{"id":"4fc1264f-e41d-4213-8ae2-8badfdef0896","arxiv_id":"2411.16496","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A USRP-based single-anchor testbed combining 5G uplink angle-of-arrival estimation with UWB ranging reports sub-10 meter positioning errors in outdoor pedestrian trials.","lead":"This paper describes a working radio testbed that locates a walking person from a single base station using 5G-style uplink signals and ultra-wideband distance measurements. It matters because real-world positioning experiments with 5G signals are rare, and this setup gives researchers a reusable way to measure how well angle-of-arrival and ranging methods work outdoors.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Described calibration scheme appears incompatible with the claimed 3-element ULA: injecting a tone into one channel of each N310 daughterboard pair leaves at most two clean SRS channels, or contaminates an array element if one such channel is reused.","rationale":"The reader's weakest assumption about calibration stability is plausible and relevant, but the more fundamental issue is whether the calibration architecture as written can support the claimed three-element ULA at all. If the splitter occupies one channel per daughterboard pair, then only two channels remain for the SRS; if a calibration channel is also an array element, the received SRS is contaminated. Both are internal inconsistencies in the system description rather than external assumptions. This directly undermines the central claim because the AoA estimates—and therefore the position errors—depend on the actual array configuration. I still view the paper as a plausible engineering contribution: the inconsistency may be a documentation error, and a simple clarification of the RF routing or a fallback analysis with the correct array size would resolve it. Therefore the appropriate verdict remains CONDITIONAL, and I mark agreement as partial because the reader's concern is related but not identical. The paper's positive aspects—a genuinely deployed dual-band USRP testbed, explicit acknowledgment of UWB dropout at long ranges, and the use of standard high-resolution AoA estimators—deserve credit; the concern is about reproducibility and internal consistency, not about the authors' competence or honesty.","tokens_in":7075,"tokens_out":11545,"duration_ms":118370,"concrete_test":"Request or inspect the RF cabling/channel map of the N310 to count how many Rx channels simultaneously receive ULA antenna signals while the calibration tone is active. If only two SRS channels exist, re-run a static known-angle measurement with a two-element array and compare AoA RMSE and the resulting position-error CDF. If one array channel also carries the calibration tone, repeat the same measurement with that channel excluded from the array, and check whether the reported sub-10 m percentages change by more than a few points.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section II.C.5 states that three of the N310's four Rx channels feed a three-element ULA for AoA. Section III.A then describes a real-time calibration in which a common signal is injected into one channel of each daughterboard pair via a 1-2 splitter. That injection occupies one channel per pair, leaving only two channels total for SRS reception. The text claims this only 'limit[s] the ULA utilization to three channels,' but the described routing leaves two, not three, usable channels. If instead the calibration tone is injected into a channel that also carries an array element, then the SRS on that channel is overlaid with an unmodulated calibration tone; the paper does not describe filtering, frequency offsetting, or time-duplexing that would separate the two. In that case the phase and amplitude of that array element are corrupted, so the three-element array is not cleanly realized. The reported CDFs and the headline 'sub-10 meter accuracy in 88% and 90% of cases' depend on the actual array size and per-channel SNR. A two-element array would yield worse angular resolution than a three-element array, and a contaminated channel would bias the AoA estimate; either could materially change the position-error statistics. No block diagram with RF routing or a per-channel mapping is provided to resolve this ambiguity.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes a single-anchor positioning testbed that combines an Ettus E312 transmitter, an Ettus N310 receiver with a three-element uniform linear array for angle-of-arrival estimation from 5G NR uplink sounding reference signals, and ultra-wideband ranging. Field trials with a pedestrian user at 2.4 GHz and 5.8 GHz produce position-error cumulative distribution functions; the authors report sub-10 m accuracy in 88% and 90% of cases, respectively, and sub-4.6 m / sub-3.4 m errors at the 90th percentile for distances below 40 m. The manuscript also describes a real-time phase-calibration procedure for the N310's two daughterboard channel pairs, snapshot-based IQ capture, and an MQTT-based control architecture.","tokens_in":7330,"tokens_out":7866,"duration_ms":66336,"significance":"If the reported measurements are reliable, the paper contributes a useful and relatively rare experimental testbed for 5G single-anchor positioning, combining uplink SRS-based AoA with UWB ranging in outdoor pedestrian conditions. Strengths include the use of real hardware, independent