In this paper, we investigate delay-Doppler (DD) sensing in a communication-centric integrated sensing and communication (ISAC) framework based on Zak transform-based orthogonal time frequency space (Zak-OTFS) modulation. Specifically, we consider target sensing with communication waveforms and propose a novel backpropagation (BP) algorithm for multi-target DD parameter estimation. We formulate the radar sensing task as a maximum likelihood parameter estimation problem, which is highly non-convex. By exploiting the structural analogy between parameter estimation and neural network training, the BP algorithm treats the DD parameters as tunable network weights and efficiently computes their gradients via the chain rule, enabling accurate and parallelized estimation. To facilitate the algorithm implementation, a successive interference cancellation method based on DD domain twisted convolution is developed to obtain coarse DD estimates. Furthermore, a constant false alarm rate based dynamic merging strategy is introduced to adaptively estimate the number of targets during the BP process. Comprehensive theoretical analyses are conducted, including the derivation of the Cramér–Rao bound (CRB) for Zak-OTFS systems and performance evaluation under various challenging sensing scenarios. Simulation results demonstrate that the proposed algorithm achieves high estimation accuracy and validates the theoretical analysis.

This paper presents a quasi-neural network (Quasi-NN) approach for parametric channel estimation in multi-input multi-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. A unified Quasi-NN framework is proposed for both the single-user (SU) and the multi-user (MU) scenarios, enabling the joint estimation of the direction of arrival, direction of departure, time delay, complex gain, and the number of multipaths. The artificial neural network-like structure of the Quasi-NN allows the application of the backpropagation algorithm while requiring only real-time pilot signals for online network training. Furthermore, considering the issue of high pilot overhead in MU MIMO-OFDM systems, we develop a location assisted Quasi-NN (LA-QNN) that utilizes a location-parameter database to reduce the pilot overhead while maintaining accurate channel estimation. Simulation results show that the proposed Quasi-NN approaches the Cramér Rao bound, and the proposed LA-QNN scheme provides a better balance between channel estimation performance and pilot overhead in the MU scenario.

This paper studies machine learning-assisted optimal receivers for a communication system with memory, which can be modelled by a trellis diagram. The prerequisite of the optimal receiver is to obtain the likelihoods of the received samples under different state transitions. We propose to learn the trellis diagram real-time using an artificial neural network (ANN) trained by a pilot sequence. This approach, termed as the online learning trellis diagram (OLTD), requires neither the channel state information (CSI) nor statistics of the noise, and can be incorporated into the classic Viterbi and the BCJR algorithm. In a channel with non-Gaussian noise, the OLTD method can significantly outperform the model-based methods that use Gaussian assumption. It requires much less training overhead than the state-of-the-art ANN-assisted methods. As an illustrative example, the OLTD-based BCJR is applied to a Bluetooth low energy (BLE) receiver trained only by a 256-sample pilot sequence. Moreover, the OLTD-based BCJR can accommodate for turbo equalization. As an interesting by-product, we propose an enhancement to the BLE standard by introducing a bit interleaver to its physical layer; the resultant improvement of the receiver sensitivity can make it a better fit for some Internet of Things (IoT) communications.