{"id":"0419b5e4-cfdb-48c9-a271-bfb73ad59985","arxiv_id":"2412.01634","paper_version":1,"verdict":"CONDITIONAL","confidence":"LOW","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"An experimental lab study reports that the OAIBox 5G core draws far more power (about 6.9 Wh idle, +9.6 Wh during downlink) than the USRP radio unit (under 1 Wh total) in a private 5G setup.","lead":"This paper measures how much electricity the core, radio, and modem of a small private 5G test network consume at different stages, from idle to active file transfer. It gives a first, concrete breakdown of where energy goes in a private 5G setup, useful for designing greener 6G networks.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reported Wh deltas are unnormalized: without phase durations, the 9.58 Wh downlink increase cannot be compared to the 1.68 Wh activation increase, so the central 'core dominates' claim may reflect test length rather than power draw.","rationale":"The reader's lowest-confidence point was the undocumented Netio sampling process, and the paper indeed omits sampling interval and integration details. My stress-test sharpens this into a more specific and arguably more damaging issue: the reported per-step energy deltas are not normalized by the duration of each phase. Since energy is the integral of power over time, a comparison of Wh increments across steps implicitly assumes equal or otherwise controlled observation windows. The manuscript provides no phase durations, no explicit statement that all steps were equal-length, and no raw data from which durations could be recovered. If, for example, the downlink iperf transfer ran for several minutes while the modem-activation step lasted only seconds, the 9.58 Wh figure would be an artifact of longer measurement time rather than higher power consumption. This directly undermines the headline quantitative claims and the conclusion that downlink traffic dominates OAIBox energy. The fix is straightforward: report elapsed time per step and mean power, or publish the timestamped Netio samples so that independent reanalysis is possible. This does not change the reader's CONDITIONAL verdict, because the concern is about missing documentation and comparability, not about internal contradiction or obvious fraud. The experimental setup is plausible, and the authors' acknowledgment of limited generality is honest, but the central numbers cannot be evaluated as stated.","tokens_in":3691,"tokens_out":2768,"duration_ms":26544,"concrete_test":"Ask the authors to release the InfluxDB timestamps or a CSV of Netio samples with phase markers. For each step, compute elapsed time Δt and mean power P̄ = ΔE/Δt. If the durations differ, replot all steps on a per-minute or per-hour basis. If the downlink's mean power is not substantially larger than the active step's mean power, the central claim that downlink causes a 9.58 Wh increase loses support.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claim depends on per-step energy increments (e.g., §4: OAIBox +9.58 Wh downlink, +1.68 Wh active, +1.93 Wh core start). Energy is time-integrated power, so a Wh increment is meaningful only with the duration of each step. The paper never states how long steps 1–9 lasted, whether durations were equal, or how Wh values were accumulated from Netio's MQTT V/I/pf samples. If the downlink iperf phase ran much longer than the modem-activation phase, the 9.58 Wh figure could simply reflect a longer observation window, not higher power draw. Similarly, the idle baseline of 6.91 Wh is an absolute energy over an unspecified interval, so it cannot be compared across steps. All high-level conclusions about which component or condition dominates rest on these unnormalized deltas, making this the most load-bearing weakness. The missing sampling interval and integration method compound the problem: without knowing the Netio publication rate and how instantaneous power was converted to energy, the reported Wh values are not reproducible from the manuscript.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports an experimental energy profiling study of the components of a private 5G network built from an OAIBox (5G core plus gNB) and a USRP B210 (radio unit), with a Quectel modem as the UE. Using a Netio PowerBox 4KF and an MQTT/InfluxDB/Grafana pipeline, the authors measure energy consumption under idle, core start, gNB start, UE attach, modem activation, iperf downlink, and iperf uplink conditions at 20 MHz bandwidth. The headline results are an idle OAIBox consumption of about 6.91 Wh, USRP consumption of 0.23 Wh, and downlink-related increases of about 9.58 Wh (OAIBox) and 0.30 Wh (USRP), with similar uplink values. The paper concludes that the core, rather than the radio, dominates the additional energy cost of user-plane traffic and calls for further profiling in real-world multi-access-point deployments.","tokens_in":3997,"tokens_out":2590,"duration_ms":24293,"significance":"If the measurements are trustworthy, the paper provides