WSDP API Reference

Complete API documentation for WSDP (Wi-Fi Sensing Data Processing).

Core Pipeline

`wsdp.pipeline()`

Main training pipeline function.

from wsdp import pipeline

pipeline(
    input_path: str,
    output_folder: str,
    dataset: str,
    model_path: Optional[str] = None,
    learning_rate: Optional[float] = None,
    num_epochs: Optional[int] = None,
    batch_size: Optional[int] = None,
    config_file: Optional[str] = None,
)

Parameters:

Parameter	Type	Default	Description
`input_path`	str	required	Path to input data directory
`output_folder`	str	required	Path to output directory
`dataset`	str	required	Dataset name (widar, gait, xrf55, elderal, zte)
`model_path`	str	None	Path to custom model file
`learning_rate`	float	None	Learning rate (overrides config)
`num_epochs`	int	None	Number of epochs (overrides config)
`batch_size`	int	None	Batch size (overrides config)
`config_file`	str	None	Path to YAML config file

Data Structures

`CSIData`

Container for raw CSI data from a file.

from wsdp.structure import CSIData

data = CSIData(file_path="/path/to/file.dat")

`CSIFrame`

Standardized frame structure for processed CSI.

from wsdp.structure import CSIFrame

frame = CSIFrame(
    csi_matrix: np.ndarray,  # Shape: (T, F, A)
    timestamp: float,
    metadata: dict,
)

Readers

Reader	Dataset	Format
`WidarReader`	Widar	.dat (bfee)
`GaitReader`	Gait	.dat (bfee)
`XRF55Reader`	XRF55	.npy
`ElderALReader`	ElderAL	.csv
`ZTEReader`	ZTE	.csv

Algorithms

Denoising

`wavelet_denoise_csi()`

Wavelet-based denoising using VisuShrink thresholding.

from wsdp.algorithms import wavelet_denoise_csi

denoised = wavelet_denoise_csi(csi_tensor)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi_tensor | np.ndarray | required | CSI array of shape (T, F, A), complex |

Returns: np.ndarray — Denoised CSI with same shape and dtype.

Reference: Donoho DL, Johnstone IM. "Ideal spatial adaptation by wavelet shrinkage." Biometrika, 1994.

`butterworth_denoise()`

Butterworth low-pass filter for CSI denoising.

from wsdp.algorithms import butterworth_denoise

denoised = butterworth_denoise(csi, order=5, cutoff=0.3)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F), complex or real | | order | int | 5 | Filter order | | cutoff | float | 0.3 | Normalized cutoff frequency in (0, 1] |

Returns: np.ndarray — Denoised CSI with same shape and dtype.

Reference: Butterworth S. "On the theory of filter amplifiers." Wireless Engineer, vol. 7, 1930.

`savgol_denoise()`

Savitzky-Golay polynomial smoothing filter.

from wsdp.algorithms import savgol_denoise

denoised = savgol_denoise(csi, window_length=11, polyorder=3)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F), complex or real | | window_length | int | 11 | Window length (must be odd, >= 3) | | polyorder | int | 3 | Polynomial order (must be < window_length) |

Returns: np.ndarray — Smoothed CSI with same shape and dtype.

Reference: Savitzky A, Golay MJE. "Smoothing and differentiation of data by simplified least squares procedures." Analytical Chemistry, 1964.

Phase Calibration

`phase_calibration()` (Linear)

Standard linear phase calibration using polynomial fitting.

from wsdp.algorithms import phase_calibration

calibrated = phase_calibration(csi_data)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi_data | np.ndarray | required | 3D CSI array (T, F, A), complex |

Returns: np.ndarray — Phase-calibrated CSI with same shape.

Reference: Halperin D, Hu W, Sheth A, Wetherall D. "Predictable 802.11 packet delivery from wireless channel measurements." ACM SIGCOMM, 2010.

`polynomial_calibration()`

Polynomial phase calibration across subcarriers.

from wsdp.algorithms import polynomial_calibration

calibrated = polynomial_calibration(csi, degree=3)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A), complex | | degree | int | 3 | Polynomial degree (1=linear, 2=quadratic, 3=cubic) |

Returns: np.ndarray — Phase-calibrated CSI with same shape.

Reference: Generalization of linear calibration from Halperin et al., ACM SIGCOMM, 2010.

