In frequency division duplexing systems, downlink massive multiple-input multiple-output (MIMO) precoding algorithms rely on accurate channel state information (CSI) feedback from users. This paper investigates the tradeoff between the CSI feedback overhead and the resulting user performance in terms of achievable sum rate. Our approach consists of determining the precoding directly from the user feedback. We employ a deep learning-based design for an end-to-end precoding-oriented feedback architecture, including learned pilots, user compressors for finite-rate feedback, and base station processing to determine precoding vectors. We propose a novel loss function that maximizes the sum of achievable rates while minimizing the CSI feedback overhead. We consider both single- and multi-cell multi-user MIMO systems, analyzing the impact of intra- and inter-cell interference on the CSI feedback strategy design, as well as robustness. Simulation results demonstrate that our approach outperforms previous precoding-oriented methods and offers greater efficiency than conventional methods that separate CSI compression and precoding.

Industry and academia have been working towards the evolution from Centralized massive Multiple-Input Multiple-Output (CmMIMO) to Distributed mMIMO (DmMIMO) architectures. Instead of splitting a coverage area into many cells, each served by a single Base Station equipped with several antennas, the whole coverage area is jointly covered by several Access Points (AP) equipped with few or single antennas. Nevertheless, when choosing between deploying more APs with few or single antennas or fewer APs equipped with many antennas, one observes an inherent trade-off between the beamforming and macro-diversity gains that has not been investigated in the literature. Given a total number of antenna elements and total downlink power, under a channel model that takes into account a probability of Line-of-Sight (LoS) as a function of the distance between the User Equipments (UEs) and APs, our numerical results show that there exists a “sweet spot” on the optimal number of APs and of antenna elements per AP which is a function of the physical dimensions of the coverage area.

Downlink massive multiple-input multiple-output (MIMO) precoding algorithms in frequency division duplexing (FDD) systems rely on accurate channel state information (CSI) feedback from users. In this paper, we analyze the tradeoff between the CSI feedback overhead and the performance achieved by the users in systems in terms of achievable rate. The final goal of the proposed system is to determine the beamforming information (i.e., precoding) from channel realizations. We employ a deep learning-based approach to design the end-to-end precoding-oriented feedback architecture, that includes learned pilots, users' compressors, and base station processing. We propose a loss function that maximizes the sum of achievable rates with minimal feedback overhead. Simulation results show that our approach outperforms previous precoding-oriented methods, and provides more efficient solutions with respect to conventional methods that separate the CSI compression blocks from the precoding processing.

Delivering Gbps high user rate over long distances (~1 km) is challenging, and the abundant spectrum available in millimeter wave band cannot solve the challenge by its own due to the severe path loss and other limitations. Since it is economically challenging to deploy wired backhaul every few hundred meters, relays (e.g., wireless access points) have been proposed to extend the coverage of a base station, which has wired connection to the core network. These relays, deployed every few hundred meters, serve the users in their vicinity and are backhauled to the base station through wireless connections. In this paper, the wireless-relayed backhaul design has been formulated as a topology-bandwidth-power joint optimization problem, and the influence of path loss, angular spread, array size, and RF power limitation on the user rate has been evaluated. It has been shown that for a linear network deployed along the street at 28 GHz, when high joint directional gain (50 dBi) is available, 1 Gb/s user rate within cell range of 1 km can be delivered using 1.5 GHz of bandwidth (using single polarization antennas). The user rates drop precipitously when joint directional gain is reduced, or when the path loss is much more severe. When the number of RF chains is limited, the benefit of larger arrays will eventually be surpassed by the increased channel estimation penalty as the effective beamforming gain saturates owing to the channel angular spread.

Owing to its low spectral decay rate, conventional orthogonal frequency-division modulation (OFDM) is unsuitable for supporting asynchronous access, user-specific subcarrier spacing and symbol period, and spectrum aggregation, all of which are envisioned for fifth-generation (5G) cellular networks. This has spurred recent work on designing new multi-carrier waveforms for 5G. In this paper, a unified signal model for OFDM is presented, which encompasses variants employing either a cyclic prefix (CP) or a zero postfix (ZP) as the guard interval technique, and either windowing or filtering to increase the spectral decay rate of the transmitted signal; both windowing-based and filtering-based receivers are also accommodated. Sufficient conditions on the channel impulse response and any windowing/filtering are given for the avoidance of inter-symbol and intra-symbol interference. Numerical results are then presented to show that, in a multiuser system operating asynchronously and with user-specific subcarrier spacing and symbol period, filtering and windowing yield substantial and comparable rate gains over standard OFDM when implemented at both transmitter and receiver; these gains can be realized with no performance degradation relative to standard OFDM (at SNRs typically of interest) when the system operates synchronously and all users have identical subcarrier spacing and symbol period. Further, CP-OFDM and ZP-OFDM have near-identical rate-versus-SNR performance.

In a heterogeneous network (HetNet) obtained by overlaying low-power picocells on a conventional macrocell network, two important problems must be addressed in order to obtain significant downlink capacity gains. These are inter-cell load balancing, i.e., determining which cell must serve each user so as to utilize all cells effectively, and inter-cell interference coordination, i.e., determining when and for how long must each macrocell be silenced so as to mitigate interference caused to nearby picocell users. We formulate a centralized optimization framework for these coupled problems, and propose a simple solution to them based on convex relaxation and two applications of the Frank-Wolfe algorithm. Simulation results show that the proposed solution has near-optimal performance in a canonical “hotspot” network model, and can double both the 50th percentile and 5th percentile user rates, when compared to a baseline HetNet with no load balancing and no interference coordination. Further, both interference coordination and load balancing are necessary; each by itself yields much lower gains.

