This paper introduces a novel deep learning-based user-side feedback reduction framework, termedself-nomination. The goal of self-nomination is to reduce the number of users (UEs) feeding back channel state information (CSI) to the base station (BS), by letting each UE decide whether to feed back based on its estimated likelihood of being scheduled and its potential contribution to precoding in a multiuser MIMO (MU-MIMO) downlink. Unlike SNR- or SINR-based thresholding methods, the proposed approach uses rich spatial channel statistics and learns nontrivial correlation effects that affect eventual MU-MIMO scheduling decisions. To train the self-nomination network under an average feedback constraint, we propose two different strategies: one based on direct optimization with gradient approximations, and another using policy gradient-based optimization with a stochastic Bernoulli policy to handle non-differentiable scheduling. The framework also supports proportional-fair scheduling by incorporating dynamic user weights. Numerical results confirm that the proposed self-nomination method significantly reduces CSI feedback overhead. Compared to baseline feedback methods, self-nomination can reduce feedback by as much as 65%, saving not only bandwidth but also allowing many UEs to avoid feedback altogether (and thus, potentially enter a sleep mode). Self-nomination achieves this significant savings with negligible reduction in sum-rate or fairness.

This paper proposes a novel mixed panel architecture for channel state information (CSI) acquisition from uplink (UL) channel training in hybrid analog-digital beamforming systems with partially-connected structures in time-division duplex massive multiple-input multiple-output networks. The proposed architecture combines a few fully-digital (FD) panels with a large number of subarray-based panels for UL pilot reception, addressing CSI acquisition challenges while balancing performance and power efficiency. We then develop a unified method that can utilize measurements from both panel types to estimate the required CSI for analog precoder design. In particular, by recognizing that the dominant eigenvector of the panel covariance matrix is crucial for analog precoding, we propose an orthogonal matching pursuit-type algorithm to estimate it by exploiting channel sparsity in the angular domain. Additionally, we introduce a data-driven technique to optimize analog combiners for subarray-based panels during UL pilot reception. Numerical experiments demonstrate that our proposed method approaches the performance of an all-FD-panel architecture for UL pilot training while maintaining the low complexity of all-subarray-based-panel structures

This paper proposes and analyzes novel deep learning methods for downlink (DL) single-user multiple-input multiple-output (MIMO) and multi-user MIMO (MU-MIMO) systems operating in time division duplex mode. A motivating application is the 6G upper midbands (7-24 GHz), where the base station (BS) antenna arrays are large, user equipment array sizes are moderate, and theoretically optimal approaches are practically infeasible for several reasons. To deal with uplink (UL) pilot overhead and low signal power issues, we introduce the channel-adaptive pilot, as part of the novel analog channel state information feedback mechanism. Deep neural network (DNN)-generated pilots are used to linearly transform the UL channel matrix into lower-dimensional latent vectors. Meanwhile, the BS employs a second DNN that processes the received UL pilots to directly generate near-optimal DL precoders. The training is end-to-end which exploits synergies between the two DNNs. For MU-MIMO precoding, we propose a DNN structure inspired by theoretically optimum linear precoding. The proposed methods are evaluated against genie-aided upper bounds and conventional approaches, using realistic upper midband datasets. Numerical results demonstrate the potential of our approach to achieve significantly increased sum-rate, particularly at moderate to high signal-to-noise ratio and when UL pilot overhead is constrained.

This paper proposes a deep learning approach to a class of active sensing problems in wireless communications in which an agent sequentially interacts with an environment over a predetermined number of time frames to gather information in order to perform a sensing or actuation task for maximizing some utility function. In such an active learning setting, the agent needs to design an adaptive sensing strategy sequentially based on the observations made so far. To tackle such a challenging problem in which the dimension of historical observations increases over time, we propose to use a long short-term memory (LSTM) network to exploit the temporal correlations in the sequence of observations and to map each observation to a fixed-size state information vector. We then use a deep neural network (DNN) to map the LSTM state at each time frame to the design of the next measurement step. Finally, we employ another DNN to map the final LSTM state to the desired solution. We investigate the performance of the proposed framework for adaptive channel sensing problems in wireless communications. In particular, we consider the adaptive beamforming problem for mmWave beam alignment and the adaptive reconfigurable intelligent surface sensing problem for reflection alignment. Numerical results demonstrate that the proposed deep active sensing strategy outperforms the existing adaptive or nonadaptive sensing schemes.