`stc_calibration()`

Sanitize-then-Calibrate (STC) phase error removal.

from wsdp.algorithms import stc_calibration

calibrated = stc_calibration(csi)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A), complex |

Returns: np.ndarray — Phase-calibrated CSI with same shape.

Reference: Xie Y, Li Z, Li M. "Precise Power Delay Profiling with Commodity WiFi." IEEE Transactions on Wireless Communications (TWC), 2019.

`robust_phase_sanitization()`

Robust phase sanitization using median-based approach.

from wsdp.algorithms import robust_phase_sanitization

calibrated = robust_phase_sanitization(csi)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A), complex |

Returns: np.ndarray — Phase-sanitized CSI with same shape.

Reference: Wang G, Zou Y, Zhou Z, Wu K, Ni LM. "FIMD: Fine-grained Device-free Motion Detection." IEEE ICPADS, 2012.

Amplitude Processing

`normalize_amplitude()`

Normalize CSI amplitude along the time axis.

from wsdp.algorithms import normalize_amplitude

normalized = normalize_amplitude(csi, method='z-score')

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F) | | method | str | 'z-score' | 'z-score' or 'min-max' |

Returns: np.ndarray — Normalized CSI with same shape.

Reference: Ma Y, et al. "PhaseFi: Phase Fingerprinting for Indoor Localization with a Deep Learning Approach." IEEE GLOBECOM, 2015.

`remove_outliers()`

Remove or clip outlier amplitudes in CSI data.

from wsdp.algorithms import remove_outliers

cleaned = remove_outliers(csi, method='iqr', factor=1.5)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F) | | method | str | 'iqr' | 'iqr' (IQR-based) or 'z-score' (std-dev based) | | factor | float | 1.5 | Detection threshold multiplier |

Returns: np.ndarray — CSI with outliers clipped, same shape.

Interpolation

`interpolate_grid()`

Interpolate CSI data to a target number of subcarriers.

from wsdp.algorithms import interpolate_grid

interpolated = interpolate_grid(csi, target_K=30, method='cubic')

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) | | target_K | int | 30 | Target number of subcarriers | | method | str | 'cubic' | 'linear', 'cubic', or 'nearest' |

Returns: np.ndarray — Interpolated CSI with shape (T, target_K, A).

Reference: Bianchi V, et al. "Indoor Localization by Interpolation of Radio Maps." Sensors, 2020.

Feature Extraction

`doppler_spectrum()`

Compute Doppler spectrum from CSI time series using STFT.

from wsdp.algorithms import doppler_spectrum

spectrum = doppler_spectrum(csi, n_fft=64, hop_length=32)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F), complex | | n_fft | int | 64 | FFT size for STFT | | hop_length | int | 32 | Hop size between STFT windows |

Returns: np.ndarray — Doppler spectrogram of shape (n_freq, n_time, F[, A]).

Reference: Ali K, et al. "Keystroke Recognition Using WiFi Signals." ACM MobiCom, 2015.

`entropy_features()`

Compute information entropy features from CSI amplitude distribution.

from wsdp.algorithms import entropy_features

entropy = entropy_features(csi, bins=50)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F) | | bins | int | 50 | Number of histogram bins for entropy estimation |

Returns: np.ndarray — Entropy values of shape (F,) or (F, A).

Reference: Shannon CE. "A Mathematical Theory of Communication." Bell System Technical Journal, 1948.

`csi_ratio()`

Compute CSI ratio between antenna pairs.

from wsdp.algorithms import csi_ratio

ratio = csi_ratio(csi, antenna_pairs=[(0, 1), (1, 2)])

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A), A >= 2 | | antenna_pairs | list | None | List of (ant1, ant2) tuples. Default: consecutive pairs |

Returns: np.ndarray — CSI ratios with shape (T, F, n_pairs).

Reference: Halperin D, et al. "Tool release: Gathering 802.11n traces with channel state information." ACM SIGCOMM CCR, 2011.

`tensor_decomposition()`

Decompose CSI tensor using CP or Tucker decomposition.

from wsdp.algorithms import tensor_decomposition

decomposed = tensor_decomposition(csi, rank=10, method='cp')

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) | | rank | int | 10 | Rank of decomposition | | method | str | 'cp' | 'cp' (Canonical Polyadic) or 'tucker' |

Returns: np.ndarray — Reconstructed low-rank CSI tensor, same shape.

