{"id":"daf9756e-ec56-4a0b-9a65-8544273fea7a","arxiv_id":"1908.03418","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Standard-compliant LTE/5G NR downlink waveforms, combined with null-subcarrier interpolation and RF/digital self-interference cancellation, can support monostatic OFDM radar sensing of static and moving targets.","lead":"This paper shows that standard 4G LTE and 5G NR base station signals can be used as radar waveforms for detecting and measuring the range and speed of drones and cars, if empty subcarriers are interpolated and transmitter leakage is cancelled. It also reports measured self-interference cancellation of about 100 dB and field demonstrations with static and moving targets.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Measured ROC in Fig. 8 is computed from a 3×3 window around the known drone position, not over the full range–Doppler search space, so it does not validate system-level false-alarm control; the claimed static-target detection performance may be optimistic.","rationale":"I considered the reader's stated weakest assumption, namely that the three-tap RF canceller and 5+5-tap digital canceller generalize to real base-station coupling; that is a legitimate external-validity concern. However, the measured ROC procedure is more directly load-bearing because it affects whether the quantitative detection evidence in the paper is valid even for the demonstrated setup. Section II-B defines system-level PFA,tot over the search space, while Section IV-C evaluates empirical threshold behavior on a 3x3-pixel window around the known target. This is an internal inconsistency between the paper's own detection criterion and the measurement analysis, not merely a question of extrapolation to other hardware. The concrete test is feasible from the existing recordings and would settle whether the claimed ROC survives a full-surface false-alarm definition. The paper still provides a plausible engineering demonstration: the range-Doppler images show a clear peak at the drone, the SI isolation measurements are detailed, and the cancellation design is reasonable. Therefore the correction would not overturn the paper's central concept, but it does justify keeping the verdict CONDITIONAL rather than ACCEPT. The same verdict is retained, so no adjustment to the reader's outcome is needed.","tokens_in":19352,"tokens_out":6591,"duration_ms":76114,"concrete_test":"Using the same stored 100 H0 and 100 H1 measurement frames that produced Fig. 8, recompute the ROC by thresholding the full range-Doppler surface: for each image and each threshold, record a false alarm if any cell outside a small guard window around the true drone position exceeds the threshold, and record a detection if the cell inside the guard window exceeds it. Plot PD against PFA,tot over the entire search space and compare with Fig. 8. If the system-level PD at PFA,tot = 10% is substantially lower than the curve shown, the measured detection performance is overstated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central measurement-based claim rests on the ROC in Fig. 8, but that ROC is not computed with the detector defined in Section II-B. There, the false-alarm probability is defined over the entire search space as PFA,tot = 1 - (1-PFA)^|Omega_A|. In Section IV-C, however, the empirical H0 and H1 distributions are built 'covering 3x3 pixels of the periodograms at and around the drone location', and the threshold is varied on those local distributions. Thus Fig. 8 reports a per-cell or local-window trade-off, not a system-level ROC. A threshold with a good local PFA can still generate frequent false alarms elsewhere in the range-Doppler map, from SI sidelobes, the 10 m building reflection visible in Fig. 7(d), or other clutter. The radar images show a visible 40 m peak, but the quantitative claim that the cancellers provide a specified detection probability at a specified false-alarm rate is not substantiated by the reported procedure. This weakens the abstract's assertion that the cancellation solutions are 'shown through RF measurements' to enable static-target detection.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes and evaluates a monostatic OFDM radar concept that reuses standard-compliant LTE and 5G NR downlink transmissions. It formulates frequency-domain radar processing based on division by the transmit grid, introduces linear interpolation across OFDM symbols to handle null subcarriers, and derives target detection and range/velocity estimation via a periodogram. Simulation results for NR bandwidths of 20, 40, and 100 MHz show detection down to very low SNR and estimation RMSE approaching the resolution limits. The paper then addresses self-interference in shared-antenna operation, describing a three-tap RF canceller and a nonlinear digital canceller with a self-orthogonalizing update rule, and reports RF measurements at 2.4 GHz with a 40 MHz NR waveform: about 100 dB total isolation, detection of a static drone at 40 m, and detection of three moving