PaperlayerView exact revision on arXiv

Source-linked AI summary

System-Level Analysis of Joint Sensing and Communication based on 5G New Radio

Lorenzo Pucci, Enrico Paolini, Andrea Giorgetti

arXiv:2201.12017v1eess.SP

TL;DR

The paper addresses limited system-level evidence on the sensing accuracy of JSC systems using current technology. It analyzes a multibeam 5G NR system that scans for targets while sharing transmitted power between sensing and communication. Its reported findings cover DoA-related position error, frequency-band trade-offs, and multi-target localization and detection under suitable sensing power.

  • Problem

    Few works, to the authors’ knowledge, have evaluated JSC target-parameter estimation accuracy with current technology, motivating a sensing-focused system-performance analysis.

    Source: §I · p. 2

  • Method

    The paper analyzes a monostatic 5G NR multibeam system that estimates target range, speed, and DoA, prunes phantom targets, and evaluates multi-target performance using OSPA.

    Source: §I · p. 2

  • Results

    DoA estimation is the primary source of target-position error; sub-6 GHz detects targets at higher distances but with lower accuracy, while tens of targets can be localized with sub-meter accuracy when sensing power ensures reliable DoA estimation.

    Source: §VI · p. 11

  • Takeaways & Limitations

    The analysis identifies sensing power and DoA estimation reliability as important to the reported target-localization performance.

    Source: §VI · p. 11

Abstract

from arXiv · show

This work investigates a multibeam system for joint sensing and communication (JSC) based on multiple-input multiple-output (MIMO) 5G new radio (NR) waveforms. In particular, we consider a base station (BS) acting as a monostatic sensor that estimates the range, speed, and direction of arrival (DoA) of multiple targets via beam scanning using a fraction of the transmitted power. The target position is then obtained via range and DoA estimation. We derive the sensing performance in terms of probability of detection and root mean squared error (RMSE) of position and velocity estimation of a target under line-of-sight (LOS) conditions. Furthermore, we evaluate the system performance when multiple targets are present, using the optimal sub-pattern assignment (OSPA) metric. Finally, we provide an in-depth investigation of the dominant factors that affect performance, including the fraction of power reserved for sensing.

I. INTRODUCTION

The introduction motivates JSC as a way to share radio hardware and spectrum for communication and sensing, and identifies limited system-level evidence on sensing accuracy with current technology. The paper analyzes a 5G NR multibeam system and evaluates target estimation, power allocation, and multi-target performance.

I. INTRODUCTION

  • JSC shares a hardware platform and frequency band between radar and communication, using communication-network waveforms for sensing.The introduction situates this approach amid spectrum congestion and growing interest in RF sensing applications.Source: §I · p. 1
  • Prior multibeam JSC work separates communication and sensing beams, with sensing scanning the environment while communication serves the user equipment.This contrasts with single-beam approaches that point sensing and communication in the same direction.Source: §I · p. 1
  • Few prior works, to the authors’ knowledge, evaluate JSC target-parameter estimation accuracy with current technology; this work addresses that gap for a 5G NR multibeam system.The stated analysis examines key aspects governing system performance from the sensing perspective.Source: §I · p. 2
  • The paper analyzes monostatic base-station estimation of multiple targets’ range, speed, and DoA, proposes pruning phantom targets, and studies single- and multi-target performance.Single-target analyses include position RMSE and radial-speed accuracy; multi-target evaluation uses OSPA for localization and detection.Source: §I · p. 2

II. SYSTEM MODEL

  • The system model considers a MIMO 5G NR OFDM waveform transmitted to a user equipment through transmitter and receiver antenna arrays.The transmitter and receiver arrays are modeled as uniform linear arrays with half-wavelength element spacing.Source: §II · p. 2

A. Joint waveform

The waveform model splits available transmit power between communication and sensing beams. The two beams use array steering vectors, while the sensing-power share trades sensing capability against communication coverage.