a useful first look at component-level energy breakdowns in private 5G systems, which is relevant to O-RAN and green 6G research. The chosen open-source testbed (OAI, USRP, Netio) is well suited to this purpose, and the data-collection pipeline is described in enough architectural detail to be reimplemented. However, the significance is currently limited because the reported energy values lack the measurement metadata necessary for verification: step durations, sampling rates, integration method, and uncertainty are all absent. In its present form, the central quantitative conclusions about which component dominates under traffic are not reproducible, and the paper should not be accepted without addressing these omissions.","major_comments":[{"comment":"The reported step-wise energy increments (e.g., OAIBox +9.58 Wh in downlink, +1.68 Wh in modem activation, +1.93 Wh in core start) are unnormalized by phase duration. Since Wh is time-integrated power, these values are comparable only if each step lasted the same length of time, which the manuscript does not report. If the downlink iperf phase ran longer than the modem-activation phase, the 9.58 Wh figure would reflect the longer observation window rather than a higher power draw. The manuscript must provide the duration of each step and either the average power per step or the energy normalized to a common time base, such as Wh per minute.","section":"Section 4, Figure 2"},{"comment":"The methodology does not specify the Netio PowerBox sampling interval, the MQTT publication rate, or how the instantaneous power values from Eq. (1) are accumulated into the reported Wh values. Without this information, the energy numbers in Section 4 are not reproducible, and transient power spikes during modem activation or the start of iperf bursts could be systematically missed if the sampling rate is too low. The authors should state the publication frequency, the averaging/integration window, and how phase boundaries were defined and synchronized with the network events.","section":"Section 3, Eq. (1)"},{"comment":"No repetition count or measurement uncertainty is provided for any reported value. Several claimed effects are very small (for example, 0.02 Wh and 0.03 Wh increments at the USRP during gNB start and UE attach), so without error bars or repeated trials it is impossible to tell whether these deltas exceed the measurement noise of the Netio PowerBox. The authors should report the number of repeated runs, the standard deviation or confidence intervals, and the accuracy specification of the power meter.","section":"Section 4"}],"minor_comments":[{"comment":"The paper repeatedly says this is the first or a missing aspect of energy profiling for private 5G components, but Section 2 cites only four related works; the novelty claim would be more credible with a broader literature review covering other experimental energy-profiling studies of base stations, cores, and radio units.","section":"Abstract / Section 1"},{"comment":"The phrase \"roughly\" is used for all measurements, but no numeric precision or rounding convention is defined; please specify the level of precision or provide exact averaged values.","section":"Section 4"},{"comment":"The figure caption and the figure itself appear duplicated in the text, with the labels \"continoues energy monitoring process\" appearing twice; please remove the duplication and fix the typo to \"continuous\".","section":"Figure 1"},{"comment":"The phrase \"real-time sugeries\" should be corrected to \"real-time surgeries\".","section":"Section 1"},{"comment":"The reference list is very short for a survey-oriented introduction; some claims, such as the security implications of O-RAN stated in Section 2, would benefit from additional citations beyond the four included works.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"I see the manuscript as a compact extended-abstract style paper. The experimental setup is relevant and the topic is timely for the green-networking community, but the central quantitative claims currently rest on energy values that cannot be interpreted without phase durations and integration details. These are fixable with a revised methodology description and a small additional measurement-effort table, so I recommend major revision rather than rejection. I would also ask the editor to check whether the 'first' novelty claim should be verified against related work published at venues like IEEE ICC, Globecom, or the 5G/6G energy-efficiency workshops."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it does something real: it measures the energy draw of the 5G core (OAIBox) and the radio (USRP B210) separately across idle, core start, gNB start, modem connect, active, downlink, and uplink stages. That fills a modest gap, since most prior work focuses on public network optimization or system-level energy efficiency. Second, the reported numbers are not yet credible as published, because they are absolute Wh increments without any statement of how long each phase lasted. That is the load-bearing weakness.