vehicles with range/velocity estimates consistent with the scenario.","tokens_in":19599,"tokens_out":5309,"duration_ms":56551,"significance":"Assuming the central claims hold, this is a valuable end-to-end demonstration of joint communications and sensing using standard-compliant LTE/NR waveforms. The paper combines a standard frequency-domain OFDM radar processing model with a concrete solution to the self-interference problem: a three-tap RF canceller and a nonlinear digital canceller with self-orthogonalizing adaptation, measured to provide about 100 dB total TX-RX isolation. The simulations provide falsifiable predictions for detection and estimation performance across carrier bandwidths, and the outdoor measurements with a drone and vehicles are a rare hardware demonstration in this area. The radar-specific tailoring that preserves target echoes beyond a few meters is a clear distinction from generic full-duplex self-interference cancellation work. The paper is well written and the results are internally consistent; the most important weakness is the mismatch between the detector definition in Section II-B and the empirical ROC evaluation in Section IV-C.","major_comments":[{"comment":"The measured ROC is not computed with the detector defined in Section II-B. In Section IV-C, the H0 and H1 empirical distributions are built from a 3x3 pixel window at and around the drone location, and the threshold is varied on those local distributions, whereas the detector in Eq. (6) operates on the full search space Omega_A and the text in Section II-B defines PFA,tot = 1 - (1-PFA)^{|Omega_A|}. Consequently, Fig. 8 reports a local per-cell trade-off, not a system-level ROC: a threshold with acceptable local PFA can still produce frequent false alarms elsewhere in the range-Doppler map, e.g., from SI sidelobes or the 10 m reflection visible in Fig. 7(d). The abstract's statement that the cancellation solutions are 'shown through RF measurements' to enable static-target detection at a specified PD/PFA is therefore not quantitatively substantiated. Please recompute the ROC over the full Omega_A or present map-level CFAR statistics over all range-Doppler bins, and adjust the claims accordingly.","section":"Section IV-C, Fig. 8"},{"comment":"The simulation curves in Fig. 2 are presented without confidence intervals and without a stated number of Monte Carlo realizations. Because the detection probability and RMSE values are averaged over randomly drawn target distances and velocities (uniform over 20-200 m and -40 to 40 m/s), the reader cannot judge whether the sharp transitions in Fig. 2(a) or the RMSE floors in Fig. 2(b)-(c) are statistically stable. Please report the number of trials per SNR point and add confidence intervals or standard-error bars to the curves.","section":"Section II-C, Fig. 2"},{"comment":"The cancellation design assumes that the self-interference channel is sufficiently modeled by three RF taps with delays up to 10 ns (Eq. (9)) plus a 5+5 tap digital canceller at 240 MHz (Eq. (11)), and the paper argues that this preserves target echoes beyond roughly 1.5-3 m. The measurements validate this assumption for the particular anechoic-chamber and outdoor coupling environments, but the paper does not characterize the delay spread of the measured coupling channel nor test scenarios with longer-delay multipath in the SI path. Since a base-station deployment may have coupling reflections beyond 10 ns, the general claim in Section V that these cancellers are key ingredients for practical deployments is stronger than the evidence supports. I recommend either adding a coupling-channel delay-spread measurement or explicitly bounding the applicability of the cancellation parameterization.","section":"Section III-B and IV-B"}],"minor_comments":[{"comment":"The word 'probablity' should be 'probability' in the paragraph following Fig. 2.","section":"Section II-C"},{"comment":"The ROC procedure is underspecified: please state whether the 100 H0 and 100 H1 measurements are independent, how the 3x3 pixel window is selected, and how the threshold sweep is mapped to PFA values. With only 100 images per hypothesis, confidence intervals on the ROC curves would also be informative.","section":"Section IV-C"},{"comment":"The integral in Eq. (10) is written with dt over sampled I/Q signals; please clarify whether this is a discrete-time accumulation and how the integration window is chosen.","section":"Eq. (10)"},{"comment":"In Eq. (12), the correlation matrix C is said to be precomputed; please state whether regularization is used for the inversion and how C is estimated from the basis function samples.","section":"Eq. (12)"},{"comment":"The caption lists only NR cases; the text states that LTE 20 MHz is omitted because it is essentially identical, but the caption should also note this omission for clarity.","section":"Fig. 2 caption"}],"recommendation":"major_revision","confidential_remarks":"This is a solid systems paper with a clear demonstration value. The main revision request concerns the statistical rigor of the measured ROC and the simulation reporting; the underlying processing model and hardware results appear sound. I do not see a fundamental flaw that would require rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a solid engineering paper. The new thing is the specific combination — 3GPP-standard LTE/NR resource grids, interpolation across null subcarriers, and a radar-tailored analog/RF plus nonlinear digital canceller that suppresses only the direct self-interference — and they back it up with real measurements of a static drone and moving cars. I think the central claim holds: with that cancellation, static and slow targets become visible in the range-Doppler map.