Enabling Emergency Communication via Semantic Radar-Centric ISAC

Technical Summary: Enabling Emergency Communication via Semantic Radar-Centric ISAC

Problem Statement
Modern military platforms (e.g., combat aircraft, unmanned systems) typically rely on two co-located RF assets: a radar for sensing and a dedicated radio for communication. In contested environments, the dedicated communication link is highly vulnerable to jamming, blockage, or physical damage, whereas the radar often remains operational. While repurposing the radar aperture for emergency communication is a logical fallback, a fundamental conflict exists: radar is engineered for sensing, not communication. Loading information onto a radar waveform perturbs the beampattern, distorts the matched-filter response, and degrades sensing performance. The core challenge is to enable emergency communication via a radar-centric Integrated Sensing and Communication (ISAC) link while strictly preserving the radar's primary sensing function. The paper posits that in emergency scenarios, platforms rarely need to transmit raw sensor data; instead, they need to convey small amounts of task-relevant information (e.g., object labels, relative geometry).

Methodology
The authors propose a Semantic Radar-Centric ISAC framework that couples a learned semantic encoder with a conservative radar transmission strategy. The system operates under the premise that the communication payload must be compressed to a few tens or hundreds of bits to fit within a "sensing-safe" budget.

1. Semantic Compression (Encoder Side):

   • The system employs a learned vector-quantized (VQ) semantic encoder. Instead of transmitting raw data, the encoder compresses the application payload (e.g., SAR images or text) into a compact, discrete representation (a sequence of tokens).
   • This representation is designed to be "task-sufficient," meaning it retains enough information for the downstream task (e.g., classification) while minimizing the bit count.
   • The output is a discrete bitstream sized to fit within the radar's safe transmission capacity.

2. Radar-Centric Transmission (Channel Side):

   • The framework utilizes a wideband phased-MIMO radar (based on a Uniform Planar Array).
   • Communication is achieved not by modulating the waveform amplitude or phase directly (which would distort the beam), but through spatial-spectral index modulation.
   • The system defines a codebook of transmission configurations (tuples of active subarray, orthogonal waveform, and selected subcarrier).
   • Sensing-Preserving Constraint: A subset of these configurations, denoted as $G_{safe}$, is selected based on strict constraints:
     • Mainlobe Coverage: The beam must maintain coverage over the sensing region.
     • Sidelobe Leakage: Sidelobes must remain bounded to avoid interfering with other sensing tasks.
     • User Gain: The selected configuration must provide sufficient gain to the friendly receiver.
   • Only configurations satisfying these constraints are admitted into $G_{safe}$. The semantic bits are mapped to indices within this safe subset.

3. Joint Optimization:

   • The system is formulated as a joint optimization problem. The semantic bit budget ($B_z$) is upper-bounded by the radar-safe rate ($R_{radar}$), which is itself determined by the size of $G_{safe}$ and the sensing-degradation budget ($\epsilon$).
   • The objective is to minimize task loss (classification error) subject to the constraint that the semantic payload size fits within the radar-safe rate and the sensing degradation remains below a threshold.

Key Contributions

• Novel Framework: The paper introduces the first framework that couples task-oriented semantic communication with sensing-preserving radar-centric constraints specifically for communication-denied emergency scenarios.
• Conservative ISAC Design: Unlike prior ISAC works that trade sensing performance for communication throughput, this approach prioritizes sensing integrity. It achieves communication by selecting from a conservative subset of spatial-spectral configurations rather than aggressive waveform perturbation.
• Vector-Quantized Semantic Encoding: The use of a learned VQ bottleneck allows the system to compress diverse payloads (images and text) into discrete, serializable tokens that fit the strict bit budget of the radar channel.
• End-to-End Validation: The authors provide a complete system evaluation, linking the semantic compression efficiency, the physical layer transmission reliability, and the final downstream task accuracy.

Experimental Results
The framework was evaluated on two modalities: SAR target classification (MSTAR dataset) and text classification (AG News dataset).

• Semantic Compression Efficiency: Discrete semantic representations (64, 128, and 256 bits) achieved task accuracy comparable to or exceeding that of continuous 2048-bit latent representations. For instance, on AG News, a 128-bit semantic payload achieved 88.30% accuracy, matching the performance of the larger latent space.
• Transmission Reliability: The radar-centric link demonstrated reliable recovery of short semantic payloads. At moderate-to-high Signal-to-Noise Ratios (SNR $\ge$ 20 dB), exact payload recovery reached 100% for both datasets. Shorter payloads (e.g., 64 bits) achieved reliable recovery at lower SNRs compared to longer payloads.
• End-to-End Task Accuracy: Despite channel errors at lower SNRs, the end-to-end task accuracy (classification) remained robust. At 15 dB SNR, the system achieved near-Genie (error-free) accuracy levels for the downstream task, even when exact bit recovery was not 100%, because token errors often fell within task-equivalent codebook regions.
• Sensing Preservation: The "conservative" operating mode introduced negligible sensing degradation. Compared to a sensing-only baseline, the mainlobe loss was minimal (approx. -0.002 dB), and sidelobe perturbation was bounded to a fraction of a decibel, while still providing sufficient gain for the receiver.

Significance and Claims
The paper claims that its framework demonstrates that sensing-critical radar can provide a practical emergency communication link without compromising its primary mission. By leveraging semantic compression to reduce the payload to a few hundred bits and restricting transmission to a sensing-safe configuration subset, the system enables reliable communication for downstream tasks (like target identification or situational awareness) while keeping the radar beampattern almost unchanged. The authors emphasize that this approach is specifically tailored for emergency fallback scenarios where the dedicated radio is down, offering a viable path to maintain coordination without sacrificing the radar's sensing capabilities.