Interference-Aware Constellation Design for Z-Interference Channels with Imperfect CSI

Xinliang Zhang, Mojtaba Vaezi, Lizhong Zheng

Abstract

A deep autoencoder (DAE)-based end-to-end communication over the two-user Z-interference channel (ZIC) with finite-alphabet inputs is designed in this paper. The design is for imperfect channel state information (CSI) where both estimation and quantization errors exist. The proposed structure jointly optimizes the encoders and decoders to generate interference-aware constellations that adapt their shape to the interference intensity in order to minimize the bit error rate. A normalization layer is designed to guarantee an average power constraint in the DAE while allowing the architecture to generate constellations with nonuniform shapes. This brings further shaping gain compared to standard uniform constellations such as quadrature amplitude modulation. The performance of the DAE-ZIC is compared with two conventional methods, i.e., standard and rotated constellations. The proposed structure significantly enhances the performance of the ZIC. Simulation results confirm bit error rate reduction in all interference regimes (weak, moderate, and strong). At a signal-to-noise ratio of 20dB, the improvements reach about two orders of magnitude when only quantization error exists, indicating that the DAE-ZIC is highly robust to the interference compared to the conventional methods.

1. Introduction

In modern wireless communication, interference remains a primary bottleneck for reliability and throughput. Standard systems typically rely on fixed constellations like Quadrature Amplitude Modulation (QAM). While effective for point-to-point links, these "interference-oblivious" shapes do not adapt to the surrounding signal environment. This paper addresses the Z-interference channel (ZIC), an asymmetric scenario where one user's transmission interferes with another, but not vice versa. We propose a Deep Autoencoder (DAE) framework to design interference-aware constellations that are robust to both channel interference and imperfect Channel State Information (CSI).

2. System Model and Challenges

The ZIC model consists of two transmitter-receiver pairs. In this asymmetric setup, User 2's signal leaks into User 1's receiver, while User 1's signal does not affect User 2. A significant challenge in practical implementations is the lack of perfect CSI. Receivers often face estimation noise and quantization errors when feeding channel data back to transmitters. Traditional methods, such as rotating standard constellations, often fail when these feedback errors occur. The goal is to bridge the gap between theoretical Gaussian signals and real-world finite-alphabet inputs.

3. Proposed Deep Autoencoder Architecture

We employ an end-to-end DAE architecture to jointly optimize the transmitter (encoder) and receiver (decoder). The system learns the best way to map bits to signals directly from data, rather than relying on fixed mathematical formulas.

The proposed DAE-ZIC architecture, showing the two-sub-network transmitter and the joint optimization of encoders and decoders.

As illustrated in the architecture diagram, each transmitter consists of two sub-networks. Sub-network 1 generates the basic constellation shape using fully connected neural network (FCNN) layers. Sub-network 2 functions as a power controller, scaling the signal to meet average power constraints while allowing for non-uniform shapes. A novel normalization layer ensures that the average power remains constant but allows the In-phase and Quadrature (I/Q) components to be shaped flexibly, providing significant shaping gain over standard square QAM.

4. Performance and Visual Analysis

The DAE-ZIC system was evaluated against two baseline methods: standard QAM (Baseline-1) and rotated QAM (Baseline-2). The study used synthetically generated data based on Rayleigh fading models, varying interference intensity ($\alpha$) across weak, moderate, and strong regimes.

Visual comparison of signal constellations for Baseline-1, Baseline-2, and the proposed DAE-ZIC at an interference level of alpha=0.5.

The most striking insight is "constellation morphing." As shown in the constellation maps, the DAE-ZIC learns to adapt its signal points to remain distinguishable even under heavy interference. While standard QAM points overlap and become unreadable, the DAE-designed points maintain a geometry that maximizes distance in the presence of the interfering signal. At a Signal-to-Noise Ratio (SNR) of 20dB, the DAE-ZIC achieves up to two orders of magnitude improvement in Bit Error Rate (BER) when quantization errors are present. On average, the proposed method reduces BER by approximately 40% compared to conventional methods.

5. Conclusion

This work demonstrates that joint optimization of encoders and decoders using deep learning can significantly enhance communication reliability in interference-limited channels. By moving away from fixed, interference-oblivious constellations toward learned, adaptive shapes, we can achieve substantial gains even with imperfect channel knowledge. Future research will explore scaling this architecture to multi-user MIMO environments and incorporating peak-to-average power ratio (PAPR) constraints into the training process to better align with real-world hardware limitations.

Summary updated

Abstract

A deep autoencoder (DAE)-based end-to-end communication over the two-user Z-interference channel (ZIC) with finite-alphabet inputs is designed in this paper. The design is for imperfect channel state information (CSI) where both estimation and quantization errors exist. The proposed structure jointly optimizes the encoders and decoders to generate interferenceaware constellations that adapt their shape to the interference intensity in order to minimize the bit error rate. A normalization layer is designed to guarantee an average power constraint in the DAE while allowing the architecture to generate constellations with nonuniform shapes. This brings further shaping gain compared to standard uniform constellations such as quadrature amplitude modulation. The performance of the DAE-ZIC is compared with two conventional methods, i.e., standard and rotated constellations. The proposed structure significantly enhances the performance of the ZIC. Simulation results confirm bit error rate reduction in all interference regimes (weak, moderate, and strong). At a signal-to-noise ratio of 20dB, the improvements reach about two orders of magnitude when only quantization error exists, indicating that the DAE-ZIC is highly robust to the interference compared to the conventional methods.
BibTeX
@misc{zhang2023interference,
  author = {Xinliang Zhang and Mojtaba Vaezi and Lizhong Zheng},
  title = {{Interference-Aware Constellation Design for Z-Interference Channels with Imperfect CSI}},
  year = {2023},
  eprint = {2303.08312},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2303.08312},
}
Figure