We propose a novel method for interpreting neural networks, focusing on convolutional neural network-based receiver model. The method identifies which unit or units of the model contain most (or least) information about the channel parameter(s) of the interest, providing insights at both global and local levels—with global explanations aggregating local ones. Experiments on link-level simulations demonstrate the method’s effectiveness in identifying units that contribute most (and least) to signal-to-noise ratio processing. Although we focus on a radio receiver model, the method generalizes to other neural network architectures and applications, offering robust estimation even in high-dimensional settings.

For 6G, the use of AI/ML is one of the key technologies and its application to the air interface is being widely studied. As an effective use of AI/ML for air interface, a deep learning-based pilotless transmission scheme has been proposed where AI/ML is applied to functions such as channel estimation, equalization and demodulation. Several papers have shown through simulation evaluations that the application of AI/ML to the air interface can improve spectral efficiency. However, in all AI/ML applications, it is important to verify how well they perform in real environments. In this paper, a proofof-concept for the deep learning-based pilotless transmission scheme is constructed and experimental results in real indoor environments are shown. The results of the conducted experiments at fixed and walking speeds verify that the scheme improves throughput performance by 6.5-18.2 %.

Recent research shows that integrating artificial intelligence (AI) into wireless communication systems can significantly improve spectral efficiency. However, most AI-based receiver studies rely on simulated radio channel data for both training and validation, raising concerns about real-world generalization, which is vital for ensuring reliable field performance. In this study, we train DeepRx, a convolutional neural network (CNN)-based OFDM receiver, under various simulated channel scenarios and validate its performance over- the-air (OTA) using software-defined radio (SDR) technology in a small cell-type setup. To enhance receiver training, we investigate a randomized 3GPP TS38.901 channel model to diversify the training data, thereby improving performance over conventional receivers and matching or exceeding the performance of receivers trained on narrowly targeted channel models. These results demonstrate DeepRx's robust generalization capability and suggest that narrowly scoped, individual TS38.901 models can compromise both training and validation, underscoring the need for tailored channel models, careful training strategies, and OTA testing in learned receiver development.

The deep learning (DL)-based channel state information (CSI) feedback methods have been intensively explored in recent years. Most existing works are trained offline based on the prestored datasets. However, in real-world deployment, the pretrained model may not fit the field environment due to the wireless environment changes. In a previous study, a supervised learning approach has been introduced to deal with this CSI feedback transferability problem by finetuning the model. Nevertheless, the transmission for the original uncompressed CSI data will cause intense traffic in air interface. In this paper, a novel unsupervised transfer learning framework named Discrepancy-based Domain Adaptation Network (DDA-Net) is proposed to solve this problem. By minimizing the discrepancy between the CSI codeword datasets from the pretrained environment and field environment, the encoder and decoder are finetuned to extract the common features in both environments, so that the DL-based CSI feedback model can also work properly in a drifted environment without transmitting any original uncompressed CSI data. Simulation results show that the DDA approach can augment CSI feedback reconstruction accuracy and combat overfitting problems in the deployment environment. Compared to the existing supervised learning approach, the DDA approach can achieve similar CSI recovery performance without transmitting the original uncompressed CSI data, reducing considerable transmission traffic.

Deep learning (DL) based receiver (DeepRx) has been proven to be able to greatly improve the data transmission performance compared to conventional OFDM (orthogonal frequency-division multiplexing) receivers. In order to accommodate DFT-s-OFDM (discrete Fourier transform-spread OFDM), which is a widely used single carrier waveform for uplink transmission characterized with low peak to average power ratio (PAPR), the internal structure of DeepRx needs to be redesigned. In this paper, we propose a novel deep neural network based receiver, referred to as IDFT-deprecoding embedded deep receiver (iDeepRx) customized for DFT-s-OFDM. By embedding an untrainable functionality of IDFT-deprecoding between two trainable neural network structures, the proposed iDeepRx can support DFT-s-OFDM transmission and mitigate implicit link and hardware-induced channel impairments. We demonstrate a single-layer 100 Gb/s error-free data transmission using iDeepRx over 20 GHz bandwidth at 220 GHz carrier frequency and observe 2 dB+ relative performance gain over the conventional receiver. Moreover, intermediate output of iDeepRx has been extracted with constellation-like patterns for visualization and performance monitoring during training and inference, which helps in making the internal working mechanism of iDeepRx more explainable.