Reference: Kolda TG, Bader BW. "Tensor Decompositions and Applications." SIAM Review, 2009.

Activity Detection

`detect_activity()`

Detect activity using sliding window variance analysis.

from wsdp.algorithms import detect_activity

activity = detect_activity(csi, window=32, threshold=0.1)

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F) | | window | int | 32 | Sliding window size in samples | | threshold | float | 0.1 | Detection threshold for normalized variance |

Returns: np.ndarray — Boolean array of shape (T,) indicating activity.

Reference: Zhou Z, et al. "Device-Free Passive Localization for Human Activity Recognition." IEEE Communications Magazine, 2013.

`change_point_detection()`

Detect change points in CSI time series.

from wsdp.algorithms import change_point_detection

change_points = change_point_detection(csi, method='bayesian')

Parameters: | Parameter | Type | Default | Description | |-----------|------|---------|-------------| | csi | np.ndarray | required | CSI array of shape (T, F, A) or (T, F) | | method | str | 'bayesian' | 'bayesian', 'cusum', or 'variance' |

Returns: np.ndarray — Array of time indices where change points detected.

Reference: Adams RP, MacKay DJC. "Bayesian Online Changepoint Detection." arXiv:0710.3742, 2007.

Visualization

`plot_csi_heatmap()`

Plot CSI amplitude as a time-frequency heatmap.

from wsdp.algorithms.visualization import plot_csi_heatmap

fig = plot_csi_heatmap(csi_data, antenna_idx=0, save_path='heatmap.png')

`plot_denoising_comparison()`

Plot before/after denoising comparison.

from wsdp.algorithms.visualization import plot_denoising_comparison

fig = plot_denoising_comparison(original, denoised, antenna_idx=0)

`plot_phase_calibration()`

Plot phase before and after calibration.

from wsdp.algorithms.visualization import plot_phase_calibration

fig = plot_phase_calibration(original, calibrated, antenna_idx=0)

Unified API

The unified API provides a consistent interface for all algorithm categories.

`denoise()`

from wsdp.algorithms import denoise

denoised = denoise(csi, method='wavelet', **kwargs)

Method	Source Function	Key Parameters
`'wavelet'`	`wavelet_denoise_csi()`	—
`'butterworth'`	`butterworth_denoise()`	`order`, `cutoff`
`'savgol'`	`savgol_denoise()`	`window_length`, `polyorder`

`calibrate()`

from wsdp.algorithms import calibrate

calibrated = calibrate(csi, method='linear', **kwargs)

Method	Source Function	Key Parameters
`'linear'`	`phase_calibration()`	—
`'polynomial'`	`polynomial_calibration()`	`degree`
`'stc'`	`stc_calibration()`	—
`'robust'`	`robust_phase_sanitization()`	—

`normalize()`

from wsdp.algorithms import normalize

normalized = normalize(csi, method='z-score')

Method	Source Function	Key Parameters
`'z-score'`	`normalize_amplitude()`	—
`'min-max'`	`normalize_amplitude()`	—

`interpolate()`

from wsdp.algorithms import interpolate

result = interpolate(csi, target_K=30, method='cubic')

Method	Source Function	Key Parameters
`'linear'`	`interpolate_grid()`	`target_K`
`'cubic'`	`interpolate_grid()`	`target_K`
`'nearest'`	`interpolate_grid()`	`target_K`

`extract_features()`

from wsdp.algorithms import extract_features

features = extract_features(csi, features=['doppler', 'entropy'])

Feature	Source Function	Key Parameters
`'doppler'`	`doppler_spectrum()`	`n_fft`, `hop_length`
`'entropy'`	`entropy_features()`	`bins`
`'ratio'`	`csi_ratio()`	`antenna_pairs`
`'decomposition'`	`tensor_decomposition()`	`rank`, `method`

Pluggable Architecture

WSDP provides a pluggable algorithm architecture built on the Registry Pattern, allowing users to switch algorithms, add custom implementations, configure via files, and use preset pipelines.

