The compression of encoded sources in a large multimodal model (LMM) can be theoretically analyzed within Wyner-Ziv coding to limit communication overhead, as the incorporation of multimodalities at the decoder is highly analogous to combining side information with a source. In this paper, after dividing the coding framework of LMM into physical and semantic levels, we study fundamental information-combining approaches within Wyner-Ziv scheme in terms of performance-complexity tradeoff, namely incorporation of sources 1) at the beginning (for best performance) and 2) at the later layers (for fast inference) of a decoder and examine them in terms of rate-distortion function and semantic efficiency. Precisely, the theoretical closed-form rate-distortion functions of these two methods are derived to understand them better. Then, a novel compression algorithm based on indexing is developed to investigate the resultant semantic similarity of these combining approaches. The results indicate that the compression rate for the fast inference is disturbed less for bad (noisy/low throughput) channels with respect to the best performance case, and the semantic similarity can be moderately preserved under certain circumstances. Additionally, the performance drop is negligible after some compression ratios for both approaches.

Adapting task-oriented communication to the distributed edge artificial intelligence (AI) is a promising paradigm. On the other hand, the proliferation of diverse intelligent applications in edge AI makes employing a specialized, dedicated model for each task infeasible. This paper studies a unified framework that enables loading a single scalable network/tensor graph into memory for multi-task edge AI. The key features of the developed architecture are the injection of parallelism to serve multiple tasks and the control of these tasks via scalar parameters, ensuring dynamic switching between tasks. The structure of the proposed model is grounded in a computational analysis to avoid inference latency. The working principle of the unified framework is explained with a theoretical subspace analysis based on singular value decomposition (SVD) and Grassmann distance. To support joint source-channel coding (JSCC) transmission for the proposed model, we develop a novel adaptability algorithm by adding binary masks. The numerical results demonstrate that the proposed unified framework achieves nearly the same performance as a dedicated model per task, while offering significant computational complexity savings and improved memory and storage efficiency. Additionally, the immense superiority of our framework is showcased over the simple approach of devising a backbone with multiple heads.

This paper presents a novel compressed sensing (CS) approach to high dimensional wireless channel estimation by optimizing the input to a deep generative network. Channel estimation using generative networks relies on the assumption that the reconstructed channel lies in the range of a generative model. Channel reconstruction using generative priors outperforms conventional CS techniques and requires fewer pilots. It also eliminates the need of a priori knowledge of the sparsifying basis, instead using the structure captured by the deep generative model as a prior. Using this prior, we also perform channel estimation from one-bit quantized pilot measurements, and propose a novel optimization objective function that attempts to maximize the correlation between the received signal and the generator's channel estimate while minimizing the rank of the channel estimate. Our approach significantly outperforms sparse signal recovery methods such as Orthogonal Matching Pursuit (OMP) and Approximate Message Passing (AMP) algorithms such as EM-GM-AMP for narrowband mmWave channel reconstruction, and its execution time is not noticeably affected by the increase in the number of received pilot symbols.

Communication at high carrier frequencies such as millimeter wave (mmWave) and terahertz (THz) requires channel estimation for very large bandwidths at low SNR. Hence, allocating an orthogonal pilot tone for each coherence bandwidth leads to excessive number of pilots. We leverage generative adversarial networks (GANs) to accurately estimate frequency selective channels with few pilots at low SNR. The proposed estimator first learns to produce channel samples from the true but unknown channel distribution via training the generative network, and then uses this trained network as a prior to estimate the current channel by optimizing the network's input vector in light of the current received signal. Our results show that at an SNR of -5 dB, even if a transceiver with one-bit phase shifters is employed, our design achieves the same channel estimation error as an LS estimator with SNR = 20 dB or the LMMSE estimator at 2.5 dB, both with fully digital architectures. Additionally, the GAN-based estimator reduces the required number of pilots by about 70% without significantly increasing the estimation error and required SNR. We also show that the generative network does not appear to require retraining even if the number of clusters and rays change considerably.

