In frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) communication systems, the base station (BS) requires channel state information (CSI) reported from user equipment (UE) for downlink precoding, which brings in significant feedback overhead. In this paper, we propose a position-based precoding method, where the precoder at the BS is directly derived from the location information of UE, without relying on channel measurement and CSI feedback. To achieve this, we devise a novel neural network (NN) structure called P2P-Net (Position-to-Precoder Net), which includes a position encoding module, an adaptive combination weight, and a refining module based on self-attention mechanism. With deep learning techniques, P2P-Net is able to learn the information about scatterers in the signal propagation environment, thereby realizing the mapping from position to precoder. Simulation results demonstrate the superiority of the proposed positionbased precoding method compared with existing feedback-based solutions in terms of spectral efficiency and communication overhead.

In frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) communication systems, the base station (BS) requires channel state information (CSI) reported from user equipment (UE) for downlink precoding, which brings in significant feedback overhead. In this paper, we propose a position-based precoding method with sensing assistance to realize MIMO downlink transmission without CSI feedback from UE. By exploiting the location of UE and the information of the propagation environment provided by wireless sensing techniques, the BS is able to derive the precoder for downlink transmission. To achieve this, we devise a novel neural network (NN) structure called PS-Net (Position-based-precoding with Sensing-assistance Network), which includes an environmental feature extractor, a weight generation module, an adaptive position encoder, and a position-to-precoder mapper. Using the PS-Net, information about the scatters in the propagation environment can be extracted and fused with the UE’s location to realize position-based precoding for time-varying channels. We also propose a dedicated data augmentation method called random phase shifting to enhance the training data diversity, thus improving the generalization ability of PS-Net. Simulation results demonstrate the superiority of the proposed PS-Net compared with the existing feedback-based solutions and other position-based approaches in terms of spectral efficiency and communication overhead.

Downlink channel state information (CSI) feedback typically results in an unacceptable overhead in frequencydivision-duplex (FDD) massive multiple-input multiple-output (MIMO) systems. To deal with this challenge, several deep learning (DL) based CSI compression and recovery approaches have been developed, which follow an auto-encoder architecture and aim at minimizing CSI reconstruction error. Different from the mainstream methodology mentioned above, in this letter, we follow a goal-oriented design philosophy. That is, instead of minimizing the reconstruction error, we train a deep neural network (NN) to compress the CSI such that the precoder using the compressed CSI as input can optimize the downlink transmission performance, i.e., minimize the bit error rate (BER) at the UEs. A two-stage training method is developed to train the NN. Experimental results demonstrate that the proposed scheme outperforms the existing solutions in terms of signal-tointerference-plus-noise ratio (SINR) and BER at terminal users

In this paper, an uplink signal detection problem is considered for Internet-of-Things (IoT) networks. Owing to the imperfections of IoT devices including I/Q imbalance and amplifier non-linearity, exact end-to-end channel models and accurate channel state information (CSI) are typically unavailable at the receiver, which obstructs the application of traditional model-based signal detection algorithms. A consensus has been reached recently that Deep learning (DL) is a promising tool to cope with this problem. However, for the IoT scenarios under consideration, devices typically transmit data using short packets with few pilot symbols, the amount of which is insufficient for each device to individually train a detector. In order to combat the data scarcity barrier and enable few-shot learning, a novel training paradigm is proposed where pilot symbols from different devices are aggregated in an intelligent manner to train a universal signal detector. Specifically, this paper devises a knowledge-driven signal detector architecture following the modular design methodology typically used in classical communication system receivers. Under this framework, three neural networks (NNs), a signal classifier, a channel feature extractor, and a signal feature extractor are created to form decision statistics and produce estimates of the transmitted symbols. Furthermore, borrowing ideas from domain adaptation, a novel component referred to as a link discriminator is integrated into the architecture to improve its generalizability. The proposed signal detector exploits pilot symbols from various IoT devices to train a universal detector that can be applied to different channel conditions without retraining, including those not seen in the training phase. Simulation results verify the superiority of the proposed knowledge-driven detector compared with existing solutions in the sense that it enjoys higher detection accuracy and can be well trained with less data.

Due to the openness of wireless medium and the public structure of pilot signals, wireless sensing procedure is vulnerable to eavesdropping, which causes privacy concerns. In this paper, a novel physical layer obfuscation solution termed as multi-antenna signal masking is proposed to realize privacy-preserving sensing. The privacy protection is realized via controlling the phase difference between the sensing signals of different antennas. Considering the fact that the channel state information (CSI) variation caused by changes of the physical environment typically slowly varies with time, the phase difference is designed as a slowly-varying function with temporal-correlation such that the real variation pattern in CSI is masked and the eavesdropper is thus unable to perform sensing based on the measured CSI. Furthermore, we also devise a round-trip transmission method to avoid secret information exchange between legitimate users, hence realizing privacy-protection without additional overhead. Simulation results demonstrate the superiority of the proposed method in terms of sensing accuracy and privacy-protection capability compared with existing works.