A. Joint waveform

  • The transmit beamformer splits total OFDM signal power between communication and sensing directions, controlled by the fraction parameter ρ.The sensing and communication beamformer vectors are denoted wT,s and wT,c, respectively.Source: A. Joint waveform · p. 2
  • The communication and sensing beamformers use steering vectors for their respective directions, based on the uniform linear array response.The model assumes half-wavelength antenna spacing; the steering vector is specified for the array elements.Source: A. Joint waveform · p. 2
  • Reserving power for sensing reduces communication coverage, with MPL and MIL reduced by 10log10(ρ) dB.The paper identifies this as a trade-off between communication and sensing performance.Source: A. Joint waveform · p. 3

B. Received signal

The received-signal model represents echoes from multiple point targets in the MIMO channel. Each target contributes delay, Doppler, direction, and complex path amplitude, alongside self-interference and additive noise.

B. Received signal

  • The received signal model includes self-interference from imperfect transmitter–receiver isolation and additive white Gaussian noise.The noise entries are modeled as independent, identically distributed circularly symmetric zero-mean Gaussian variables.Source: B. Received signal · p. 3
  • For L point-target reflections, the received channel is represented by a channel matrix over OFDM symbols and subcarriers, with target-specific delay, Doppler, DoA, and complex amplitude.The target coefficients can also be organized in a diagonal matrix, with transmit and receive steering matrices describing target directions.Source: B. Received signal · p. 3
  • The signal-to-self-interference ratio is defined for the echo generated by each target at each receiving antenna.The self-interference model treats its contribution as a static target with approximately zero delay and Doppler.Source: B. Received signal · p. 3

C. Beam-scanning

The sensing beam sequentially scans different directions while the communication beam points to the user equipment. Collected OFDM symbols at each scan direction support target range, Doppler, and DoA estimation.

C. Beam-scanning

  • The multibeam scheme keeps a beam pointed at the user equipment while directing a sensing beam sequentially across scan directions.During the scan, the sensing direction of departure and direction of arrival are equal.Source: C. Beam-scanning · p. 4
  • For each sensing direction, the receiver collects OFDM symbols, which are used to estimate target range, Doppler, and DoA.The scan spans from −θ0 to θ0 in steps of ∆θs across Ndir directions.Source: C. Beam-scanning · p. 4

D. Sensor-target-sensor path

Under LOS, the sensor-target-sensor path depends on target and array-response factors. The system estimates target count and DoA with covariance-based model selection and MUSIC for each sensing direction.

D. Sensor-target-sensor path

  • Under LOS, received power from a target is proportional to |α_l|^2 and depends on its RCS, distance, receiver gain, and transmit-array alignment.Alignment is represented by γ_l, which equals 1 when the target DoA matches the sensing direction.Source: D. Sensor-target-sensor path · p. 4

III. ESTIMATION OF TARGET PARAMETERS AND DETECTION

  • For each sensing-beam step, the system collects M_s OFDM symbols and uses MUSIC for DoA estimation and a periodogram for range and velocity.Source: §III · p. 4

A. Estimation of the number of targets and DoAs

  • The target count is estimated using MDL applied to the sample covariance matrix, considering targets illuminated in the current sensing direction.The sample covariance matrix substitutes for the unknown covariance matrix.Source: A. Estimation of the number of targets… · p. 4
  • MUSIC identifies DoA estimates from pseudo-spectrum peaks in a local angle range, and assumes fewer targets than receiving-array elements.The local search range depends on array beamwidth.Source: A. Estimation of the number of targets… · p. 4

B. Detection and range-Doppler estimation

The method removes unwanted data symbols from received OFDM signals and forms a range-Doppler periodogram. A threshold test detects targets, whose range and speed are estimated from periodogram peaks.