\n\nThe setup is straightforward and honestly described: Netio PowerBox 4KF sampling voltage, current, and power factor, published over MQTT, stored in InfluxDB, visualized in Grafana. The authors acknowledge the single-access-point limitation. The observation that the core dominates radio energy is plausible and consistent with common sense, but the paper never establishes it rigorously. The 9.58 Wh downlink increase over the OAIBox, compared to 1.68 Wh for modem activation, could simply mean the downlink test ran longer. The idle baseline of 6.91 Wh is meaningless without a time interval. The paper also omits the Netio sampling rate, how the MQTT stream was integrated into energy, and how phase boundaries were set. No repetition, no error bars, no statistical analysis. None of this is fatal to the underlying idea, but it means the central empirical claims are not reproducible from the manuscript text.\n\nTo their credit, the authors do not overclaim. They call it an experimental analysis, mention future work in a real hospital, and note the lab's simplicity. The literature review is brief but correctly identifies the missing component-level profiling in private networks. The measurement infrastructure is a reasonable template for further studies.\n\nFor a serious referee, I would send this in, but the revision expectations should be explicit: report the duration of each phase, the sampling interval and accuracy of the Netio setup, the integration method, and ideally raw data or a public repo. With those additions, this becomes a useful baseline for anyone building or simulating private 5G energy models. Without them, it is an anecdote. I would not cite it in my own work until the numbers are reproducible.","headline":"A genuinely useful per-component energy profile of a private 5G lab network, but the headline Wh numbers are uninterpretable without phase durations and sampling details.","tokens_in":4426,"tokens_out":1764,"would_cite":false,"duration_ms":17315,"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":"The paper reports the first experimental energy profiling of individual private 5G components, showing the software-based core is the dominant energy consumer while the radio front-end costs a fraction of a watt-hour.","keywords":["private 5G","energy profiling","network energy consumption","Open Air Interface","O-RAN","5G core network","radio access network","6G sustainability"],"falsifier":"Repeat the staged experiment with a high-frequency wattmeter (1 Hz or faster) logging the OAIBox and USRP power continuously, and compare the integrated energy of the downlink phase against the paper's 9.58 Wh core and 0.30 Wh radio figures; if the integrated totals differ by more than the meter's rated accuracy, the reported profiles are artifacts of the measurement window.","tokens_in":3483,"feed_emoji":"⚡","tokens_out":8217,"duration_ms":65154,"temperature":0.7,"pith_summary":"This paper attempts to establish a first experimental energy breakdown for a private 5G network by profiling the network core and the radio unit separately under idle, attach, and data-transfer conditions. Using a socket-level power meter, it reports that the software-based core (an OAIBox server) is the dominant consumer, drawing about 6.91 Wh at idle and an additional 9.58 Wh during a downlink burst, while the radio (a USRP B210) draws about 0.23 Wh at idle and 0.30 Wh in the same burst. The point of the measurement is to show where energy actually goes in a small standalone private network, so that sustainability engineering for 5G and 6G can target the right component. A careful reader would care because private 5G is entering hospitals and factories, where both energy cost and carbon footprint matter.","feed_headline":"Private 5G's energy hog is the core, not the radio","feed_subtitle":"Idle core draws 6.91 Wh; a downlink burst adds 9.58 Wh while the radio adds only 0.30 Wh","key_machinery":"The work is carried by a socket-level power-measurement chain plus a staged network-state protocol. A Netio PowerBox 4KF samples voltage, current, and true power factor for each device and publishes the values over MQTT; an Ubuntu VM subscribes, computes instantaneous power as $P = V \\cdot I \\cdot pf$, and stores the time series in InfluxDB, from which Grafana renders the curves. In parallel, the experimenter steps the private 5G network through defined states (core off, core on, gNB on, UE attached, modem active, iPerf downlink, iPerf uplink) so that the energy difference between consecutive steps can be attributed to a single component or action. The OAIBox Max provides the 5G core and gNB software, the USRP B210 acts as the radio front-end, a Quectel modem is the UE, and iPerf generates the traffic.","core_discovery":"On the paper's own terms, the discovery is that per-component energy profiling of a private 5G network is practical and yields a clear rank order: the 5G core, running Open Air Interface software