{"id":"cf995bdb-3d5e-4604-accb-0e6821726b72","arxiv_id":"2502.02889","paper_version":2,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":0.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A reading report that synthesizes prior AI/ML work in spectrum sensing and Open RAN, presenting no new results.","lead":"This paper is a reading report that surveys AI and machine learning methods for dynamic spectrum sensing and Open RAN networks, covering frameworks like DeepSense, DeepSweep, and ORAN xApps. It offers a synthesis of existing research rather than new experiments or models.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Citation-to-claim mismatches (refs [2]/[3] swapped; [4] used for Stitching's pipeline) undermine the evidence base for the report's favorable synthesis of AI/ML DSS and ORAN.","rationale":"The reader's weakest assumption correctly identifies the reported performance figures (e.g., DeepSweep's sub-millisecond latency, 10x complexity reduction, 98% accuracy) as the most load-bearing unsupported element. I agree, and I add that the problem is broader: the report's own internal citation mappings are inconsistent and, in at least two places, demonstrably wrong. This matters because the paper makes no original experimental contribution; its entire value is as a faithful digest of prior work. If the digest misattributes authors, swaps references, and attaches specific numbers to the wrong papers, then the central conclusion about practical impact is not trustworthy. The paper is still best characterized as unverdictable as a research contribution rather than rejectable, since it does not claim novel results; however, as a literature review it would need correction before being used as a reliable reference. Therefore, the reader's UNVERDICTED verdict stands, with the caveat that the review's accuracy is a genuine concern rather than a stylistic issue.","tokens_in":6107,"tokens_out":3715,"duration_ms":34030,"concrete_test":"Run a source audit: for every quantitative claim in Section III.B and every named-author attribution in Sections II-IV, open the arXiv or DOI source and record the exact sentence supporting it. Specifically check whether arXiv:2401.04805 (DeepSweep) reports sub-millisecond latency, 10x complexity reduction, and 98% narrowband interference accuracy; whether arXiv:2402.03465 (Stitching) describes non-local blocks and an OTA-plus-synthetic dataset pipeline; and which authors actually wrote references [8] and [9]. If any claimed metric is absent or the attributed authors differ, the review's favorable conclusions are unsupported as written.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The report's central claim is a synthesis: DeepSense, DeepSweep, and Wideband Signal Stitching, combined with ORAN digital twins and xApps, enable efficient, self-optimizing 5G/6G networks. Since this is a reading report with no original measurements, every supporting quantitative statement must be traceable to the cited papers. That traceability fails in multiple places. In Section II, 'Stitching the Spectrum' is cited as [2] and DeepSweep as [3], but the reference list assigns [2] to DeepSweep and [3] to Stitching; Section III.B uses the opposite, correct mapping. Section III.B then supports Wideband Signal Stitching's dataset pipeline with [4], which is DeepLab, not the stitching paper. The headline metrics in Section III.B — DeepSweep's sub-millisecond latency, 10x complexity reduction, and 98% narrowband interference accuracy — are asserted without page or table references, so a reader cannot determine which experiments produced them or whether they transfer to other bands or platforms. The ORAN section shows similar attribution drift: xDevSM is credited to 'Melodia et al.' [8] (actual first author Feraudo), and AERPAW to 'Mandal et al.' [9] (actual first author Moore). These are not cosmetic: the report's persuasive force rests entirely on fidelity to its sources, and the mismatches mean a cautious reader cannot use it as a reliable map to the literature.