This paper considers a massive random access problem in which a large number of sporadically active devices wish to communicate with a base station (BS) equipped with massive multiple-input multiple-output (MIMO) antennas. Each device is preassigned a unique signature sequence, and the BS identifies the active devices by detecting which sequences are transmitted. This device activity detection problem can be formulated as a maximum likelihood estimation (MLE) problem for which the sample covariance matrix of the received signal is a sufficient statistic. The goal of this paper is to characterize the feasible set of problem parameters under which this covariance based approach is able to successfully recover the device activities in the massive MIMO regime. Through an analysis of the asymptotic behaviors of MLE via its associated Fisher information matrix, this paper derives a necessary and sufficient condition on the Fisher information matrix to ensure a vanishing probability of detection error as the number of antennas goes to infinity, based on which a numerical phase transition analysis is obtained. This condition is also examined from a perspective of covariance matching, which relates the phase transition analysis to a recently derived scaling law. Further, we provide a characterization of the distribution of the estimation error in MLE, based on which the error probabilities in device activity detection can be accurately predicted. Finally, this paper studies a random access scheme with joint device activity and data detection and analyzes its performance in a similar way.

This paper proposes a deep learning approach to channel sensing and downlink hybrid beamforming for massive multiple-input multiple-output systems operating in the time division duplex mode and employing either single-carrier or multicarrier transmission. The conventional precoding design involves a two-step process of first estimating the high-dimensional channel, then designing the precoders based on such estimate. This two-step process is, however, not necessarily optimal. This paper shows that by using a learning approach to design the analog sensing and the hybrid downlink precoders directly from the received pilots without the intermediate high-dimensional channel estimation, the overall system performance can be significantly improved. Training a neural network to design the analog and digital precoders simultaneously is, however, difficult. Further, such an approach is not generalizable to systems with different number of users. In this paper, we develop a simplified and generalizable approach that learns the uplink sensing matrix and downlink analog precoder using a deep neural network that decomposes on a per-user basis, then designs the digital precoder based on the estimated low-dimensional equivalent channel. Numerical comparisons show that the proposed methodology results in significantly less training overhead and leads to an architecture that generalizes to various system settings.

This paper proposes a deep learning approach to the adaptive and sequential beamforming design problem for the initial access phase in a mmWave environment with a single-path channel model. In particular, for a single-user scenario where the problem is equivalent to designing the sequence of sensing beamformers to learn the angle of arrival (AoA) of the dominant path, we propose a novel deep neural network (DNN) that designs a sequence of adaptive sensing vectors based on the available information so far at the base station (BS). By recognizing that the posterior distribution of the AoA provides sufficient statistic for solving the initial access problem, we consider the AoA posterior distribution as the main component of the input to the proposed DNN for designing the adaptive beamforming strategy. However, computing the AoA posterior distribution can be computationally challenging when the fading coefficient is unknown. To address this issue, this paper proposes to use the minimum mean squared error (MMSE) estimate of the fading coefficient to compute an approximation of the posterior distribution. Numerical results demonstrate that as compared to the existing adaptive beamforming schemes utilizing predesigned hierarchical codebooks, the proposed deep learning-based adaptive beamforming achieves a higher AoA detection performance.

This paper proposes a computationally efficient algorithm to solve the joint data and activity detection problem for massive random access with massive multiple-input multiple-output (MIMO). The BS acquires the active devices and their data by detecting the transmitted preassigned nonorthogonal signature sequences. This paper employs a covariance based approach that formulates the detection problem as a maximum likelihood estimation (MLE) problem. To efficiently solve the problem, this paper designs a novel iterative algorithm with low complexity in the regime where the device activity pattern is sparse - a key feature that existing algorithmic designs have not previously exploited for reducing complexity. Specifically, at each iteration, the proposed algorithm focuses on only a small subset of all potential sequences, namely the active set, which contains a few most likely active sequences (i.e., transmitted sequences by all active devices), and performs the detection for the sequences in the active set. The active set is carefully selected at each iteration based on the current detection result and the first-order optimality condition of the MLE problem. Simulation results show that the proposed active set algorithm enjoys significantly better computational efficiency (in terms of the CPU time) than the state-of-the-art algorithms.