Optimizing a hybrid beamforming transmitter is a non-convex problem and requires channel state information, leading in most cases to nontrivial feedback overhead. We propose a methodology relying on the principles of deep generative models and unfolding to achieve near-optimal hybrid beamforming with reduced feedback and computational complexity. We first represent the channel as a low-dimensional manifold via a generative adversarial network (GAN) and search the optimum digital and analog precoders in this low-dimensional space. To decrease the search complexity, we find an iteration rule by formulating hybrid beamforming as a bi-level optimization problem and then unfold each iteration as a neural layer. This results in a novel model-based deep neural network that incorporates domain knowledge. Our results show that this method (i) approaches the capacity-achieving spectral efficiency, (ii) provides a superior energy and spectral efficiency tradeoff, (iii) decreases feedback overhead, and (iv) reduces the complexity significantly, by optimizing a single low-dimensional vector per channel coherence time, with the neural network itself trained offline. The achieved spectral efficiency is robust when tested with realistic 3GPP channel models, even if the offline training relies on a simple geometric channel model.

System level simulations of large 5G networks are essential to evaluate and design algorithms related to network issues such as scheduling, mobility management, interference management, and cell planning. In this paper, we look back to the idea of spatial indexing and its advantages, applications, and future potentials in accelerating large 5G network simulations. We introduce a multi-level inheritance based architecture which is used to index all elements of a heterogeneous network (HetNet) on a single geometry tree. Then, we define spatial queries to accelerate searches in distance, azimuth, and elevation. We demonstrate that spatial indexing can accelerate location-based searches by 3 orders of magnitude. Further, the proposed design is implemented as an open source platform freely available to all.

This paper proposes a novel deep learning-based error correction coding scheme for AWGN channels under the constraint of one-bit quantization in receivers. Specifically, it is first shown that the optimum error correction code that minimizes the probability of bit error can be obtained by perfectly training a special autoencoder, in which “perfectly” refers to converging the global minima. However, perfect training is not possible in most cases. To approach the performance of a perfectly trained autoencoder with a suboptimum training, we propose utilizing turbo codes as an implicit regularization, i.e., using a concatenation of a turbo code and an autoencoder. It is empirically shown that this design gives nearly the same performance as to the hypothetically perfectly trained autoencoder, and we also provide a theoretical proof of why that is so. The proposed coding method is as bandwidth efficient as the integrated (outer) turbo code, since the autoencoder exploits the excess bandwidth from pulse shaping and packs signals more intelligently thanks to sparsity in neural networks. Our results show that the proposed coding scheme at finite block lengths outperforms conventional turbo codes even for QPSK modulation. Furthermore, the proposed coding method can make one-bit quantization operational even for 16-QAM.

This paper proposes a deep learning-based channel estimation method for multi-cell interference-limited massive MIMO systems, in which base stations equipped with a large number of antennas serve multiple single-antenna users. The proposed estimator employs a specially designed deep neural network (DNN) based on the deep image prior (DIP) network to first denoise the received signal, followed by conventional least-squares (LS) estimation. We analytically prove that our LS-type deep channel estimator can approach minimum mean square error (MMSE) estimator performance for high-dimensional signals, while avoiding complex channel inversions and knowledge of the channel covariance matrix. This analytical result, while asymptotic, is observed in simulations to be operational for just 64 antennas and 64 subcarriers per OFDM symbol. The proposed method also does not require any training and utilizes several orders of magnitude fewer parameters than conventional DNNs. The proposed deep channel estimator is also robust to pilot contamination and can even completely eliminate it under certain conditions.