laser-based ground-truth landmarks, and the absence of any fitting loop between the measured angles/ranges and the final position errors. However, the central quantitative claims are currently undermined by an ambiguity in the RF calibration routing and by missing-data issues in the CDF computation; these need to be resolved before the accuracy percentages can be taken at face value.","major_comments":[{"comment":"The real-time calibration procedure as written is incompatible with the stated three-element ULA. Section II.C.5 says that three of the four Rx channels feed a three-element ULA for AoA estimation. Section III.A then states that a common signal is injected into one channel of each daughterboard pair via a 1-2 splitter. That injection occupies two channels, leaving only two channels available for SRS reception, not three. If instead one of the calibration-injected channels is also used as an array element, then the calibration tone is superimposed on the SRS, and the paper does not describe any frequency offset, filtering, or time-duplexing that would separate the two. Either reading changes the array size or element quality used for AoA, and both could materially affect the position-error statistics in Fig. 6 and the headline 88%/90% accuracy claims. Please provide a per-channel RF routing diagram and state explicitly which physical Rx channels carry the ULA elements, which carry the calibration tone, and how the calibration signal and SRS are separated in frequency or time.","section":"Section III.A (with Section II.C.5)"},{"comment":"The reported CDFs are computed on a subset of the trajectory. The text states that beyond 40 m the UWB ranging becomes less reliable and 'occasionally fail[s] to provide a position estimate every second', with the failure rate increasing with distance. The percentages 'sub-10 meter accuracy in 88% and 90% of cases' are therefore conditional on successful UWB ranging, not representative of the full trajectory. The paper should report the total number of position estimates, the number and spatio-temporal distribution of dropped UWB measurements, and provide either complete-case and all-track CDFs or distance-stratified results. In addition, ground truth is established by interpolation between stationary landmarks (Section IV-B), which can smooth out genuine errors between landmarks; an estimate of the interpolation error or a continuously surveyed reference would strengthen the validation. No trial counts or error bars are provided for the reported percentages, so the uncertainty of the 88%/90% figures is unknown.","section":"Section IV-C and Section IV-B"}],"minor_comments":[{"comment":"The y-axis label 'Cumulative Density Function' should be 'Cumulative Distribution Function'.","section":"Fig. 6"},{"comment":"The use of the LCMV beamformer to select among multiple AoA candidates when the estimated channel order is greater than one is not explained; the selection rule should be stated.","section":"Section III.D, step 7"},{"comment":"The trajectory includes points up to 90 m, but all quantitative accuracy claims are separated only by the 40 m threshold; a distance-binned error table would help the reader understand where the errors occur.","section":"Section IV.A"},{"comment":"The paper does not state how many independent pedestrian runs were performed; if the CDFs are from a single run, this should be stated explicitly.","section":"Section IV.C"},{"comment":"The abstract emphasizes 'real-time calibration', but it is not clear whether the calibration is repeated during the pedestrian motion or only once at startup; please clarify the calibration schedule.","section":"Section III.A"}],"recommendation":"major_revision","confidential_remarks":"To the editor: The calibration routing issue in Section III.A is the main technical concern. If the authors clarify that only two channels were used for SRS, the reported results would need to be re-interpreted with a two-element array; if a contaminated channel was used, the AoA estimates could be biased. This needs to be resolved before publication. The manuscript would also benefit from reporting sample sizes and missing-data handling, but those are standard revision requests. The paper fits the testbed/experimental scope of the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real field testbed paper, not a simulation, and that alone is useful. But the calibration write-up in Section III-A does not add up as printed, and the headline accuracy numbers are computed on a subset of the trajectory. Both are fixable, and the paper deserves a careful review, but not a free pass.\n\nWhat is genuinely new: the combination of 5G NR uplink SRS for AoA on an Ettus N310 with a three-element ULA, plus UWB ranging, field-tested with a pedestrian. Prior testbeds used LTE, downlink mmWave, or different SDRs. The snapshot capture to handle 61.44 MSps is sensible, and the authors are honest that the barometer did not work and drop it from the analysis. Ground truth from laser-measured landmarks with UWB timestamps is a reasonable validation method.\n\nSoft spots, in order:\n- The calibration description is ambiguous. Injecting a common signal into one channel of each daughterboard pair via a 1-2 splitter should occupy two of the four Rx channels, leaving two for SRS, not three. The paper says this \"limits the ULA utilization to three channels.