发表机构
Illinois Institute of Technology; Fermi National Accelerator Laboratory（伊利诺伊理工学院; 费米国家加速器实验室）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文研究超导量子比特色散读取中，实时自适应FIR滤波（LMS算法）相比boxcar平均能否提升判别保真度，发现固定FIR滤波存在boxcar极限，但优化窗口和逐样本加权可提升保真度，并给出工作点记录。
AI 中文摘要
在超导量子比特的色散读取中，常见的基线方法是boxcar平均器：在固定窗口内进行均匀加权平均，随后进行基于阈值的态分配。我们探究何时额外的数字处理能提高保真度。我们在FPGA量子控制器上评估了一种由最小均方（LMS）算法实时训练的自适应有限脉冲响应（FIR）滤波器，并离线分析了测量的读取样本，以比较冻结滤波器与boxcar平均，并测试逐样本加权和序贯检测。对于所测试的固定FIR滤波器，将滤波后的样本求和得到的是boxcar和乘以一个复数，除了一个小的边缘修正（核坍缩）。这使态分离和噪声等比例缩放，因此判别能力无法提高（boxcar极限）。冻结的LMS滤波器降低了迹线噪声，但在所测试的读取长度下未解决保真度提升问题。在相同的样本上离线重放，每个滤波后的样本在2D transmon量子比特（Q1）的8微秒工作点携带boxcar样本判别信噪比的99.4%。在Q1上，优化读取窗口的起始时间和持续时间，在四种读取长度下将对比度读取保真度提高了+2.4%至+5.6%。逐样本加权在全窗口积分基础上最多增加+3.0%，而在8微秒时相对于调优窗口仅增加+0.33%。其余结果来自3D辅助量子比特（Q2），它位于Q1所处的早期决策交叉之上。硬件扫描将交叉窗口定位在约1微秒附近。离线时，序贯测试大多提前做出决策，保真度与boxcar相差在0.5个百分点以内。每类约30个标记样本构建的匹配滤波器模板几乎与数百个样本构建的一样好。这些结果构成了一个工作点记录，将波形去噪与决策改进区分开来。
英文摘要
Control and readout systems for superconducting quantum computing support real-time control and high-fidelity qubit-state discrimination. We examine when additional digital processing improves readout fidelity. A common baseline for dispersive readout uses a boxcar averager to compute a uniformly weighted average over a fixed window, followed by threshold-based state assignment. In this paper, we evaluated an adaptive finite-impulse-response (FIR) filter trained in real time using the least-mean-squares (LMS) algorithm on a field-programmable gate array (FPGA) quantum controller. We also analyzed measured readout shots offline to compare the frozen filter with boxcar averaging and to evaluate per-sample weighting and sequential detection. For the fixed FIR filters tested, we show summing the filtered samples gives the boxcar sum multiplied by one complex number, apart from a small edge correction. We call this the kernel collapse. The multiplication scales the state separation and noise equally without improving state discrimination, and we call this the boxcar limit. The frozen LMS filter reduces trace noise, but same-shot offline comparisons show no resolved improvement in readout fidelity at the tested operating points. On a 2D transmon, tuning the integration window accounts for much of the observed improvement over full-window integration, highlighting the importance of an optimized boxcar baseline when evaluating per-sample weighting. We then examine how changing the decision method can reduce calibration effort and decision time. On a higher-fidelity 3D ancilla readout, model-based template estimation reaches the same fidelity target as direct averaging with about 30 rather than 300 labeled shots per state. Offline sequential replay assigns most shots before the acquisition window ends, with contrast fidelity within an absolute +/-0.5% tolerance of the boxcar on the primary...
Comments16 pages, 12 figures, 2 tables, 27 equations