The interactions between the cells, most notably due to their coupled interference and the large number of users, render the optimization of antenna parameters prohibitively complex. To cope with this problem, we propose a novel practical deep learning (DL) based reinforcement learning (RL) algorithm to jointly optimize antenna tilt angle and vertical and horizontal half-power beamwidths of the macrocells in a heterogeneous cellular network (HetNet). In the proposed algorithm, DL is used to extract the features by learning the locations of users, and mean field RL is used to learn the average interference values for different antenna settings. Our results illustrate that the proposed deep RL algorithm can approach the optimum weighted sum rate with hundreds of online trials, as opposed to millions of trials for standard Q-learning, assuming relatively low environmental dynamics. Furthermore, the proposed algorithm is compact and implementable, and empirically appears to provide a performance guarantee regardless of the amount of environmental dynamics.

This paper proposes a deep learning-based error correction coding for AWGN channels under the constraint of one-bit quantization in receivers. An autoencoder is designed and integrated with a turbo code that acts as an implicit regularization. This implicit regularizer facilitates approaching the Shannon bound for the one-bit quantized AWGN channels even if the autoencoder is trained suboptimally, since one-bit quantization stymies ideal training. Our empirical results show that the proposed coding scheme gives better results at finite block lengths than conventional turbo codes even for QPSK modulation, which can achieve the Shannon bound at infinite block length despite one-bit quantization. Furthermore, the proposed coding method makes one- bit quantization operational even for 16-QAM, which is unprecedented.

This paper develops novel deep learning-based architectures and design methodologies for an orthogonal frequency division multiplexing (OFDM) receiver under the constraint of one-bit complex quantization. Single bit quantization reduces greatly the complexity and power consumption but makes accurate channel estimation and data detection difficult. This is particularly true for multicarrier waveforms that have high peak-to-average power ratio in the time domain and fragile subcarrier orthogonality in the frequency domain. The severe distortion for one-bit quantization typically results in an error floor even at moderately low signal-to-noise-ratio (SNR) such as 5 dB. For channel estimation (using pilots), we design a novel generative supervised deep neural network that can be trained with a reasonable number of pilots. After channel estimation, a neural network-based receiver-specifically, an autoencoder-jointly learns a precoder and decoder for data symbol detection. Since quantization prevents end-to-end training, we propose a two-step sequential training policy for this model. With synthetic data, our deep learning-based channel estimation can outperform least squares channel estimation for unquantized (full-resolution) OFDM at average SNRs up to 14 dB. For data detection, our proposed design achieves lower bit error rate (BER) in fading than unquantized OFDM at average SNRs up to 10 dB.

In this paper, a clustering algorithm is proposed that dynamically determines the locations of fog nodes in 5G wireless networks, which are upgraded from small cells, in order to maximize throughput assuming the number of fog nodes and small cells are given as a priori information. The proposed algorithm dynamically clusters the small cells around the fog nodes. The approach is based on a soft clustering model where one small cell can be connected to many fog nodes. The numerical results demonstrate that the proposed clustering algorithm significantly enhances the throughput and lowers the latency with respect to the distance-based K- means hard clustering algorithm or Voronoi tessellation model.

This paper proposes a new medium access control (MAC) protocol for Internet of Things (IoT) applications incorporating pure ALOHA with power domain non-orthogonal multiple access (NOMA) in which the number of transmitters are not known as a priori information and estimated with multihypothesis testing. The proposed protocol referred to as ALOHA-NOMA is not only scalable, energy efficient and matched to the low complexity requirements of IoT devices, but it also significantly increases the throughput. Specifically, throughput is increased to 1.27 with ALOHA-NOMA when 5 users can be separated via a SIC (Successive Interference Cancellation) receiver in comparison to the classical result of 0.18 in pure ALOHA. The results further show that there is a greater than linear increase in throughput as the number of active IoT devices increases.

Non-orthogonal multiple access (NOMA) that exploits the differences of transmission powers of users or power domain NOMA is an attractive technology to meet the increasing demands of 5G network users. In this paper, a multiuser diversity gain analysis for power domain NOMA is introduced to compare the reliability of various power transmission methods in case of uplink communications. Accordingly, a power transmission scheme that ensures a certain quality of user experience (QoE) for all users is compared with the power back-off transmission scheme inherited from LTE. The results show that the QoE based power transmission can exploit the diversity gain and has much lower symbol error rate than the power back-off method for both ideal successive interference cancellation (SIC) and practical imperfect SIC at the receiver.

