[BabaSanfour]

Addresses item 6 of #137.

ZapLine subtracts the whole time course of each removed component. Those components also carry broadband activity, so subtracting them can raise power away from the line on single channels (up to +4.8 dB in the benchmark below). With line_bandwidth, only the removed components' part within line_bandwidth Hz of the targeted harmonics is subtracted, and the spectrum outside these bands is unchanged.

• line_bandwidth=None (default) keeps the current behaviour; the output is bit-identical to main.
• ZapLine selects the same components as without it, in standard and adaptive mode. In adaptive mode the segments select and subtract as by default, and the band is applied once to the removed signal of the whole recording.
• The FFT band mask moves to _filtering.line_band_mask and is shared with interpolate_spectrum (output unchanged).
• Documented limitation: on short inputs, or epochs that are not contiguous, part of the line falls outside the band and is not removed.

Benchmark

66 real recordings at their native rate, cropped to 180 s (CTF, MEGIN, 4D/BTi, KIT, EEG, mobile EEG, ECoG; 160 Hz to 6 kHz), and 96 synthetic cases with a known clean signal (8–272 channels, 50/60 Hz, one to many line sources, stable and broadened lines), on the Alliance fir cluster, with line_bandwidth=1.0:

	standard, default	standard, line-only	adaptive, default	adaptive, line-only
Worst-channel change away from the line, max (real)	4.78 dB	0.00 dB	4.46 dB	0.00 dB
Recordings with a band > 0.5 dB higher on a channel (real)	15	0	19	0
Residual line peak, median (real)	+0.33 dB	+0.03 dB	+0.17 dB	−0.05 dB
Line removal, median (real)	9.19 dB	9.19 dB	9.39 dB	9.39 dB
SNR against the clean signal, median (synthetic)	21.7 dB	26.8 dB	21.0 dB	26.7 dB

Line-only keeps more than 1 dB more line than the default in 1 (standard) and 2 (adaptive) real recordings, all clinical EEG, at most +1.7 dB. Its SNR is higher in 75/96 (standard) and 77/96 (adaptive) synthetic cases; it is lower for some broadened lines (better in 27/48), whose power extends beyond the band.

On 1,237 clinical EEG recordings (19 channels, 200 Hz, standard mode with at most 3 components), recordings with a channel more than 0.5 dB higher in 1–45 Hz drop from 134 to 0, and line removal at 59–61 Hz is a median 0.02 dB weaker.

Related, not changed here: #149 (ZapLine's period smoother is causal). With line_bandwidth, it does not change power away from the line.

Claude Code was used to implement and benchmark these changes.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 94.11765% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.53%. Comparing base (94f47e1) to head (4216833).

Files with missing lines	Patch %	Lines
mne_denoise/_filtering.py	71.42%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #148      +/-   ##
==========================================
+ Coverage   85.47%   85.53%   +0.05%     
==========================================
  Files          77       77              
  Lines       10719    10747      +28     
  Branches     1888     1891       +3     
==========================================
+ Hits         9162     9192      +30     
+ Misses        931      929       -2     
  Partials      626      626              

Flag	Coverage Δ
unittests	85.53% <94.11%> (+0.05%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
mne_denoise/spectrum_interpolation.py	83.19% <100.00%> (+2.86%)	⬆️
mne_denoise/zapline/core.py	85.08% <100.00%> (+1.12%)	⬆️
mne_denoise/_filtering.py	83.33% <71.42%> (-16.67%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.