Supporting Probabilistic Constellation Shaping in 5G-NR

Abstract

<p>In this paper, we consider applying probabilistic constellation shaping (PCS) to the fifth-generation new-radio (5G-NR) system. We propose a practical PCS transceiver design which is fully backward compatible to the existing 5G-NR system, and can bring significant improvements in block-error rate (BLER) and throughput. Further, we derive the properties of average-power, entropy-loss, and peak-to-average power-ratio (PAPR) increment in connection to the proposed PCS design. We show that an effective PCS can set the variance of a zero-mean Gaussian distribution for shaping to be around 0.5,making both the PAPR-increment and the entropy-loss to be at acceptable levels. Furthermore, we derive a necessary condition for PCS to attain a higher throughput than a uniform constellation. It shows that the normalized entropy-loss with PCS must be smaller than the BLER attained from transmissions with a uniform constellation. This is a critical observation and provides an important insight for the PCS design. Moreover, we also illustrate that the proposed PCS scheme can provide additional flexibilities in adjusting the effective code-rate without affecting the encoder. As a result, transitions among different MCSs can be smoother even with an adaptive modulation and coding scheme (MCS), and a better throughput-envelope can be obtained as signal-to-noise ratio (SNR) changes.</p>