The synergistic combination of Diversity and Network Coding (DC-NC) was previously introduced to provide very low end-to-end latency in recovering from a link failure and improve the throughput for a wide variety of network architectures. This paper is directed towards further improving DC-NC to be able to tolerate multiple, simultaneous link failures with less computational complexity. In this way, reliability will be maximized and the recovery time from multiple link or node failures is reduced in 5G fronthaul wireless networks. This is accomplished by modifying Triangular Network Coding (TNC) to create enhanced DC-NC (eDC-NC) that is applied to 5G wireless Fog computing-based Radio Access Networks (Fog-RAN). Our results show that using eDC-NC coding in Fog-RAN fronthaul network will provide ultra-reliability and enable near-instantaneous fault recovery while retaining the throughput improvement feature of DC-NC. In addition, the scalability of eDC-NC coding is demonstrated. Furthermore, it is shown that the redundancy percentage for complete protection is always less than 50% for the practical cases that were evaluated.

This paper incorporates fog networking into heterogeneous cellular networks that are composed of a high power node (HPN) and many low power nodes (LPNs). The locations of the fog nodes that are upgraded from LPNs are specified by modifying the unsupervised soft-clustering machine learning algorithm with the ultimate aim of reducing latency. The clusters are constructed accordingly so that the leader of each cluster becomes a fog node. The proposed approach significantly reduces the latency with respect to the simple, but practical, Voronoi tessellation model, however the improvement is bounded and saturates. Hence, closed-loop error control systems will be challenged in meeting the demanding latency requirement of 5G systems, so that open-loop communication may be required to meet the 1ms latency requirement of 5G networks.

This paper is directed towards improving both throughput and reliability of 5G wireless Cloud Radio Access Networks (C-RANs) by the synergistic combination of Diversity Coding and Network Coding (DC-NC). In this paper, we directly apply the concept of DC-NC coding to two network scenarios: first, remote radio heads in a CRAN connected to the baseband unit in two hierarchical tiers with optical and wireless fronthaul links, second, most remote radio heads are connected directly to the baseband unit via wireless links. Our results show that the combination of Diversity and Network Coding increases the throughput of fronthaul networks for downlink broadcasting or multicasting applications, while enabling reliable networking with near-instant latency in fault recovery by using forward error control across spatially diverse paths. Moreover, the number of redundant links inherent in Diversity Coding can be decreased using the proposed scheme.

Faster than symbol rate (FTSR) sampling is necessary once practical non-adaptive analog front-end filter and excess bandwidth are employed in case of unequally spaced channel taps. Based on this a novel FTSR sampled mismatched minimum mean square error (MMSE) time domain equalization (TDE) is proposed in response to cyclic prefix (CP) based block transmissions. Our results show that the proposed MMSE TDE implementation nearly introduces 3dB signal-to-noise ratio (SNR) advantage with respect to conventional symbol rate (SR) sampled CP based MMSE single carrier frequency domain equalization (SC-FDE) for an error rate of 10^-3 and 16 quadrature amplitude modulation (QAM).

Nanoscale communications is an appealing domain in nanotechnology. Novel nanoscale communications techniques are currently being devised inspired by some naturally existing phenomena such as the molecular communications governing cellular signaling mechanisms. Among these, neuro-spike communications, which governs the communications between neurons, is a vastly unexplored area. The ultimate goal of this paper is to accurately investigate nanoscale neuro-spike communications characteristics through the development of a realistic physical channel model between two neurons. The neuro-spike communications channel is analyzed based on the probability of error and delay in spike detection at the output. The derived communication theoretical channel model may help designing novel artificial nanoscale communications methods for the realization of future practical nanonetworks, which are the interconnections of nanomachines.