Algorithm Registry

`register_algorithm()`

Register a custom algorithm.

from wsdp.algorithms import register_algorithm

def my_denoise(csi, strength=1.0, **kwargs):
    return csi * strength

register_algorithm('denoise', 'my_method', my_denoise)

Parameters: | Parameter | Type | Description | |-----------|------|-------------| | category | str | Category: 'denoise', 'calibrate', 'normalize', 'interpolate', 'extract_features', 'detect', 'outliers' | | name | str | Algorithm name (used as method= parameter) | | func | Callable | Algorithm implementation |

`get_algorithm()`

Get an algorithm function by category and name.

from wsdp.algorithms import get_algorithm

func = get_algorithm('denoise', 'butterworth')
result = func(csi, order=5)

`list_algorithms()`

List available algorithms.

from wsdp.algorithms import list_algorithms

# All categories
>>> list_algorithms()
{'denoise': ['wavelet', 'butterworth', 'savgol'], 'calibrate': [...]}

# Specific category
>>> list_algorithms('denoise')
{'wavelet': 'wsdp.algorithms.denoising:wavelet_denoise_csi', ...}

`is_registered()`

Check if an algorithm is registered.

>>> is_registered('denoise', 'wavelet')
True

`algorithm_info()`

Get detailed information about an algorithm.

>>> algorithm_info('denoise', 'butterworth')
{'name': 'butterworth', 'category': 'denoise', 'module': '...', ...}

`unregister_algorithm()`

Remove a custom algorithm (built-in algorithms cannot be unregistered).

unregister_algorithm('denoise', 'my_method')

Configuration Files

WSDP supports YAML and JSON configuration files for algorithm selection.

YAML Format

# algorithms_config.yaml
denoise:
  method: butterworth
  params:
    order: 5
    cutoff: 0.3

calibrate:
  method: polynomial
  params:
    degree: 3

normalize:
  method: z-score

JSON Format

{
  "denoise": {
    "method": "butterworth",
    "params": {"order": 5, "cutoff": 0.3}
  },
  "calibrate": {
    "method": "polynomial",
    "params": {"degree": 3}
  }
}

Preset Reference

preset: high_quality

`load_config()`

Load algorithm configuration from a file.

from wsdp.algorithms import load_config

config = load_config('algorithms_config.yaml')

`save_config()`

Save algorithm configuration to a file.

from wsdp.algorithms import save_config, apply_preset

steps = apply_preset('high_quality')
save_config(steps, 'my_config.yaml', format='yaml')

Pipeline Presets

Built-in Presets

Preset	Denoise	Calibrate	Normalize	Use Case
`high_quality`	butterworth (order=5)	stc	z-score	Maximum accuracy
`fast`	savgol (w=7)	linear	min-max	Speed-optimized
`robust`	wavelet	robust	z-score	Noisy environments
`gesture_recognition`	butterworth (order=4)	stc	z-score	Gesture tasks
`activity_detection`	savgol (w=11)	polynomial (deg=2)	z-score	HAR tasks
`localization`	wavelet	robust	z-score	Localization tasks

`apply_preset()`

Get a preset pipeline configuration.

from wsdp.algorithms import apply_preset

steps = apply_preset('high_quality')
# {'denoise': {'method': 'butterworth', 'order': 5, ...}, ...}

`execute_pipeline()`

Execute a processing pipeline on CSI data.

from wsdp.algorithms import apply_preset, execute_pipeline

steps = apply_preset('high_quality')
processed = execute_pipeline(csi, steps)

`register_preset()`

Register a custom preset.

from wsdp.algorithms import register_preset

register_preset('my_preset', {
    'denoise': {'method': 'butterworth', 'order': 3},
    'calibrate': {'method': 'linear'},
})

`list_presets()`

List all available presets.

>>> list_presets()
{'high_quality': ['denoise', 'calibrate', 'normalize'], 'fast': [...], ...}

Custom Algorithms

Users can register their own algorithm implementations:

from wsdp.algorithms import register_algorithm, denoise, calibrate

# 1. Define your algorithm
def my_denoise(csi, strength=1.0, **kwargs):
    """Custom denoising: simple smoothing."""
    from scipy.ndimage import uniform_filter1d
    result = np.empty_like(csi)
    for f in range(csi.shape[1]):
        for a in range(csi.shape[2]):
            real_part = uniform_filter1d(np.real(csi[:, f, a]), size=int(strength * 5))
            imag_part = uniform_filter1d(np.imag(csi[:, f, a]), size=int(strength * 5))
            result[:, f, a] = real_part + 1j * imag_part
    return result

