AI-Empowered Integrated Sensing and Communications

TL;DR

This paper surveys AI-enabled ISAC, arguing that unified waveform and beamforming design face fundamental sensing–communication trade-offs that are hard to solve with traditional model-based methods alone. It contrasts traditional multi-objective optimization with learning-based approaches, including supervised, unsupervised, multi-task, and model-based techniques, and demonstrates two case studies: an unsupervised learning–driven ISAC waveform design and a learned-optimizer for hybrid beamforming. Through these case studies, it showcases how data-driven losses, Lambda layers, and algorithm unrolling can balance sensing accuracy, data rate, and computational complexity, achieving performance close to optimization-based baselines with lower complexity. The paper also outlines future directions, such as end-to-end ISAC, transformer-based design, and digital-twin aided architectures, to push AI-enabled ISAC toward practical 6G deployment.

Abstract

Integrating sensing and communication (ISAC) can help overcome the challenges of limited spectrum and expensive hardware, leading to improved energy and cost efficiency. While full cooperation between sensing and communication can result in significant performance gains, achieving optimal performance requires efficient designs of unified waveforms and beamformers for joint sensing and communication. Sophisticated statistical signal processing and multi-objective optimization techniques are necessary to balance the competing design requirements of joint sensing and communication tasks. Since model-based analytical approaches may be suboptimal or overly complex, deep learning emerges as a powerful tool for developing data-driven signal processing algorithms, particularly when optimal algorithms are unknown or when known algorithms are too complex for real-time implementation. Unified waveform and beamformer design problems for ISAC fall into this category, where fundamental design trade-offs exist between sensing and communication performance metrics, and the underlying models may be inadequate or incomplete. This article explores the application of artificial intelligence (AI) in ISAC designs to enhance efficiency and reduce complexity. We emphasize the integration benefits through AI-driven ISAC designs, prioritizing the development of unified waveforms, constellations, and beamforming strategies for both sensing and communication. To illustrate the practical potential of AI-driven ISAC, we present two case studies on waveform and beamforming design, demonstrating how unsupervised learning and neural network-based optimization can effectively balance performance, complexity, and implementation constraints.

Figures (9)

• Figure 1: Mutual information ($I$ in nats per channel-use) and MMSE versus SNR for an AWGN channel with three different input signals: Gaussian, QPSK, and BPSK. While Gaussian signaling achieves the maximum information rate, it also results in the highest (worst) MMSE.
• Figure 2: A classification of different types of radars.
• Figure 3: Comparison of single-task and multi-task learning for ISAC systems. Left: single-task learning trains separate models for each task independently. Right: Multi-task learning leverages shared layers to learn common features before branching into task-specific layers.
• Figure 4: A coherent processing interval of $T$ seconds duration with $M$ ISAC frames. Each frame contains a fixed-length preamble of $P$ symbols and a varying length data segment. The length of each frame is an integer multiple, $q_m$, of the Nyquist sampling interval in the Doppler domain, $T_{\rm D}$.
• Figure 5: An unsupervised learning-based DNN for ISAC waveforms Ddassanayake2025unsupervised.

Theorems & Definitions (2)

• Remark 1
• Remark 2