{"id":"3e229412-3c13-4abf-b896-953826a3cb6c","arxiv_id":"2505.24763","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"In standardized 3GPP urban simulations, 5G PRS signals used as a monostatic radar can detect drones, with up to 16% missed detections in dense UMi at 25 m altitude and position errors within 4 m (UMi) and 8 m (UMa).","lead":"This paper simulates using standard 5G cell tower signals (Positioning Reference Signals) as a radar to detect drones in urban environments. It finds detection is hardest in dense urban microcells at low altitude, and that position estimates stay within 4 to 8 meters depending on scenario.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (9) subtracts the temporal mean over PRS symbols, which cancels a zero- or low-Doppler target along with the static clutter; the paper never states the UAV velocity/Doppler used, so the headline detection rates may not apply to hovering or slow-moving UAVs.","rationale":"The reader's weakest_assumption correctly identified Eq. (9) and the perfect-cancellation assumption, but framed it mainly as an issue of unrealistic static clutter (NRP=3). My concern sharpens the same equation: the cancellation is not merely imperfect in real clutter; it also removes the target when the target's Doppler phase change across the averaging window is small. Since the paper does not specify the UAV velocity or Doppler shift, the reported detection and positioning numbers are under-determined. This is a load-bearing concern because the central claim is about detecting UAVs generally, and hovering or slow-drifting UAVs are common operational cases. The concern is addressable: the authors can report the Doppler used and evaluate a range of velocities, including zero. It does not by itself refute the method for fast-moving targets, so the appropriate verdict remains CONDITIONAL, consistent with the reader's original verdict. I selected UNCHANGED because my read does not move the verdict away from CONDITIONAL; it strengthens the conditions (disclose Doppler, test low-Doppler regime) rather than changing the overall assessment. I partially agree with the reader because we both point at Eq. (9), but the reader's framing misses the target-cancellation failure mode, which is arguably the more severe issue.","tokens_in":8612,"tokens_out":6336,"duration_ms":75706,"concrete_test":"Re-run the simulations with the UAV radial velocity set to 0, 1, 5, and 10 m/s (Doppler at 30 GHz: 0, 200, 1000, and 2000 Hz) while keeping all other parameters fixed; report missed-detection probability and position error for each velocity and for both UMi and UMa at 25 m altitude. If the 0 m/s case yields near-100% missed detections, or the 1–5 m/s cases are substantially worse than the headline 16%/3%/1% rates, then the central feasibility claim must be explicitly scoped to moving targets and the paper must disclose the Doppler used to produce Fig. 5. Alternatively, an analytical check of the residual target amplitude after mean subtraction as a function of f_D L would settle the severity of the cancellation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The clutter suppression step in Eq. (9) computes eH = H - E_l[H], where l indexes PRS symbols transmitted with the same beam. For a target with complex amplitude a*exp(j2π f_D l) (Eq. 7), subtracting the empirical mean over L symbols leaves a residual of order a·π f_D (L-1) when f_D L is small, and exactly zero when f_D = 0. Thus a hovering UAV—arguably a primary use case—would be suppressed together with the static background. The simulation setup in Section IV-A specifies carrier frequency, bandwidth, PRS comb, array size, EIRP, RCS, and NRP=3, but omits any target velocity or Doppler shift. It also does not state the coherent integration time over which the mean is taken (LPRS=4 symbols per occasion vs. NPRS=256 occasions aggregated). At 30 GHz, a 10 m/s radial velocity gives f_D ≈ 2 kHz; over four 120 kHz-subcarrier OFDM symbols (~30–35 μs), the phase advance is only ~0.4 rad, so most target energy remains in the subtracted mean and detection would be severely degraded. The reported miss rates of 16% (UMi, 25 m), 3% (UMi, 50 m), and ~1% (UMa, 200 m) are therefore uninterpretable without knowing the Doppler assumed. If a large Doppler (e.g., a fast jet) was used, the results do not support the general claim of 'UAV detection.' This is more load-bearing than the sparse clutter model because it affects the core detection mechanism itself, not just the environmental realism.