PaperlayerView exact revision on arXiv

Source-linked AI summary

Multibeam for Joint Communication and Sensing Using Steerable Analog Antenna Arrays

J. Andrew Zhang, Xiaojing Huang, Y. Jay Guo, Jinhong Yuan, Robert W. Heath

arXiv:1810.04105v1eess.SPcs.NI

TL;DR

The paper addresses the mismatch between stable communication beams and scanning sensing beams in JCAS. It proposes a two-array steerable-analog multibeam framework with fixed communication and packet-varying sensing subbeams, together with beamforming and estimation methods. Simulations validate the framework, multibeam generation, and sensing algorithms, with compressive sensing improving distance-AoA resolution and accuracy over DFT.

  • Problem

    JCAS is difficult because communication requires stable, accurately pointed beams while sensing requires time-varying scanning beams; single-beam schemes restrict sensing to the communication direction.

  • Method

    The paper uses two steerable analog arrays and multibeams combining fixed communication subbeams with packet-varying sensing subbeams in TDD packet communication.

  • Results

    Simulations validate the proposed framework, multibeam generation, and sensing algorithms; the CS algorithm provides significantly improved distance-AoA resolution and accuracy compared with DFT.

  • Takeaways & Limitations

    The framework makes it feasible to integrate sensing into standard TDD packet communication systems with OFDM modulation.

Abstract

from arXiv · show

Beamforming has great potential for joint communication and sensing (JCAS), which is becoming a demanding feature on many emerging platforms such as unmanned aerial vehicles and smart cars. Although beamforming has been extensively studied for communication and radar sensing respectively, its application in the joint system is not straightforward due to different beamforming requirements by communication and sensing. In this paper, we propose a novel multibeam framework using steerable analog antenna arrays, which allows seamless integration of communication and sensing. Different to conventional JCAS schemes that support JCAS using a single beam, our framework is based on the key innovation of multibeam technology: providing fixed subbeam for communication and packet-varying scanning subbeam for sensing, simultaneously from a single transmitting array. We provide a system architecture and protocols for the proposed framework, complying well with modern packet communication systems with multicarrier modulation. We also propose low-complexity and effective multibeam design and generation methods, which offer great flexibility in meeting different communication and sensing requirements. We further develop sensing parameter estimation algorithms using conventional digital Fourier transform and 1D compressive sensing techniques, matching well with the multibeam framework. Simulation results are provided and validate the effectiveness of our proposed framework, beamforming design methods and the sensing algorithms.

I. INTRODUCTION

JCAS must reconcile communication’s need for stable, high-gain beams with sensing’s need for time-varying directional scanning. The paper proposes a two-array multibeam framework that integrates both functions into packet-based TDD communication.

  • Motivation: JCAS is motivated by platforms such as unmanned aerial vehicles and smart cars, where communication and radio sensing could share hardware and signal processing.The integration can reduce cost, size, and weight while supporting information sharing between functions.
  • Challenge: High-frequency systems require stable, accurately pointed communication beams but time-varying directional scanning beams for sensing.Existing single-beam JCAS schemes limit sensing to the communication-node direction.
  • Framework: The proposed multibeam uses fixed-direction subbeams for communication and packet-varying scanning subbeams for sensing from analog arrays.This enables sensing over different directions while maintaining communication beams toward the target node.
  • Architecture: The architecture uses two spatially separated analog arrays primarily to suppress transmitter leakage so the receiver can remain operational while switching between communication and sensing.The framework reuses conventional TDD timeslots and supports communication and sensing transitions.
  • Contributions: The paper develops beamforming design, multibeam generation and updating, and sensing-parameter estimation methods, including 1D compressive-sensing solutions.The methods target flexibility for time-varying requirements and compatibility with packet communication systems.
  • Protocol: In the proposed protocol, a transmitting array forms one subbeam toward the communication node and another adapted to sensing, while the second array senses the scanning direction.Each complete cycle contains Communication Transmission and Active Sensing and Communication Reception and Passive Sensing stages.