on a commodity server, uses roughly 6.91 Wh while idle and grows by about 9.58 Wh during a downlink iPerf burst, whereas the USRP B210 radio uses about 0.23 Wh idle and adds only 0.30 Wh in the same burst. The authors step through nine network states — idle, core start, gNB start, UE plug-in, modem activation, server start, downlink, return, and uplink — and attribute the meter readings to the OAIBox and USRP separately. They conclude that the core is the energy bottleneck in a laboratory-scale private 5G network and that this component-level view, which existing optimization studies do not provide, is a necessary input for designing energy-efficient private 5G and 6G architectures.","pith_inferences":["I would expect the core's dominance to shrink in a real multi-cell deployment with more radios, but the per-cell figures here suggest the core-server share stays substantial because the radio is so cheap per unit.","The reported Wh values are phase-integrated, not power readings; normalizing to average watts or energy per transferred bit would allow direct comparison with other testbeds and could be done from the stored InfluxDB data.","A natural next experiment is to test whether the core's idle draw can be reduced by container-level scaling or sleep modes, since the 1.93 Wh jump at core start and 0.81 Wh at UE plug-in indicate which state transitions cost energy."],"forward_implications":["If these profiles hold, energy-optimization work for private 5G should focus on the core/gNB software: its idle draw of 6.91 Wh is the largest fixed cost in the small setup.","Traffic is the main variable cost on the core, with a downlink burst adding about 9.58 Wh and uplink a similar 9.63 Wh, so traffic shaping or burst scheduling on the core could yield measurable savings.","The radio front-end's tiny delta (0.30 Wh downlink, 0.28 Wh uplink) suggests that radio-side optimization alone will not move the total bill much in a single-cell private network.","The same measurement chain can be applied to additional components, such as edge nodes or UEs, to extend the breakdown toward a full private-network energy budget."],"supporting_citations":[{"why":"Supplies the massive MIMO energy-efficiency background that the paper contrasts with its component-level approach.","marker":"[1]"},{"why":"Motivates deeper energy profiling by linking energy harvesting and wireless power transfer to network elements.","marker":"[2]"},{"why":"Establishes that O-RAN security mechanisms add energy consumption, making element-wise profiling relevant for green security.","marker":"[3]"},{"why":"Represents the adaptive massive MIMO optimization approach that treats the network as a whole without per-component breakdown.","marker":"[4]"}],"fun_headline_variants":["Core dominates private 5G energy profile","5G core uses 6.91 Wh idle, radios add little","Energy profiling: 5G core is the main draw","Private 5G power: core vs radio measured"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central numbers stand on the unstated assumption that the power meter's sampling and the phase timing capture the true energy of each step; the paper reports neither the sampling interval nor how the Wh values were integrated, so missed transients could shift the deltas.","fun_headline_variants_meta":{"raw":{"variants":["Core dominates private 5G energy profile","5G core uses 6.91 Wh idle, radios add little","Energy profiling: 5G core is the main draw","Private 5G power: core vs radio measured"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00021,"raw_usage":{"total_tokens":1374,"prompt_tokens":872,"completion_tokens":502,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":488,"completion_tokens_details":{"reasoning_tokens":436}},"tokens_in":488,"tokens_out":502,"duration_ms":4955,"temperature":1.0,"reasoning_tokens":436,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T04:13:06.169311+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Repeat the staged experiment with a high-frequency wattmeter (1 Hz or faster) logging the OAIBox and USRP power continuously, and compare the integrated energy of the downlink phase against the paper's 9.58 Wh core and 0.30 Wh radio figures; if the integrated totals differ by more than the meter's rated accuracy, the reported profiles are artifacts of the measurement window.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the massive MIMO energy-efficiency background that the paper contrasts with its component-level approach."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Motivates deeper energy profiling by linking energy harvesting and wireless power transfer to network elements."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes that O-RAN security mechanisms add energy consumption, making element-wise profiling relevant for green security."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Represents the adaptive massive MIMO optimization approach that treats the network as a whole without per-component breakdown."}],"review_version":1}