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies whether a single 5G NR base station can use standardized Positioning Reference Signals (PRS) as a monostatic OFDM radar to detect and localize UAVs. It builds a complete processing chain: channel and clutter modeling, angle-of-arrival estimation with temporal mean subtraction, effective channel estimation, range estimation via IDFT, peak-to-average detection, and 3D position reconstruction. Simulation results are reported for 3GPP UMi-AV and UMa-AV scenarios at 30 GHz for UAV altitudes of 25, 50, 100, and 200 m, claiming missed-detection rates up to about 16% in UMi at 25 m, negligible misses in UMi above 100 m, near-zero misses in UMa below 100 m, and position errors within about 4 m in UMi and 8 m in UMa. The authors have released the simulation platform as open-source software.","tokens_in":8980,"tokens_out":4347,"duration_ms":55039,"significance":"If the results hold, the paper would provide a useful feasibility data point for using existing 5G NR infrastructure for passive or cooperative sensing of aerial targets, which is relevant to ISAC standardization and to practical drone-detection applications. The paper is strongest where it is explicit and reproducible: it follows established OFDM radar processing, uses 3GPP TR 36.777 and TR 38.901 models, adopts a 3GPP-motivated UAV RCS value, and releases the code, which are genuine strengths. The central feasibility claim, however, is currently built on a simulation whose target Doppler is unspecified and whose clutter environment is extremely sparse; these omissions directly affect the validity of the headline detection numbers.","major_comments":[{"comment":"The clutter-suppression step subtracts the empirical mean over the symbol index l: eH = H - E_l[H]. For the target model in Eq. (7), hST_q,l = a_q exp(j2π f_D,q l), this subtraction leaves a residual that vanishes as f_D,q approaches zero and exactly cancels a zero-Doppler target. The simulation setup in Section IV-A specifies LPRS=4 symbols per occasion and NPRS=256 occasions, but it does not state the UAV radial velocity or the Doppler shift f_D used in the simulations, nor whether l in Eq. (9) runs over the four symbols within an occasion, over all 256 occasions, or both. At 30 GHz, a 10 m/s radial velocity gives f_D ≈ 2 kHz, which corresponds to only about 0.4 rad of phase advance across four 120 kHz-subcarrier OFDM symbols; under Eq. (9), most of the target energy is then removed together with the static clutter. The reported miss rates in Fig. 5a are therefore uninterpretable without knowing the assumed target Doppler, and the paper's feasibility claim does not currently cover hovering or slow-moving UAVs, which are a primary use case for drone detection.","section":"Section III-A, Eq. (9); Section IV-A"},{"comment":"The detection threshold η = 3.4 dB is selected from the same 64,000-sample simulated dataset that is then used to report the miss-detection probabilities in Fig. 5a. The reported values (about 16% in UMi at 25 m, about 3% at 50 m, about 1% in UMa at 200 m) are in-sample operating points evaluated at a threshold tuned on the same draws. The paper should provide receiver operating characteristics or PAR detection/false-alarm curves and demonstrate that the qualitative conclusions are stable under reasonable threshold variation, or use a separate calibration set and a test set. Without that, the absolute miss-rate numbers are not robust evidence for the feasibility claim.","section":"Section IV-B, Fig. 5"},{"comment":"The background clutter is generated from only NRP=3 random reference points, sampled from a scenario-dependent distribution and combined as in Eq. (5). This is a very sparse model of an urban scattering environment. The paper attributes the higher UMi miss rates to 'severe clutter,' but with only three reference points the clutter statistics are unlikely to represent dense urban environments, and no sensitivity analysis with respect to NRP or the RP density is provided. The conclusion that the system 'demonstrates feasibility in realistic urban propagation environments' is therefore not yet supported; the authors should either justify the choice of NRP=3 or show that the detection and positioning results are robust to the number and placement of clutter reference points.","section":"Section II-B, Eq. (5); Section IV-A"}],"minor_comments":[{"comment":"The sentence '5G NR radars exhibits the highest missed detection rate' has a subject-verb agreement error ('radars exhibits' should be 'radars exhibit').","section":"Abstract"},{"comment":"The phrase 'σM,q is a the mean RCS value' contains a typo ('a the'); it should read 'is the mean RCS value'.","section":"Section II-B, Eq. (6)"},{"comment":"The symbol index l is dropped in the PAR definition of Eq. (15); the authors should clarify whether the detection statistic is computed per PRS symbol, averaged over the LPRS symbols in an occasion, or averaged over the NPRS occasions, since this affects the interpretation of the detection