\" For that to be right, the calibration tone must share a ULA channel and be separated from SRS (e.g., placed on an empty subcarrier, since SRS uses a comb). The paper never says that. If the tone is not separated, the array is either two-element or has a contaminated element, and the AoA results change. A block diagram of the RF routing and the tone's frequency plan is essential.\n- The CDFs are computed only on positions where UWB returned a range. The paper admits UWB becomes unreliable beyond 40 m, so the longer-range, likely worse, points drop out. That biases the \"88%/90% sub-10m\" headline upward. The authors should report the dropout rate and show results including a missing-data handling.\n- No error bars, no trial counts. For a testbed paper that is common, but with only one run it is hard to separate systematic bias from noise.\nMinor: ground truth interpolation between 30-s stationary landmarks is fine but should be stated as an assumption.\n\nBottom line: the paper is for people building single-anchor or SDR-based positioning testbeds. It is a useful data point, not a breakthrough. I would send it to peer review with a request for major revision: clarify calibration, add statistics, and address the selection effect. The calibration issue is the one that could sink the results if it turns out the array was not clean.","headline":"A useful field-testbed paper with a plausible integration of 5G uplink AoA and UWB ranging, but the calibration description is ambiguous enough that the headline accuracy needs a clarifying revision before the numbers can be trusted.","tokens_in":7918,"tokens_out":6415,"would_cite":false,"duration_ms":59135,"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 testbed that combines 5G uplink angle-of-arrival from a single USRP base station with UWB ranging places a walking pedestrian within 10 meters in 88-90% of measurements, and within 4.6 meters at short range.","keywords":["5G positioning","angle of arrival","USRP testbed","sounding reference signal","UWB ranging","single-anchor localization","pedestrian localization"],"falsifier":"Measure the calibration phase offset immediately before and after a pedestrian run; if the offset shifts by more than a small fraction of a wavelength, or if rerunning the same trajectory with the test tone switched off changes the reported angles, the calibration assumption fails.","tokens_in":6907,"feed_emoji":"📡","tokens_out":5519,"duration_ms":46353,"temperature":0.7,"pith_summary":"This paper argues that a single-anchor positioning system built from commercial off-the-shelf software-defined radios can localize a moving pedestrian outdoors without GNSS. The testbed measures the angle of arrival of 5G New Radio uplink sounding reference signals with a three-element antenna array on an Ettus N310, and fuses that angle with ultra-wideband ranging to the base station. The authors report that 88% of position estimates at 2.4 GHz and 90% at 5.8 GHz are within 10 meters of ground truth over a trajectory reaching 90 meters, and that for distances under 40 meters the 90th-percentile error drops to 4.6 meters and 3.4 meters respectively. If these numbers hold, single-node positioning becomes a practical complement to GNSS in indoor and urban canyon settings.","feed_headline":"Single 5G base station locates walkers within 10 m 90% of time","feed_subtitle":"AoA from 5G uplink plus UWB ranging hits 90% errors under 4.6 m inside 40 m range","key_machinery":"The load-bearing mechanism is the phase-alignment calibration for the Ettus N310's two independent local-oscillator domains. Because the two daughterboards each serve a pair of receive channels with a random phase offset between runs, the testbed injects a common continuous tone through a 1-2 splitter into one channel of each pair, estimates the differential phase by cross-correlation, and compensates it in real time. This makes three of the four receive channels coherent so that a three-element uniform linear array can feed ESPRIT for angle-of-arrival estimation; a snapshot capture scheme keeps the large IQ files manageable, and UWB ranging provides the distance that turns the angle into a position.","core_discovery":"The central discovery is that a single base station can track a pedestrian by combining two complementary measurements: an angular estimate from 5G uplink SRS, obtained with a three-element uniform linear array and the ESPRIT algorithm, and a distance estimate from UWB two-way ranging. The enabling step is a real-time calibration of the Ettus N310 receiver, which injects a common tone into one channel of each daughterboard pair to measure and remove the phase offset caused by the two separate local oscillators. With this alignment, the testbed achieves sub-10-meter errors in 88% of cases at 2.4 GHz and 90% at 5.8 GHz, and 90th-percentile errors below 4.6 meters and 3.4 meters respectively for ranges under 40 meters.","pith_inferences":["Beyond the paper: the same calibration and ESPRIT pipeline could be applied to the vehicular configurations in Table I; the open question is whether the phase alignment remains stable under Doppler and vibration.","Beyond the paper: a two-dimensional array or adding elevation estimation could lift the system to 3D positioning, especially because the barometer readings were too unstable to use.","Beyond the paper: reporting the calibration phase offset measured before and after each run would provide a