PaperlayerView exact revision on arXiv

Source-linked AI summary

Enabling Emergency Communication via Semantic Radar-Centric ISAC

Mohaimin Al Barat, Chaoyu Zhang, Hexuan Yu, Kai Zeng, Y. Thomas Hou, Wenjing Lou

arXiv:2609.06801v1cs.NI

TL;DR

When a dedicated communication link is disrupted, a working radar offers a possible emergency channel, but carrying information can impair its sensing function. The paper couples task-oriented semantic compression with conservative radar configurations and reports reliable task communication with the beampattern almost unchanged.

  • Problem

    How can a radar provide emergency communication when the dedicated link is unavailable while preserving the radar’s primary sensing function?

    Source: §I · p. 1

  • Method

    A vector-quantized encoder compresses task-relevant content into a few tens to a few hundred bits, matched to a conservative spatial-spectral radar subset under a bounded sensing-degradation budget.

    Source: §I · pp. 1–2, 1) Spatial-Spectral Configuration… · p. 3, 3) Representation-to-Configuration… · p. 3

  • Results

    Experiments on SAR target and text classification show compact semantic payloads retain high end-to-end accuracy with negligible beampattern distortion.

    Source: §V · p. 6

  • Takeaways & Limitations

    The results demonstrate that sensing-critical radar can provide a practical emergency link without compromising its primary mission.

    Source: §V · p. 6

    Evaluation covers SAR target and text classification; mission-specific waveforms, hardware impairments, and field deployment remain future work.

    Source: §V · p. 6

Abstract

from arXiv · show

Combat aircraft and other modern military platforms field two co-located RF assets: a radar for sensing and a dedicated radio for communication. In contested battlefield environments, however, the dedicated communication path is highly likely to be targeted and disrupted by jamming, blockage, or physical damage, while the radar remains fully operational. A natural fallback is to repurpose the surviving radar aperture as an emergency communication channel. The difficulty is that radar is purposely engineered for sensing rather than communication, so loading any information onto its waveform perturbs the beampattern and inevitably erodes sensing performance: the more bits we push, the more sensing we lose. We identify the core problem as enabling emergency communication via a radar-centric integrated sensing and communication (ISAC) link while preserving the radar's primary sensing function. Therefore, we propose a semantic radar-centric ISAC framework that pairs a learned vector-quantized semantic encoder, which compresses task-relevant content into a few tens to a few hundred bits, with a wideband phased-MIMO radar restricted to a conservative spatial-spectral configuration subset that respects a bounded sensing-degradation budget. Experiments on SAR target classification and text classification show that our framework supports reliable emergency communication for downstream tasks while keeping the radar beampattern almost unchanged.

I. INTRODUCTION

The paper asks how a radar can provide emergency communication when a dedicated link is disrupted without sacrificing its primary sensing function. It proposes coupling task-oriented semantic compression with a sensing-preserving radar configuration subset, addressing a gap in prior work that assumes a conventional radio link.

I. INTRODUCTION

  • Jamming, blockage, or battle damage can sever dedicated links while onboard radar continues local sensing, motivating the radar aperture as a fallback communication resource.The introduction describes this as an operational risk for platforms coordinating with headquarters and wingmen.Source: §I · p. 1
  • Radar communication competes with sensing for spatial, spectral, and waveform degrees of freedom, so stronger or denser modulation can distort the beampattern and degrade sensing –.The paper frames its central question as enabling emergency communication while preserving the radar’s primary sensing function.Source: §I · p. 1
  • The proposed framework compresses task-relevant content into a few tens to a few hundred bits and couples that budget to a radar-safe rate under bounded sensing degradation –.A vector-quantized encoder is paired with a conservative spatial-spectral subset whose beamforming vectors meet mainlobe, sidelobe, and user-gain constraints.Source: §I · pp. 1–2

II. RELATED WORK

  • Prior radar-centric ISAC embeds bits in radar emissions through methods such as precoding, sidelobe modulation, and index modulation, but communication remains constrained by sensing preservation,,,.The related-work discussion identifies payload compression, rather than more aggressive modulation, as the alternative pursued here.Source: §II · p. 2
  • Task-oriented semantic communication compresses data into task-relevant representations, but prior joint sensing and task-oriented communication assumes a conventional radio rather than a sensing-preserving emergency radar channel –.The paper positions coupling semantic compression with radar-centric sensing constraints as the gap it targets.Source: §II · p. 2

III. SYSTEM DESIGN

The design couples a compact semantic bit budget to a conservative radar-safe configuration rate. Its system model uses a shared radar emission for sensing and communication, built on a wideband phased-MIMO platform.

