脑机接口 / BCI

Wireless foundation models (WFMs) have recently emerged as a promising paradigm for AI-native 6G networks, enabling universal channel representations adaptable to diverse communication and sensing tasks. Existing WFMs are predominantly built upon the Transformer architecture, which delivers superior performance but incurs computational complexity proportional to the square of the input sequence length, posing a significant barrier to their deployment under stringent inference latency constraints. To address this issue, in this paper, we propose ConsisFormer, a compute-efficient Transformer design based on short-term consistency of wireless channels, as a WFM backbone. By utilizing the observation that adjacent time or frequency instances share similar clusters of scatterers and thus exhibit similar channel characteristics, we develop an adaptive token aggregation (ATA) module to dynamically merge neighboring channel state information (CSI) tokens, thereby reducing the length of the token sequence involved in self-attention calculations to lower the computational cost. Furthermore, we propose a feature sequence interpolation (FSI) method to recover the full CSI representation based on the sparse feature sequence outputted from the Transformer blocks, thus keeping the performance unaffected while ensuring low complexity. Moreover, we propose an aggregated auto-encoder (AAE) pre-training paradigm for WFMs, enabling robust channel representation learning from sparsified CSI tokens via compression and recovery. Simulation results show that the proposed design reduces the computational complexity of WFM by over $83\%$ with negligible performance loss on various tasks including channel prediction, LoS/NLOS classification, beam prediction, and localization.

Recent advances in neuromorphic signal processing have introduced time encoding machines as a promising alternative to conventional uniform sampling for low-power communication receivers. In this paradigm, analog signals are converted into event timings by an integrate-and-fire circuit, allowing information to be represented through spike times rather than amplitude samples. While event-driven sampling eliminates the need for a fixed-rate clock, receivers equipped with integrate-and-fire time encoding machines, called time encoding receivers, often assume perfect symbol synchronization, leaving the problem of symbol timing recovery unresolved. This paper presents a joint timing recovery and data detection framework for integrate-and-fire time encoding receivers. The log-likelihood function is derived to capture the dependence between firing times, symbol timing offset, and transmitted sequence, leading to a maximum likelihood formulation for joint timing estimation and sequence detection. A practical two-stage receiver is developed, consisting of a timing recovery algorithm followed by a zero-forcing detector. Simulation results demonstrate accurate symbol timing offset estimation and improved symbol error rate performance compared to existing time encoding receivers.