decode-wfs — GitHub

Decode binary .WFS (Waveform Stream) files produced by Physical Acoustics AEwin64 / EasyAE acoustic emission software into numpy arrays.

The AEwin64 manual describes a format using message ID 173 (0xAD), but real streaming captures use ID 174 (0xAE) with a slightly different layout. This library handles the empirically observed streaming format and has been verified against real captures. Voltage values are recovered with the calibration factor used by AEwin64's own ASCII waveform exporter (raw_counts / 16384).

Installation

pip install decode-wfs

Requires Python 3.10+ and numpy.

Quick start

decode_wfs() — full access to every record

from decode_wfs import decode_wfs

wfs = decode_wfs("capture.wfs")

print(f"Sample rate : {wfs.sample_rate_hz:,} Hz")
print(f"Channels    : {wfs.channels()}")
print(f"Records     : {len(wfs.waveforms)}")

# Iterate individual waveform records
for rec in wfs.waveforms:
    print(rec.channel, rec.samples.shape, rec.samples.max())

# Time axis for each record (includes pretrigger offset when present)
t = wfs.waveforms[0].time_axis_s   # 1-D float64 array in seconds

# Stack all records from channel 1 into a 2-D array (records × samples)
arr = wfs.to_array(channel=1)      # shape (n_records, n_samples), float64 volts

wfs_to_numpy() — one-liner for a 2-D array

from decode_wfs import wfs_to_numpy

# shape: (n_waveforms, n_samples_per_waveform), dtype float64, units volts
arr = wfs_to_numpy("capture.wfs", channel=1)

# Optionally return the per-waveform time axis as well
arr, t = wfs_to_numpy("capture.wfs", channel=1, return_time_axis=True)

load_continuous() — single concatenated time series

Reconstructs the continuous stream from all records, preserving absolute timing when the file contains per-record position pointers.

from decode_wfs import load_continuous

samples, time, sr = load_continuous("capture.wfs", channel=1)
# samples: 1-D float64 array, volts
# time   : 1-D float64 array, seconds (negative at pretrigger)
# sr     : sample rate in Hz

# Example: compute a power spectrum
import numpy as np
freqs = np.fft.rfftfreq(len(samples), d=1 / sr)
psd   = np.abs(np.fft.rfft(samples)) ** 2

API summary

Logging

decode_wfs uses the standard logging module under the logger name decode_wfs. To see progress messages:

import logging
logging.basicConfig(level=logging.INFO)

CLI

decode-wfs <file.wfs> [--channel N] [--max-records N]

Example:

$ decode-wfs capture.wfs --channel 1

File          : capture.wfs
Sample rate   : 1,000,000 Hz  (1.000 MSPS)
Sample format : ADT=2  (16-bit signed samples)
Setup blocks  : 2
Pretrigger    : -256 samples
Total records : 512
Channels      : [1, 2]

  [0] ch=1  n=4096 samples  min=-0.03125000 V  max=0.04394531 V  std=0.00201416 V
  [1] ch=1  n=4096 samples  min=-0.02832031 V  max=0.03955078 V  std=0.00198364 V
  ...

numpy array shape : (256, 4096)  dtype=float64
Done.

Notes

• Voltage calibration — ADC counts are divided by 16384 to match the voltages in AEwin64's ASCII waveform exports. This factor was determined empirically and may not hold for all hardware configurations.
• Pretrigger — when a pretrigger window is configured, time_axis_s and the time axis from load_continuous start at a negative time offset so that t=0 corresponds to the trigger point.
• Multi-channel files — records from all channels are interleaved in the file. Use the channel parameter to filter, or iterate WFSFile.waveforms and check rec.channel.
• Large files — pass max_records=N to any function to stop after reading N waveform records, useful during development.