Decoding brain signals can not only reveal Metaverse users' expectations but also early detect error-related behaviors such as stress, drowsiness, and motion sickness. For that, this article proposes a pioneering framework using wireless/over-the-air Brain-Computer Interface (BCI) to assist creation of virtual avatars as human representation in the Metaverse. Specifically, to eliminate the computational burden for Metaverse users' devices, we leverage Wireless Edge Servers (WES) that are popular in 5G architecture and therein URLLC, enhanced broadband features to obtain and process the brain activities, i.e., electroencephalography (EEG) signals (via uplink wireless channels). As a result, the WES can learn human behaviors, adapt system configurations, and allocate radio resources to create individualized settings and enhance user experiences. Despite the potential of BCI, the inherent noisy/fading wireless channels and the uncertainty in Metaverse users' demands and behaviors make the related resource allocation and learning/classification problems particularly challenging. We formulate the joint learning and resource allocation problem as a Quality-of-Experience (QoE) maximization problem that takes into the latency, brain classification accuracy, and resources of the system. To tackle this mixed integer programming problem, we then propose two novel algorithms that are (i) a hybrid learning algorithm to maximize the user QoE and (ii) a meta-learning algorithm to exploit the neurodiversity of the brain signals among multiple Metaverse users. The extensive experiment results with different BCI datasets show that our proposed algorithms can not only provide low delay for virtual reality (VR) applications but also can achieve high classification accuracy for the collected brain signals.

翻译：解码脑信号不仅能揭示元宇宙用户的期望，还能早期检测与错误相关的行为（如压力、困倦和晕动症）。为此，本文提出了一种开创性框架，利用无线/空中脑机接口辅助创建虚拟化身作为元宇宙中的人类表征。具体而言，为消除元宇宙用户设备的计算负担，我们利用了5G架构中广泛使用的无线边缘服务器及其超可靠低时延通信与增强宽带特性，通过上行无线信道获取并处理大脑活动（即脑电图信号）。由此，无线边缘服务器可学习人类行为、调整系统配置并分配无线电资源，以创建个性化设置并增强用户体验。尽管脑机接口具有潜力，但无线信道固有的噪声/衰落特性以及元宇宙用户需求与行为的不确定性，使得相关的资源分配和学习/分类问题极具挑战性。我们将联合学习与资源分配问题建模为体验质量最大化问题，该问题综合考虑延迟、脑信号分类准确率及系统资源。为解决这一混合整数规划问题，我们提出两种新颖算法：（i）一种混合学习算法以最大化用户体验质量，（ii）一种元学习算法以利用多位元宇宙用户脑信号的神经多样性。基于不同脑机接口数据集的广泛实验结果表明，所提算法不仅能降低虚拟现实应用的延迟，还能对采集的脑信号实现高分类准确率。