This paper considers a wireless federated learning (FL) system, where the parameters of neural networks (NNs) from distributed users are transmitted to the base station (BS) periodically via wireless links for global aggregation. Due to random fading, users experiencing deteriorated channel conditions are unable to upload their NN parameters successfully, which lowers the convergence rate and degrades the accuracy of the NN model. In order to mitigate the influence of channel fading and accelerate convergence, we propose UC-FL, a user cooperation framework for wireless FL. Unlike the traditional FL paradigm where only “vertical” connections (i.e., users-to-BS) are supported, in the UC-FL framework, “horizontal” connections (i.e., users-to-users) are also introduced to enable user cooperation. In this manner, users with good channel conditions help those experiencing deep fading channels to upload their NN parameters, which provides more opportunities for distributed users to participate in global aggregation. Moreover, a novel global aggregation weight design is proposed by taking into account the channel conditions, to further improve the performance. Simulation results demonstrate the superiority of the proposed UC-FL compared with the classic FedAvg counterpart in terms of model accuracy and convergence rate.

This paper considers a signal detection problem for uplink transmission in Internet-of-Things (IoT) networks. Due to the non-idealities of IoT devices such as amplifier’s non-linearity, the exact end-to-end channel model as well as the accurate channel state information (CSI) is not available at the receiver (i.e., base station), which precludes the possibility of using classical model-based signal detection methods. Deep learning (DL) techniques have been recognized recently as an effective tool to deal with this challenge. However, for IoT scenarios, devices typically transmit data using short packets with few pilot symbols. The amount of training data is insufficient to train a detector for each device individually. In order to combat the data scarcity barrier and enable few-shot learning, this paper proposes to aggregate pilot symbols from different devices in an intelligent manner to train a universal signal detector which is applicable to all possible channel conditions. To be specific, a knowledge-driven signal detector architecture is devised following the modular design methodology for classical communication system receivers. Under this framework, three neural networks (NN), termed as channel feature extractor, signal feature extractor, and signal classifier, respectively, are employed to form decision statistics and make estimates on the transmitted symbols. Furthermore, borrowing the ideas in domain adaptation, a novel component termed as link discriminator is integrated into the architecture to improve the generalization capability of the detector. Simulation results demonstrate that the proposed knowledge-driven detector outperforms the existing solutions in the sense that it enjoys higher detection accuracy and can be well trained with less training data.

The Viterbi algorithm is widely adopted in digital communication systems because of its capability of realizing maximum-likelihood signal sequence detection. However, implementation of the Viterbi algorithm requires instantaneous channel state information (CSI) to be available at the receiver. This is difficult to satisfy in some emerging communication systems such as molecular communications, underwater optical communications, etc, where the underlying channel models are highly complex or completely unknown. ViterbiNet, developed in the prior literature, is a promising framework to cope with this challenge, where deep learning (DL) techniques are combined with the Viterbi Algorithm to enable near-optimal signal detection without CSI. This paper offers a non-trivial variation of ViterbiNet based on generative adversarial networks (GAN). Specifically, a novel architecture using GAN is designed to directly learn the channel transition probability (CTP) from receiver observations, which is the only part of the Viterbi algorithm that is channel-dependent. With the learned CTP, the classical Viterbi algorithm can be implemented without modifications. To make the proposed architecture applicable to time-varying channels, we further develop two methods to fine-tune the learned CTP online. In the first method, pilots within each frame are exploited to update the CTP learning network; In the second method, a decision-directed approach is devised to generate training data in real-time, which is utilized to re-train the learning network. By combining these two approaches, the receiver is able to track the dynamic channel conditions without being trained from scratch. Numerical simulations demonstrate the superiority of the proposed design compared to existing methods.

In this letter, a deep-learning based method is proposed for signal sequence detection. A novel neural network (NN) architecture, in communications systems called Cooperative and Time-varying Bidirectional Recurrent Neural Network (CTBRNN), is developed, which learns from the training data and estimates the transmitted signal sequence without knowing the underlying channel model. Furthermore, we develop a chemical communication experimental platform to collect real data, which is used to train the NN and evaluate the performance of the developed detector. Experimental results demonstrate that, the proposed detection method outperforms the existing NN-based and NN-free candidate solutions in terms of the detection accuracy.