A. Overview

  • The semantic bit budget is bounded by the radar-safe rate, which is itself bounded by the sensing-degradation budget.Communication is carried by selecting configurations rather than injecting arbitrary symbols, while semantic compression fits mission-relevant messages within the safe rate.Source: A. Overview · p. 2

B. System Model and Assumptions

  • The target and friendly listener observe the same radar emission, with the models differing in beam steering and the received quantity.The radar emission is the shared physical basis for both sensing and communication.Source: B. System Model and Assumptions · p. 2

1) Phased-MIMO Radar Platform:

  • The platform is a wideband near-field phased-MIMO radar with M transmit subarrays, each an Nx × Ny array; its total bandwidth W is split into Q subcarriers.Each subcarrier uses M orthogonal sequences of length P.Source: 1) Phased-MIMO Radar Platform · p. 2

2) Sensing Model:

The sensing model describes the intercepted signal produced by the radar emission and identifies beamformers as the shared resource for target sensing and listener illumination.

2) Sensing Model:

  • The sensing model’s transmission produces an intercepted signal with an added noise term.The supplied signal description gives the received signal as m(p, k) plus n_s(p, k).Source: 2) Sensing Model · p. 2
  • The sensing signal combines subarray beamformed waveforms across subcarriers, with propagation loss and near-field steering represented in the model.The steering vector is retained because wideband near-field operation invalidates planar approximations.Source: 2) Sensing Model · p. 2
  • The beamforming vectors are shared: they focus energy on the target for sensing and are reused, within limits, to illuminate the communication listener.Source: 2) Sensing Model · p. 2

3) Communication Receiver Model:

The communication receiver is modeled as a friendly platform using its existing radar receive chain, while the payload is a learned discrete semantic representation. Radar-side communication transmits information by selecting configurations rather than modulating the waveform.

3) Communication Receiver Model:

  • The receiver is a friendly formation platform with the same class of phased-MIMO radar, modeled as a single-antenna listen point at a known location with dominant line of sight.Its existing radar chain recovers the payload using matched filtering and spectral-index detection, requiring no extra hardware.Source: 3) Communication Receiver Model · p. 2

4) Emergency Semantic Payload:

C. Radar-Centric Communication Design

  • Communication bits are conveyed by selecting radar configurations: the design enumerates choices, retains sensing-safe configurations, and counts the bits those choices can carry.Each admissible configuration acts as one symbol.Source: C. Radar-Centric Communication Design · p. 3

1) Spatial-Spectral Configuration Codebook:

The codebook jointly indexes spatial and spectral resources, while communication uses only configurations that meet sensing-preservation and receiver-gain constraints.

1) Spatial-Spectral Configuration Codebook:

  • Each configuration Γi = [mi, si, qi] indexes an active transmit subarray, its assigned orthogonal waveform, and a selected subcarrier.The spatial-spectral codebook combines subarray–waveform pairing with subcarrier selection steered toward the receiver.Source: 1) Spatial-Spectral Configuration… · p. 3

2) Sensing-Preserving Constraint:

  • A configuration enters Gsafe only when its beamformer preserves sensing-region mainlobe coverage, bounds sidelobe leakage, and supplies sufficient gain at the receiver.Configurations that violate any of these conditions are excluded.Source: 2) Sensing-Preserving Constraint · p. 3

3) Representation-to-Configuration Mapping and Rate:

The semantic bitstream is mapped to sensing-safe radar configurations, whose subset size sets the conservative rate; the receiver then decodes configurations back into the semantic representation.

3) Representation-to-Configuration Mapping and Rate:

  • The conservative radar rate is ⌊log2 |Gsafe|⌋/Tsym, and the unconstrained rate is available only if every configuration is sensing-safe.The design offsets the smaller safe subset's rate loss through semantic compression rather than stronger modulation.Source: 3) Representation-to-Configuration… · p. 3

4) Radar Receiver Decoding:

  • The receiver separates frequency components with a Q-point DFT, matched-filters the orthogonal waveforms, and recovers each configuration from its dominant spectral index and phase structure.It remaps the decoded configurations to a bitstream and deserializes that stream to recover z.Source: 4) Radar Receiver Decoding · p. 3
  • When the recovered bitstream equals the transmitted bitstream, the radar link is bittransparent and remaining end-to-end loss comes solely from semantic compression.Source: 4) Radar Receiver Decoding · p. 3

D. Semantic Representation Learning Design

The semantic encoder is designed to fit the radar's tight bit budget with a compact, discrete representation that preserves downstream task accuracy.

1) Discrete Representation and Bit Budget:

  • The representation is a sequence of L discrete tokens from K possible values, with payload size Bz = L⌈log2 K⌉ bits.Discrete tokens are exactly serializable, avoiding floating-point quantization noise.Source: 1) Discrete Representation and Bit… · p. 3

2) Model Architecture:

  • The vector-quantized bottleneck maps each of L continuous encoder features to its nearest entry in a learned K-entry codebook before decoding the token sequence.Binary and learned-tokenization bottlenecks are also compatible with the design.Source: 2) Model Architecture · pp. 3–4

3) Training Objective and Radar-Aware Constraint:

4) Joint Design Objective:

The joint objective couples semantic learning and radar embedding to find a task-sufficient representation within payload, latency, and sensing-preservation constraints.