In this paper, we provide an information-theoretic analysis of a Gaussian multiple-input multiple-output multiple access channel (MIMO MAC) with imperfect channel knowledge at the receiver. In particular, we derive inner and outer bounds for the MIMO MAC rate region when the inputs are Gaussian. We then apply these bounds to a Gaussian interference network with receiver cooperation, in which a central processor with incomplete channel state information must jointly decode all the received signals. Then, in the case where the channel knowledge at the receiver is obtained through training signals, we derive the structure of the optimum training signals for all users under a definite and semidefinite rank constraint. Numerical results show that the bounds we derive can be quite tight, confirming the asymptotic analysis conducted for the finite case. Finally, we also investigate the low-SNR and high-SNR regimes, specifically analyzing the minimum required energy per information bit and the wideband slope region in the first case, and the high-SNR slope in the second.

In this work we provide an information-theoretic analysis of a Gaussian multiple-input multiple-output multiple access channel (MIMO MAC) with imperfect channel knowledge at the receiver. In particular we derive inner and outer bounds for the rate region of the MIMO MAC when the inputs are Gaussian. In the case where the channel knowledge at the receiver is obtained through training signals, we derive the structure of the optimum training signals for all users under a definite and semi-definite rank constraint. Numerical results show that the bounds we derive can be quite tight. Finally, we also investigate the low-SNR regime, specifically analyzing the minimum required energy per information bit and the wideband slope region.

To overcome the spectral efficiency limits imposed by cochannel interference on the uplink of a cellular network, we propose coordinating each base station with several of its neighbors in the reception of user signals, and decoding each user at one such cluster of base stations. We evaluate the impact on the user rate distribution due to such coordination when there is 1 user per base station antenna in the network, each user has 1 transmitting antenna, all users transmit at maximum power, and the data rates of the users assigned to each cluster are subject to a proportional fairness criterion. We highlight the dependence of the user rate distribution on the number of rings of neighbors with which each base station is coordinated, as well as the underlying signal-to-noise ratio (SNR) distribution in the network. Our results point to the possibility of increasing cell- edge user rates almost tenfold, while simultaneously doubling the overall system throughput, in high-SNR conditions.

We propose an ambitious approach towards lifting the limits imposed by cochannel interference on the uplink spectral efficiency of a cellular network, viz., coordinating several base stations in the reception of users within their coverage area, and suppressing interference between users by means of coherent linear beamforming across the base stations. We evaluate by simulation the potential gain in spectral efficiency from such coordination, when there is 1 user per base station antenna in the network, and all users (but for a small fraction in outage) must be served at a constant and common data rate. We highlight the dependence of the spectral efficiency gain on the number of rings of neighbors with which each base station is coordinated, as well as the underlying signal-to-noise ratio (SNR) distribution in the network. Results from this study point to the possibility of doubling the uplink spectral efficiency with 1-ring coordination and nearly quadrupling it with 4-ring coordination, under high-SNR conditions.

It has been observed through simulations of some specific scheduling algorithms that multiuser diversity gains in packet data systems with channel-aware scheduling can be reduced in the presence of any form of link diversity, such as transmit antenna diversity or wideband multipath diversity. We establish that asymptotically, in the limit of large number of transmit antennas and users, the maximum throughput achieved by any optimal scheduling algorithm in the presence of transmit diversity under signal-to-noise-ratio-only feedback can be infinitely worse than that of a system with no diversity. Our results are general and are independent of any particular scheduling algorithm.

This paper describes a power, speed and area efficient VLSI implementation of a noise whitening algorithm for a 4x4 MIMO channel. The architecture combines innovative use of Hermitian matrices to streamline the iterative calculations, with a 4x1 matrix row-column multiplier as the core component. The optimisations in the datapath reduce the power and latency needed to implement the algorithm. The Booth recoded complex multipliers use logic sharing to reduce power and complexity, and incorporate low-power sleep logic that does not increase the critical path. The design has been successfully synthesised in a 0.18μm, 1.8V CMOS technology, and has the potential to be adapted to other applications requiring matrix multiplication.

We generalize our previous results on generating common randomness at two terminals to a situation where any finite number of agents, interconnected by an arbitrary network of independent, point-to-point, discrete memoryless channels, wish to generate common randomness by interactive communication over the network. Our main result is an exact characterization of the common randomness capacity of such a network, i.e., the maximum number of bits of randomness that all the agents can agree on per step of communication. As a by-product, we also obtain a purely combinatorial result, viz., a characterization of (the incidence vectors of) the spanning arborescences rooted at a specified vertex in a digraph, and having exactly one edge exiting the root, as precisely the extreme points of a certain unbounded convex polyhedron, described by a system of linear inequalities.

We study the following problem: two agents Alice and Bob are connected to each other by independent discrete memoryless channels. They wish to generate common randomness, i.e. agree on a common random variable, by communicating interactively over the two channels. Assuming that Alice and Bob are allowed access to independent external random sources at rates (in bits per step of communication) of H/sub A/ and H/sub B/, respectively, we show that they can generate common randomness at a rate of max{min[H/sub A/+H(W|Q),I(P;V)]+min[H/sub B/+H(V|P), I(Q;W)]} bits per step, by exploiting the noise on the two channels. Here, V is the channel from Alice to Bob, and W is the channel from Bob to Alice. The maximum is over all probability distributions P and Q on the input alphabets of V and W, respectively. We also prove a strong converse which establishes the above rate as the highest attainable in this situation.

We determine the region of all identification and transmission rate-pairs achievable over a discrete memoryless channel (DMC) with perfect and instantaneous feedback, for both randomized and deterministic encoding. As a by-product, we also have a new proof of Kemperman's (1973) strong converse to Shannon's coding theorem for DMC's with feedback.