WAVEFORM DESIGN FOR MONOSTATIC DOWNLINK INTEGRATED SENSING AND COMMUNICATION SYSTEMS

Abstract

With the opportunities provided by higher frequencies, larger bandwidths, and intelligence, Integrated Sensing and Communication (ISAC) is widely recognized as one of the new applications that will drive the development of future generations of wireless networks. This thesis focuses on dual-function radar-communication systems, in which a single waveform is synthesized to achieve both the communication and sensing functions. The thesis develops design techniques for that waveform, aiming to jointly optimize communication and sensing performance, under practical constraints that facilitate implementation. A progressive three-part framework is proposed, covering robust extended linear precoding, hybrid linear-nonlinear precoding, and unstructured direct waveform design.

First, to address the sensitivity of the extended linear precoding scheme to imperfect knowledge of the environment, in Chapter 2 we develop a robust design that seeks jointly optimal transmit and radar receive beamformers in the presence of uncertainty. The method maximizes the worst-case Signal-to-Interference-plus-Noise Ratio (SINR) in the radar return signal, while ensuring communication users meet their SINR targets with a given probability of outage. Numerical results show that our proposed method can achieve better performance than approaches that are based on heuristic modifications of designs that assume perfect knowledge of the environment.

The extended linear precoding architecture used in the robust design facilitates design techniques that are based on statistical models for the communication symbols. That is sufficient for design objectives that are functions of the transmit covariance. However there are several important sensing objectives and implementation constraints that are functions of the waveform itself, and not simply its covariance. In scenarios where those objectives and constraints are important, nonlinear precoding has the potential to provide significantly better performance. The existing approaches to nonlinear precoding take a block-by-block symbol-dependent design approach, and may require adaptation of the communication receivers in each symbol block. Therefore, in Chapter 3 we develop a design technique for hybrid linear-nonlinear precoders (HLNP) that fuses the operational simplicity of statistics-based design of linear precoders and the degrees of design freedom provided by symbol-dependent nonlinear precoding. We evaluate this design approach using a problem that seeks to minimize a simplified Cramer-Rao bound on angle estimation of multiple point targets. Our experimental results show that the proposed method achieves essentially the same performance as an existing symbol-dependent hybrid linear-nonlinear precoding, while being able to directly control the transmitted waveform and maintaining receiver adaption on the time scale of environment coherence time.

Finally, in Chapter 4 we introduce a new approach to unstructured direct waveform design. Unlike existing approaches, which require the receivers to have fixed equalizers, or to update the equalizers for every data block, the proposed design allows the equalizer at each receiver to be adapted at the scale of the coherence time, using the conventional dedicated downlink training, while maintaining the ability to explicitly control the transmitted waveform. In an example employing sensing objectives obtained from a Bayesian Cramer-Rao bound, the proposed approach demonstrates performance that is close to the methods with equalizer adaption at the time scale of the data block, better than the methods with constant equalizers, and better than symbol-dependent linear precoding techniques.