{"id":"ef2c2027-837e-43a5-8002-a291129b5a87","arxiv_id":"2502.00242","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Choosing which cells and beams stay on during idle hours through a digital twin can cut simulated radio access network energy use by up to 44% in a 28 GHz urban deployment.","lead":"This paper proposes several strategies to reduce cell and beam activity in 5G networks during low-traffic hours, and uses a digital twin simulation of downtown Philadelphia to estimate up to 44% energy savings. A smart generalist might care because radio access network energy is a large share of telecom operators' costs and carbon footprint, so safe idle-mode savings matter for 6G sustainability.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The coverage constraint in (5)/(10) is downlink-only; without an uplink PRACH feasibility constraint, the claimed guarantee of idle-mode coverage and the 44% savings are not established.","rationale":"I read the paper as a framework-and-evaluation contribution: the optimization formulations are coherent, and Table III is internally consistent with the linear energy model (inferred cstatic ~ 31.15, m ~ 0.426 reproduce the reported energy values). The reader's weakest-assumption identification (digital-twin accuracy) is explicitly acknowledged by the authors in Section VI.B, and it is a standard limitation of simulation studies; a field validation would be desirable but is often beyond a system-level study. The more specific and load-bearing concern is that the coverage constraint models only downlink SINR. Idle UEs must also be able to transmit PRACH, and in mmWave the uplink is often the bottleneck. The paper's own Section VII.1 admits that PRACH transmissions may be degraded by the proposed optimizations, yet the optimization problem does not include any uplink feasibility constraint. Therefore, even with a perfect digital twin, the claimed 'guarantee' of coverage for all idle UEs is not supported by the mathematics. This does not invalidate the entire contribution—it suggests a concrete additional constraint and a revised evaluation—so the verdict should remain CONDITIONAL, with the condition being that the authors either add an uplink coverage constraint or explicitly scope the claim to downlink-only coverage and discuss the potential for PRACH access failures. I therefore partially agree with the reader: the DT-accuracy concern is real but is a calibration issue, while the uplink gap is a structural modeling issue in the optimization itself.","tokens_in":9640,"tokens_out":8173,"duration_ms":85193,"concrete_test":"Re-run the global joint optimization (10) with an additional uplink coverage constraint. Build a second connectivity matrix A_ul from the same ray-tracing data, where A_ul[i,j] = 1 if UE j's transmitted PRACH to beam i of cell c achieves the required uplink SINR given UE transmit power (e.g., 23 dBm), gNB noise figure, body loss, and the PRACH detection threshold. Require each UE to be covered by at least one active beam that satisfies both downlink SSB SINR and uplink PRACH SINR, i.e., replace A_beam^T x >= 1 with (A_dl .* A_ul)^T x >= 1 (or a joint constraint). If the optimum energy cost increases materially, or the problem becomes infeasible for any UE, the 44% savings figure overstates the feasible savings under true idle-mode coverage. Report the new savings and the number of UEs that would have downlink-only coverage.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the global joint cell-and-beam optimization in (10) maintains coverage for all 46,884 idle UEs while achieving up to 44% energy savings. The coverage constraint is A_beam^T x >= 1, where A_beam is built from the downlink SSB SINR condition SINR_ij > SINR_th in (5). This captures only the UE's ability to receive SSB/SIB/paging on the downlink. It does not capture the UE's ability to transmit PRACH on the uplink. In idle mode, a UE must be able to perform random access to transition to connected mode. In mmWave systems, uplink coverage is typically more limited than downlink because UE transmit power is far below gNB transmit power, and the model already includes 8 dB UE body loss and a 10 dB gNB noise figure. A UE can therefore satisfy the downlink SSB SINR threshold yet still be unable to reach the cell for PRACH, especially after being pushed to a more distant cell whose downlink SINR is above threshold but whose uplink budget is insufficient. The paper itself acknowledges this in Section VII.1: deactivated cells/beams may force UEs to use links with lower quality, and UEs 'may need multiple attempts to successfully receive a DL signal ... or transmit an UL signal (such as PRACH).' However, the optimization never constrains uplink feasibility. Thus the solution is not guaranteed to maintain end-to-end idle-mode operation, and the reported 'guaranteed' savings may overstate the savings that are actually achievable under a realistic coverage requirement. A secondary internal inconsistency (169 cells in Section V vs 212 cells in Table III vs 234 cells in Section VI.C) reinforces the need for a clean, reproducible evaluation, but the uplink gap is the more load-bearing technical issue: even with a perfectly calibrated digital twin and corrected cell counts, the formulated problem does not ensure idle-mode coverage.