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript is a survey/reading report that synthesizes recent work in AI/ML-driven dynamic spectrum sensing (DeepSense, DeepSweep, Wideband Signal Stitching) and Open RAN (digital twins, xApps, AERPAW). It concludes that bridging these two areas can enable efficient, resilient, and self-optimizing 5G/6G networks. The paper contains no original measurements or derivations; its contribution is the synthesis and the reported performance figures drawn from the cited literature.","tokens_in":6432,"tokens_out":4382,"duration_ms":39521,"significance":"If the citations are reliable, the survey would be a useful compact entry point connecting the DeepSense line of work with recent ORAN xApp and digital-twin research. Strengths include its breadth of topics, the reproduced figures from the original papers, and the explicit focus on recent arXiv/conference literature. However, because the contribution is exclusively a synthesis, its value depends entirely on accurate representation of the cited sources; the attribution errors identified below undermine this value and make the paper, in its current form, an unreliable map to the literature.","major_comments":[{"comment":"The citation mapping in this section is reversed with respect to the reference list. The text attributes 'Stitching the Spectrum' to [2] and 'DeepSweep' to [3], but the reference list assigns [2] to DeepSweep and [3] to Stitching. Section III.B uses the opposite, correct mapping. This inconsistency prevents a reader from tracing the descriptions of semantic segmentation and parallelized spectrum sensing to the intended papers, and it must be corrected throughout.","section":"Section II"},{"comment":"The dataset generation pipeline of Wideband Signal Stitching is attributed to reference [4], which is the DeepLab paper on semantic image segmentation. The described combination of over-the-air (OTA) signals with synthetic interference and noise is presented in the Stitching paper (reference [3]), not in DeepLab. This is a load-bearing misattribution because it is the only methodological detail given for the stitching framework.","section":"Section III.B"},{"comment":"The ORAN section contains multiple author-citation mismatches: 'Wang et al. [5]' (the first author of [5] is Amiri, not Wang), 'Allen et al. [6]' (the first author of [6] is Yungaicela-Naula), 'Hyodis et al. [7]' (the first author of [7] is Hoydis), 'Melodia et al. [8]' (the first author of [8] is Feraudo), and 'Mandal et al. [9]' (the first author of [9] is Moore). In addition, Section II cites AERPAW as [6] and xDevSM as [7], whereas [6] is the misconfiguration paper and [7] is Sionna. Since the persuasive force of this survey lies entirely in faithful reporting of the cited works, these errors are not cosmetic and must be fixed.","section":"Section IV"},{"comment":"The headline quantitative metrics for DeepSweep—sub-millisecond latency, 10x complexity reduction, and 98% narrowband interference accuracy—are asserted without any specific table, page, or figure reference to the source paper. Because the manuscript has no original experiments, a reader cannot determine which experimental setup produced these numbers, under what assumptions, or whether they generalize beyond that setup. The authors should either provide precise pointers into the DeepSweep paper or explicitly qualify the numbers as reported by that paper for a specific configuration.","section":"Section III.B"},{"comment":"The abstract and conclusion claim that the report 'bridges' AI-based DSS methodologies with ORAN's open architecture, but the body does not support this bridge. Section III discusses only spectrum sensing and Section IV discusses only ORAN; there is no substantive discussion of how xApps, digital twins, or ORAN control loops consume or act on the spectrum-sensing outputs from DeepSense-family systems. If the synthesis is the central claim, a dedicated section or at least a bridging discussion is needed to make it credible.","section":"Section V"}],"minor_comments":[{"comment":"The abbreviation 'UA V' appears with a stray space in several places (e.g., the abstract and Section IV.B); it should be 'UAV'.","section":"Abstract and passim"},{"comment":"Reference [4] is listed as 'Deeplab' but the canonical title is 'DeepLab'; please correct the capitalization in the title.","section":"Reference list"},{"comment":"The captions for the segmentation model and pipeline figures are vague ('Semantic Spectrum Segmentation Model' and 'Scalable and Portable Pipeline') and would benefit from a one-sentence description of what is shown, especially since the figures appear to be adapted from the cited papers.","section":"Figures 6 and 7"}],"recommendation":"major_revision","confidential_remarks":"For a survey whose only contribution is traceability to the original literature, the density of citation mismatches (reversed numbering, wrong