\n\nWhat it does well: the system model is standard and the processing chain is described cleanly. The interpolation over unused subcarriers is simple and effective, and their simulation results for 20/40/100 MHz NR are plausible and match the expected processing gain. The measured ~100 dB TX-RX isolation is a genuinely useful result, and the comparison against building reflections gives the drone detection some credibility. The moving-car experiment with estimated velocities is a nice practical validation.\n\nSoft spots, in proportion: the stress-test note is right. The ROC in Fig. 8 is computed from empirical distributions over a 3x3 pixel window around the known drone location, not over the full detection space. That means it is a local per-cell trade-off, not the PFA_tot defined in Section II-B. A threshold with a good local false-alarm rate can still produce spurious detections elsewhere, e.g., from the 10 m building reflection visible in Fig. 7(d). The paper does not explicitly claim full-surface CFAR in the text, but the abstract's \"shown through RF measurements\" is a bit strong. This is a moderate weakness, not a fatal one — the images themselves are convincing evidence that the drone was detectable.\n\nAlso: no code or data released, so independent reproduction is limited. The measurement is at 2.4 GHz with +20 dBm and a horn antenna, so scaling to a macro base station (+46 dBm, more complex coupling) is an extrapolation, not a demonstration. The 10 ns RF canceller delay and 5+5 digital taps imply targets inside ~1.5–3 m are suppressed — they mention this, so it's not hidden, but it is worth keeping in mind for claim of general deployment. The simulation curves in Fig. 2 lack confidence intervals and trial counts — minor.\n\nBottom line: this paper is for people working on joint communications and sensing or full-duplex radar. It deserves a serious referee, but I would ask for a revised ROC that uses the full search surface and ideally a statement about false alarms outside the target bin. If that is too hard, soften the abstract's claim. I would cite the paper as a useful experimental reference, and I would bring it to reading group to discuss the ROC methodology.","headline":"Solid full-duplex OFDM radar demonstration with real measurements, but the measured ROC is computed in a local window around the known target and does not support the system-level false-alarm claim.","tokens_in":20091,"tokens_out":2478,"would_cite":true,"duration_ms":25634,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Standard-compliant LTE and 5G NR downlink waveforms can serve as monostatic OFDM radar signals when unused subcarriers are interpolated and transmitter self-interference is actively cancelled.","keywords":["OFDM radar","5G New Radio","LTE","self-interference cancellation","full-duplex radio","joint communications and sensing","range-Doppler estimation","RF convergence"],"falsifier":"Take a shared-antenna base-station front end in an environment with a strong reflector 5--10 m from the antenna, such as a mast, building edge, or vehicle, and measure the residual self-interference after the 10 ns three-tap RF canceller and 5+5-tap digital canceller; if a static target at that range is still masked by leakage sidelobes, or if the measured residual coupling exceeds the roughly 100 dB isolation shown here, the central claim fails for realistic multipath. A complementary check is to place a target at 2 m: if the canceller suppresses it along with the leakage, the minimum-detectable-range design trade-off is confirmed.","tokens_in":19161,"feed_emoji":"📡","tokens_out":5577,"duration_ms":54115,"temperature":0.7,"pith_summary":"This paper argues that standard-compliant LTE and 5G NR downlink waveforms can be used as the transmit signal of a monostatic OFDM radar, so a base station can sense its surroundings with the very signal it is already broadcasting. The two obstacles are the null subcarriers in the LTE/NR resource grid and the self-interference that leaks from the transmitter into the shared receiver. The paper supplies an interpolation step for the empty subcarriers and a cascade of RF and digital cancellation for the self-interference, then validates the combination with RF measurements. If the claim holds, ordinary base stations gain a radar function