This paper proposes a deep learning approach to the adaptive and sequential beamforming design problem for the initial access phase in a mmWave environment with a single-path channel. For a single-user scenario where the problem is equivalent to designing the sequence of sensing beamformers to learn the angle of arrival (AoA) of the dominant path, we propose a novel deep neural network (DNN) that designs the adaptive sensing vectors sequentially based on the available information so far at the base station (BS). By recognizing that the AoA posterior distribution is a sufficient statistic for solving the initial access problem, we use the posterior distribution as the input to the proposed DNN for designing the adaptive sensing strategy. However, computing the posterior distribution can be computationally challenging when the channel fading coefficient is unknown. To address this issue, this paper proposes to use an estimate of the fading coefficient to compute an approximation of the posterior distribution. Further, this paper shows that the proposed DNN can deal with practical beamforming constraints such as the constant modulus constraint. Numerical results demonstrate that compared to the existing adaptive and non-adaptive beamforming schemes, the proposed DNN-based adaptive sensing strategy achieves a significantly better AoA acquisition performance.

This paper shows that deep neural network (DNN) can be used for efficient and distributed channel estimation, quantization, feedback, and downlink multiuser precoding for a frequency-division duplex massive multiple-input multiple-output system in which a base station (BS) serves multiple mobile users, but with rate-limited feedback from the users to the BS. A key observation is that the multiuser channel estimation and feedback problem can be thought of as a distributed source coding problem. In contrast to the traditional approach where the channel state information (CSI) is estimated and quantized at each user independently, this paper shows that a joint design of pilots and a new DNN architecture, which maps the received pilots directly into feedback bits at the user side then maps the feedback bits from all the users directly into the precoding matrix at the BS, can significantly improve the overall performance. This paper further proposes robust design strategies with respect to channel parameters and also a generalizable DNN architecture for varying number of users and number of feedback bits. Numerical results show that the DNN-based approach with short pilot sequences and very limited feedback overhead can already approach the performance of conventional linear precoding schemes with full CSI.

In this paper, we study the channel estimation problem in microwave correlated massive multiple-input-multiple-output systems with reduced number of radio-frequency chains. We exploit the knowledge of the transmit and receive correlation between the antennas. Leveraging the fact that the channel entries are uncorrelated in its eigen-domain, we seek to estimate the channel in this domain. Due to reduced number of radio-frequency chains, channel estimation is performed in multiple time slots. Under a total energy budget, we aim to optimally design the hybrid precoder and combiner in each training time slot, in order to estimate the channel using the minimum mean squared error criterion. We show that the optimal precoder and combiner in each time slot are aligned to transmitter and receiver eigen-directions, respectively. The energy allocation of each eigen-direction determines the significance of each eigen-direction; more energy is allocated to the stronger eigen-directions. At low training energy budget, only significant part of the channel needs to be estimated. At high training energy budget, the energy is equally distributed among all eigen-directions. Simulation results show that the proposed channel estimation scheme can efficiently estimate correlated massive multiple-input-multiple-output channels within a few training time slots.

This paper proposes an autoencoder-based symbol-level precoding (SLP) scheme for a massive multiple-input multiple-output (MIMO) system operating in a limited-scattering environment. By recognizing that only imperfect channel state information (CSI) is available in practice, the goal of the proposed approach is to design the down-link SLP system robust to such imperfect CSI. Toward this goal, this paper leverages the concept of autoencoder wherein the end-to-end communications system is modeled by a deep neural network. By end-to-end training the proposed autoencoder, this paper shows that the downlink symbol-level precoder as well as the receivers' decision rule can be jointly designed in ways that are robust to channel uncertainty. Moreover, this paper introduces a novel two-step training procedure to design a robust precoding scheme for conventional modulations such as quadrature amplitude modulation (QAM) and phase shift keying (PSK). Numerical results indicate that the proposed autoencoder-based framework, either trained by the end-to-end approach in which the receive constellation is a design variable or by the proposed two-step training approach with QAM constellation, can efficiently design a SLP scheme for massive MIMO system which is robust to channel uncertainty.

