ 62 ]CSI [ 129 ]ContrastiveDual-branchBeam mgmt. – CSI [ 129 ]Frozen FTPartial
ICWLM [ 53 ]CSI [ 130 ]Task-drive SupervisionTransformerPrecoding – CSI [ 130 ]Zero-shotPartial
ContraWiMAE [ 93 ]CSI [ 129 ]HybridTransformerBeam mgmt. – CSI [ 129 ]Frozen FTPartial
TABLE VIII: Overview of WFMs for beam management tasks.
ModelPT Modality/DataPT ObjectiveBackboneFT Task/DataAdaptationEval. Shift
SSRadio [ 28 ]Spectrogram ∗Masked Rec.RNNSpectrum forecast. – Spectrogram ∗ Spectrum seg. – Spectrogram ∗Frozen FTID + Partial
6G-RadioFM [ 89 ]Spectrogram ∗Masked Rec.ViTSpectrum seg. – Spectrogram ∗Frozen FTPartial
Multimodal WFM [ 103 ]Spectrogram ∗ IQ ∗Masked Rec.ViTInterference det./cls. – IQ [ 144 ]Frozen FT Partial FT PEFTOOD
SpectrumFM [ 48 ]IQ [ 125 , 122 ] IQ ∗HybridCNN-TransformerSpectrum sensing – IQ [ 120 ] Anomaly det. – IQ ∗PEFT Frozen FTPartial
FM-RME [ 64 ]PSD map ∗ Spectrum map ∗Masked Rec.Geometry-aware TransformerRadio map est. – PSD map ∗ + Spectrum map ∗Zero-shotPartial
FARM [ 110 ]RSS map ∗ Aerial radio map ∗HybridDiffusion TransformerRadio map est. – Aerial radio map ∗Zero-shotOOD
TABLE IX: Overview of WFMs for spectrum sensing and monitoring tasks.
More in this topic

Wireless Foundation Models

Wireless foundation models learn reusable representations of radio channels, signals, network traces, and telemetry for transfer across wireless tasks and settings.

See all 32 papers

Explore similar work

CardsList
Jun 9, 2026

A Unified Adaptive Feature Composition Framework for Multi-Task Generalization in Wireless Foundation Models

Though wireless foundation models (WFMs) have shown strong potential in learning universal channel representations, their adaptation to various downstream tasks remains constrained by existing paradigms. Fine-tuning strategies introduces substantial computational and storage overhead, while frozen feature extraction leads to sub-optimal performance across diverse downstream tasks. To address this issue, we propose a unified adaptive feature composition framework for multitask generalization in WFMs, where the key component is the Routing Adapter for Feature Composition (RAFC). Instead of extracting only the final-layer output, this router treats the hidden states from different Transformer depths as a reusable pool of multi-level hidden features, and employs a lightweight task-driven feature composition network to generate layer-wise aggregation weights, then adaptively combine hierarchical representations through weighted summation. This design enables each downstream task to access suitable mixture of low-, mid-, and high-level wireless features without modifying the pretrained backbone. Extensive experiments on four representative wireless tasks demonstrate that RAFC consistently outperforms conventional adaptation baselines while introducing fewer than 50K additional parameters. Moreover, the learned routing weights provide interpretable evidence of task-specific layer preferences, making the proposed framework a low-complexity, scalable, and explainable interface for adapting WFMs to diverse downstream scenarios.

Yuxuan Shi, Tingting Yang, Kangning Ma +4

Department of Broadband Communication, Pengcheng Laboratory, Shenzhen, 518000, China

Wireless Foundation ModelsAdaptive Model RoutingMulti-Task LearningParameter-Efficient Fine-Tuning
AskPDF
Jun 4, 2026

LatentWave: JEPA Pretraining for Wireless Foundation Models

Wireless foundation models have emerged as a promising alternative to building separate models for each wireless task. However, existing approaches rely on masked input reconstruction, which can bias representations toward low-level signal details. In this paper, we propose LatentWave, a wireless foundation model pretrained using a Joint-Embedding Predictive Architecture (JEPA) on diverse wireless spectrograms and channel state information (CSI). By predicting masked regions in latent space, LatentWave learns representations that are more transferable out of the box across diverse downstream tasks. The proposed architecture employs per-channel patch embeddings with stochastic channel sampling during pretraining, allowing it to process variable antenna counts and improving usability across heterogeneous wireless configurations. We evaluate LatentWave on four downstream tasks: RF signal classification, 5G NR positioning, beam prediction, and LoS/NLoS classification, comparing against a masked-modeling baseline (WavesFM) pretrained on the same data. Additionally, we show that the masking geometry introduces a task-dependent inductive bias: frequency masking strongly favors channel-related tasks such as positioning and beam prediction, while region masking better preserves discriminability for signal classification.

Ahmed Mohamed, Ahmed Aboulfotouh, Hatem Abou-Zeid

Department of Electrical and Software Engineering, University of Calgary, Canada

Wireless Foundation ModelsSelf-Supervised Pre-TrainingJoint-Embedding Predictive ArchitectureWireless Communications
AskPDF
May 5, 2026

RFPrompt: Prompt-Based Expert Adaptation of the Large Wireless Model for Modulation Classification

Automatic modulation classification (AMC) in real-world deployments demands robustness to distribution shifts arising from hardware impairments, unseen propagation environments, and recording conditions never encountered during training. Although wireless foundation models offer a promising starting point for robust RF representation learning, an important open question is how to adapt them efficiently to out-of-distribution (OOD) downstream tasks without overwriting the structure learned during large-scale pre-training. In this paper, we investigate prompt-based adaptation as a general mechanism for OOD transfer in wireless foundation models. We propose RFPrompt, a parameter-efficient framework that introduces learnable deep prompt tokens while keeping the pretrained backbone frozen, enabling task-specific adaptation with minimal trainable parameters. We instantiate and evaluate this approach on the Large Wireless Model (LWM), a mixture-of-experts wireless foundation model, and study its behavior under both standard and OOD modulation-classification settings. Results show that prompt-based adaptation consistently improves robustness under distribution shift and limited supervision, particularly on real-world over-the-air IQ data, while preserving strong parameter efficiency. These findings suggest that prompt learning is a practical and effective strategy for adapting wireless foundation models to challenging downstream RF environments.

Md Raihan Uddin, Tolunay Seyfi, Fatemeh Afghah

Department of Electrical and Computer Engineering, Clemson University, Clemson, SC, USA

Wireless Foundation ModelsAutomatic Modulation ClassificationOOD GeneralizationPrompt TuningParameter-Efficient Fine-Tuning
AskPDF