Machine learning algorithms have recently been considered for many tasks in the field of wireless communications. Previously, we have proposed the use of a deep fully convolutional neural network (CNN) for receiver processing and shown it to provide considerable performance gains. In this study, we focus on machine learning algorithms for the transmitter. In particular, we consider beamforming and propose a CNN which, for a given uplink channel estimate as input, outputs downlink channel information to be used for beamforming. The CNN is trained in a supervised manner considering both uplink and downlink transmissions with a loss function that is based on UE receiver performance. The main task of the neural network is to predict the channel evolution between uplink and downlink slots, but it can also learn to handle inefficiencies and errors in the whole chain, including the actual beamforming phase. The provided numerical experiments demonstrate the improved beamforming performance.

In this article, we propose multiple machine learning (ML) based physical-layer receiver solutions for demodulating orthogonal frequency-division multiplexing (OFDM) signals that are subject to high level of nonlinear distortion. Specifically, three novel deep learning based convolutional neural network receivers are devised, containing layers in time- and/or frequency-domains, allowing to demodulate and decode the transmitted bits reliably despite the high error vector magnitude (EVM) in the transmit signal. Applicable training procedures are also described, such that the learned layers in the receiver processing properly generalize over different nonlinear distortion and multipath channel characteristics. Extensive set of numerical results is provided, in the context of 5G NR uplink (UL) incorporating also measured terminal power amplifier (PA) characteristics. The obtained results show that the proposed receiver systems are able to clearly outperform the classical linear minimum mean-squared error (LMMSE) receiver as well as the existing ML receiver approaches, especially when the EVM is high compared to modulation order. This is particularly so when the devised ML receiver is of hybrid nature with layers both in time and frequency. The proposed ML receivers can thus facilitate pushing the terminal PA systems deeper into saturation, and thereon improve the terminal power-efficiency, radiated power and network coverage. Through combining the obtained radio link performance results with link budget calculations, all carried out at the 28 GHz mmWave band, it is shown that the proposed ML receivers can enhance the network coverage in terms of maximum UL link distances by close to 100%, when compared to classical LMMSE receiver based networks.

This paper studies communications service availability for industrial wireless control systems. We consider a motion controller with a continuous closed-loop control link to a group of actuator devices on a factory floor. The goal is to satisfy end-to-end latency for each packet and to guarantee that the communication service will not be un-available for longer than a survival time. We propose to decouple the scheduling operation between the normal and survival modes of operation, enabling a dual-mode ultra-reliable and low-latency communications (URLLC) scheduler. Scheduler strategies for the survival mode are presented, targeting link adaptation and signal to interference and noise ratio (SINR) estimation in presence of temporal and spatial channel correlation. Through numerical examples, we investigate the impact of channel correlation on the schedulers ability to target the required reliability for each mode. We further present our findings on system-level performance evaluation of such scheduling strategies by adopting a realistic system setup and channel model to obtain insights with high level of realism. Extensive simulation results are presented which demonstrate significant reduction in resource utilization with the proposed dual-mode scheduler when compared to single-mode URLLC scheduling. Specifically, our results demonstrate that the scheduler should target moderate packet error rate (PER) for normal mode of operation and very low PER for the survival mode; the latter guarantees service availability while the former saves radio resources.

The time-varying and non-stationary channel characteristics caused by high Doppler effects is a significant challenge for modern wireless communication systems. Aiming at the high-speed scenarios, a constellation image analysis aided channel estimation using a hybrid neural network (NN) structure, named PR-SRNN, is proposed to combat Doppler effects. Only 1 demodulation reference signal (DMRS) pilot, rather than multiple pilots as in previous reports, is needed by the proposed PR-SRNN in high-speed scenarios. The impacts from Doppler effects are analyzed and compensated by an intelligent pattern recognition neural network (PRNN) based on the contaminated constellation images. The output layer of PRNN is then merged with a super resolution NN (SRNN) which preliminarily reconstructs the channel estimation in frequency domain. Adaptive features learning against Doppler effects and synergic compensation against multipath fading are learnt jointly. It is the first time to our knowledge that a pattern recognition NN on constellation images is introduced to the physical layer, functioning as expert-like analysis system. The simulation results based on both 3GPP statistical channel models and ray tracing show that PR-SRNN exhibits robustness against diverse degrees of Doppler effects beyond the pretrained scope. Amongst different framework candidates of super resolution (SR) NNs, residual convolution SRNN with channel attention has been selected regarding its performance superiority in terms of loss and convergence speed. Furthermore, the cross validation between PR-SRNN and our previously proposed SubSRNN which takes in extra semantic information as a 2ndinput proves that PR-SRNN is effective for Dopplers effects without extra side information to be reported.