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies network energy savings (NES) during low-traffic hours in an mmWave urban deployment. It formulates three optimization strategies at different granularities: local per-cell SSB codebook optimization (Eq. (7)), global cell-level activation (Eq. (8)), and global joint cell-and-beam activation (Eqs. (9)-(10)). The optimization is performed over a digital twin of downtown Philadelphia with 169 cells and 49,876 candidate UE locations, using a 3GPP-based energy cost model approximated by a linear function of the number of active beams. The reported results show energy savings of 25.9% for local beam optimization, 23.0% for global cell-level optimization, and 44.0% for the global joint optimization, while claiming SSB coverage for all 46,884 covered UEs. The paper also discusses practical implementation aspects, complexity, and the impact on UE operation such as link SNR, coverage diversity, and cell search.","tokens_in":10002,"tokens_out":2696,"duration_ms":29763,"significance":"The paper addresses a timely and industrially relevant problem: reducing RAN energy consumption in idle mode, a topic currently under study in 3GPP for 5G-Advanced and 6G. The MILP formulations are clearly stated and the optimization framework is sensible. A strength is the use of a detailed ray-tracing digital twin with realistic parameters, which goes beyond purely analytical studies. If the results are confirmed, the proposed joint cell-and-beam activation strategy could serve as a system-level benchmark for NES feature evaluation. However, the central quantitative claim of 44% energy savings with a coverage guarantee is not fully established: the coverage constraint is downlink-only, the digital twin is not validated against field data, and the SINR threshold used to define coverage is not specified. These issues affect the reproducibility and the strength of the conclusions, but they appear addressable within the scope of a revision.","major_comments":[{"comment":"The coverage constraint AT_beam x ≥ 1 in Eq. (10) enforces only downlink SSB SINR coverage, as defined in Eq. (5). It does not constrain uplink feasibility for idle UEs, e.g., PRACH transmission. The manuscript itself acknowledges in Section VII.1 that after cell/beam deactivation UEs \"may need multiple attempts to ... transmit an UL signal (such as PRACH).\" Since idle-mode operation requires that a UE can perform random access to transition to connected mode, the claimed guarantee that the optimized configuration maintains coverage for all 46,884 UEs is too strong. The claim in Section VI.A that the global strategy \"guarantees the most energy savings\" while maintaining SSB coverage should be softened or, preferably, the optimization should incorporate an uplink feasibility constraint (e.g., a PRACH uplink SINR or link-budget threshold). Without this, the reported 44% savings may not be achievable under realistic end-to-end idle-mode requirements.","section":"§IV.C.2, Eq. (10); §VI.A; §VII.1"},{"comment":"The digital twin is the basis for the connectivity matrix A used in all optimizations, but the paper provides no calibration or validation of the digital twin against field measurements or a reference simulator. Section VI.B states that global optimization \"relies on a key assumption that an accurate DT is available,\" yet the manuscript does not report any error analysis, sensitivity study, or comparison to a validated data set. A miscalibrated twin could produce coverage holes or overestimate link SINR, directly affecting the computed savings. Please add a validation section or, at minimum, a quantitative sensitivity analysis showing how the energy savings and coverage guarantees change with plausible errors in the ray-tracing predictions or in the assumed UE distribution.","section":"§V; §VI.B"},{"comment":"The SSB coverage definition in Eq. (5) uses an SINR threshold SINRth, and the deployment problem in Eq. (4) uses SINRth(K), but the numerical value of SINRth is never specified anywhere in the manuscript. This is a key parameter: the connectivity matrix A, and hence all optimization results, depend directly on it. Without this value, the results in Table III are not reproducible, and the reader cannot assess whether the chosen threshold is realistic for SSB/SIB/paging reception. Please report the exact threshold value(s) used in the simulations and, if applicable, the mapping from target data rate to SINR threshold.","section":"§IV.A, Eq. (5); §V"},{"comment":"The energy cost model is a linear fit C(Nb) ≈ 1{Nb>0} c_static + m Nb, with no reported fit error, confidence intervals, or sensitivity. Since the optimization objective in Eq. (10) is exactly this cost function, the 44% savings figure in