International audience

We address the problem of data detection for the multiple-input multiple-output (MIMO) channel employing one-bit quantizers at the receiver, taking into account different settings of channel state information (CSI) at the receiver (CSIR). In the first part and under perfect CSI conditions, we propose a two-step low-complexity data detection algorithm that reduces the maximum likelihood (ML) search space. The key idea is based on constructing a list of constellation points exploiting the Hessian matrix of the log-likelihood function. We convert the original detection problem under binary observations into the classical integer least-squares optimization enabling direct use of efficient sphere-decoding algorithms. This method is then extended to the multi-bit case along with an assessment of the computational complexity. In the second part, we focus on a real channel model and assume only the availability of statistical CSIR. We formulate the optimal detection metric under a pilot training scheme and present the main challenges in its evaluation, then employ the Laplace method to retrieve an approximation in closed form. We demonstrate through numerical experiments near-optimality of our proposed solutions in terms of vector error rates with respect to oracle lower bounds on their corresponding ML metric. Finally, we also investigate the performance in practical spatially correlated massive MIMO channels.

We address in this article the uplink transmission in a multiple-input multiple-output channel employing 1-bit analog-to-digital converters at the base station, assuming no a priori channel state information. In particular, we investigate under the original "probit" statistical model, the problems of channel estimation and data detection by first formulating them as binary classification procedures based on the cross-entropy loss. Under perfect CSI, the proposed data detection scheme relaxes the exhaustive search requirement to a convex problem with a box boundary constraint that can be solved using gradient descent methods. Achievable mismatched rates of proposed metrics are evaluated with the generalized mutual information and symbol error rates are presented. Simulation results show that the proposed channel estimation scheme under the probit model does not exhibit any instability with imperfect CSI in comparison with the Bussgang linear minimum mean square error estimator.

Signals satisfying low Peak-to-Average-Power Ratio (PAPR) properties are desirable in order to ensure the efficient operation of power amplifiers (PA) and to enhance the uplink (UL) coverage for user equipment. Building on the Discrete-Fourier-Transform spread Orthogonal Frequency Division Multiplexing (DFT-s-OFDM) waveform, we investigate in this paper a novel communication scheme capable of achieving significant PAPR gains for higher order modulations based on controlling the transition probabilities between time domain symbols. The design draws its inspiration from probabilistic amplitude shaping (PAS) with the use of a distribution matcher (DM) to generate sequences of symbols approximating the desired transitions. The performance is evaluated according to the PAPR and cubic metric (CM) as measures for a 16-QAM modulation order. The bit-error-rate (BER) performance curves are also assessed in comparison with the 16-QAM baseline scheme.