Current fourth-generation (4G) long-term evolution (LTE) base-stations (BSs) that are not capable of performing elevation beamforming are down-tilted in order to optimize the throughput of ground user equipment (GUE). As a consequence of this down-tilt, unmanned aerial vehicles (UAVs), especially those at higher altitudes, are forced to connect with the BS side-lobes, resulting in increased transmit powers and interference to other GUEs and UAVs. In this paper, we propose using passive reflectors placed close to the BS as a low-cost alternative to elevation beamforming to compensate for the reduced antenna gain, and consequently, the increased pathloss between the BS and UAVs. The effectiveness of such reflectors in lowering the interference and increasing throughputs is tested through extensive simulations using the recently introduced third-generation partnership project aerial channel models.

Recently, deep learning has been proposed as a potential technique for improving the physical layer performance of radio receivers. Despite the large amount of encouraging results, most works have not considered spatial multiplexing in the context of multiple-input and multiple-output (MIMO) receivers. In this paper, we present a deep learning-based MIMO receiver architecture that consists of a ResNet-based convolutional neural network, also known as DeepRx, combined with a so-called transformation layer, all trained together. We propose two novel alternatives for the transformation layer: a maximal ratio combining-based transformation, or a fully learned transformation. The former relies more on expert knowledge, while the latter utilizes learned multiplicative layers. Both proposed transformation layers are shown to clearly outperform the conventional baseline receiver, especially with sparse pilot configurations. To the best of our knowledge, these are some of the first results showing such high performance for a fully learned MIMO receiver.

In this paper, we propose a machine learning (ML) based physical layer receiver solution for demodulating OFDM signals that are subject to a high level of nonlinear distortion. Specifically, a novel deep learning based convolutional neural network receiver is devised, containing layers in both time- and frequency domains, allowing to demodulate and decode the transmitted bits reliably despite the high error vector magnitude (EVM) in the transmit signal. Extensive set of numerical results is provided, in the context of 5G NR uplink incorporating also measured terminal power amplifier characteristics. The obtained results show that the proposed receiver system is able to clearly outperform classical linear receivers as well as existing ML receiver approaches, especially when the EVM is high in comparison with modulation order. The proposed ML receiver can thus facilitate pushing the terminal power amplifier (PA) systems deeper into saturation, and thereon improve the terminal power-efficiency, radiated power and network coverage.

In this work, we propose a reinforcement learning (RL) based solution for managing inter-user-interference in wireless full-duplex networks. In particular, the RL algorithm is trained to divide half-duplex capable UEs into two groups such that a full-duplex capable base station can serve one group in uplink while the other one is in downlink. Moreover, it can do the grouping based on only the realized data rates of the UEs, without any explicit knowledge of the mutual interference patterns. With proper training, it is shown that the RL algorithm can identify groups where the interference from the uplink UEs to the downlink UEs is minimized. The results indicate that such a scheme can achieve 92% of the optimal sum data rate, thereby providing over a 60% throughput gain in comparison to a corresponding half-duplex network. A purely random UE grouping will only yield a 48% gain over a half-duplex network.

