1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
use crate::filter::design::{FilterError, FilterSpec, Sos};
use crate::core::EdgeMode;
use crate::filter::fir::gaussian::{gaussian_radius, GAUSSIAN_DEFAULT_EDGE, GAUSSIAN_TRUNCATE};
use crate::filter::non_linear::{MEDIAN9_RADIUS, MEDIAN_DEFAULT_EDGE, TEAGER_KAISER_DEFAULT_EDGE};
use crate::spatial::{SpatialWhitening, SurfaceLaplacian};
/// Individual processing stage within an in-VRAM DSP pipeline.
#[derive(Debug, Clone, PartialEq)]
pub enum PipelineStage {
/// Linear scaling and offset: $y = \alpha x + \beta$
Scale { alpha: f32, beta: f32 },
/// Constant baseline subtraction: $y = x - \text{baseline}$ (in the signal's unit)
SubtractBaseline { baseline: f32 },
/// Sample clamping / rectification: $\text{clamp}(x, \min, \max)$
Clamp { min: f32, max: f32 },
/// IIR filter (Butterworth of any order and band, notch, or explicit sections), run forward or
/// forward-backward. Build with [`PipelineStage::bandpass`], [`PipelineStage::highpass`], ….
Filter(FilterSpec),
/// Common Average Referencing across all channels
CommonAverageReference,
/// Common median reference: the median over all channels subtracted at every sample
CommonMedianReference,
/// Spatial whitening ($\mathbf{W}_{\text{ZCA}}$ or Local $K$-NN) across channels
SpatialWhitening(SpatialWhitening),
/// 2D Surface Laplacian (double-differential spatial filter) across channels
SurfaceLaplacian(SurfaceLaplacian),
/// Zero-phase 1D Gaussian temporal smoothing with standard deviation `sigma_samples`, cut at
/// `GAUSSIAN_TRUNCATE` σ; samples past the ends come from `edge`.
GaussianSmooth { sigma_samples: f32, edge: EdgeMode },
/// Running median over an odd `width` (≤ `MAX_MEDIAN_WIDTH`); width 9 runs the branch-free
/// `med9` network. Samples past the ends come from `edge`.
Median { width: usize, edge: EdgeMode },
/// Discrete Teager-Kaiser Energy Operator: `Ψ[xₜ] = xₜ² − xₜ₋₁·xₜ₊₁`; the neighbours
/// of the end samples come from `edge`.
TeagerKaiser { edge: EdgeMode },
}
impl PipelineStage {
/// Butterworth band-pass, default order 5, zero phase.
pub fn bandpass(low_hz: f64, high_hz: f64) -> Self {
Self::Filter(FilterSpec::bandpass(low_hz, high_hz))
}
/// Butterworth high-pass, default order 5, zero phase (e.g. `highpass(300.0)` for spikes).
pub fn highpass(cutoff_hz: f64) -> Self {
Self::Filter(FilterSpec::highpass(cutoff_hz))
}
/// Butterworth low-pass, default order 5, zero phase (e.g. `lowpass(300.0)` for LFP).
pub fn lowpass(cutoff_hz: f64) -> Self {
Self::Filter(FilterSpec::lowpass(cutoff_hz))
}
/// Butterworth band-stop, default order 5, zero phase.
pub fn bandstop(low_hz: f64, high_hz: f64) -> Self {
Self::Filter(FilterSpec::bandstop(low_hz, high_hz))
}
/// Second-order notch (`iirnotch`), zero phase.
pub fn notch(freq_hz: f64, q: f64) -> Self {
Self::Filter(FilterSpec::notch(freq_hz, q))
}
/// Common Average Referencing across all channels.
pub fn common_average_reference() -> Self {
Self::CommonAverageReference
}
/// Common median reference across all channels.
pub fn common_median_reference() -> Self {
Self::CommonMedianReference
}
/// Explicit second-order sections, zero phase.
pub fn sos(sos: Sos) -> Self {
Self::Filter(FilterSpec::sos(sos))
}
/// Zero-phase Gaussian smoothing with reflected edges (`scipy.ndimage.gaussian_filter1d`).
pub fn gaussian_smooth(sigma_samples: f32) -> Self {
Self::GaussianSmooth { sigma_samples, edge: GAUSSIAN_DEFAULT_EDGE }
}
/// Running median over an odd `width` with zero-padded edges (`scipy.signal.medfilt`).
pub fn median(width: usize) -> Self {
Self::Median { width, edge: MEDIAN_DEFAULT_EDGE }
}
/// 9-point running median (`scipy.signal.medfilt(x, 9)`).
pub fn median9() -> Self {
Self::median(2 * MEDIAN9_RADIUS + 1)
}
/// Teager-Kaiser energy with reflected neighbours at the ends.
pub fn teager_kaiser() -> Self {
Self::TeagerKaiser { edge: TEAGER_KAISER_DEFAULT_EDGE }
}
/// `(left, right)` samples of context this stage needs around a chunk at `sample_rate` Hz so
/// the chunk interior matches whole-recording processing.
///
/// - `Filter`: from the designed filter's pole radii ([`FilterSpec::settling`]); forward-backward
/// needs both sides.
/// - `GaussianSmooth`: the kernel radius ([`gaussian_radius`] at [`GAUSSIAN_TRUNCATE`]) each side.
/// - `Median`: `width / 2` each side; `TeagerKaiser`: 1 each side.
/// - Pointwise / spatial stages: none.
///
/// # Errors
///
/// [`FilterError`] when a filter stage's design is invalid at `sample_rate`.
pub fn settling(&self, sample_rate: f64) -> Result<(usize, usize), FilterError> {
Ok(match self {
PipelineStage::Filter(spec) => spec.settling(sample_rate)?,
PipelineStage::GaussianSmooth { sigma_samples, .. } => {
let r = gaussian_radius(*sigma_samples, GAUSSIAN_TRUNCATE);
(r, r)
}
PipelineStage::Median { width, .. } => (width / 2, width / 2),
PipelineStage::TeagerKaiser { .. } => (1, 1),
PipelineStage::Scale { .. }
| PipelineStage::SubtractBaseline { .. }
| PipelineStage::Clamp { .. }
| PipelineStage::CommonAverageReference
| PipelineStage::CommonMedianReference
| PipelineStage::SpatialWhitening(_)
| PipelineStage::SurfaceLaplacian(_) => (0, 0),
})
}
}