科研 Skill

神经记录噪声诊断

neuro-noise-detective

Diagnoses noise and artifacts in neural recordings from spectra or raw traces and names the likely source and fix. Use when a recording shows hum, drift or unexplained interference.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Noise detective

Start from the data, not guesses: compute spectra per channel, then match the pattern to a source. Run the bundled script first; it does the measurements so you only interpret.

python scripts/spectrum_report.py --data raw.npy --fs 1000 [--names names.json] [--line 50]

Input: array of shape (channels, samples) in volts or microvolts (.npy, or any file MNE can read via --mne-file). Output: line frequency and harmonics with their size above the noise floor, odd/even harmonic balance, other narrow peaks, broadband high-frequency and slow-drift power per channel, flat and saturated channels, and whether each problem is common to all channels or channel-specific.

Pattern to source

PatternLikely sourceFix
50/60 Hz on all channels, similar sizemains pickup; poor ground or reference contact, unshielded cablesfix ground/reference impedance, twist and shorten leads, move mains cables; then notch or ZapLine
50/60 Hz on a few channelshigh impedance on those electrodesre-prep those electrodes
strong odd harmonics (150/180, 250/300 Hz)non-linear loads: switching supplies, dimmers, motorsunplug or move the equipment; battery-power the amplifier
narrow peaks not at line multiplesmonitor refresh, LED or motor drivers, Wi-Fi/BT duty cycles, stimulatorswitch devices off one at a time; check the stimulator clock
broadband rise above about 30 Hz, temporal/frontalmuscle (EMG)relax jaw/neck, reject or ICA
very slow drift below 1 Hz, largesweat, electrode gel drying, cable movementcooler room, re-prep, secure cables; high-pass
same waveform on all channelsreference or ground electrode problemfix reference/ground
step changes or spikeselectrode pops, movement, cable knocksmark and reject segments
flat channeldisconnected, shorted or bridged electrodecheck connection; see neuro-channel-map-verifier for bridging
clipped valuesamplifier range exceeded, DC offset too largelower gain, re-prep, check DC coupling
periodic about 1 Hz sharp peaksECGICA or regression with ECG channel

Workflow

  1. Run the script on a few minutes of raw (unfiltered) data.
  2. Read whether the issue is common (reference/ground/environment) or channel-specific (that electrode).
  3. Match the pattern above; propose the hardware fix first, software fix second (software can remove noise but cannot recover saturated data).
  4. If the user can change the setup, suggest one change at a time and re-measure.

Output

A short diagnosis: symptom with numbers from the report, most likely source, hardware fix, software fix, and how to confirm.

示例提示词

我的EEG记录有50Hz工频干扰和漂移,帮我诊断噪声来源并给修复建议。

所属插件

神经科学面向神经科学研究:EEG/MEG、fMRI、BIDS、spike sorting、连接组、脑机接口解码评估、神经刺激安全与公开神经数据检索。

安装该插件会一并获得同领域的 45 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

神经记录噪声诊断能做什么?

Diagnoses noise and artifacts in neural recordings from spectra or raw traces and names the likely source and fix. Use when a recording shows hum, drift or unexplained interference。

如何安装神经记录噪声诊断?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「神经记录噪声诊断」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

神经记录噪声诊断需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

神经记录噪声诊断经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。