In this paper, new digital predistortion (DPD) solutions for power amplifier (PA) linearization are proposed, with particular emphasis on reduced processing complexity in future 5G and beyond wideband radio systems. The first proposed method, referred to as the spline-based Hammerstein (SPH) approach, builds on complex spline-interpolated lookup table (LUT) followed by a linear finite impulse response (FIR) filter. The second proposed method, the spline-based memory polynomial (SMP) approach, contains multiple parallel complex spline-interpolated LUTs together with an input delay line such that more versatile memory modeling can be achieved. For both structures, gradient-based learning algorithms are derived to efficiently estimate the LUT control points and other related DPD parameters. Large set of experimental results are provided, with specific focus on 5G New Radio (NR) systems, showing successful linearization of multiple PA samples as well as a 28 GHz active antenna array, incorporating channel bandwidths up to 200 MHz. Explicit performance-complexity comparisons are also reported between the SPH and SMP DPD systems and the widely-applied ordinary memory-polynomial (MP) DPD solution. The results show that the linearization capabilities of the proposed methods are very close to that of the ordinary MP DPD, particularly with the proposed SMP approach, while having substantially lower processing complexity.

In this paper, we present a novel nonlinear digital self-interference canceller algorithm, its implementation details on a software-defined radio (SDR) platform, and performance results of real-time full-duplex experiments on both device and link level. The canceller algorithm is based on an augmented Hammerstein model, with a nonlinear part modeling the transmitter non-idealities followed by a linear filter to model the self-interference (SI) channel. The nonlinear part includes a spline-based model for the nonlinear power amplifier, a polynomial model for baseband nonlinearities, as well as models for I/Q mismatch and LO leakage. The canceller is implemented on an FPGA as a part of an OFDM transceiver testbed for real-time measurements. Extensive real-time measurements show excellent performance: (1) the digital canceller, together with an RF isolator, can suppress the SI to within 1-2 dB's of the receiver noise floor, with total SI suppression of up to 103 dB; (2) digital cancellation of up to 46 dB is evidenced, which is among the highest real-time cancellations in literature; (3) system-level measurements with OFDM signals demonstrate the benefit of utilizing the proposed canceller in a two-way communication scenario, showing up to 90 % increase in sum-rate compared to half-duplex communication.

Deep learning has solved many problems that are out of reach of heuristic algorithms. It has also been successfully applied in wireless communications, even though the current radio systems are well-understood and optimal algorithms exist for many tasks. While some gains have been obtained by learning individual parts of a receiver, a better approach is to jointly learn the whole receiver. This, however, often results in a challenging nonlinear problem, for which the optimal solution is infeasible to implement. To this end, we propose a deep fully convolutional neural network, DeepRx, which executes the whole receiver pipeline from frequency domain signal stream to uncoded bits in a 5G-compliant fashion. We facilitate accurate channel estimation by constructing the input of the convolutional neural network in a very specific manner using both the data and pilot symbols. Also, DeepRx outputs soft bits that are compatible with the channel coding used in 5G systems. Using 3GPP-defined channel models, we demonstrate that DeepRx outperforms traditional methods. We also show that the high performance can likely be attributed to DeepRx learning to utilize the known constellation points of the unknown data symbols, together with the local symbol distribution, for improved detection accuracy.

Inband full-duplex (IBFD) technology enables radios to simultaneously transmit and receive (STAR) on the same frequencies with the benefit of, e.g., enhanced spectral efficiency in non-military communications. In addition, there is significant potential in the IBFD concept in military applications as currently conventional time-or frequency-division half-duplex radios are used in all military applications. A military full-duplex radio (MFDR) would be capable of simultaneous integrated tactical communication and electronic warfare operations. This paper presents an application where an MFDR enables the user to successfully detect an electronic attack, i.e., jamming from an adversary, while simultaneously transmitting tactical transmissions to an ally on the same frequency channel. Successful detection enables the MFDR to gather intelligence and take countermeasures against the jamming, e.g., switching to a different carrier frequency. The experimental results reported herein prove that the radio is able to reliably detect the presence of jamming for received jamming signal powers down to -95 dBm while simultaneously transmitting to an ally at 10-dBm power level. Therefore, the full-duplex radio can give armed forces a significant technical lead over an enemy by detecting enemy jamming even when the adversary only transmits jamming during friendly transmissions.