without a separate radar waveform, enabling environmental sensing from the existing mobile network.","feed_headline":"Base stations can sense targets with their own LTE and 5G signals","feed_subtitle":"Filling in empty subcarriers and canceling self-interference turns standard downlink waveforms into working radar.","key_machinery":"The engine of the argument is subcarrier-domain radar processing with the channel-estimation-style ratio $\\bar{G}^{\\mathrm{CH}}_{p,q}=Y_{p,q}/X_{p,q}$ on the LTE/NR transmit-receive grid, followed by a two-stage IFFT/FFT periodogram that maps delay to range and Doppler to velocity. Because LTE and NR leave some passband subcarriers unused, the paper inserts linear interpolation along the OFDM-symbol axis at those frequency positions before forming the periodogram. The companion hardware machinery is a three-tap RF canceller, whose maximum 10 ns delay intentionally keeps true target echoes beyond about 1.5 m intact, and a digital canceller using memory-polynomial basis functions $|x(n)|^{p-1}x(n)$ with a self-orthogonalizing update rule that avoids explicit basis orthogonalization. Together these carry the argument by converting an unreadable self-interference-dominated image into one in which real targets appear.","core_discovery":"On the paper's own terms, the discovery is that the normal downlink transmission of an LTE or NR base station, processed in the frequency domain by element-wise division of the received grid by the known transmitted grid, yields a clean range-Doppler image once the gaps left by unused subcarriers are filled by linear interpolation across OFDM symbols. The transmitter's own leakage appears in that image as an extremely strong static target at zero range, and its sidelobes mask real static and slow-moving targets. The paper shows that a three-tap RF canceller with delays up to 10 ns, followed by a nonlinear memory-polynomial digital canceller with five precursor and five postcursor taps at 240 MHz, suppresses the leakage by about 100 dB in measurements while preserving echoes from targets beyond roughly 3 m. With this cancellation, a static drone at 40 m and three moving cars at 51--102 m are detected and their ranges and velocities estimated from a 40 MHz NR waveform at 2.44 GHz.","pith_inferences":["If the same cancellation chain scales to the 24--40 GHz NR bands, where 400 MHz contiguous carriers are already standardized, the distance resolution would drop below 0.5 m, making micro-Doppler signatures such as the drone-propeller effect seen in the measurements a practical sensing feature rather than a curiosity.","A natural testable extension is to run the radar processing while an uplink signal is received inband; the cancellation chain would then need to separate the uplink user signal from the downlink echoes, which is a different interference geometry than the one measured.","The static-drone experiment already shows propeller micro-Doppler sidebands at the target's range, suggesting that the same waveform could support classification of targets by their micro-motion, though the paper does not pursue classification.","Deploying this in a live network would require the base station to know the transmitted resource grid exactly and to keep the radar receiver coherent with the transmitter; the measurements validate this on a controlled setup, so the main open engineering question is integration with real schedulers and MIMO antenna arrays."],"forward_implications":["At 100 MHz NR bandwidth with 30 kHz subcarrier spacing, the paper's simulations give distance resolution of about 1.5 m and reliable detection at input SNRs below $-30$ dB, implying that sub-6 GHz 5G base stations can support useful short-to-medium-range sensing without new spectrum.","Because the transmitter self-interference behaves as a zero-range static target, moving targets are detectable even with weaker isolation, whereas static targets require the full RF-plus-digital cancellation chain; this orders the practical deployment difficulty.","With the cancellers active, measured range profiles reveal building reflections at 120 m and 180 m, showing the dynamic range needed for urban sensing is attainable in a real front end.","The 10 ns RF canceller delay and 5+5 digital taps set a minimum detectable target distance around 3 m, an explicit design trade-off between cancelling close coupling and preserving close echoes."],"supporting_citations":[{"why":"Establishes the OFDM radar processing principle and the channel-estimation-like division approach the paper adopts.","marker":"[3]"},{"why":"Supplies the frequency-domain radar processing, resolution formulas, and estimation framework that the paper adapts to LTE/NR grids.","marker":"[4]"},{"why":"Provides the OFDM joint radar-communications concept and element-wise division processing basis.","marker":"[6]"},{"why":"The authors' own earlier LTE-waveform radar work, which this paper extends to 5G NR and to self-interference cancellation.","marker":"[10]"},{"why":"Justifies the periodogram peak