B. Detection and range-Doppler estimation

  • Removing unwanted data symbols leaves two complex sinusoids per target, with frequencies related to Doppler frequency and delay.Source: B. Detection and range-Doppler… · p. 5
  • The periodogram is used to estimate target range and speed from the received-signal frequencies.The estimates are given by the peak indices scaled by the corresponding 5G NR parameters.Source: B. Detection and range-Doppler… · p. 5
  • A hypothesis test detects a target when the range-Doppler periodogram exceeds a threshold chosen for a predefined false-alarm probability.Source: B. Detection and range-Doppler… · p. 5
  • Range and velocity resolutions depend on 5G NR parameters, while zero-padding factor F_p is used to improve speed-estimation resolution.The periodogram computation uses FFTs and IFFTs with zero-padded lengths.Source: B. Detection and range-Doppler… · p. 5
  • Algorithm 1 sorts the target-point matrix by its third-column values and outputs a pruned matrix with its estimated target count.Source: B. Detection and range-Doppler… · p. 5

C. Pruning redundant target points

The system prunes repeated detections that can arise when beam sidelobes trigger detections in adjacent directions. It retains the strongest range-speed pair among estimates deemed redundant.

C. Pruning redundant target points

  • Local MUSIC searches reduce computational cost but can produce multiple detections for one target because of sidelobes and inaccurate DoA estimates.Source: C. Pruning redundant target points · p. 5
  • Detected peaks and estimates are collected into a matrix and sorted by descending periodogram value so stronger range-speed pairs are retained.Source: C. Pruning redundant target points · p. 5
  • Two points are treated as repetitions when both their range and radial-velocity estimates fall within uncertainty bounds ε_r and ε_v of an already retained point.The retained matrix has at most as many rows as the number of detections.Source: C. Pruning redundant target points · p. 6

IV. PERFORMANCE EVALUATION IN THE PRESENCE OF MULTIPLE TARGETS

The OSPA metric evaluates multi-target estimation by comparing true and estimated target sets. It combines localization and cardinality errors, with its parameters controlling their relative contributions.

IV. PERFORMANCE EVALUATION IN THE PRESENCE OF MULTIPLE TARGETS

  • OSPA measures multi-target miss-distance by accounting for both localization error and differences in the number of true and estimated targets.Source: §IV · p. 6
  • The estimated target count can differ from the true count because detection is conditioned on sensing direction and transmit beamforming spatially filters targets.Source: §IV · p. 6
  • OSPA matches estimates to true positions using an optimal subset assignment and caps each paired distance at cutoff c̄; unmatched estimates incur the cutoff.The metric uses Euclidean distance between estimated and true positions.Source: §IV · p. 6
  • The order q controls sensitivity to outliers, while cutoff c̄ balances localization error against cardinality error.Smaller c̄ emphasizes localization error, whereas larger c̄ emphasizes cardinality error.Source: §IV · p. 6

V. SYSTEM-LEVEL ANALYSIS

The system-level analysis uses simulated 5G NR signals to assess sensing in single-target and multi-target scenarios. It examines how sensing power and scan settings relate to estimation performance.

V. SYSTEM-LEVEL ANALYSIS

  • The simulations use 5G NR signals compliant with the 3GPP specification, with the main signal-generation parameters summarized in Table I.The OFDM signals use QPSK; the scan covers [−θ0, θ0] with direction spacing chosen in relation to array beamwidth.Source: §V · pp. 6–7
  • For single-target RMSE analysis versus SNR, the target is aligned with the sensing beam; distance-varying tests use specified target and receiver parameters.The analysis also varies SSIR to study its impact on DoA, distance, and speed RMSE.Source: §V · p. 7
  • The multi-target evaluation uses 10 point targets, including the UE, and compares sensing-power fractions ρ = 0.1 and ρ = 0.3 at 28 GHz with 50 transmit and receive antennas.Performance is summarized with OSPA using q = 2 and ¯c = 10 m.Source: §V · p. 7

A. RMSE and detection probability vs SNR and SSIR

The analysis finds that sensing errors depend on both self-interference suppression and SNR. Range and speed estimation retain processing gains, while DoA estimation becomes the main limitation on radar performance.

