The future Six-Generation (6G) envisions massive access of wireless devices in the network, leading to more serious interference from concurrent transmissions between wireless devices in the same frequency band. Existing interference mitigation approaches takes the interference signals as Gaussian white noise, which cannot precisely estimate the non-Gaussian interference signals from other devices. In this paper, we present IntLearner, a new interference mitigation technique that estimates and mitigates the impact of interference signals with only physical-layer (PHY) information available in base-station (BS) and user-equipment (UE), including channel estimator and constellations. More specifically, IntLearner utilizes the power of AI to estimate the features in interference signals, and removes the interference from the interfered received signal with neural network (NN). IntLearner's NN adopts a modular NN design, which takes the domain knowledge of BS and UE PHY as the guidance to NN design for minimizing training confusion and NN complexity. Simulation results show IntLearner increases Uplink (UL) channel estimation accuracy up to 7.4x, and reduces the Downlink (DL) Signal to Interference Ratio plus Noise Ratio (SINR) requirement to achieve the same Block Error Rate (BLER) by 1.5dB in a conventional multi-cell scenario.

翻译：未来第六代（6G）通信网络设想海量无线设备的接入，这将导致同一频段内多设备并发传输引发的干扰问题日益严重。现有干扰抑制方法将干扰信号视为高斯白噪声，无法精确估计来自其他设备的非高斯干扰信号。本文提出IntLearner——一种新型干扰抑制技术，仅利用基站（BS）和用户设备（UE）可获取的物理层（PHY）信息（包括信道估计器和星座图），即可实现干扰信号的估计与影响抑制。具体而言，IntLearner利用人工智能技术提取干扰信号特征，并通过神经网络（NN）从受扰接收信号中消除干扰。其NN采用模块化设计，以BS和UE物理层领域知识为指导，最小化训练混淆程度并降低网络复杂度。仿真结果表明，在传统多小区场景中，IntLearner可将上行链路（UL）信道估计精度提升高达7.4倍，并使下行链路（DL）实现相同误块率（BLER）所需的信干噪比（SINR）门限降低1.5dB。