Big generative models have significantly impacted several domains like natural language processing, computer vision, and drug discovery. These developments, large-scale generative foundation models exemplified by architectures such as GPT-4 have emerged to address a broad spectrum of generalized tasks. These models are trained to capture the underlying general correlations from a large training dataset. Although big generative AI techniques have been explored in various wireless communication applications, such as channel generation, the integration of generalized foundation models into this domain remains limited. A core component of these models is the masked autoencoder. This work investigates the suitability of masked autoencoders for massive multiple-input multiple-output (MIMO) systems, focusing on their capacity to capture spatial and temporal correlations in massive MIMO channels. To this end, a massive MIMO scenario with user mobility is considered, where the channel state information (CSI) varies with both spatial and temporal correlation. A masked autoencoder is trained in a self-supervised manner using channel state information (CSI) from multiple users. The trained model is then tested for its performance in tasks of feedback compression, channel interpolation, and channel prediction. Experimental results demonstrate that masked autoencoders effectively capture inherent correlations within massive MIMO channels, underscoring their potential to advance foundational model-based approaches in wireless communications.

Recent transition in mobile communication standards suggests massive multiple-input multiple-output (MIMO) to be an integral part of the foreseeable future. However, as antenna elements increase to hundreds in the fifth-generation (5G) and beyond, traditional signal processing methods become prone to significant hardware impairments compound from multiple chains, leading to a substantial performance degradation. This paper explores the effectiveness of generative artificial intelligence (AI) techniques in addressing these challenges within massive MIMO systems. For this purpose, the conditional generative adversarial network (CGAN), a special class of generative AI algorithms, is employed to enhance the accuracy of channel state information (CSI) estimation in a hardware-impaired transceiver setup. This problem is treated as an image-denoising task, where the noise is introduced by the hardware impairments and LS estimation error. Through simulations conducted across various antenna array sizes, the potential of generative AI to improve CSI estimation accuracy under hardware impairments is demon-strated. This highlights its capacity to address critical signal processing challenges in the next-generation wireless systems.

Estimation of downlink (DL) channel state information (CSI) is necessary in massive multiple-input multiple-output (MIMO) systems to enable precoding and achieve high spectral efficiency. However, CSI estimation (for both the uplink (UL) and DL) is challenging in an environment with highly-mobile users due to rapidly-varying fading. The estimation becomes even more challenging when UL CSI is incomplete due to system constraints. In this work, we combine two machine learning techniques to tackle the twofold problem of predicting upcoming DL CSI from earlier UL CSI estimates and estimating full UL CSI from its incomplete form. For the first sub-problem, we employ long short-term memory (LSTM) to capture the spatio-temporal correlation between CSI at different time instances and user positions. For the second sub-problem, we use a conditional generative adversarial network (CGAN) to estimate the full UL CSI from varying amounts of incomplete CSI. We examine the normalized mean square error performance of the proposed CGAN-LSTM method and compare the spectral efficiency of the system with what is maximally achievable with complete up-to-date CSI. Furthermore, we extend our machine learning methodology to directly estimate precoding matrices from partial CSI and similarly compare the performance with that achievable using complete up-to-date CSI.

For implementation of massive multiple-input multiple-output (MIMO) cellular systems in frequency division duplex (FDD) mode, accurate estimation of downlink channel state information (CSI) is necessary, but full radio channel reciprocity between the uplink and downlink does not exist in that mode. Existing work on estimating downlink CSI in FDD massive MIMO systems has considered such approaches as angle-of-arrival reciprocity, compressive sensing, using second-order channel statistics (particularly the channel covariance matrix (CCM)), and machine learning using deep neural networks (DNNs). Typical DNN-based approaches are unsuitable for this problem because DNNs require large datasets, thousands of training epochs, and are susceptible to environmental variations. To overcome these shortcomings, we develop a conditional generative adversarial network (CGAN) approach to uplink-to-downlink mapping of both CCMs and CSI. To apply this method, we convert the uplink and downlink CCMs/CSI to images and employ CGAN techniques previously applied to image translation. The normalized mean square error performance of the proposed CGAN is evaluated for several array sizes for both CCM and CSI mapping. For uplink-to-downlink CSI mapping, we also examine the spectral efficiency performance of our CGAN-based method, as well as the impact of pilot reuse; both simulated and measured CSI data are considered. Our results demonstrate performance improvement over existing algorithms.

Conventional massive multiple-input multiple-output (MIMO) systems provide high spectral efficiency, throughput, and energy efficiency, but suffer from high inter-cell interference and poor cell-edge coverage. Cell-free massive MIMO addresses these shortcomings by geographically distributing access points (APs), each with one or more antennas, that form a virtual massive MIMO array instead of co-locating all antennas at a base station. This distributed AP placement significantly reduces the average distance between a user equipment (UE) and an AP. Existing work mainly considers the ideal canonical case where each UE is served by all APs, which is impractical because of limited fronthaul capacity and finite computational resources. Therefore, to make the network scalable to an arbitrary size, a user-centric approach should be adopted, where each UE is served by a personalized cluster of nearby APs. However, the clustering problem is combinatorially complex and may be too time consuming to solve optimally when the UEs are in motion. Therefore, in this work, we develop a multi-agent reinforcement learning (MARL) algorithm for AP selection and clustering, where each AP is an agent in the MARL algorithm and trained to near-optimally select for itself which UEs to serve. Simulation results demonstrate our MARL algorithm outperforms typical AP selection algorithms such as greedy selection as well as more sophisticated ones from the literature, and also provides performance comparable to the ideal canonical case.