This paper considers a grant-free random access scenario for massive machine-type communications (mMTC) in which the devices are sporadically active with small payloads. Each active device transmits the identification information as well as the data symbol by selecting a sequence from a pre-assigned sequence set, and the base-station (BS) detects both the device activity and the data by detecting which sequences are transmitted. This paper makes an observation that in the massive multiple-input multiple-output (MIMO) regime, where the BS is equipped with a large number of antennas, a covariance based detection scheme that solves a maximum likelihood estimation problem is more effective than the approximate message passing (AMP) based compressed sensing approach for sequence detection. A main contribution of this paper is an analytic framework capable of accurately predicting the performance of the proposed scheme in terms of the probabilities of false alarm and missed detection. The analysis is based on the asymptotic properties of the maximum likelihood estimator under a nonstandard condition. Simulation results validate the analysis, and demonstrate that as compared to the AMP based approach, the covariance based approach achieves lower error probabilities, especially when the sequence length is short, as is often the case for low-latency mMTC.

This paper tackles the problem of millimeter-wave (mmWave) channel estimation in massive MIMO communication systems. A new Bayes-optimal channel estimator is derived using recent advances in the approximate belief propagation Bayesian inference paradigm. By leveraging the inherent sparsity of the mmWave MIMO channel in the angular domain, we recast the underlying channel estimation problem into that of reconstructing a compressible signal from a set of noisy linear measurements. Then, the generalized approximate message passing (GAMP) algorithm is used to find the entries of the unknown mmWave MIMO channel matrix. Unlike all the existing works on the same topic, we model the angular-domain channel coefficients by Laplacian distributed random variables. Furthermore, we establish the closed-form expressions for the various statistical quantities that need to be updated iteratively by GAMP. To render the proposed algorithm fully automated, we also develop an expectation-maximization (EM) based procedure that can be easily embedded within GAMP's iteration loop in order to learn all the unknown parameters of the underlying Bayesian inference problem. The computer simulations show that the proposed combined EM-GAMP algorithm under a Laplacian prior exhibits improvements both in terms of channel estimation accuracy, achievable rate, and computational complexity, as compared to the Gaussian mixture prior that has been advocated in the recent literature. In addition, it is found that the Laplacian prior speeds up the convergence time of GAMP over the entire signal-to-noise ratio range.

This paper considers sparse device activity detection for cellular machine-type communications with non-orthogonal signatures using the approximate message passing algorithm. This paper compares two network architectures, massive multiple-input-multiple-output (MIMO) and cooperative MIMO, in terms of their effectiveness in overcoming inter-cell interference. In the massive MIMO architecture, each base station (BS) detects only the users from its own cell while treating inter-cell interference as noise. In the cooperative MIMO architecture, each BS detects the users from neighboring cells as well; the detection results are then forwarded in the form of a log-likelihood ratio (LLR) to a central unit where final decisions are made. This paper analytically characterizes the probabilities of false alarm and missed detection for both architectures. The numerical results validate the analytic characterization and show that as the number of antennas increases, a massive MIMO system effectively drives the detection error to zero, while as the cooperation size increases, the cooperative MIMO architecture mainly improves the cell-edge user performance. Moreover, this paper studies the effect of LLR quantization to account for the finite-capacity fronthaul. The numerical simulations of a practical scenario suggest that in specific case cooperating three BSs in a cooperative MIMO system achieves about the same cell-edge detection reliability as a non-cooperative massive MIMO system with four times the number of antennas per BS.

Sparse device activity detection for machine-type communications has attracted increasing attention in recent studies. However, most of the previous works focus on the single-cell case. This paper studies the impact of the inter-cell interference on the device activity detection problem with non-orthogonal signatures in multi-cell systems by employing the computationally efficient approximate message passing algorithm (AMP). Specifically, this paper studies the impact of the inter-cell interference by either treating it as noise or recovering it, showing that it is always beneficial to recover the interference at each base station (BS). Two network architectures, namely BSs with large antenna arrays and network with multi-cell cooperation, are compared in terms of their effectiveness in overcoming inter-cell interference. This paper provides an analytical characterization of probabilities of false alarm and missed detection. Simulation results show that large-scale antenna array is effective in improving the performance of all users whereas cooperation is effective in improving the performance of cell-edge users. In terms of the detection performance of the 95-percentile users, simulation results under a typical network setting show that having twice as many antennas provides almost the same benefit as multi-cell cooperation.