# 2. Register it
register_algorithm('denoise', 'my_smooth', my_denoise)

# 3. Use it via unified API
denoised = denoise(csi, method='my_smooth', strength=2.0)

# 4. Or use it in a pipeline
from wsdp.algorithms import execute_pipeline
steps = {
    'denoise': {'method': 'my_smooth', 'strength': 1.5},
    'calibrate': {'method': 'stc'},
}
result = execute_pipeline(csi, steps)

Models

Unified Model API

from wsdp.models import create_model, list_models, get_model

# Create any model
model = create_model(name, num_classes, input_shape, **kwargs)

# List models
all_models = list_models()                          # All models
baselines = list_models("baseline")                 # By category

# Get model class
model = get_model("ResNet1D", num_classes=10, input_shape=(20, 30, 3))

Parameters: - name: Model name (case-insensitive) - num_classes: Number of output classes - input_shape: Tuple of (T, F, A) — time steps, frequency bins, antennas - **kwargs: Model-specific hyperparameters

Model Registry

All models are stored in MODEL_REGISTRY and can be accessed by category:

Category	Models
`baseline`	MLPModel, CNN1DModel, CNN2DModel, LSTMModel
`mainstream`	ResNet1D, ResNet2D, BiLSTMAttention, EfficientNetCSI
`sota`	VisionTransformerCSI, MambaCSI, GraphNeuralCSI, CSIModel

Baseline Models

MLPModel

Fully-connected baseline, flattens input through MLP.

model = create_model("MLPModel", num_classes=10, input_shape=(20, 30, 3),
                      hidden_dims=[512, 256], dropout=0.3)

CNN1DModel

1D convolution extracting temporal features.

model = create_model("CNN1DModel", num_classes=10, input_shape=(20, 30, 3),
                      base_channels=64, num_layers=4)

CNN2DModel

2D convolution processing spectral representations per time step.

model = create_model("CNN2DModel", num_classes=10, input_shape=(20, 30, 3),
                      base_channels=32, num_layers=3)

LSTMModel

LSTM for temporal sequence modeling.

model = create_model("LSTMModel", num_classes=10, input_shape=(20, 30, 3),
                      hidden_size=128, num_layers=2, dropout=0.3)

Mainstream Models

ResNet1D

1D residual network with 3 residual blocks.

model = create_model("ResNet1D", num_classes=10, input_shape=(20, 30, 3),
                      base_channels=64)

ResNet2D

2D residual network for spatial feature extraction.

model = create_model("ResNet2D", num_classes=10, input_shape=(20, 30, 3),
                      base_channels=32)

BiLSTMAttention

Bidirectional LSTM with multi-head self-attention.

model = create_model("BiLSTMAttention", num_classes=10, input_shape=(20, 30, 3),
                      hidden_size=128, num_layers=2, num_heads=4, dropout=0.3)

EfficientNetCSI

Efficient CNN with configurable width and depth multipliers.

model = create_model("EfficientNetCSI", num_classes=10, input_shape=(20, 30, 3),
                      width_mult=1.0, depth_mult=1.0, base_channels=16)

SOTA Models

VisionTransformerCSI

Vision Transformer treating F×A spatial patches across time steps.

model = create_model("VisionTransformerCSI", num_classes=10, input_shape=(20, 30, 3),
                      embed_dim=128, num_heads=4, num_layers=4,
                      patch_size_f=4, patch_size_a=4, dropout=0.1)

MambaCSI

State space model (Mamba) for efficient long-range temporal modeling.

model = create_model("MambaCSI", num_classes=10, input_shape=(20, 30, 3),
                      d_model=128, d_state=16, num_layers=4)

GraphNeuralCSI

Graph neural network modeling antenna/subcarrier topology.

model = create_model("GraphNeuralCSI", num_classes=10, input_shape=(20, 30, 3),
                      hidden_dim=64, num_gcn_layers=3, num_heads=4)

CSIModel

Original CNN + Transformer model (backward compatible).

model = create_model("CSIModel", num_classes=10, input_shape=(20, 30, 3),
                      base_channels=32, latent_dim=128)

Custom Model Registration

from wsdp.models import register_model
import torch.nn as nn

class MyModel(nn.Module):
    def __init__(self, num_classes, input_shape, **kwargs):
        super().__init__()
        # Your architecture

    def forward(self, x):
        # x: (B, T, F, A) — complex or real
        # return: (B, num_classes)
        ...

register_model("custom", "MyModel", MyModel)

Inference

`wsdp.predict()`

Run inference on CSI data.