probabilities.","section":"Section III-C, Eq. (12) and Eq. (15)"},{"comment":"The table entry for UMa-AV LOS probability is shown as '+' with no explanation; the authors should spell out the LoS probability formula for UMa-AV or refer explicitly to the relevant TR 36.777 table.","section":"Section IV-A, Table I"},{"comment":"The caption says '5G NR Radar operating curves,' but the two panes plot PFA versus η and PD versus η; a conventional operating curve would show PD versus PFA as η varies. The authors should relabel the figure or adjust the caption to avoid confusion.","section":"Fig. 3"}],"recommendation":"major_revision","confidential_remarks":"The most serious issue is the absence of any target Doppler specification in a processing chain whose first step is a temporal mean subtraction that cancels zero- and low-Doppler targets. This is a load-bearing omission that should have been caught by the authors; it can be fixed by rerunning the open-source simulator with explicit radial velocities (including zero velocity) and reporting how the detection rates depend on Doppler. The threshold-calibration and sparse-clutter issues also need to be addressed before the paper can support its feasibility claim. I do not see this as a reject, because the signal model and processing chain are sound and the code is available, but the current manuscript's headline numbers are not yet interpretable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a competent simulation study with an open-source release, and the 3GPP-based UMi/UMa comparison is genuinely useful. But the central numbers have a hole that matters more than the reader's report suggests: the clutter suppression in Eq. (9) subtracts the temporal mean of the channel across PRS symbols, which cancels any target whose phase is roughly constant over that window. The paper never specifies the UAV velocity or Doppler shift, and at 30 GHz a 10 m/s radial velocity gives only about 0.4 rad of phase advance over four 120 kHz symbols. So a hovering or slow-moving UAV—the primary drone-detection case—would be suppressed along with the static clutter. The reported 16% miss rate in UMi and the error CDFs say nothing about detecting the targets the paper claims to detect unless the simulation gave the UAV a large Doppler, which is not stated. This is a load-bearing gap, not a minor parameter omission.\n\nWhat the paper does well: the processing chain is standard OFDM radar but it is implemented completely and clearly, the use of 3GPP TR 36.777 aerial models with height-dependent LoS is appropriate, and the open-source code is a real contribution. The threshold ROC analysis is standard practice. I agree with the reader that the contribution is incremental—prior experimental work detected UAVs with 5G signals—but the systematic environment comparison adds value.\n\nOther soft spots: the threshold eta = 3.4 dB is chosen from the same 64,000-sample dataset that produces the reported miss rates, so those numbers are in-sample. The clutter model uses only NRP = 3 static reference points with no sensitivity analysis; denser or dynamic clutter would change the UMi results substantially. No baseline comparison to a simpler detector or to classical radar is provided, which makes the feasibility claim hard to judge.\n\nBottom line: the paper deserves a serious referee because the framework and code are useful, but as it stands the headline claim is not supported. The authors need to state the assumed velocity/Doppler, show how detection varies with it, and calibrate the threshold on independent data. If they do that, this becomes a citable engineering reference. For now I would not cite the quantitative results.","headline":"Useful simulation framework with real release value, but the headline detection rates are uninterpretable because the target's Doppler/velocity is never stated and the clutter-suppression step cancels slow-moving UAVs.","tokens_in":9513,"tokens_out":2479,"would_cite":false,"duration_ms":27807,"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":"A single 5G New Radio base station can detect and localize small drones using the standard Positioning Reference Signals already in the system, with 3D position errors within 4 meters in dense urban microcells and within 8 meters in…","keywords":["5G New Radio","Positioning Reference Signals","integrated sensing and communication","drone detection","UAV localization","OFDM radar","clutter suppression","urban propagation"],"falsifier":"Run the open-source simulator with many more background scatterers than the three used in the baseline, such as ten or twenty reference points or one moving scatterer, and observe the 25 m urban-microcell missed-detection rate: if it stays near 16% and position error near 4 m, the central claim is robust; if it