B. Formulation of Signal Model

The signal model represents communication and sensing over steerable uniform linear arrays using an OFDM waveform and a common multipath channel formulation. Sparse multipath parameters can be estimated directly for sensing.

  • Array Model: The paper assumes planar wavefronts and uniform linear arrays with elements spaced at half a wavelength.The array response is parameterized by angle of departure or arrival.
  • OFDM Signal: The OFDM model uses N subcarriers over bandwidth B, with subcarrier interval f0 = B/N and symbol period Ts = N/B + Tp.Tp denotes the cyclic-prefix period.
  • Beamforming: The transmitted signal is beamformed by a transmitter vector wt before radiation from the antenna array.The receiver applies a corresponding beamforming vector wr to the arriving signal.
  • Channel Model: A time-varying channel is modeled as L multipath components characterized by angle, complex amplitude, propagation delay, and Doppler frequency.For each path, Doppler frequency is given as fD,ℓ = 10^-8vsfc/3 in the supplied model.
  • Sensing Model: When multipath is sparse, estimating underlying path parameters directly can be simpler than estimating the channel-matrix elements.The communication and sensing channels may differ for active sensing.
  • Received Signal: The received signal includes additive white Gaussian noise, denoted z(t).

III. BEAMFORMING DESIGN

The beamforming design fixes communication beams over packets while varying sensing beams across packets and OFDM symbols. A generalized weighted least-squares formulation provides the basis for generating desired multibeam responses under power constraints.

  • Beamforming Requirements: JCAS requires stable, high-gain communication beams and direction-varying scanning beams for sensing over a large area.The communication beam must remain fixed during at least one packet to avoid complex within-packet channel tracking.
  • Transmitter Design: The transmitter beamforming vector is fixed for one packet and generates directional communication and sensing subbeams, with sensing directions changing over Nt packet periods.Subbeam width and gain can differ between communication and sensing.
  • Receiver Design: For sensing, Nr receiver beamforming vectors are interleaved across OFDM symbols to support Doppler estimation and wider angular scanning.The receiver vector remains fixed for at least one OFDM symbol so information symbols can be removed.
  • Array Combining: The communication receiver can independently design beams for the two arrays and apply a phase shift to constructively combine their signals.Jointly optimal design is challenging because the arrays differ in orientation and location.
  • Parameterization: A complete scanning sequence uses Nt packets and NtNrNd OFDM symbols, allowing these parameters and beam widths to be optimized for sensing while meeting communication requirements.
  • Generalized LS Design: The generalized weighted least-squares method designs a beamforming vector for a desired array response and can impose different accuracy requirements across waveform segments.The weighting matrix D controls these accuracy requirements.
  • Power Constraint: The unconstrained solution wLS = A†v does not establish optimality when transmitter and receiver beamforming require a power constraint.
  • Optimality: Theorem 1 states that the constrained optimum with w^Hw = 1 equals the normalized least-squares solution.The result identifies the conventional normalized LS solution as optimal for the stated constrained problem.

2) Known Magnitude Only and Iterative LS Algorithm:

When only desired beam magnitudes are specified, the unknown phases provide optimization degrees of freedom, but the resulting unit-circle problem is difficult and iterative LS methods are sub-optimal.

  • Known Magnitude Only: Only the magnitudes of the desired beamforming-vector elements may be specified, leaving their phases available for least-squares-error minimization.The phase vector provides K degrees of freedom in the optimization.
  • Optimization Challenge: The phase-constrained optimization is difficult because every phase-vector element must lie on the unit circle.
  • Iterative LS Algorithm: The two-step iterative least-squares method provides a sub-optimal solution by exploiting freedom in choosing the phase vector.Its convergence has not been proven, and it can often converge to the same beamforming vector for different magnitude specifications.