Estimating or predicting the downlink channel state information (CSI) is extremely important for practical implementation of frequency division duplex (FDD) massive MIMO. Estimation of downlink CSI from uplink CSI using second order channel statistics, namely the channel covariance matrix (CCM), is a promising approach. However, published work so far has rarely applied machine learning techniques to solve this problem using CCMs, most probably due to the unavailability of a direct mapping function or parametric model for supervised learning to convert from uplink to downlink CCMs. In this paper, we develop a conditional generative adversarial network (CGAN) method for uplink-to-downlink CCM conversion. To apply the CGAN-based method, we convert the uplink and downlink CCMs to images and use image translation techniques for CGANs. The normalized mean square error performance of the proposed CGAN is evaluated for several antenna array sizes and with both perfect and imperfect knowledge of the CCMs. Our results demonstrate performance improvement over existing algorithms.

Most current stochastic geometric modeling of heterogeneous cellular networks (HetNets) assumes independent deployment of small-cell base stations (SBSs) with respect to macro base stations (MBSs), which leads to limited enhancement in network coverage and capacity. Therefore, in this paper we propose a new HetNet model where the locations of SBSs are correlated with those of the MBSs. We place the SBSs at the vertices of each macrocell, where the macrocells are modeled by a Poisson-Voronoi tesellation with the MBSs as seeds. Theoretical analysis of this deployment scheme is studied using the tools of stochastic geometry. A novel distribution is also derived for the distance between the typical user and its closest SBS. One significant advantage of this deployment scheme is that any given SBS can serve users from multiple macrocells, thus offloading the traffic from the MBSs and also improving network coverage. The performance of the proposed model is evaluated for several network parameters and our results concretely demonstrate the improvement in coverage probability compared to other schemes in the literature.

In wireless channels, the received signal to noise ratio (SNR) can be represented as γ = βγ̅, where γ is the average SNR and β is a random variable with probability density function (PDF) f(β). In this paper, we analyze the high SNR performance of wireless channels with a logarithmic singularity. That is, f(β) = aβt+ bβμlog(β) + ··· near β = 0. This logarithmic singularity (LS) is critically important in determining the high SNR performance and appears to have been completely overlooked. Important special cases include Gamma-Gamma and Generalized-K channels. For instance, the GG has been used to model scattering, reflection, and diffraction and optical, navigation and relay channels [1]. This versatility highlights the importance of LS wireless channels. Classical asymptotic or high SNR analysis is developed by expanding f(β) = aβt+ · · · near β = 0 and expressing the diversity and coding gain as direct functions of a and t. However, as this monomial expansion does not hold for LS channels, we develop generalized asymptotic performance measures for outage and error rates. The results show significantly improved accuracy in the SNR range of 1025 dB. For this range, our new asymptotic expressions achieve much better accuracy than the conventional ones that ignore this singularity.

Caching popular multimedia content has the potential to take wireless networking to an unprecedented height in terms of user experience. Primary motif behind content caching is to give frequent access to popular content cached at local caches, such as femto access point with finite storage. Although content-caching was Initially limited to wired backbone networks, it now being developed for wireless networks. The main difference between these two cases is the potential mobility of the user. We thus investigate the impact of user mobility on the performance of content-caching wireless heterogeneous networks (HetNets). We describe the user mobility by the random waypoint model and characterize the spatial randomness of different types of nodes by using independent Poisson point processes. Using their stochastic properties, we analyze the handover probabilities and evaluate expected download delay as a function of handover probabilities.

This paper proposes a new medium access control (MAC) protocol for low power sensor devices, suitable for IoT systems. IEEE 802.15.4 standard is suitable for low power wireless personal area network (WPAN) but it does not satisfy the data rate and reliability requirements for IoT systems in a 5G wireless network. We have observed that unnecessary packet drop takes place due to beacon superframe broadcasting during data transmission and it is the primary reason for the standard's data-rate and reliability shortfall. This problem represents a scenario where data transmission takes place with the lack of available time for data transmission in that superframe duration. To overcome this lacuna, we incorporate backoff freezing mechanism, where the backoff counter freezes whenever the available time for data transmission is insufficient in that superframe duration. A novel sleep protocol is designed to reduce power consumption in idle states too. The proposed MAC protocol is modeled using a 3- dimensional Markov chain for analytical performance evaluation. Analytical results are verified with the simulation run in ns-2.35. Proposed MAC with sleep protocol significantly outperforms the existing state-of-the-art protocols.