acceptodds
Under review as a conference paper at ICLR 2027

Manifold-Regularized Latent Diffusion Model for Partition-Robust Open Set Wireless Signal Recognition

Abstract

Wireless Signal Recognition (WSR) leverages deep learning to perform blind estimation of signal attributes, serving as the core technology for spectrum monitoring and wireless security. Real-world spectrum environments are open and dynamic, which requires models to accurately classify known signals while reliably detecting unknown types. This defines the open set WSR problem. Despite its importance, robust open set WSR remains underexplored due to a fundamental barrier: partition sensitivity, where model performance and rankings fluctuate across different data partitions, undermining evaluation consistency and deployment reliability. This challenge is particularly acute for wireless signals: channel-induced perturbations cause substantial intra-class feature variation, while inherent structural similarities among signal types render inter-class boundaries fragile. To address this, we formalize the problem as Partition-Robust Risk Minimization (PRRM), which minimizes a surrogate risk that jointly optimizes expected performance and suppresses risk variance. The variance decomposes into sample-level variance and class-composition variance. Guided by PRRM, we introduce DualDiff, which employs manifold-regularized latent diffusion to construct a smooth feature manifold that reduces sensitivity to sample-level perturbations, and geometry-aware fine-tuning to stabilize class boundaries against subset composition changes. Extensive experiments demonstrate that DualDiff achieves strong open set recognition performance while maintaining consistent rankings across diverse data partitions.

Primary area:
applications to physical sciences (physics, chemistry, biology, etc.)
Keywords:
Diffusion, Wireless Signal, Open Set

Reviews

Among:
Shape:
1234this paper, Bottom 2%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.33 over 3 reviews, above 2% of the others.
Reviewer_3fWHOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper studies the changes in performance and ranking of open-set wireless signal recognition (WSR) methods under different known/unknown class partitions. The authors propose Partition-Robust Risk Minimization (PRRM). A second-order approximation is used to connect mean performance with risk variance, which is further decomposed into sample-level variance (VEV_E) and class-composition variance (VCV_C). Based on this formulation, DualDiff first pre-trains an encoder using latent diffusion noise prediction and signal reconstruction, and then fine-tunes it using ArcFace and a classifier-weight uniformity regularizer. The experiments use three synthetic signal datasets: RML2016.10A, RML2022, and RadChar. DualDiff obtains competitive average OSCR performance on these datasets.

Critical strengths

A relevant problem with a clear formal framing. Open-set WSR is relevant to spectrum monitoring, where unknown signal types appear after deployment, and the paper points out that reported performance and method rankings can change with the choice of known and unknown classes. The paper does not stop at observing this. It formulates the goal as maximizing the expected log-performance over a meta-distribution of partitions, and uses a second-order expansion of log⁡F\log F to obtain a mean–variance surrogate (Eqs. 2–4). The law-of-total-variance split into sample-level variance VEV_E and class-composition variance VCV_C (Eq. 5) gives a clear vocabulary for explaining why average performance alone is an insufficient criterion, and it provides the stated rationale for the two-stage design.

Competitive average performance and a relatively comprehensive evaluation.** DualDiff achieves the highest Overall OSCR among the compared baselines in all six main settings, by 0.70 to 6.53 points over the strongest baseline in each setting, and it has the best mean OSCR and mean rank over the 96 additional random partitions on RML2016.10A. The evaluation includes ablation studies, five runs of DualDiff with different random seeds, and analyses of hyperparameters and network depth. To separate the effect of ArcFace, the authors also add it to IQFormer and TS-DDAE, and to the baselines in the 96-partition study. They also try to connect the framework to the experiments by estimating VEV_E and VCV_C (Tables 8 and 10).

Critical weaknesses

W1. The motivation is not sufficiently supported by the experiments set and evaluation

Different modulation types have inherent differences in their physical signal structures, such as constellation geometry and symbol mapping. As a result, some combinations of known and unknown classes are intrinsically harder to distinguish than others. The experiments do not examine how much of the observed partition sensitivity comes from these inherent differences in class separability and how much comes from instability of the model. Decomposing the risk variance into VEV_E and VCV_C does not itself distinguish these two causes.

PRRM aims to reduce performance variation across different known/unknown class partitions. However, the proposed pre-training learns denoising and reconstruction, while fine-tuning optimizes classification and weight geometry for the known classes of the current partition. Neither objective directly constrains the performance differences between partitions. The paper does not clearly explain why these objectives should make the risks across different partitions more consistent, rather than only improving recognition within each setting.