4) Joint Design Objective:

  • The complete system jointly constrains the semantic encoder and decoder, discrete bottleneck, and conservative embedding subset.Source: 4) Joint Design Objective · p. 4
  • The objective targets the smallest task-sufficient representation under a sensing-preservation constraint rather than maximizing radar-centric ISAC throughput.Its stated quantities include payload size Bz, conservative radar rate Rradar(Gsafe), latency budget Tmax, and sensing-preservation budget Lsense ≤ ϵ.Source: 4) Joint Design Objective · p. 4
  • The design uses the same spatial-spectral degrees of freedom as a conservative, radar-safe emergency channel.Source: 4) Joint Design Objective · p. 4

IV. EVALUATION

The evaluation tests whether compact semantic payloads preserve classification accuracy and can be transmitted through the radar-centric link. Across SAR and text tasks, the discrete representations offer substantial compression with limited accuracy loss.

A. Experimental Setup

  • The evaluation covers four-way SAR target classification on 3,549 MSTAR chips and four-way news classification on a stratified AG News subset.MSTAR uses an 80/20 train/test split; AG News uses 32,000 training and 4,000 test examples.Source: A. Experimental Setup · p. 4
  • The radar simulator leaves four sensing-preserving spatial-spectral configurations for communication, yielding 2 bits per radar symbol across the SNR sweep.It uses a 10 GHz carrier, 500 MHz bandwidth, 64 OFDM subcarriers, 32 pulses per symbol, and four selected subcarriers.Source: A. Experimental Setup · p. 4
  • Test samples are live-encoded, transmitted through the simulator, recovered by maximum-likelihood spectral-index detection, and classified by the semantic decoder.The upstream encoder is not rerun at the receiver.Source: A. Experimental Setup · p. 4

B. Semantic Compression Efficiency (RQ1)

  • Discrete semantic payloads compress by several orders of magnitude with marginal accuracy loss; Sem-128 comes closest to int8 on MSTAR, while learned representations match or slightly exceed int8 on AG News.On MSTAR, int8 has the highest accuracy, and Sem-256 underperforms its smaller-codebook siblings; the authors attribute this to training a large codebook on a small SAR dataset.Source: B. Semantic Compression Efficiency (RQ1) · p. 4

C. Radar-Centric Semantic Transmission (RQ2)

The radar-centric link becomes increasingly reliable as SNR rises, while shorter payloads reach useful recovery earlier. End-to-end task accuracy approaches the error-free-payload reference by 20 dB on both modalities.

C. Radar-Centric Semantic Transmission (RQ2)

  • Across both datasets, BER and SER fall with rising SNR, while exact payload recovery moves from near-zero at low SNR to near-perfect at high SNR.Shorter payloads reach useful recovery earlier; at the top of the sweep, the link is essentially error-free across payloads, making payload length rather than channel quality the constraint at moderate-to-high SNR.Source: C. Radar-Centric Semantic Transmission… · p. 4, Table I · p. 5

D. End-to-End Task Accuracy After Radar Transmission (RQ3)

  • On both modalities, end-to-end accuracy matches the genie baseline by 20 dB, and at 15 dB it already nears the baseline despite substantially lower exact recovery.The passage attributes this gap to discrete-token errors falling in task-equivalent codebook regions.Source: D. End-to-End Task Accuracy After… · p. 5, Table II · p. 5

E. Sensing Preservation Under Communication Embedding (RQ4)

The conservative embedding setting supports compact semantic communication while keeping the radar beampattern almost unchanged. The conclusion identifies future evaluation of mission-specific waveforms, hardware impairments, and field deployment.

E. Sensing Preservation Under Communication Embedding (RQ4)

  • The conservative operating point has small mainlobe loss and sidelobe perturbation bounded to a fraction of a decibel, while user-side gain rises above the sensing-only reference.The aggressive setting trades additional user gain for measurable mainlobe loss and is not used in the reported link experiments.Source: E. Sensing Preservation Under… · p. 6

V. CONCLUSION

  • The authors conclude that compact semantic payloads retain high end-to-end accuracy with negligible beampattern distortion, and identify mission-specific waveforms, hardware impairments, and field deployment for future evaluation.The framework pairs a vector-quantized semantic encoder with a wideband phased-MIMO radar restricted to a conservative spatial-spectral subset under a bounded sensing-loss constraint.Source: §V · p. 6
Loading 2609.06801v1…