A. RMSE and detection probability vs SNR and SSIR

  • At SSIR ≥10 dB, DoA, range, and radial-speed RMSE reach a floor where thermal noise is the only limiting factor.At SNR = −20 dB, performance degrades quickly at lower SSIR; the figure compares these estimates across SSIR values.Source: A. RMSE and detection probability vs… · p. 7
  • Range and speed can be estimated accurately at lower SNRs than DoA because their periodogram and beamforming processing gains do not apply to MUSIC.The SNR analysis spans −65 to −15 dB; the figure includes distance, DoA, speed, position, and detection probability results.Source: A. RMSE and detection probability vs… · p. 7, Figure 3 · p. 8
  • At increasingly negative SNR, angle RMSE converges approximately to ΔΘ/2.8 because MUSIC searches only within the limited interval ΔΘ.The bound depends on antenna count through the array beamwidth.Source: A. RMSE and detection probability vs… · p. 8, Figure 3 · p. 8
  • Normalizing position RMSE by true distance removes its distance dependence; as SNR falls, DoA error causes the first performance drop and range error a second.The position estimate is formed from estimated range and DoA.Source: A. RMSE and detection probability vs… · p. 8
  • As detection probability degrades, range and velocity estimation also begin to degrade, while DoA estimation remains the main factor limiting radar performance.The detection threshold is set for PFA = 1%.Source: A. RMSE and detection probability vs… · p. 8, Figure 3 · p. 8

B. RMSE vs distance

In LOS, localization remains accurate at tens of meters with a modest sensing-power fraction, while DoA error is the main position-estimation limitation. In the multi-target analysis, greater sensing power improves performance, and the selected pruning settings yield low OSPA error while removing redundant detections in the illustrated case.

B. RMSE vs distance

  • At ρ = 0.1, position RMSE stays below 0.33 m at 3.5 GHz and 0.1 m at 28 GHz even at tens of meters in LOS.RMSE is higher at 3.5 GHz, mainly because the 10-antenna setup has wider beams than the 50-antenna setup.Source: B. RMSE vs distance · p. 8, Figure 4 · p. 9
  • At the considered ranges, DoA estimation error mainly determines position RMSE, which becomes proportional to distance when that error reaches its upper bound.Source: B. RMSE vs distance · p. 9

C. Performance analysis of multi-target scenario

  • Pruning tolerances should account for noise-driven range and velocity estimates falling in adjacent periodogram bins; velocity estimates have greater low-SNR RMSE because zero padding trades noise sensitivity for velocity resolution.The velocity tolerance has the greatest effect on pruning detection performance.Source: C. Performance analysis of… · pp. 9–10
  • At fixed pruning tolerances, multi-target localization and cardinality performance improve as ρ increases; for Ndir ≥ 40, ρ = 0.3 and ϵv = 3∆v yield fewer than one missed target on average.An ϵr greater than ∆r does not appreciably change performance.Source: Figure 5 · p. 9, C. Performance analysis of… · p. 10
  • With ρ = 0.3, ϵr = ∆r, ϵv = 3∆v, and Ndir = 60, mean OSPA is below 3 m and its 20th percentile is approximately 1 m.The analysis includes regions where DoA estimation is severely degraded; appropriate sensing-cell sizing may avoid those regions and improve performance.Source: C. Performance analysis of… · p. 10, Figure 6 · p. 10
  • The pruning algorithm removes redundant detections while retaining useful points in the illustrated 9-target, 1-UE case; repeated points cluster around the UE’s distance because its communication beam creates a strong echo.Source: C. Performance analysis of… · p. 11, Figure 7 · p. 11

VI. CONCLUSION

The paper concludes that DoA estimation is the primary source of position error, yet good LOS localization at tens of meters is possible at mmWave with relatively little sensing power. It also reports sub-meter accuracy for multiple targets when sensing power supports reliable DoA estimation.

VI. CONCLUSION

  • DoA estimation is the primary source of error when evaluating target position.Source: §VI · p. 11
  • Good localization at tens of meters is achievable in LOS at mmWave even with a relatively small fraction of power devoted to sensing.Source: §VI · p. 11
  • Tens of targets can be detected and localized with sub-meter accuracy when sensing power ensures reliable DoA estimation.Source: §VI · p. 11
Loading 2201.12017v1…