rises sharply, the reported performance is an artifact of the simulated clutter. An outdoor 30 GHz monostatic test with a small drone at 25 m in a dense urban canyon and a ground-truth tracker would settle the same question directly.","tokens_in":8436,"feed_emoji":"📡","tokens_out":9171,"duration_ms":102171,"temperature":0.7,"pith_summary":"The paper claims that a standard 5G New Radio base station can act as a radar for detecting and locating small drones, using the Positioning Reference Signals (PRS) already defined in the 5G standard rather than a dedicated radar waveform. A complete processing chain is simulated—background clutter removal, angle-of-arrival search, range estimation, and 3D position reconstruction—and tested in standard urban microcell and macrocell propagation environments. The central finding is that performance depends strongly on deployment: dense urban microcells miss up to about 16% of drones at 25 m altitude, dropping to about 3% at 50 m and negligible above 100 m, while macrocell deployments miss almost nothing below 100 m and only about 1% at 200 m. When a detection is made, the 3D position error stays within about 4 m in the microcell and 8 m in the macrocell case. This matters because, if true, the existing cellular network could supply a low-cost layer of airspace monitoring without new spectrum or dedicated radar hardware.","feed_headline":"Standard 5G signals locate drones within meters","feed_subtitle":"Simulation: one base station can detect and position small drones, with misses near zero above 100 m.","key_machinery":"The central mechanism is the PRS comb resource grid treated as an OFDM radar waveform. PRS occupies every $K$-th subcarrier, so the phase shift across the active tones encodes the round-trip delay, and the repeated PRS occasions provide the repeated time samples needed for Doppler and for background removal. The processing chain targets three operations: clutter suppression by subtracting the empirical mean over PRS symbols, which relies on the static-environment assumption; angle-of-arrival estimation by a beam-sweeping search over the analog beamforming codebook; and range estimation by an IDFT over the active subcarriers. Detection is decided by comparing the peak-to-average ratio of the range profile to a threshold, and the final 3D position is computed from the angle and range estimates.","core_discovery":"On the paper's own terms, the discovery is that standardized PRS pilot symbols—periodic comb-pattern signals already transmitted by 5G base stations—carry enough phase and delay information for a single base station to act as a monostatic radar, meaning one site that both transmits and receives, for a small drone. The base station transmits PRS, receives the round-trip echo, subtracts the time-averaged received signal to remove static urban clutter, sweeps a beam codebook to find the drone's azimuth and elevation, estimates range from the phase progression across active subcarriers via an IDFT, and declares a detection when the range profile's peak-to-average ratio exceeds a threshold. In the most difficult tested configuration, a dense urban microcell with the drone at 25 m altitude, the missed-detection probability reaches about 16%, falling to about 3% at 50 m and negligible at 100 m and above. In the urban macrocell configuration, missed detections are negligible below 100 m and about 1% at 200 m. Position error grows with target distance, staying within about 4 m in the microcell and within 8 m in the macrocell across the tested altitudes.","pith_inferences":["The reported numbers are bounded by a simulated clutter scene made of only three synthetic reference points; a real urban deployment would likely see denser and partially moving clutter, so the 16% microcell miss rate and the 4 m/8 m errors are best-case estimates rather than guarantees.","The same PRS-based processing could be tried at lower frequencies or with wider aggregated bandwidths, which would change range resolution and clutter behavior in ways the paper does not test.","The single-target peak detector would need to become a multi-peak resolver before the chain can track several drones, since the current angle and range estimators pick only the strongest reflection.","A cooperative version with several base stations sharing angle and range estimates would likely mitigate the low-altitude microcell gap and the long-distance macrocell error, because the different viewing geometries would help disambiguate non-line-of-sight paths."],"forward_implications":["Existing 5G base stations could act as a complementary drone-detection layer in cities, using only standard PRS resources and no extra spectrum or radar transmitter.","The best deployment for close, low-altitude targets is a dense urban