as the maximum-likelihood range and velocity estimator used in the paper.","marker":"[36]"},{"why":"Frames the in-band full-duplex transmitter-receiver isolation challenge and the state of the art in self-interference suppression.","marker":"[19]"},{"why":"Provides the nonlinear digital cancellation architecture and the beyond-100 dB suppression benchmark the paper builds on.","marker":"[41]"},{"why":"Supplies the multi-tap RF canceller topology and gradient-based weight control reused in the radar cancellation chain.","marker":"[47]"},{"why":"Defines the LTE base-station transmission grid and its unused subcarriers, which motivate the interpolation step.","marker":"[17]"},{"why":"Defines the 5G NR base-station transmission grid, numerologies, and bandwidths used in the simulations and measurements.","marker":"[18]"}],"fun_headline_variants":["LTE and 5G signals turn base stations into radar","Base stations use their own LTE/5G signals for radar","Filling empty subcarriers and canceling self-interference enables OFDM radar","5G NR waveforms make base stations into high-precision radar"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The claim rests on the assumption that the direct transmitter-receiver coupling can be represented by an RF canceller with only three taps and delays up to 10 ns plus a 240 MHz digital canceller with five precursor and five postcursor taps, while echoes from targets closer than about 3 m are sacrificed; if real base-station coupling contains longer-delay multipath, or if the cancellers suppress close-range target returns, the demonstrated static-target detection will not transfer to practical deployments.","fun_headline_variants_meta":{"raw":{"variants":["LTE and 5G signals turn base stations into radar","Base stations use their own LTE/5G signals for radar","Filling empty subcarriers and canceling self-interference enables OFDM radar","5G NR waveforms make base stations into high-precision radar"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000625,"raw_usage":{"total_tokens":2945,"prompt_tokens":1052,"completion_tokens":1893,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":668,"completion_tokens_details":{"reasoning_tokens":1818}},"tokens_in":668,"tokens_out":1893,"duration_ms":13260,"temperature":1.0,"reasoning_tokens":1818,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:13:26.616683+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a shared-antenna base-station front end in an environment with a strong reflector 5--10 m from the antenna, such as a mast, building edge, or vehicle, and measure the residual self-interference after the 10 ns three-tap RF canceller and 5+5-tap digital canceller; if a static target at that range is still masked by leakage sidelobes, or if the measured residual coupling exceeds the roughly 100 dB isolation shown here, the central claim fails for realistic multipath. A complementary check is to place a target at 2 m: if the canceller suppresses it along with the leakage, the minimum-detectable-range design trade-off is confirmed.","supporting_citations":[{"cited_title":"An OFDM system concept for joint radar and communications operations,","cited_arxiv_id":null,"evidence_quote":"Provides the OFDM joint radar-communications concept and element-wise division processing basis."},{"cited_title":"OFDM radar with LTE waveform: Processing and performance,","cited_arxiv_id":null,"evidence_quote":"The authors' own earlier LTE-waveform radar work, which this paper extends to 5G NR and to self-interference cancellation."},{"cited_title":"Maximum likelihood speed and distance estimation for OFDM radar,","cited_arxiv_id":null,"evidence_quote":"Justifies the periodogram peak as the maximum-likelihood range and velocity estimator used in the paper."},{"cited_title":"In-band full-duplex technology: Techniques and systems survey,","cited_arxiv_id":null,"evidence_quote":"Frames the in-band full-duplex transmitter-receiver isolation challenge and the state of the art in self-interference suppression."},{"cited_title":"Com- pact inband full-duplex relays with beyond 100 dB self-interference suppression: Enabling techniques and field measurements,","cited_arxiv_id":null,"evidence_quote":"Provides the nonlinear digital cancellation architecture and the beyond-100 dB suppression benchmark the paper builds on."},{"cited_title":"Wideband self-adaptive RF cancellation circuit for full-duplex radio: Operating principle and measurements,","cited_arxiv_id":null,"evidence_quote":"Supplies the multi-tap RF canceller topology and gradient-based weight control reused in the radar cancellation chain."},{"cited_title":"3GPP TS 36.104 v15.3.0,","cited_arxiv_id":null,"evidence_quote":"Defines the LTE base-station transmission grid and its unused subcarriers, which motivate the interpolation step."},{"cited_title":"3GPP TS 38.104 v15.4.0,","cited_arxiv_id":null,"evidence_quote":"Defines the 5G NR base-station transmission grid, numerologies, and bandwidths used in the simulations and measurements."}],"review_version":1}