Table III is the optimal value of the modeled cost, not an independent estimate of achievable energy savings. This is a methodological point, not an error, but the paper should explicitly acknowledge that the reported savings are model-consistent optima and provide a sensitivity analysis with respect to c_static and m, as well as an error bound for the linear approximation shown in Figure 1. Without such analysis, a reader cannot judge how robust the 44% claim is to plausible variations in the energy model parameters.","section":"§III.B, Eq. (3); §VI.A, Table III"}],"minor_comments":[{"comment":"There is an inconsistency in equation numbering: the global joint beam- and cell-level optimization is formulated as Eq. (10), but Table IV refers to it as \"Global joint beam- and cell-level optimization (9)\". Please correct the reference.","section":"§VI.A, Tables III and IV"},{"comment":"In Section VI.C, the text states \"With 234 active cell in initial deployment\" but the earlier section reports NC=169 active cells and 69 sites. This appears to be a typo, and the number should be corrected for consistency.","section":"§V and §VI.C"},{"comment":"The local beam-level optimization assumes the UE-cell association is fixed and given. This assumption should be stated more prominently before presenting the formulation, as it restricts the solution space and explains why the local approach cannot switch UEs to neighboring cells. A brief remark on the sensitivity of results to the association criterion (e.g., strongest SINR vs. strongest received power) would be helpful.","section":"§IV.B, Eq. (7)"},{"comment":"Figure 4a reports reductions in SSB SNR after optimization, but the y-axis and the threshold line (if any) are not described in the caption or text. Please add axis labels and a clear description of the SNR metric and its units.","section":"§VII, Fig. 4"},{"comment":"The paper would benefit from a notation table for symbols such as NU E, NCNB, NT P, Np, a, K, and α, as they are used across multiple sections and occasionally without restating their definitions.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of the journal and addresses a relevant problem. The main concern is that the headline result — up to 44% energy savings with a coverage guarantee — is not fully supported because (a) the coverage constraint ignores uplink PRACH feasibility, (b) the digital twin is not validated, and (c) the SINR threshold is unspecified. These are load-bearing but fixable issues. I do not see evidence of author misconduct or a fundamentally flawed approach. A revision that addresses these points, even if it only softens the claims and adds sensitivity analysis, would make the paper more credible. I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a serious system-level study of using a digital twin (DT) to optimize idle-mode RAN energy at cell and beam granularity. What is actually new is the application itself: prior work used DTs for CAPEX planning or applied DRL locally, but nobody has formulated the network-wide idle-mode NES problem as a set of MILPs and compared local, global cell-level, and global joint cell-beam strategies on a realistic mmWave urban deployment. The formulations are clean, the linearization of the cell-activation indicator is standard and correctly handled, and the evaluation is grounded in a ray-traced DT of downtown Philadelphia with 3GPP-style power models. The paper also earns credit for being upfront about its own limitations: Section VI.B explicitly says global optimization relies on an accurate DT, and Section VII discusses how deactivation degrades link SNR, coverage diversity, and can force extra PRACH attempts.\n\nThe soft spots are real but not fatal. The stress-test note lands: the coverage constraint in (5)/(10) only requires downlink SSB SINR above threshold. It does not check whether idle UEs can transmit PRACH on the uplink, which in mmWave is often the tighter budget because UE transmit power is low and there is 8 dB body loss. The paper itself admits in Section VII.1 that UEs may need multiple UL attempts after deactivation, yet the optimization never constrains uplink feasibility. So the claimed \"guarantee\" of idle-mode coverage is too strong; the 44% savings should be read as an upper bound under a downlink-only feasibility assumption. This is fixable by adding an uplink feasibility constraint or explicitly weakening the coverage claim.\n\nThe other issues are smaller but worth listing: the DT is not validated or calibrated anywhere, so the numerical results are conditional on the twin being accurate; the cell count is inconsistent (169 in Section V, 212 in Table III, 234 in Section VI.C); the SSB SINR threshold is never specified; and the intro promises up to 46.4% savings while the abstract says 44%. The energy cost is a linear fit with no error bounds, and the reported savings are computed with the same cost model used in the optimization, so 44% is an optimal value of the model rather than an independent prediction. That last point is not damning because the cost parameters come from 3GPP TR 38.864, but it does mean the headline number is model-bound.