IV. GENERATION AND UPDATING OF MULTIBEAM BF VECTORS

The framework generates multibeam beamforming vectors through reference-waveform construction, directional displacement, and subbeam combination. These steps are designed to support changing communication and sensing requirements with rapid updates.

  • Design goals: The design targets simultaneous communication and sensing across different beamwidths, power levels, and packet parameters.Communication and sensing subbeams may be added constructively to improve communication-link SNR.
  • Design goals: Beamforming vectors can be updated simply and rapidly to adapt to changed communication and sensing requirements in real time.
  • Generation and updating: The multibeam-generation process first creates reference beamforming waveforms for communication and sensing, then shifts them to desired directions and combines the subbeams.The process is organized as S1 reference generation, S2 directional displacement, and S3 subbeam combination.

A. Generating Basic Reference BF Waveform

Reference beamforming waveforms are generated with selected beamwidths, scanning directions, shapes, and power allocations, then displaced efficiently using phase multiplication. The displacement preserves waveform shape only in equivalent directions, with acceptable distortions in many applications.

  • Reference waveform generation: Reference communication and sensing waveforms are generated separately with the ILS algorithm, potentially using different beamwidths and shapes.They are reused when only communication or scanning directions change.
  • Scanning-beam parameters: Wider scanning beams cover larger ranges or require fewer packets, but reduce sensing range and angle resolution through lower beamforming gain.
  • Scanning-beam parameters: Scanning directions are generally spaced at the 3dB beamwidth, approximately 2arcsin(1.2/M) for a half-wavelength-spaced ULA with M omnidirectional elements.
  • Waveform shape: Small-sidelobe waveforms can be encouraged by setting the desired magnitude response to the radiation-pattern mainlobe and zeros elsewhere.
  • Power allocation: Communication and sensing range requirements impose a power-allocation compromise, because increasing one range generally reduces the other.
  • Directional displacement: Equivalent directional displacement is implemented by multiplying a pre-generated beamforming vector by a phase-shifting vector, while actual-direction displacement cannot use one direction-independent vector.The equivalent-direction method retains waveform shape there, although actual-direction distortions may occur.

C. Generating and Combining Communication and Sensing Subbeams

The paper presents separated and joint methods for combining communication and sensing subbeams, balancing waveform-shape preservation, communication beamforming gain, flexibility, and computational updating requirements.

  • Combination methods: Two multibeam-generation methods balance preservation of the desired waveform shape against combined communication beamforming gain.Method 1 combines separately generated subbeams with vector-level phase shifting; Method 2 generates them jointly.
  • Method 1: Separated design: The separated design coherently combines independently generated communication and sensing beams, with power allocation controlled through normalized beamforming vectors.
  • Method 1: Separated design: The energy-distribution parameter ρ is selected jointly from communication and sensing requirements, accounting for different active-sensing and communication path-loss factors.The paper states that active sensing has a path-loss factor of about 4, versus about 2 for communication and passive sensing.
  • Method 1: Separated design: The separated design phase-aligns subbeams at the dominant communication direction, but becomes less efficient when communication and sensing subbeams are closely aligned.
  • Method 2: Joint design: The joint design directly obtains a combined beamforming vector from one desired multibeam waveform using the ILS algorithm.

2) Method 2: Joint Design

The joint design controls the overall waveform shape, whereas the separated design generally offers higher communication gain and more flexible updates. The paper therefore mainly uses the separated method, while receiver processing scans communication and sensing directions.

  • Method 2: Joint design: The joint method forms the desired magnitude response by taking the element-wise maximum of the weighted communication and sensing responses before ILS processing.Using the maximum rather than the sum better maintains waveform shape and minimizes sidelobes.
  • Comparison of the two methods: The separated method generally achieves higher communication beamforming gain and permits individual subbeam updates in power and pointing direction.Its sidelobe variation is generally small and gain at desired directions can generally be maintained.
  • Comparison of the two methods: The joint method provides excellent waveform-shape control but generally has inferior beamforming gain because phase alignment is not applied.
  • Comparison of the two methods: Changes to the joint method’s power allocation or pointing directions require running almost the complete generation process, making beamforming updates inflexible.
  • Receiver beamforming: The receiver covers communication and scanning directions either with a receiver multibeam or with a single beam varied over time.The paper considers time-varying single-beam reception for consistency with its sensing algorithm.

