AI 中文总结
针对双基地感知中收发器时钟异步问题，开发微多普勒构建框架，通过循环差分和设计流水线抑制残留副产物，解决镜像模糊，在大规模 WiFi 手势数据集上表现出色，准确率达 90.1% - 98.2%，优于基线。
AI 中文摘要
集成感知与通信（ISAC）可将无线网络转变为普遍的感知平台，但双基地部署受收发器时钟异步影响，导致信道状态信息（CSI）出现随机相位波动。CSI 比率净化会引入非线性失真，限制多目标可扩展性并使延迟和到达角（AoA）域处理复杂化。交叉天线共轭乘法（CACC）保留线性结构，但存在镜像模糊和二阶副产物，会破坏运动引起的多普勒特征。我们开发了一个微多普勒构建框架来解决模糊度并抑制这些残留副产物。循环差分首先衰减主要的镜像分量。然后利用残留项出现在差分延迟坐标且缺乏有序 AoA 导向结构这一事实，设计了一个轻量级延迟 - AoA - 多普勒流水线，以隔离所需的运动响应，而无需在目标 bins 处相干地累积残留项。滤波后的响应被聚合为更高信噪比的微多普勒表示。消融研究证实了残留抑制和多维滤波的价值。在大规模 WiFi 手势数据集上，所提出的表示在四个域因素上具有通用性，准确率达到 90.1% - 98.2%，平均比代表性基线高出约 18 个百分点。
英文摘要
Integrated sensing and communications (ISAC) can turn wireless networks into pervasive sensing platforms, but bistatic deployments are impaired by transceiver clock asynchrony, which induces random phase fluctuations in channel state information (CSI). CSI-ratio sanitization introduces nonlinear distortion that limits multi-target scalability and complicates delay- and angle-of-arrival (AoA)-domain processing. Cross-antenna conjugate multiplication (CACC) preserves a linear structure, but leaves mirror ambiguity and second-order by-products that corrupt motion-induced Doppler signatures. We develop a micro-Doppler construction framework that resolves the ambiguity and suppresses these residual by-products. Cyclic differencing first attenuates the dominant mirror component. We then exploit the facts that residual terms occur at differenced delay coordinates and lack an ordered AoA steering structure, designing a lightweight delay-AoA-Doppler pipeline that isolates the desired kinematic response without coherently accumulating residuals at target bins. The filtered responses are aggregated into a higher-SNR micro-Doppler representation. Ablation studies confirm the value of residual suppression and multidimensional filtering. On a large-scale WiFi gesture dataset, the proposed representation generalizes across four domain factors, attaining 90.1%-98.2% accuracy and outperforming representative baselines by about 18 percentage points on average.
