Shows Superimposed-Pilot OTFS with Fractional Doppler Improves End-To-End Learning Performance

Researchers are tackling the challenges of reliable communication in fast-moving environments with a new approach to Orthogonal Time Frequency Division Multiplexing (OTFS) modulation. Yushi Lei, Yusha Liu, and Guanghui Liu, all from the University of Electronic Science and Technology of China, alongside Lei Wan of Xiamen University and Kun Yang from Nanjing University, present a modular, learning-based OTFS transceiver framework in their paper. This work is significant because it moves beyond traditionally separated signal-processing modules, instead employing trainable neural network components for tasks like pilot placement and channel estimation, offering a globally optimised system. Simulations demonstrate substantial performance gains over existing methods, even under complex fractional Doppler conditions, paving the way for more robust and practical OTFS systems for future high-mobility applications.

The research addresses limitations in conventional OTFS designs, which often rely on separately optimised signal-processing modules that hinder overall performance.

This study introduces an end-to-end OTFS transceiver framework built upon modular and trainable neural network (NN) components, including constellation mapping, superimposed pilot placement, inverse Zak and Zak transforms, and a U-Net-enhanced NN for joint channel estimation and detection (JCED). Explicitly accounting for the impact of the cyclic prefix, this physics-informed modular architecture offers flexibility for integration with existing OTFS systems and adaptability to diverse communication configurations.
The team achieved a breakthrough by creating an entirely trainable system, moving beyond isolated optimisation of individual modules. Experiments demonstrate that this design substantially outperforms baseline methods in both normalized mean squared error (NMSE) and detection reliability, even under challenging integer and fractional Doppler conditions.

The modular approach allows for interchangeable NN modules, facilitating customisation and optimisation for specific deployment scenarios. This innovative framework directly addresses the issue of fractional Doppler shifts, which cause energy leakage and interference in practical high-mobility environments, a problem often overlooked in previous OTFS designs.

This work establishes a new paradigm for OTFS transceiver design, leveraging the power of DL to optimise the entire system end-to-end. Simulations reveal significant gains in performance, highlighting the potential of DL-based optimisation to enable practical and high-performance OTFS transceivers for future networks.

The research opens avenues for deploying robust and efficient wireless communication in demanding applications such as low Earth orbit (LEO) satellite communications, massive multiple-input multiple-output (MIMO) systems, and integrated sensing and communications (ISAC). The research team developed an end-to-end system comprising trainable neural network (NN) modules, specifically addressing challenges in constellation mapping, pilot placement, and Zak transforms.

This approach moves beyond isolated optimisation of signal-processing modules, aiming for globally optimal performance. Researchers implemented a U-Net-enhanced NN tailored for joint channel estimation and detection (JCED), explicitly accounting for the impact of the cyclic prefix. The system harnesses the power of deep learning to integrate these crucial functions, improving performance in dynamic environments.

This physics-informed modular architecture allows flexible integration with existing OTFS systems and adaptation to diverse communication configurations. Experiments employed simulations to rigorously evaluate the transceiver’s performance under both integer and fractional Doppler conditions. The study pioneered a method for superimposed pilot placement, optimising signal transmission for improved channel estimation.

The team measured performance using normalized mean squared error (NMSE) and detection reliability, demonstrating significant improvements over baseline methods. The innovative transceiver design achieves robustness against fractional Doppler shifts, a key challenge in practical high-mobility scenarios.

The technique reveals that DL-based end-to-end optimisation enables practical and high-performance OTFS transceivers for next-generation networks. Results highlight a substantial advancement in handling the energy leakage and inter-Doppler interference caused by continuous motion and time-varying velocities, paving the way for reliable communication in demanding environments.

Deep learning transceiver achieves robust performance in high-mobility scenarios, even with limited data

Scientists have developed a deep learning (DL) based end-to-end OTFS transceiver framework to overcome limitations in high-mobility wireless communication systems. The research addresses performance degradation commonly encountered in conventional orthogonal frequency division multiplexing (OFDM) systems.

This new framework consists of trainable and interchangeable neural network (NN) modules, including constellation mapping/demapping, superimposed pilot placement, inverse Zak (IZak)/Zak transforms, and a U-Net-enhanced NN for joint channel estimation and detection (JCED). The team explicitly accounted for the impact of the cyclic prefix during the design process.

Experiments revealed that the proposed design significantly outperforms baseline methods in normalized mean squared error (NMSE) and detection reliability. Measurements confirm robustness under both integer and fractional Doppler conditions, crucial for applications like low Earth orbit (LEO) satellite communications and massive multiple-input multiple-output (MIMO) systems.

The work highlights the potential of DL-based end-to-end optimization to enable practical and high-performance OTFS transceivers for next-generation networks. Researchers focused on addressing the challenges posed by fractional Doppler shifts, which cause energy leakage and inter-Doppler interference.

Simulations demonstrate that the modular architecture provides flexibility for integration with conventional OTFS systems and adaptability to different communication configurations. The U-Net-enhanced NN, specifically tailored for JCED, delivers improved performance in challenging high-mobility scenarios.

Data shows that the physics-informed modular architecture allows for precise control over key parameters and facilitates efficient training of the DL modules. Tests prove that the proposed transceiver maintains accurate channel estimation and reliable data detection even with significant fractional Doppler shifts.

The breakthrough delivers a robust solution for scenarios demanding reliable links under pronounced environmental variations and continuous motion. Conventional OTFS systems utilise separate, individually optimised signal-processing modules, which can limit overall performance.

This work proposes a modular architecture comprising trainable neural network (NN) modules for tasks including constellation mapping, pilot placement, and channel estimation, explicitly considering the impact of the cyclic prefix. Simulations demonstrate that this DL-based OTFS transceiver significantly outperforms existing methods in terms of normalised mean squared error (NMSE) and detection reliability, even under challenging integer and fractional Doppler conditions.

The physics-informed modular design allows for flexible integration with conventional OTFS systems and adaptation to various communication configurations, highlighting the potential of end-to-end optimisation for practical, high-performance OTFS in next-generation wireless networks. The authors acknowledge that the performance of the proposed system is dependent on the training data and the specific network architecture chosen.

Future research could explore the application of this framework to more complex channel models and investigate methods for reducing the computational complexity of the NN modules. These findings suggest a promising pathway towards robust and efficient OTFS transceivers, though further investigation is needed to fully realise their potential in real-world deployments.