2) Passive Sensing in CRPS Stage:

Passive sensing uses receiver multibeams that remain fixed during each packet while varying with the transmit multibeam, trading angle resolution for improved SNR through measurement averaging. Relative distance is measured because transmitter and receiver timing clocks are unsynchronized.

  • Passive sensing: The receiver beam remains fixed over each packet for communication and can vary packet-by-packet with the transmit multibeam for passive sensing.Avoiding direction-varying scanning within each OFDM symbol reduces angle resolution but permits measurement averaging.
  • Passive sensing: Passive sensing measures direction and Doppler similarly to active sensing, but distance is relative because of an unknown timing offset.The offset arises from nonsynchronized transmitter and receiver timing clocks.
  • Comparison with time division: Multibeam sensing can perform at least as well as time division because its scanning subbeam has lower power but proportionally longer scanning time.The sensing mutual information is stated to be always larger for multibeam than time division.
  • Sensing parameters: The sensing task estimates object distance, direction, and speed from received signals using parameter estimation and data fusion across time, frequency, and space.The section considers direct estimation of delay, Doppler, transmit angle, and receive angle.

A. Low-resolution Low-complexity Approach: DFT Method

The low-complexity approach applies periodograms and 2D DFTs to organized multibeam measurements for coarse delay-Doppler estimation, while highlighting averaging constraints and resolution limitations. The paper therefore motivates a 1D MMV compressive-sensing alternative.

  • DFT estimation: Periodograms and 2D DFTs applied to each measurement matrix provide coarse estimates of delay and Doppler and can form a combined Delay-Doppler view.The view sums absolute 2D-DFT outputs across packets and receiver scanning directions.
  • Measurement averaging: Averaging measurements across receiver-beam indices must be cautious because different pointing directions can introduce phase differences approaching 2π.Such phase variation can prevent straightforward SNR improvement.
  • Measurement averaging: Averaging is limited to approximately four consecutive columns when Doppler phase remains small across those columns but not across the full measurement span.The stated example uses Nr = 8 and the condition fD,ℓTs ≤ 0.0072.
  • Resolution limitation: The DFT method has limited resolution, particularly for Doppler frequencies, and improving it requires larger NrNd values and therefore longer packets.Doppler frequencies are typically small because object speeds are limited.
  • Motivation for 1D CS: High-dimensional compressive-sensing models can have high complexity and large quantization errors in Doppler, AoA, and AoD when measurements are limited.The paper states that these effects make current high-dimensional CS solutions ineffective in this framework.
  • Motivation for 1D CS: The proposed 1D MMV-CS method estimates delays first and then estimates remaining parameters, while allowing measurements across packets to be combined for improved delay SNR.The on-grid model is reported to work well for continuous parameters in simulation.

1) Estimation of Delays

Delay estimation is formulated as an on-grid MMV compressive-sensing problem whose measurements share common delay support across scanning directions. Recovered nonzero rows identify delays, after which Doppler estimates can be obtained from the corresponding components.

  • MMV model: The delay-quantized measurement model is formed from Nd OFDM-symbol measurements while ignoring quantization error.The model uses the delay dictionary and organizes measurements for MMV processing.
  • Dictionary design: An overcomplete dictionary with Lp = 2N balances delay quantization error against correlation between dictionary columns.The resulting interpolated DFT matrix has dimension N × Lp.
  • Delay-bin cases: The model distinguishes delay bins containing no multipath, one multipath, or multiple multipaths with different Doppler frequencies.These cases correspond to rows of the association matrix containing zero, one, or multiple ones.
  • MMV model: Common support across measurements is necessary for MMV estimation gains, and nearby or overlapping receiver beams can observe objects with the same quantized delay.This shared support makes combining measurements across scanning directions beneficial.
  • Recovery: MMV BCS or MMV OMP recovers the shared coefficient matrix, whose nonzero rows identify quantized delays and associated Doppler components.The estimates are obtained by referring back to the individual composing matrices.
  • Delay detection: A threshold on the mean cross-correlation statistic determines whether a multipath exists at each quantized delay.The threshold can be selected according to anticipated received-signal power for that delay.