As low Earth orbit (LEO) satellite communication systems are gaining increasing popularity, new theoretical methodologies are required to investigate such networks' performance at large. This is because deterministic and location-based models that have previously been applied to analyze satellite systems are typically restricted to support simulations only. In this paper, we derive analytical expressions for the downlink coverage probability and average data rate of generic LEO networks, regardless of the actual satellites' locality and their service area geometry. Our solution stems from stochastic geometry, which abstracts the generic networks into uniform binomial point processes. Applying the proposed model, we then study the performance of the networks as a function of key constellation design parameters. Finally, to fit the theoretical modeling more precisely to real deterministic constellations, we introduce the effective number of satellites as a parameter to compensate for the practical uneven distribution of satellites on different latitudes. In addition to deriving exact network performance metrics, the study reveals several guidelines for selecting the design parameters for future massive LEO constellations, e.g., the number of frequency channels and altitude.

We analyze an inband full-duplex access node that is serving mobile users while simultaneously connecting to a core network over a wireless backhaul link, utilizing the same frequency band for all communication tasks. Such wireless self-backhauling is an intriguing option for the next generation wireless systems since a wired backhaul connection might not be economically viable if the access nodes are deployed densely. In particular, we derive the optimal transmit power allocation for such a system in closed form under quality-of-service (QoS) requirements, which are defined in terms of the minimum data rates for each mobile user. For comparison, the optimal transmit power allocation is solved also for two reference scenarios: a purely half-duplex access node, and a relay-type full-duplex access node. Based on the obtained expressions for the optimal transmit powers, we then show that the systems utilizing a full-duplex capable access node have a fundamental feasibility boundary, meaning that there are circumstances under which the QoS requirements cannot be fulfilled using finite transmit powers. This fundamental feasibility boundary is also derived in closed form. The feasibility boundaries and optimal transmit powers are then numerically evaluated in order to compare the different communication schemes. In general, utilizing the purely full-duplex access node results in the lowest transmit powers for all the communicating parties, although there are some network geometries under which such a system is not capable of reaching the required minimum data rates. In addition, the numerical results indicate that a full-duplex capable access node is best suited for relatively small cells.

In this paper, a self-backhauling radio access system is studied and analyzed. In particular, we consider a scenario where a full-duplex access node is serving mobile users simultaneously in uplink and downlink, while also maintaining a wireless backhaul connection. The full-duplex capability of the access node, together with large antenna arrays, allows it to do all of this using the same center frequency. The minimum transmit powers for such a system are solved in a closed form under the condition that certain Quality of Service (QoS) requirements, defined in terms of minimum uplink and downlink data rates, are fulfilled. It is demonstrated with numerical results that, by using the derived expressions for the optimal transmit powers, the probability of fulfilling the QoS requirements is greatly increased, while simultaneously the overall transmit power usage of the system is significantly reduced when compared to a benchmark scheme.

This article investigates novel adaptive self-interference cancellation solutions and the total integrated cancellation performance of a mobile single-antenna inband full-duplex transceiver. First, novel self-adaptive digital self-interference cancellation algorithms are described, with an emphasis on tracking of time-varying self-interference coupling channel in a mobile device as well as on structural ability to suppress also nonlinear self-interference with highly nonlinear mobile power amplifiers. This leads to an advanced self-adaptive nonlinear digital canceller which utilizes a novel orthogonalization procedure for nonlinear basis functions, together with low-cost LMS-based parameter learning. The achievable self-interference cancellation performance is then evaluated with actual RF measurements using mobile device scale RF components, in particular a highly nonlinear PA. The measurements also incorporate a novel self-adaptive RF cancellation circuit in order to realistically assess the total integrated cancellation performance. The reported results show that highly efficient self-interference cancellation can be achieved also in a mobile device, despite a heavily nonlinear PA and limited computing and hardware resources. The proposed cancellation solutions, when integrated together, show that 100 dB of self-interference can be cancelled using a 20 MHz LTE waveform, while the SI can be attenuated by over 110 dB with a narrower bandwidth of 1.4 MHz, all measured at 2.4 GHz ISM band. Furthermore, these results are achieved using a highly nonlinear transmitter power amplifier and fully adaptive canceller structures which can track a rapidly changing coupling channel in a mobile full-duplex device.