The channel-related motivation has not been directly evaluated through experiments on unseen channel conditions. The paper reports recognition performance at different SNR levels, but this does not by itself demonstrate generalization to unseen SNR conditions. A more direct test would train the model at one SNR level (or within a specified SNR range) and test the same model at other SNR levels excluded from training. The experiments also do not separately examine the effects of specific channel impairments, such as multipath fading or carrier-frequency offsets. Therefore, the reported results mainly compare performance across different signal conditions, rather than directly demonstrating robustness to channel impairments not seen during training.

W2. Limited wireless-specific methodological contribution

As described in Sections 3.2–3.3, Gaussian noise is added to the encoded latent features rather than generated through a physical model of channel impairments on the received IQ signal. The paper does not sufficiently establish why such latent perturbations can represent structured wireless effects such as multipath fading, frequency offsets, or timing errors. The central-limit argument in Section 3.2 alone is not sufficient to justify this connection. The wireless-specific contribution of this training mechanism therefore remains unclear.

The method mainly uses general representation-learning objectives, with very limited integration of domain-specific wireless signal characteristics, such as constellation geometry, symbol timing, modulation structure, and physical channel transformations, into its design and analysis. The IQ/spectrum fusion and the TS-Net denoiser come from previous WSR work. The authors relate angular margins to modulation geometry and weight uniformity to channel-induced rotations, but these links are not supported by physical modeling or targeted experiments.

W3. The theoretical guarantees are not sufficiently established

Appendix A aims to show that diffusion pre-training reduces VEV_E and classifier-weight uniformity reduces VCV_C, but the key links are not sufficiently justified. For VCV_C, Eqs. (26)–(28) assume that an ideal ETF arrangement of classifier weights gives the same expected risk for every known-class subset. However, different subsets can contain classes with different recognition difficulty. Similar classifier-weight geometry alone does not guarantee equal risks. Under this assumption, Eq. (28) would even imply VC=0V_C=0 if the weights of all subsets matched the reference configuration, although their actual risks could still differ.

For VEV_E, Eqs. (17)–(20) assume that the diffusion loss is bounded below by a positive multiple of the expected squared within-class feature deviation. This relationship is not sufficiently established: a low denoising loss can still coexist with substantial variation of same-class features along the learned manifold. In addition, OSCR describes performance over a set of test signals, while the derivation treats risk as a function of an individual signal without explaining the connection. Therefore, even as qualitative bounds, these results do not sufficiently establish that optimizing the proposed objectives necessarily reduces the partition-level risk variances defined by PRRM.

W4. Synthetic-only evaluation and incomplete open-set comparisons

RML2016.10A, RML2022, and RadChar are synthetic datasets, and there is no evaluation on measured wireless signals. This limits the support for claims about robustness in real wireless environments. Apart from FSOS-AMC, the evaluation does not include representative general-purpose open-set recognition methods such as OpenMax and ARPL, or the wireless-specific Open-ICL method cited by the authors.

Questions for authors

Q1. Can the authors quantify intrinsic class-pair or partition difficulty, for example with confusion analysis or difficulty-stratified OSCR? Please also test trained models without retraining, on SNR levels excluded from training and on data with controlled channel impairments (e.g., multipath fading, carrier-frequency offsets, or phase shifts), to verify robustness to unseen conditions.

Q2. Which parts of DualDiff are specifically motivated by wireless signal properties? Please clarify the exact noise injection point and scale, and the meaning of the temporal/spectral noise parameters in Table 5. Please provide supporting experiments or relevant references to justify treating latent Gaussian noise as a surrogate for physical wireless channel impairments rather than merely a generic feature perturbation.

Q3. Please justify the two key assumptions behind the variance bounds: why the diffusion loss controls within-class feature deviation (Eqs. (17)–(20)), and why an ideal ETF configuration would imply the same expected risk across different class subsets (Eqs. (26)–(30)). Please provide rigorous arguments or relevant references and clarify how the per-input analysis is linked to partition-level OSCR risk.

Q4. Could the authors include a real measured wireless signal dataset, rather than relying entirely on synthetic datasets, to strengthen the credibility of the evaluation? Please also compare with representative open-set recognition approaches under comparable protocols, such as OpenMax (Bendale and Boult, 2016), ARPL (Chen et al., 2022), and the wireless-specific Open-ICL (Yang et al., 2025).

References

[1] Bendale, Abhijit, and Terrance E. Boult. "Towards open set deep networks." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.

[2] G. Chen, P. Peng, X. Wang and Y. Tian, "Adversarial Reciprocal Points Learning for Open Set Recognition," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 11, pp. 8065-8081, 1 Nov. 2022, doi: 10.1109/TPAMI.2021.3106743.