2) Estimation of Doppler Frequency:

Doppler estimation uses spectrum analysis or phase-based processing after delay recovery, with simulations evaluating two multibeam combining methods. The results show shape-control benefits for joint design and a communication-power advantage for coherent combining.

  • Doppler estimation: After MMV recovery, Doppler frequencies can be estimated from each recovered delay component using 1D ESPRIT, with at most Nd/2 resolvable multipaths.This approach does not distinguish single- and multiple-multipath cases.
  • Doppler limitation: Separating single- and multiple-multipath cases using a variance threshold was unreliable and remains an open problem.The proposed distinction relies on a variance that is zero only in the noiseless case.
  • Doppler estimation: When each scanning direction and delay bin contains one multipath, Doppler frequency can be estimated directly from the phase of λq,nd.The small phase increment condition leaves only one possible Doppler value.
  • Simulation setup: The simulations use Nt = 8 packets, Nr = 5 receiver beams, and Nd = 12 OFDM symbols, with a complete sensing cycle lasting about 0.768 ms.The setup uses two 16-element ULAs and a node speed of 20 m/s.
  • Simulation setup: The eight sensing subbeams scan approximately −54.3, −37.8, −24.4, −12.3, 10.8, 22.8, 35.9, and 51.9 degrees over a −60-to-60-degree range.They are spaced uniformly in equivalent directions, producing nonuniform actual directions.
  • Beamshape comparison: Method 2 controls the overall beamshape better, while beamwidth increases with distance from the zero-degree communication direction.Narrower beamwidth generally provides better directional resolution.
  • Power comparison: Method 1 yields approximately 6% higher received power than Method 2 across scanning directions because it uses coherent combining phase.Power is normalized to single-beam communication-only reception.

B. Sensing Results

The simulations evaluate sensing in a 12-scatter environment with continuous, off-grid distances, speeds, and angles. The proposed CS method improves distance–AoA resolution and accuracy over DFT, while most relative-speed estimates are accurate.

  • Simulation setup: The evaluation uses 12 point-source scatters distributed up to 30 m and across AoAs from -60 to 60 degrees, with continuous off-grid parameters.The sensing setup assumes free-space pathloss with path-loss factor 4, 25 dBm transmission power, and no radar cross-section information.
  • Estimation methods: For CS, an interpolated DFT dictionary with interpolation factor 2 enables fast Fourier transforms, and BCS solves the MMV problem.The CS configuration uses the proposed MMV approach with an on-grid model that is evaluated on continuous parameters.
  • DFT results: 1D IDFT produces distance–AoA plots before and after pathloss compensation, normalization, and thresholding, making obstacle locations clearer but still low-resolution.The post-processing sets estimates below -10 dB to -10 dB; the resulting locations remain insufficiently accurate because of DFT’s low resolution.
  • CS results: The proposed CS algorithm yields significantly improved distance–AoA resolution and accuracy compared with DFT, despite some scattered estimates around true locations.Figure 9 shows both direct location estimates and post-processed location and normalized-power results.
  • Speed results: Most relative-speed estimates achieve good accuracy when aligned with the distance estimates, although some estimates remain farther from the true values.The relative-speed results are presented against estimated distance.
  • Overall validation: Simulation results validate the effectiveness of the proposed system, multibeam generation, and sensing algorithms.The conclusion presents the framework as feasible for integrating sensing into standard TDD packet communication systems with OFDM modulation.
Loading 1810.04105v1…