paper attributions, incorrect first-author names) is unusually high. I would recommend that the editor require the authors to re-verify every in-text citation, author name, and reported metric against the listed references before any further consideration. The manuscript is not suitable for publication in its current form, but the errors appear to be correctable within the scope of a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this is a reading report, not a research preprint. It re-describes the DeepSense → DeepSweep → Wideband Signal Stitching line and then surveys ORAN digital twins, xApps, and AERPAW. There are no new equations, measurements, or frameworks. If you need a one-page orientation to that slice of the spectrum-sensing/ORAN literature, the prose is clear and the chosen papers are largely the right ones. The DeepSense-to-stitching arc is the actual trajectory in the literature, and the ORAN half hits the main themes: digital twins, xDevSM, AERPAW, and misconfiguration analysis.\n\nThe soft spots are real, and they matter for a report whose whole purpose is fidelity to sources. The reference mapping is internally inconsistent. Section II labels Stitching as [2] and DeepSweep as [3], but the reference list assigns [2] to DeepSweep and [3] to Stitching; Section III.B uses the opposite, correct mapping. Section II calls \"Big Data Goes Small\" [4] and \"DeepLab\" [5], but the references assign [4] to DeepLab and [5] to the VNF-splitting paper. Section III.B credits Wideband Signal Stitching's dataset pipeline to [4] (DeepLab), which is not the stitching paper. The ORAN attributions drift the same way: xDevSM is credited to \"Melodia et al.\" [8] when the first author is Feraudo; AERPAW is credited to \"Mandal et al.\" [9] when it is Moore et al. The headline performance numbers — sub-1 ms latency, 10× complexity reduction, 98% narrowband accuracy — are asserted without page or table pointers, so a reader cannot tell which experiment produced them or whether they generalize.\n\nNone of this makes the central narrative wrong: DeepSense-style CNN sensing is a real direction, and ORAN xApps are a real application. But the report's value is as a map, and a map with mislabeled streets is not safe. I would treat it as a student's reading report: useful for a first pass, not citable for specifics.\n\nMy recommendation: desk reject. There is no research claim to referee, and the citation errors need a full rewrite before it could serve as a dependable survey. If the authors fix the attributions and add exact table/figure references for the metrics, it could become a decent arXiv note or workshop paper.","headline":"A readable but citation-sloppy reading report; useful as a first orientation, not as a reliable map to the literature.","tokens_in":6865,"tokens_out":2551,"would_cite":false,"duration_ms":22477,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"AI spectrum sensing plus Open RAN could make 5G/6G self-optimizing","keywords":["dynamic spectrum sensing","DeepSense","DeepSweep","wideband signal stitching","Open RAN","xApps","digital twins","5G/6G networks"],"falsifier":"Reproduce DeepSweep on an independent wideband over-the-air dataset and measure latency and accuracy; if narrowband interference detection falls far below 98% or inference exceeds one millisecond at the claimed complexity, the report's central case for AI-native real-time spectrum sensing is weakened. On the ORAN side, a decisive negative result would be digital-twin predictions of VNF splitting or misconfiguration effects diverging from physical testbed measurements under the same traffic load.","tokens_in":5940,"feed_emoji":"📡","tokens_out":8427,"duration_ms":76376,"temperature":0.7,"pith_summary":"This reading report argues that two research lines—AI/ML dynamic spectrum sensing and Open RAN—are converging into a single software-defined infrastructure for 5G/6G. It presents DeepSense as the foundational step of feeding raw I/Q radio samples into convolutional neural networks, and reads DeepSweep and Wideband Signal Stitching as proof that the approach can scale to real-time, crowded spectrum environments. It then claims ORAN's digital twins and xApps give operators a safe way to simulate, deploy, and self-heal AI-driven network decisions. A sympathetic reader would care because these technologies target exactly the latency, scalability, and resilience constraints that autonomous vehicles, UAVs, and industrial automation will impose on next-generation