\n\nWho should read this: anyone working on 5G-Advanced/6G NES, operators evaluating cell/beam shutdown strategies, and people building DTs for network optimization. It deserves a serious referee. I would send it out with a request for revision, not desk-reject it. The review should ask for an uplink feasibility check or a clear caveat, DT calibration details, and a pass over the inconsistent numbers.","headline":"Solid industry study of DT-assisted idle-mode energy optimization, but the 44% savings figure is an upper bound because the coverage constraint is downlink-only and the digital twin is unvalidated.","tokens_in":10541,"tokens_out":2298,"would_cite":true,"duration_ms":24495,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"By optimally choosing which cells and SSB beams stay active during low-traffic hours, the paper demonstrates a 44% reduction in idle-mode energy consumption in a dense urban mmWave network.","keywords":["network energy savings","digital twin","idle mode","SSB beam optimization","cell activation","mixed integer linear programming","mmWave","6G"],"falsifier":"Apply the global joint optimization's cell and beam sleep set in a live network during low-traffic hours and drive-test SSB SINR over the same UE-location grid; if the fraction of locations meeting the SSB SINR threshold falls below the 94% used in the paper, or the measured idle-mode energy reduction is well below 44%, the twin's coverage predictions are wrong.","tokens_in":9475,"feed_emoji":"🌙","tokens_out":7355,"duration_ms":64741,"temperature":0.7,"pith_summary":"During off-peak hours, most of a cellular network's energy goes not to data, but to keeping cells and broadcast beams on for idle devices. This paper argues that this idle-mode energy can be sharply reduced by solving a set-cover-style optimization that picks a minimal set of active cells and SSB beams while still covering every modeled UE location. Using a ray-traced digital twin of a dense urban mmWave deployment, it evaluates the idea at three granularities and reports that the global joint cell-and-beam optimization achieves a 44% reduction in idle-mode energy, from 9.494e3 to 5.286e3 relative units. If this holds in practice, operators could cut operating cost and carbon footprint in low-traffic hours without deploying new hardware.","feed_headline":"Digital twin cuts idle network energy by up to 44%","feed_subtitle":"A digital twin of downtown Philadelphia shows which cells and beams can sleep at night, cutting idle-mode energy.","key_machinery":"The load-bearing object is a binary connectivity matrix $A$, built from ray-traced link estimates in the digital twin, whose entry is 1 when a cell or beam delivers SSB SINR above threshold to a UE location. The cell energy cost is modeled as $C(N_b)=\\mathbf{1}_{\\{N_b>0\\}}c_{\\mathrm{static}} + m N_b$, so cutting beams and cells directly cuts cost. The three optimizations are set-cover integer programs: local beam selection (Equation 7), global cell selection (Equation 8), and joint selection (Equation 10). To keep the joint problem linear, the nonlinear indicator $\\mathbf{1}_{\\{B^T x>0\\}}$ is replaced by an auxiliary active-cell vector $x_c$ with the constraint $N_B x_c \\ge B^T x$, yielding a MILP that the paper solves in 34 seconds for the 169-cell, 49,876-location instance.","core_discovery":"The central claim is that a global joint optimization of active cells and their SSB beam codebooks, formulated as a mixed-integer linear program, yields the largest idle-mode energy savings of the strategies considered: 44% in the paper's digital-twin study, versus 25.9% for local beam-level optimization and 23.0% for global cell-level optimization. The optimizer is allowed to turn off 23% of the cells and 92.8% of the 6,784 baseline SSB beams while preserving SSB coverage for all 46,884 modeled UE locations. The paper presents this as the natural next step from the standardized idle-mode network energy savings (NES) features: implementation details and system-level evaluation, rather than another proposal for what those features should be. It also provides practical considerations—centralized versus distributed control, cluster sizing, slow-timescale operation—and quantifies the impact on idle UEs, including a reduction in the cell search burden.","pith_inferences":["A miscalibrated or stale digital twin would likely overstate savings: if the twin's predicted idle-UE locations are wrong, the chosen sleep set could create real coverage holes; a testable extension is to run the same optimizations under perturbed user distributions and quantify how quickly savings and coverage degrade.","The set-cover formulation is not tied to SSB beams; the same machinery could be applied to other periodic broadcasts (SIB, PRACH) or to multi-layer networks, where lower-frequency cells could sleep while higher-frequency cells maintain coverage.","The paper's trade-off