This paper presents a novel RF circuit architecture for self-interference cancellation in inband full-duplex radio transceivers . The developed canceller is able to provide wideband cancellation with waveform bandwidths in the order of 100 MHz or beyond and contains also self-adaptive or self-healing features enabling automatic tracking of time-varying self-interference channel characteristics. In addition to architecture and operating principle descriptions, we also provide actual RF measurements at 2.4 GHz ISM band demonstrating the achievable cancellation levels with different bandwidths and when operating in different antenna configurations and under low-cost highly nonlinear power amplifier. In a very challenging example with a 100 MHz waveform bandwidth, around 41 dB total cancellation is obtained while the corresponding cancellation figure is close to 60 dB with the more conventional 20 MHz carrier bandwidth. Also, efficient tracking in time-varying reflection scenarios is demonstrated.

This paper investigates the achievable throughput of a multi-antenna two-hop relay link under hybrid full/half-duplex operation. The analysis is facilitated by realistic waveform simulations, which explicitly model all the essential circuit impairments occurring in the relay transceiver together with degrading channel estimation and self-interference cancellation. The obtained results indicate that pure full-duplex operation does not ensure optimal performance but additional half-duplex transmission periods are usually needed to maximize the end-to-end throughput. Especially, it is shown that the estimation of the self-interference channel within the relay should be performed when the source is not transmitting anything while also the source should be allowed to transmit alone to avoid making the first hop a bottleneck. These findings form a solid basis for optimizing the full-duplex MIMO relay deployments in future mobile networks.

This paper addresses the modeling and cancellation of self-interference in full-duplex direct-conversion radio transceivers, operating under practical imperfect radio frequency (RF) components. First, detailed self-interference signal modeling is carried out, taking into account the most important RF imperfections, namely, transmitter power amplifier nonlinear distortion as well as transmitter and receiver IQ mixer amplitude and phase imbalances. The analysis shows that after realistic antenna isolation and RF cancellation, the dominant self-interference waveform at the receiver digital baseband can be modeled through a widely linear transformation of the original transmit data, opposed to classical purely linear models. Such widely linear self-interference waveform is physically stemming from the transmitter and receiver IQ imaging and cannot be efficiently suppressed by classical linear digital cancellation. Motivated by this, novel widely linear digital self-interference cancellation processing is then proposed and formulated, combined with efficient parameter estimation methods. Extensive simulation results demonstrate that the proposed widely linear cancellation processing clearly outperforms the existing linear solutions, hence enabling the use of practical low-cost RF front ends utilizing IQ mixing in full-duplex transceivers.

Despite the intensive recent research on wireless single-channel full-duplex communications, relatively little is known about the transceiver chain nonidealities of full-duplex devices. In this paper, the effect of nonlinear distortion occurring in the transmitter power amplifier (PA) and the receiver chain is analyzed, beside the dynamic range requirements of analog-to-digital converters (ADCs). This is done with detailed system calculations, which combine the properties of the individual electronics components to jointly model the complete transceiver chain, including self-interference cancellation. They also quantify the decrease in the dynamic range for the signal of interest caused by self-interference at the analog-to-digital interface. Using these system calculations, we provide comprehensive numerical results for typical transceiver parameters. The analytical results are also confirmed with full waveform simulations. We observe that the nonlinear distortion produced by the transmitter PA is a significant issue in a full-duplex transceiver and, when using cheaper and less linear components, also the receiver chain nonlinearities become considerable. It is also shown that, with digitally intensive self-interference cancellation, the quantization noise of the ADCs is another significant problem.

This paper addresses the analysis of oscillator phase-noise effects on the self-interference cancellation capability of full-duplex direct-conversion radio transceivers. Closed-form solutions are derived for the power of the residual self-interference stemming from phase noise in two alternative cases of having either independent oscillators or the same oscillator at the transmitter and receiver chains of the full-duplex transceiver. The results show that phase noise has a severe effect on self-interference cancellation in both of the considered cases, and that by using the common oscillator in upconversion and downconversion results in clearly lower residual self-interference levels. The results also show that it is in general vital to use high quality oscillators in full-duplex transceivers, or have some means for phase noise estimation and mitigation in order to suppress its effects. One of the main findings is that in practical scenarios the subcarrier-wise phase-noise spread of the multipath components of the self-interference channel causes most of the residual phase-noise effect when high amounts of self-interference cancellation is desired.