microcell, where position error stays near 4 m, but it pays for that accuracy with a 16% missed-detection rate at 25 m altitude.","The best deployment for reliable detection is a macrocell, where missed detections are negligible below 100 m but position error grows to about 8 m as target distance increases.","The 3.4 dB peak-to-average detection threshold gives a single operating point with a specific false-alarm and detection trade-off across all tested scenarios and altitudes.","Because the range estimate comes from a fixed-width IDFT over the active PRS subcarriers, the same chain can be reconfigured by changing the comb spacing $K$ and the PRS occasion length, trading range ambiguity for detection quality."],"supporting_citations":[{"why":"Establishes the approach this work builds on: using 5G PRS as a sensing reference signal for joint communication and sensing.","marker":"[2]"},{"why":"Specifies the PRS waveform, comb subcarrier mapping, and Gold-sequence generation used in the transmit signal.","marker":"[10]"},{"why":"Provides the standardized urban path-loss and shadow-fading models that form the channel backbone.","marker":"[11]"},{"why":"Defines the UAV deployment scenarios, base station heights, and height-dependent line-of-sight probability used in the evaluation.","marker":"[12]"},{"why":"Supplies the OFDM radar processing framework, including range IDFT and peak-based detection.","marker":"[14]"},{"why":"Introduces the reference-point scattering model used to synthesize the static background clutter.","marker":"[15]"}],"fun_headline_variants":["5G towers detect drones with 4m accuracy","One 5G base station locates drones within 4m","Standard 5G signals find drones, 16% miss at low altitude","5G radar: single base station, 4m drone position"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole result rests on the assumption that the urban background is static enough that subtracting the time-averaged received signal removes it completely, and that the simulated three-point clutter scene stands in for a real city.","fun_headline_variants_meta":{"raw":{"variants":["5G towers detect drones with 4m accuracy","One 5G base station locates drones within 4m","Standard 5G signals find drones, 16% miss at low altitude","5G radar: single base station, 4m drone position"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000281,"raw_usage":{"total_tokens":1675,"prompt_tokens":968,"completion_tokens":707,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":633}},"tokens_in":584,"tokens_out":707,"duration_ms":7871,"temperature":1.0,"reasoning_tokens":633,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:13:27.066740+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the open-source simulator with many more background scatterers than the three used in the baseline, such as ten or twenty reference points or one moving scatterer, and observe the 25 m urban-microcell missed-detection rate: if it stays near 16% and position error near 4 m, the central claim is robust; if it rises sharply, the reported performance is an artifact of the simulated clutter. An outdoor 30 GHz monostatic test with a small drone at 25 m in a dense urban canyon and a ground-truth tracker would settle the same question directly.","supporting_citations":[{"cited_title":"5G prs-based sensing: A sensing reference signal approach for joint sensing and communication system,","cited_arxiv_id":null,"evidence_quote":"Establishes the approach this work builds on: using 5G PRS as a sensing reference signal for joint communication and sensing."},{"cited_title":"NR; Physical channels and modulation (Release 17),","cited_arxiv_id":null,"evidence_quote":"Specifies the PRS waveform, comb subcarrier mapping, and Gold-sequence generation used in the transmit signal."},{"cited_title":"Study on channel model for frequencies from 0.5 to 100 GHz (Release 16),","cited_arxiv_id":null,"evidence_quote":"Provides the standardized urban path-loss and shadow-fading models that form the channel backbone."},{"cited_title":"Study on Enhanced LTE Support for Aerial Vehicles (Release 15),","cited_arxiv_id":null,"evidence_quote":"Defines the UAV deployment scenarios, base station heights, and height-dependent line-of-sight probability used in the evaluation."},{"cited_title":"thesis, Karlsruhe, Karlsruher Institut f ¨ur Technologie (KIT), Diss., 2014, 2014","cited_arxiv_id":null,"evidence_quote":"Supplies the OFDM radar processing framework, including range IDFT and peak-based detection."},{"cited_title":"A novel approach to model the scattering environment in channel modeling for integrated sensing and communications,","cited_arxiv_id":null,"evidence_quote":"Introduces the reference-point scattering model used to synthesize the static background clutter."}],"review_version":1}