Radio environment maps are geographical maps with overlaid performance information of radio communication systems, that can be considered as one of the key enablers for self-organizing networks. After the introduction of Minimization of Drive Tests (MDT) to the standards, operators are interested to generate radio environment maps based on sparse sets of MDT measurements gathered from user equipment. In this paper, we consider the construction of the Reference Signal Received Power (RSRP) map, which is an example of a radio environment map using such a sparse MDT measurement set from user equipment. In particular, we propose a two-step algorithm for RSRP map generation which is shown to bring substantial performance improvement in terms of the mean absolute error of the predicted RSRP map as compared to the existing baseline methods, i.e., environmental-based regression and inverse distance weighting interpolation.

Hybrid analog and digital beamforming is a promising candidate for large-scale millimeter wave (mmWave) multiple-input multiple-output (MIMO) systems because of its ability to significantly reduce the hardware complexity of the conventional fully digital beamforming schemes while being capable of approaching the performance of fully digital schemes. Most of the prior work on hybrid beamforming considers frequency-flat channels. However, broadband mmWave systems are frequency-selective. In broadband systems, it is desirable to design common analog beamformer for the entire band while employing different digital (baseband) beamformers in different frequency sub-bands. This paper considers the hybrid beamforming design for systems with orthogonal frequency division multiplexing modulation. First, for a single-user MIMO (SU-MIMO) system where the hybrid beamforming architecture is employed at both transmitter and receiver, we show that hybrid beamforming with a small number of radio frequency (RF) chains can asymptotically approach the performance of fully digital beamforming for a sufficiently large number of transceiver antennas due to the sparse nature of the mmWave channels. For systems with a practical number of antennas, we then propose a unified heuristic design for two different hybrid beamforming structures, the fully connected and the partially connected structures, to maximize the overall spectral efficiency of an mmWave MIMO system. Numerical results are provided to show that the proposed algorithm outperforms the existing hybrid beamforming methods, and for the fully connected architecture, the proposed algorithm can achieve spectral efficiency very close to that of the optimal fully digital beamforming but with much fewer RF chains. Second, for the multiuser multiple-input single-output case, we propose a heuristic hybrid percoding design to maximize the weighted sum rate in the downlink and show numerically that the proposed algorithm with practical number of RF chains can already approach the performance of fully digital beamforming.

Large-scale multiple-input multiple-output (MIMO) systems enable high spectral efficiency by employing large antenna arrays at both the transmitter and the receiver of a wireless communication link. In traditional MIMO systems, full digital beamforming is done at the baseband; one distinct radio-frequency (RF) chain is required for each antenna, which for large-scale MIMO systems can be prohibitive from either cost or power consumption point of view. This paper considers a two-stage hybrid beamforming structure to reduce the number of RF chains for large-scale MIMO systems. The overall beamforming matrix consists of analog RF beamforming implemented using phase shifters and baseband digital beamforming of much smaller dimension. This paper considers precoder and receiver design for maximizing the spectral efficiency when the hybrid structure is used at both the transmitter and the receiver. On the theoretical front, bounds on the minimum number of transmit and receive RF chains that are required to realize the theoretical capacity of the large-scale MIMO system are presented. It is shown that the hybrid structure can achieve the same performance as the fully-digital beamforming scheme if the number of RF chains at each end is greater than or equal to twice the number of data streams. On the practical design front, this paper proposes a heuristic hybrid beamforming design strategy for the critical case where the number of RF chains is equal to the number of data streams, and shows that the performance of the proposed hybrid beamforming design can achieve spectral efficiency close to that of the fully-digital solution.

We consider the problem of power allocation for the multiple-input single-output (MISO) downlink with uncertain channel state information at the transmitter. The uncertainty is modeled probabilistically and the receivers specify quality-of-service (QoS) constraints in terms of a target signal-to-interference-and-noise ratio that is to be achieved with a given outage probability. The proposed approach is based on a deterministic characterization of the outage probability, and mildly conservative approximations thereof. Although the resulting optimization problems are not convex, the good solutions that we obtain using straightforward coordinate update algorithms provide significantly better performance than the existing convex approaches because the approximations are less conservative.