wireless.","feed_headline":"AI spectrum sensing plus Open RAN could make 5G/6G self-optimizing","feed_subtitle":"A survey ties DeepSense-style CNN sensing to ORAN's digital twins and xApps, framing software-defined wireless.","key_machinery":"The carrying mechanism is the radio-to-decision loop that starts with raw in-phase/quadrature (I/Q) samples and turns them into spectrogram-like images that CNNs classify, instead of passing signals through protocol-specific feature extractors. DeepSense establishes this loop; DeepSweep parallelizes it by chopping the spectrum into chunks processed by lightweight CNNs; Wideband Signal Stitching adds semantic segmentation at the I/Q-sample level, with non-local blocks and a synthetic-plus-over-the-air data augmentation pipeline that helps the network separate overlapping signals. On the ORAN side, the analogous mechanism is the closed control loop formed by digital twins, which simulate network configurations before deployment, and xApps, which run on the RAN Intelligent Controller to make real-time resource, scheduling, and healing decisions. The report's argument is that these two loops can be coupled into one intelligent infrastructure.","core_discovery":"The report's central claim is that the combination of AI/ML dynamic spectrum sensing and ORAN's open, vendor-neutral architecture is the enabling condition for efficient, resilient, self-optimizing 5G/6G networks. It treats DeepSense as the foundational demonstration that CNNs can classify wideband spectrum directly from raw I/Q samples, then reads DeepSweep and Wideband Signal Stitching as evidence that the approach scales: DeepSweep's parallel 'chunk and process' CNN design reaches sub-millisecond inference with 10x lower complexity and 98% narrowband interference accuracy, while Stitching uses semantic segmentation with non-local blocks to handle fragmented and overlapping signals. On the network side, it argues that ORAN digital twins allow operators to simulate VNF splitting and misconfigurations before deployment, and that AI/ML xApps developed under frameworks like xDevSM and tested on AERPAW bring real-time resource scheduling and self-healing to live RANs. The synthesis claim is that these two lines are converging into a single software-defined intelligent infrastructure.","pith_inferences":["An implication the report leaves implicit is that DeepSweep-level sensing could turn spectrum awareness into a shared network service: a spectrum-sensing xApp that multiple tenants query, much like compute or storage today.","A testable extension is to benchmark the same semantic segmentation model on an independent public corpus of overlapping OFDM and narrowband emitters, checking whether the reported complexity and accuracy hold outside the original over-the-air setup.","The synthesis also suggests a stronger claim than the report states explicitly: the RAN Intelligent Controller could close the loop between sensing and action, using DeepSense-style predictions as direct inputs to xApp resource scheduling rather than as monitoring aids.","A concrete next experiment would compare an RL-trained xApp against the static CNN baseline on the AERPAW testbed, injecting misconfigurations to see whether autonomous optimization actually reduces downtime relative to the self-healing claims."],"forward_implications":["If DeepSense-style sensing reaches sub-millisecond latency in deployed systems, autonomous vehicle and industrial automation links can act on spectrum changes fast enough to prevent missed collision alerts or coordination failures.","If semantic segmentation separates overlapping wideband signals as claimed, operators can exploit spectral holes more aggressively and detect narrowband interference that bounding-box methods miss.","If ORAN digital twins predict the effects of VNF splitting and AI/ML misconfigurations before deployment, operators can test self-optimizing changes without risking live network stability.","If xApps developed under frameworks like xDevSM and validated on AERPAW generalize beyond testbeds, UAV and emergency-response networks gain real-time resource scheduling and self-healing on vendor-neutral hardware.","Taken together, the convergence of DSS and ORAN implies that network intelligence can be updated in software rather than replaced in hardware, matching the service-based design goals of 5G/6G."],"supporting_citations":[{"why":"Foundational DeepSense framework that feeds raw I/Q samples into CNNs for