[3] C. Yang, Z. Feng, S. Yang and Q. Pan, "Open-ICL: Open-Set Modulation Classification via Incremental Contrastive Learning," in IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 5, pp. 8853-8866, May 2025, doi: 10.1109/TNNLS.2024.3414942.

Reviewer_Q5vhOpenReview ↗
Rating
1
Confidence
4
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper investigates partition sensitivity in open-set wireless signal recognition (WSR). The authors formulate the problem as Partition-Robust Risk Minimization (PRRM), decomposing risk variance into sample-level variance and class-composition variance. Based on this formulation, they propose DualDiff, a two-stage framework that combines manifold-regularized latent diffusion pre-training with geometry-aware fine-tuning using ArcFace and uniformity regularization.

Critical strengths

1.Robust recognition of known and unknown wireless signals under changing class partitions is important for practical open-set wireless signal recognition.

2.The authors conduct experiments on multiple datasets and compare against a relatively large number of baselines.

Critical weaknesses

1.The theoretical justification is not convincing. The proposed PRRM formulation is primarily derived from a second-order Taylor approximation and the law of total variance, yet the actual training objectives do not directly optimize the proposed partition-level risk variance. More importantly, the proofs in Appendix A rely on assumptions that are not sufficiently justified. For example, Appendix A.2 assumes a direct lower bound relating diffusion loss to intra-class feature deviations without establishing that this property holds for the learned diffusion model.

2.The presentation and scientific narrative need significant improvement. The paper repeatedly interprets better OSCR performance as evidence of reduced partition variance, without clearly separating these two properties. Several important theoretical assumptions are introduced without sufficient justification, and the distinction between partition-level risk, sample-level perturbations, and feature-space geometry is not consistently established. As a result, the relationship between the proposed theory, the implemented method, and the experimental conclusions remains unclear.

3.The methodological novelty is limited. The framework primarily combines existing techniques, including latent diffusion, reconstruction-based pre-training, ArcFace, and feature uniformity regularization. The proposed PRRM formulation does not lead to a clearly new optimization algorithm, and the connection between the theoretical variance decomposition and the actual training procedure remains weak. Consequently, it is difficult to identify a sufficiently substantial methodological contribution beyond integrating established components.

Questions for authors

1.How can the authors rigorously justify the connection between the diffusion objective and the proposed upper bound on sample-level risk variance? In addition, how does ArcFace resolve global rotational ambiguity when its objective is invariant to a joint rotation of the features and classifier weights?

2.Can the authors compare DualDiff against stronger baselines equipped with the same ArcFace and uniformity regularization? How much improvement can be specifically attributed to manifold-regularized latent diffusion?

3.How do the authors explain that DualDiff does not consistently achieve lower cross-partition OSCR variance than the baselines?

4.All of the aforementioned Weaknesses.

Reviewer_Dk5wOpenReview ↗
Rating
1
Confidence
3
Informativeness
1
Presentation
2
Soundness
1
Read review
Paper summary

This paper addresses partition sensitivity in open-set wireless signal recognition and proposes a two-stage method, DualDiff, which first pretrains with a latent diffusion model and then fine-tunes with ArcFace plus a uniformity loss on the classifier weights. The problem is important, but the method seems to be a combination of existing modules, and the experiments provide weak support for the effectiveness of the core design.

Critical strengths

1.Open-set wireless signal recognition and its robustness are important and practically relevant problems.

2.The paper highlights the partition sensitivity phenomenon, where performance varies sharply depending on how classes are split into known and unknown.

3.Experiments cover three datasets and multiple partitions.

Critical weaknesses

1.DualDiff does not appear to be a novel framework. It looks like a direct combination of IQFormer’s feature fusion, TS-DDAE’s denoising network, and the ArcFace loss. Could the authors clarify the essential novelty beyond these components?

2.Existing components seem to account for much of the performance gain. In the ablation (Table 9), simply adding ArcFace to IQFormer improves OSCR on RadChar by about 20 pp. Although the authors provide pretraining ablations, the incremental benefit of the proposed diffusion design over strong ArcFace-based baselines could be further clarified.

3.The link between PRRM and the actual training objective is not well established. The training objectives do not explicitly minimize cross-partition variance. Could the authors clarify how these surrogate losses relate to the variance terms defined in PRRM?

4.The open-set mechanism is not closed. The paper claims that the diffusion loss concentrates known-class features on the manifold and places unknown classes in low-density regions, but the density or denoising error from the diffusion model is not used at inference. How do the authors connect low-density regions to rejecting unknown classes?

Questions for authors

Please refer to the weaknesses.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.