direct check on the stability assumption, and would make the accuracy numbers reproducible.","Beyond the paper: the decreasing UWB reliability beyond 40 meters suggests that a fusion with 5G timing-based ranging could extend the reliable range without new hardware."],"forward_implications":["A single base station, using only uplink 5G signals and UWB ranging, can provide outdoor pedestrian localization where GNSS is unavailable or degraded.","The sub-4.6-meter 90th-percentile error below 40 meters is accurate enough for pedestrian navigation and asset tracking applications in that range.","The phase-calibration trick removes the need for an external clock or a four-channel fully synchronized array, lowering the hardware cost of coherent angle-of-arrival estimation.","The 5.8 GHz band gives slightly better accuracy than 2.4 GHz, suggesting that higher-frequency 5G bands are favorable for single-anchor positioning.","The snapshot capture method keeps storage and processing feasible at up to 61.44 MSps without sacrificing positioning accuracy."],"supporting_citations":[{"why":"Defines the 5G NR uplink SRS structure and its positioning-oriented configuration that the testbed transmits and processes.","marker":"[12]"},{"why":"Documents the pathway to coherent phase acquisition in multi-channel USRP SDRs, motivating the need for the calibration procedure.","marker":"[14]"},{"why":"Prior work by the same group on static AoA estimation with SRS in 5G NR uplink, which the pedestrian trial extends.","marker":"[15]"},{"why":"Introduces the MUSIC algorithm, one of the candidate AoA estimators evaluated before choosing ESPRIT for the field trials.","marker":"[19]"},{"why":"Introduces ESPRIT, the shift-invariance algorithm selected as the preferred AoA estimator in the field trials.","marker":"[20]"},{"why":"Describes an earlier single-base-station ToA-AoA localization testbed using LTE signals, providing a baseline for SDR-based positioning experiments.","marker":"[9]"},{"why":"Reports initial findings from single-base-station 5G mmWave positioning, highlighting the gap between theoretical assumptions and real-world behavior.","marker":"[11]"}],"fun_headline_variants":["Single base station uses 5G AoA and UWB to locate walkers under 5 m","5G uplink AoA plus UWB ranging: single-station pedestrian fix under 5 m","USRP testbed: single 5G base station locates walkers with AoA and UWB","One base station, two measurements: 5G AoA and UWB give sub-5m fixes","Pedestrian tracking: single-station 5G AoA + UWB achieves sub-5 m errors"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The positioning accuracy depends on the assumption that the phase difference between the two receiver channel pairs, measured at setup by injecting a common test tone, stays unchanged throughout the walk and does not contaminate the 5G signal measurement.","fun_headline_variants_meta":{"raw":{"variants":["Single base station uses 5G AoA and UWB to locate walkers under 5 m","5G uplink AoA plus UWB ranging: single-station pedestrian fix under 5 m","USRP testbed: single 5G base station locates walkers with AoA and UWB","One base station, two measurements: 5G AoA and UWB give sub-5m fixes","Pedestrian tracking: single-station 5G AoA + UWB achieves sub-5 m errors"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000849,"raw_usage":{"total_tokens":3687,"prompt_tokens":936,"completion_tokens":2751,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":552,"completion_tokens_details":{"reasoning_tokens":2622}},"tokens_in":552,"tokens_out":2751,"duration_ms":16171,"temperature":1.0,"reasoning_tokens":2622,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:02:21.948993+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the calibration phase offset immediately before and after a pedestrian run; if the offset shifts by more than a small fraction of a wavelength, or if rerunning the same trajectory with the test tone switched off changes the reported angles, the calibration assumption fails.","supporting_citations":[{"cited_title":"3GPP TS 38.211: NR Physical channels and modulation (Re- lease 18),","cited_arxiv_id":null,"evidence_quote":"Defines the 5G NR uplink SRS structure and its positioning-oriented configuration that the testbed transmits and processes."},{"cited_title":"Pathway to Coherent Phase Acquisition in Multi-Channel USRP SDRs for Direction of Arrival Estimation,","cited_arxiv_id":null,"evidence_quote":"Documents the pathway to coherent phase acquisition in multi-channel USRP SDRs, motivating the need for the calibration procedure."},{"cited_title":"Angle of Arrival Estimation Using SRS in 5G NR Uplink Scenarios,","cited_arxiv_id":null,"evidence_quote":"Prior work by the same group on static AoA estimation with SRS in 5G NR uplink, which the pedestrian trial extends."},{"cited_title":"Performance Evaluation of Single Base Station ToA-AoA Localization in an LTE Testbed,","cited_arxiv_id":null,"evidence_quote":"Describes an earlier single-base-station ToA-AoA localization testbed using LTE signals, providing a baseline for SDR-based positioning experiments."},{"cited_title":"Experimental Validation of Single Base Station 5G mm Wave Positioning: Initial Findings,","cited_arxiv_id":null,"evidence_quote":"Reports initial findings from single-base-station 5G mmWave positioning, highlighting the gap between theoretical assumptions and real-world behavior."}],"review_version":1}