from wsdp import predict

predictions = predict(csi_data, model_path, num_classes=6)

CLI

`wsdp run`

wsdp run ./data/elderAL ./output elderAL --lr 0.001 --epochs 50

`wsdp download`

wsdp download elderAL ./data --email you@example.com --password yourpassword

`wsdp list`

wsdp list -V

Algorithm Library Overview

Category	Algorithm	Function	Key Reference
Denoising	Wavelet	`wavelet_denoise_csi()`	Donoho & Johnstone, 1994
	Butterworth	`butterworth_denoise()`	Butterworth, 1930
	Savitzky-Golay	`savgol_denoise()`	Savitzky & Golay, 1964
Phase Calibration	Linear	`phase_calibration()`	Halperin et al., 2010
	Polynomial	`polynomial_calibration()`	Extension of linear
	STC	`stc_calibration()`	Xie et al., IEEE TWC 2019
	Robust	`robust_phase_sanitization()`	Wang et al., IEEE ICPADS 2012
Normalization	Z-Score	`normalize_amplitude()`	Standard statistical
	Min-Max	`normalize_amplitude()`	Standard statistical
Interpolation	Linear/Cubic/Nearest	`interpolate_grid()`	de Boor, 1978
Features	Doppler	`doppler_spectrum()`	Ali et al., MobiCom 2015
	Entropy	`entropy_features()`	Shannon, 1948
	CSI Ratio	`csi_ratio()`	Halperin et al., 2011
	Tensor Decomposition	`tensor_decomposition()`	Kolda & Bader, SIAM 2009
Detection	Activity	`detect_activity()`	Zhou et al., 2013
	Change Point	`change_point_detection()`	Adams & MacKay, 2007

For more examples, see the examples/ directory.

WSDP API Reference

Table of Contents

Core Pipeline

wsdp.pipeline()

Data Structures

CSIData

CSIFrame

Readers

Algorithms

Denoising

wavelet_denoise_csi()

butterworth_denoise()

savgol_denoise()

Phase Calibration

phase_calibration() (Linear)

polynomial_calibration()

stc_calibration()

robust_phase_sanitization()

Amplitude Processing

normalize_amplitude()

remove_outliers()

Interpolation

interpolate_grid()

Feature Extraction

doppler_spectrum()

entropy_features()

csi_ratio()

tensor_decomposition()

Activity Detection

detect_activity()

change_point_detection()

Visualization

plot_csi_heatmap()

plot_denoising_comparison()

plot_phase_calibration()

Unified API

denoise()

calibrate()

normalize()

interpolate()

extract_features()

Pluggable Architecture

Algorithm Registry

register_algorithm()

get_algorithm()

list_algorithms()

is_registered()

algorithm_info()

unregister_algorithm()

Configuration Files

YAML Format

JSON Format

Preset Reference

load_config()

save_config()

Pipeline Presets

Built-in Presets

apply_preset()

execute_pipeline()

register_preset()

list_presets()

Custom Algorithms

Models

Unified Model API

Model Registry

Baseline Models

MLPModel

CNN1DModel

CNN2DModel

LSTMModel

Mainstream Models

ResNet1D

ResNet2D

BiLSTMAttention

EfficientNetCSI

SOTA Models

VisionTransformerCSI

MambaCSI

GraphNeuralCSI

CSIModel

`wsdp.pipeline()`

`CSIData`

`CSIFrame`

`wavelet_denoise_csi()`

`butterworth_denoise()`

`savgol_denoise()`

`phase_calibration()` (Linear)

`polynomial_calibration()`

`stc_calibration()`

`robust_phase_sanitization()`

`normalize_amplitude()`

`remove_outliers()`

`interpolate_grid()`

`doppler_spectrum()`

`entropy_features()`

`csi_ratio()`

`tensor_decomposition()`

`detect_activity()`

`change_point_detection()`

`plot_csi_heatmap()`

`plot_denoising_comparison()`

`plot_phase_calibration()`

`denoise()`

`calibrate()`

`normalize()`

`interpolate()`

`extract_features()`

`register_algorithm()`

`get_algorithm()`

`list_algorithms()`

`is_registered()`

`algorithm_info()`

`unregister_algorithm()`

`load_config()`

`save_config()`

`apply_preset()`

`execute_pipeline()`

`register_preset()`

`list_presets()`

`wsdp.predict()`

`wsdp run`

`wsdp download`

`wsdp list`