foysalhaque/CSI-BFI-HAR-Dataset

CSI-BFI-HAR Dataset

This repository contains the dataset, structure and usage of the CSI-BFI-HAR dataset of the corresponding dataset paper:

Please download the dataset either from huggingface or IEEE dataport:

• —https://huggingface.co/datasets/foysalhaque/CSI-BFI-HAR-Dataset
• —https://ieee-dataport.org/documents/csi-bfi-har-wi-fi-datasets-human-activity-recognition

Dataset Structure

The dataset is organized into two subsets:

• —Dataset-1: single-subject HAR (HAR-1 to HAR-6)
• —Dataset-2: simultaneous multi-subject HAR (HAR-7 to HAR-12)

Each HAR-* folder contains two subfolders:

• —CSI/: channel state information traces
• —BFI/: beamforming feedback information traces

CSI-BFI-HAR-Dataset/
├── Dataset-1/
│   ├── HAR-1/
│   │   ├── BFI/
│   │   └── CSI/
│   ├── HAR-2/
│   │   ├── BFI/
│   │   └── CSI/
│   ├── HAR-3/
│   │   ├── BFI/
│   │   └── CSI/
│   ├── HAR-4/
│   │   ├── BFI/
│   │   └── CSI/
│   ├── HAR-5/
│   │   ├── BFI/
│   │   └── CSI/
│   └── HAR-6/
│       ├── BFI/
│       └── CSI/
└── Dataset-2/
    ├── HAR-7/
    │   ├── BFI/
    │   └── CSI/
    ├── HAR-8/
    │   ├── BFI/
    │   └── CSI/
    ├── HAR-9/
    │   ├── BFI/
    │   └── CSI/
    ├── HAR-10/
    │   ├── BFI/
    │   └── CSI/
    ├── HAR-11/
    │   ├── BFI/
    │   └── CSI/
    └── HAR-12/
        ├── BFI/
        └── CSI/

HAR Folder Mapping

Trace Naming Convention

Each trace filename follows:

X_D_YY_ZZ

Where:

• —X: activity code (A to T)
• —D: day index (1 to 6)
• —YY: sensing device ID
• —ZZ: subject ID

Device IDs:

• —CSI devices: M1, M2, M3
• —BFI devices: M1, M2, M3

Subject IDs:

• —P1 to P6 for Dataset-1
• —P1 to P3 for Dataset-2

Activity code map:

• —A: jogging
• —B: clapping
• —C: push forward
• —D: boxing
• —E: writing
• —F: brushing teeth
• —G: rotating
• —H: standing
• —I: eating
• —J: reading a book
• —K: waving
• —L: walking
• —M: browsing phone
• —N: drinking
• —O: hands-up-down
• —P: phone call
• —Q: side bends
• —R: check wrist (watch)
• —S: washing hands
• —T: browsing laptop

Examples:

• —A_3_M1_P2 -> jogging, day 3, CSI device M1, subject P2
• —N_6_M2_P1 -> drinking, day 6, BFI device M2, subject P1

CSI Extraction

CSI traces are extracted from Nexmon PCAP files using:

• —CSI-Extraction/Extract_CSI.m

Prerequisites

• —MATLAB (or GNU Octave with compatible MEX support)
• —Files in CSI-Extraction/:
• —Extract_CSI.m
• —readpcap.m
• —plotcsi.m (optional visualization)
• —unpack_float.mexa64 (for Broadcom float unpacking)

Input

• —A Nexmon CSI capture file (.pcap), for example:
• —D_1_M1_P2_short.pcap

Set the file path at the top of Extract_CSI.m:

FILE = '../CSI-Samples/D_1_M1_P2_short.pcap';

Configuration

In Extract_CSI.m, set:

• —CHIP (e.g., 4366c0)
• —BW in MHz (e.g., 80)

For BW = 80, the script keeps valid data subcarriers and removes null/pilot bins, producing 242 CSI subcarriers per packet.

Run

From repository root:

cd CSI-Extraction
matlab -batch "Extract_CSI"

Or run Extract_CSI.m directly from the MATLAB editor while your current folder is CSI-Extraction.

Output

The script saves a .mat file next to the input PCAP, with the same base filename:

• —Input: ../CSI-Samples/D_1_M1_P2_short.pcap
• —Output: ../CSI-Samples/D_1_M1_P2_short.mat

Saved variables:

• —csi: complex CSI matrix of size [num_packets x num_subcarriers] (242 subcarriers for 80 MHz)
• —seq_num: sequence identifier extracted per packet
• —core_num: RF core identifier extracted per packet

BFI Extraction

BFI traces are extracted with Wi-BFI using:

• —Wi-BFI/main.py

Prerequisites

• —Python environment with numpy and pyshark
• —tshark installed and available in PATH (required by pyshark)
• —Wi-BFI files in Wi-BFI/ (main.py, bfi_angles.py, vmatrices.py, utils.py)

Optional setup (from Wi-BFI project files):

cd Wi-BFI
conda env create -f wi-bfi.yml
conda activate wi-bfi

Run (your current command)

From Wi-BFI/:

python main.py ./traces/11ac_MU_3x1_80.pcapng AC MU 3x1 80 b0:b9:8a:63:55:9c 200 V_ac_mu_3x1_80 bfa_ac_mu_3x1_80

Argument meaning

python main.py <file_name> <standard> <mimo> <config> <bw> <MAC> <num_packet_to_process> <saved_vmatrices> <saved_angles>

• —file_name: input pcap/pcapng trace path
• —standard: AC or AX
• —mimo: SU or MU (for this dataset use MU with AC)
• —config: antenna setup (3x1 in your command)
• —bw: bandwidth in MHz (80 in your command)
• —MAC: target beamformee MAC address to filter frames
• —num_packet_to_process: number of packets to parse (must be <= available packets after MAC/filter match)
• —saved_vmatrices: output filename prefix for reconstructed V matrices
• —saved_angles: output filename prefix for extracted beamforming angles

Output

main.py stores two NumPy files (.npy) in the current directory:

• —V_ac_mu_3x1_80.npy: reconstructed V matrices
• —bfa_ac_mu_3x1_80.npy: extracted beamforming feedback angles

For the shown example command with 200 packets and AC MU 3x1 at 80 MHz, output shapes are:

• —V_ac_mu_3x1_80.npy: (200, 234, 3, 1) complex array
• —bfa_ac_mu_3x1_80.npy: (200, 234, 4) integer array

In the 3x1 case, the 4 angles per subcarrier correspond to:

• —phi_11, phi_21, psi_21, psi_31

For any question or query, please contact Foysal Haque (haque.k@northeastern.edu)