Signal Processing and Feature Extraction

This tutorial explains the main EMG signal processing utilities in intan.processing, including filtering, rectification, normalization, feature extraction, and dimensionality reduction. All functions are compatible with multi-channel EMG arrays loaded from .rhd or .dat files.

Filtering and Preprocessing

The intan.processing._filters module provides robust EMG filtering routines, including bandpass, lowpass, notch filtering, and rectification.

Example: Bandpass and Notch Filtering

from intan.processing import bandpass_filter, notch_filter, rectify

# Assume emg_data shape is (channels, samples), fs is the sampling rate in Hz
filtered = bandpass_filter(emg_data, lowcut=20, highcut=450, fs=fs)
filtered = notch_filter(filtered, fs=fs, f0=60)  # Remove 60 Hz noise

# Rectify EMG (absolute value)
rectified = rectify(filtered)

Windowed and Smoothed RMS

RMS (Root Mean Square) is often used to measure the envelope of EMG signals.

from intan.processing import window_rms, calculate_rms

# Windowed RMS across each channel (e.g., 400-sample window)
windowed_rms = window_rms(emg_data, window_size=400)

# Non-overlapping RMS for feature extraction (returns (channels, windows))
rms_features = calculate_rms(emg_data, window_size=400)

Envelope Extraction

Use the analytic Hilbert envelope for precise EMG envelope estimation:

from intan.processing import envelope_extraction

# Returns same shape as input
envelope = envelope_extraction(emg_data, method='hilbert')

Normalization and Referencing

Z-score normalization: Standardizes each channel to zero mean/unit variance.

from intan.processing import z_score_norm

normalized = z_score_norm(emg_data)

Common Average Referencing (CAR): Remove global artifacts across all channels.

from intan.processing import common_average_reference

car_data = common_average_reference(emg_data)

Dimensionality Reduction (PCA)

Reduce channel count and noise with Principal Component Analysis (PCA):

from intan.processing import apply_pca

# Reduce to 8 principal components
pca_data, explained_var = apply_pca(emg_data, num_components=8)
print("Explained variance ratio:", explained_var)

Sliding Windows for Segmentation

Split continuous EMG into overlapping windows for further analysis:

from intan.processing import sliding_window

# Example: 250 ms windows with 50 ms step size
window_size = int(0.25 * fs)
step_size = int(0.05 * fs)
windows = sliding_window(emg_data, window_size, step_size)
# Each entry is (channels, window_size)

Feature Extraction

The intan.processing._features module provides standard time-domain EMG features:

Mean Absolute Value (MAV)
Zero Crossings (ZC)
Slope Sign Changes (SSC)
Waveform Length (WL)
Root Mean Square (RMS)

Extract all features for each channel in a window:

from intan.processing import extract_features

# Example: compute all features for every channel in a segment
feats = extract_features(emg_window)
print("Feature vector shape:", feats.shape)

You can also specify which features to extract:

feature_list = ['mean_absolute_value', 'zero_crossings', 'waveform_length']
feats = extract_features(emg_window, feature_fns=feature_list)

Full Preprocessing Pipeline Example

The following combines filtering, rectification, and RMS envelope smoothing in a single pipeline:

from intan.processing import bandpass_filter, notch_filter, rectify, window_rms

# Bandpass filter
emg_filtered = bandpass_filter(emg_data, lowcut=20, highcut=450, fs=fs)
# Notch filter for line noise
emg_notched = notch_filter(emg_filtered, fs=fs, f0=60)
# Rectify
emg_rectified = rectify(emg_notched)
# Windowed RMS smoothing
emg_smooth = window_rms(emg_rectified, window_size=400)

Noise Characterization Example

Compare baseline RMS noise between two conditions (e.g., dry vs. wet electrodes):

from intan.processing import bandpass_filter, notch_filter
import numpy as np

# Load EMG for each condition
emg_dry = ...   # (channels, samples)
emg_wet = ...   # (channels, samples)
fs = 4000

def compute_rms(data):
    return np.sqrt(np.mean(data ** 2, axis=1))

emg_dry_filt = notch_filter(bandpass_filter(emg_dry, fs=fs), fs=fs)
emg_wet_filt = notch_filter(bandpass_filter(emg_wet, fs=fs), fs=fs)
rms_dry = compute_rms(emg_dry_filt)
rms_wet = compute_rms(emg_wet_filt)

print("Dry mean RMS:", np.mean(rms_dry))
print("Wet mean RMS:", np.mean(rms_wet))

Additional Tools

Downsampling: Reduce sampling rate for long recordings.

from intan.processing import downsample

downsampled = downsample(emg_data, sampling_rate=fs, target_fs=1000)

Grid Averaging: Compute averages across spatially grouped channels.

from intan.processing import compute_grid_average

grid_avg = compute_grid_average(emg_data, grid_spacing=8)

Summary

These processing tools form the foundation for all subsequent feature extraction and classification tasks using EMG data collected with Intan systems. For details on each function, see the API Reference.