between energy savings and coverage diversity suggests a natural multi-objective version—maximize savings subject to a minimum diversity level—which the authors explicitly leave for future work.","Since only outdoor UE locations are modeled, adding indoor hotspots as coverage constraints would likely reduce the number of cells that can safely sleep; quantifying this would bound the real-world savings."],"forward_implications":["Local beam optimization alone reduces the number of active SSB beams by 85.2% (from 6,784 to 1,002) and saves 25.9% of idle-mode energy without changing which cells are on.","Global cell-level optimization deactivates 23% of cells and saves 23.0%, showing that most of the gain comes from beam-level dormancy rather than cell shutdown.","The joint global strategy leaves at most 9 beams active per cell, cutting the idle UE's cell search and measurement window by up to 3x, a potential UE-side energy benefit.","Because the decisions are made on a slow timescale from historical or statistical data, the MILP's NP-hard complexity is not an obstacle in practice: the studied network solves in about 34 seconds.","Coverage of all modeled UE locations is preserved by constraint, but UEs do experience lower SSB SNR and fewer candidate cells after optimization."],"supporting_citations":[{"why":"Supplies the gNB energy consumption model with static and dynamic power and sleep levels that defines the cost function being minimized.","marker":"[15]"},{"why":"Provides the MILP-based cell deployment method and SINR-threshold selection used to construct the baseline network.","marker":"[14]"},{"why":"Supplies the partial set cover relaxation used to linearize the joint cell-and-beam objective.","marker":"[16]"},{"why":"The mixed-integer programming presolve and solve technology used to solve the optimization problems.","marker":"[18]"},{"why":"Calibration of ray tracing for wireless digital twinning, supporting the fidelity of the digital twin used to build connectivity matrices.","marker":"[7]"}],"fun_headline_variants":["Digital twin finds 44% energy savings in idle networks","Global optimizer sleeps 92% of beams, saves 44% energy","Network digital twin cuts low-traffic energy by 44%","Joint cell-beam sleep optimization yields 44% savings","Digital twin schedules idle cells and beams for 44% cut"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The digital twin correctly predicts where idle UEs are and how strong their links are; if that prediction is off, the optimizer may switch off cells or beams that are actually needed, creating coverage holes or inflating the reported savings.","fun_headline_variants_meta":{"raw":{"variants":["Digital twin finds 44% energy savings in idle networks","Global optimizer sleeps 92% of beams, saves 44% energy","Network digital twin cuts low-traffic energy by 44%","Joint cell-beam sleep optimization yields 44% savings","Digital twin schedules idle cells and beams for 44% cut"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000242,"raw_usage":{"total_tokens":1501,"prompt_tokens":897,"completion_tokens":604,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":513,"completion_tokens_details":{"reasoning_tokens":519}},"tokens_in":513,"tokens_out":604,"duration_ms":5712,"temperature":1.0,"reasoning_tokens":519,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T19:41:22.035939+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply the global joint optimization's cell and beam sleep set in a live network during low-traffic hours and drive-test SSB SINR over the same UE-location grid; if the fraction of locations meeting the SSB SINR threshold falls below the 94% used in the paper, or the measured idle-mode energy reduction is well below 44%, the twin's coverage predictions are wrong.","supporting_citations":[{"cited_title":"Study on network energy savings for NR,","cited_arxiv_id":null,"evidence_quote":"Supplies the gNB energy consumption model with static and dynamic power and sleep levels that defines the cost function being minimized."},{"cited_title":"A mixed-integer linear programming ap- proach to deploying base stations and repeaters,","cited_arxiv_id":null,"evidence_quote":"Provides the MILP-based cell deployment method and SINR-threshold selection used to construct the baseline network."},{"cited_title":"On Approximating Partial Set Cover and Generalizations","cited_arxiv_id":"1907.04413","evidence_quote":"Supplies the partial set cover relaxation used to linearize the joint cell-and-beam objective."},{"cited_title":"Presolve reductions in mixed integer programming,","cited_arxiv_id":null,"evidence_quote":"The mixed-integer programming presolve and solve technology used to solve the optimization problems."},{"cited_title":"Calibrating wireless ray tracing for digital twinning using local phase error esti- mates,","cited_arxiv_id":null,"evidence_quote":"Calibration of ray tracing for wireless digital twinning, supporting the fidelity of the digital twin used to build connectivity matrices."}],"review_version":1}