real-time wideband spectrum classification.","marker":"[1]"},{"why":"Parallelized CNN architecture behind the sub-millisecond latency, 10x complexity reduction, and 98% narrowband interference accuracy claims.","marker":"[2]"},{"why":"Semantic spectrum segmentation with wideband stitching that addresses granularity and dataset-diversity limits.","marker":"[3]"},{"why":"Adapted semantic-image-segmentation techniques (atrous convolution, CRFs) that the spectrum segmentation approach builds on.","marker":"[4]"},{"why":"Shows digital twins used to optimize VNF splitting in Open-RAN, grounding the digital-twin optimization claims.","marker":"[5]"},{"why":"Analyzes AI/ML misconfiguration impact in O-RAN, grounding the resilience and pre-deployment testing claims.","marker":"[6]"},{"why":"Provides the xDevSM framework for developing xApps on O-RAN's E2 interface, grounding the intelligent-applications claims.","marker":"[8]"},{"why":"Demonstrates O-RAN-enabled UAV experimentation on the AERPAW testbed, grounding the adaptability claims.","marker":"[9]"}],"fun_headline_variants":["AI sensing plus Open RAN for self-optimizing 5G/6G","DeepSense to Open RAN: AI-driven spectrum sensing scales","CNNs and Open RAN converge for self-optimizing networks","From DeepSense to Open RAN: AI spectrum sensing scales up"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything the report concludes depends on the assumption that the performance numbers it quotes from the underlying studies—sub-millisecond latency, 10x complexity reduction, 98% narrowband interference accuracy, and dependable digital-twin predictions—hold outside the specific experimental setups in which they were measured.","fun_headline_variants_meta":{"raw":{"variants":["AI sensing plus Open RAN for self-optimizing 5G/6G","DeepSense to Open RAN: AI-driven spectrum sensing scales","CNNs and Open RAN converge for self-optimizing networks","From DeepSense to Open RAN: AI spectrum sensing scales up"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000261,"raw_usage":{"total_tokens":1610,"prompt_tokens":979,"completion_tokens":631,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":595,"completion_tokens_details":{"reasoning_tokens":553}},"tokens_in":595,"tokens_out":631,"duration_ms":6250,"temperature":1.0,"reasoning_tokens":553,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T10:43:50.155263+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Reproduce DeepSweep on an independent wideband over-the-air dataset and measure latency and accuracy; if narrowband interference detection falls far below 98% or inference exceeds one millisecond at the claimed complexity, the report's central case for AI-native real-time spectrum sensing is weakened. On the ORAN side, a decisive negative result would be digital-twin predictions of VNF splitting or misconfiguration effects diverging from physical testbed measurements under the same traffic load.","supporting_citations":[{"cited_title":"Uvaydov, S","cited_arxiv_id":null,"evidence_quote":"Foundational DeepSense framework that feeds raw I/Q samples into CNNs for real-time wideband spectrum classification."},{"cited_title":"DeepSweep: Parallel and Scalable Spectrum Sensing via Convolutional Neural Networks","cited_arxiv_id":"2401.04805","evidence_quote":"Parallelized CNN architecture behind the sub-millisecond latency, 10x complexity reduction, and 98% narrowband interference accuracy claims."},{"cited_title":"Stitching the Spectrum: Semantic Spectrum Segmentation with Wideband Signal Stitching","cited_arxiv_id":"2402.03465","evidence_quote":"Semantic spectrum segmentation with wideband stitching that addresses granularity and dataset-diversity limits."},{"cited_title":"Amiri, N","cited_arxiv_id":null,"evidence_quote":"Shows digital twins used to optimize VNF splitting in Open-RAN, grounding the digital-twin optimization claims."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Analyzes AI/ML misconfiguration impact in O-RAN, grounding the resilience and pre-deployment testing claims."},{"cited_title":"Feraudo, S","cited_arxiv_id":null,"evidence_quote":"Provides the xDevSM framework for developing xApps on O-RAN's E2 interface, grounding the intelligent-applications claims."},{"cited_title":"Prototyping O-RAN Enabled UAV Experimentation for the AERPAW Testbed","cited_arxiv_id":"2411.04027","evidence_quote":"Demonstrates O-RAN-enabled UAV experimentation on the AERPAW testbed, grounding the adaptability claims."}],"review_version":1}