High-Capacity and Low-PAPR BICM-OFDM Systems Using Non-Equiprobable and Non-Uniform Constellation Shaping With Clipping and Filtering

TL;DR

The paper tackles achieving high spectral efficiency in BICM-OFDM with low PAPR by designing non-equiprobable non-uniform (NENU) constellations shaped via a truncated Gaussian and a two-parameter merging process. The constellations, ${\cal A}(M,N,\rho)$, are optimized for systems employing clipping and filtering (CAF) at the transmitter and clipping-noise cancellation (CNC) at the receiver, exploiting a one-dimensional PAM structure to keep demapping complexity low. Performance analyses show that careful merging shifts BMI from lower- to higher-order bits, reducing the loss from the Shannon limit at high SNR, and that the proposed shaping achieves substantial PAPR reduction with CNC recovering much of the distortion. Simulation results on AWGN and frequency-selective fading channels demonstrate that NENU with CAF/CNC outperforms DFT-precoded BICM-OFDM and approaches the performance of higher-complexity shaping schemes, while maintaining practical transmitter complexity and scalability to higher modulation orders.

Abstract

We address a design of high-capacity and low-peak-to-average power ratio (PAPR) orthogonal frequency-division multiplexing (OFDM) systems based on bit-interleaved coded modulation (BICM) utilizing non-equiprobable and non-uniform (NENU) constellations as well as clipping and filtering (CAF). The proposed constellations are generated using a truncated Gaussian distribution and the merging of constellation points, where the former creates a non-uniform constellation (NUC), and the latter adjusts the number of signal points for further improving the total bit-wise mutual information (BMI). Unlike other exhaustive search-based approaches, the proposed constellations are uniquely determined by only two parameters associated with NUC and cardinality. Due to this property of limited degrees of freedom, the complexity required for the numerical optimization process can be significantly low. We focus on the constellation design based on one dimension, i.e., pulse amplitude modulation (PAM), which facilitates the reduction of demapping complexity for the BICM receiver. The use of CAF at the transmitter can efficiently reduce the PAPR of OFDM signals; however, it introduces clipping noise that may degrade error rate performance, making the application of clipping noise cancellation (CNC) at the receiver essential. Therefore, we optimize the NENU constellations in the presence of CAF and CNC. Simulation results demonstrate that the combination of constellation shaping with CAF and CNC enables BICM-OFDM systems to simultaneously achieve low PAPR and high spectral efficiency over additive white Gaussian noise (AWGN) as well as frequency-selective fading channels. Furthermore, comparative studies confirm that the proposed system significantly outperforms the single-carrier counterpart (i.e., DFT-precoded BICM-OFDM) in terms of PAPR and bit error rate (BER) performance over fading channels.

Figures (13)

• Figure 1: Constellation point selection based on equiprobable partitioning of the Gaussian CDF with $M=32$. Due to the symmetry, only the positive region is shown. (a) is the case with $\rho=1.0$ (i.e., without truncation), while (b) corresponds to the case with $\rho=0.5$
• Figure 2: Constellation point selection based on non-equiprobable partitioning of the Gaussian CDF ($\rho=1.0$) with $M=32$ and $N=10$.
• Figure 3: Comparison between the proposed QAM consisting of two (32,20)-PAM constellations (shown as green circles) and the equivalent one-dimensional NUC adopted in ATSC 3.0 (shown as blue diamonds). Both constellations are optimized for BICM with an information rate of 6.0 bits per symbol (code rate of $3/5$). The size of the circles reflects their probability of occurrence.
• Figure 4: Loss in achievable information rate from the Shannon limit for the BICM systems using the proposed NENU constellations with various modulation orders.
• Figure 5: Loss from the Shannon limit for the BICM systems with several different modulation schemes.