File size: 6,021 Bytes
c89047a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
130
131
132
133
134
135
136
137
138
139
140
141
142
"""
Continuous-Time Photonic Time-Stretch ADC (CTS-PADC) Dataset
Data inspection, loading, and visualization script.
Repository: https://huggingface.co/datasets/OCEAN-Lab/CTS-PADC_Dataset
"""

import json
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

# =====================================================================
# Constants & Acquisition Specifications
# =====================================================================
VECTOR_LEN = 400         # Sequence length (temporal samples per vector)
STRETCH_FACTOR = 3.0     # Nominal photonic stretch factor (S ≈ 3)
SAMPLE_RATE_GSPS = 1.0   # Digitizer sampling rate (1 GSa/s post-stretch)


# =====================================================================
# 1. Data Parsing Utilities
# =====================================================================
def parse_waveform_series(series: pd.Series, length: int = VECTOR_LEN) -> np.ndarray:
    """
    Decodes string-encoded JSON arrays from CSVs or extracts lists from Parquet/Hugging Face.
    """
    first_item = series.iloc[0]
    if isinstance(first_item, str):
        return np.array([json.loads(row)[:length] for row in series], dtype=np.float32)
    elif isinstance(first_item, (list, np.ndarray)):
        return np.array([row[:length] for row in series], dtype=np.float32)
    else:
        raise TypeError(f"Unexpected data type in series: {type(first_item)}")


# =====================================================================
# 2. Loading Methods
# =====================================================================
def load_via_huggingface(split: str = "train"):
    """
    Downloads and loads the dataset directly using the Hugging Face `datasets` library.
    Install dependency: pip install datasets
    """
    try:
        from datasets import load_dataset
    except ImportError:
        raise ImportError("Run `pip install datasets` to fetch data directly from Hugging Face.")

    print(f"Loading '{split}' partition from Hugging Face (OCEAN-Lab/CTS-PADC_Dataset)...")
    dataset = load_dataset("OCEAN-Lab/CTS-PADC_Dataset", split=split)
    df = pd.DataFrame(dataset)
    return df


def load_via_local_csv(csv_path: str):
    """
    Loads dataset directly from a local CSV file.
    """
    print(f"Loading data from local path: {csv_path}...")
    df = pd.read_csv(csv_path)
    return df


# =====================================================================
# 3. Diagnostic Visualization (Time & Frequency Domain)
# =====================================================================
def plot_sample_diagnostics(df: pd.DataFrame, sample_index: int = 0, save_path: str = "sample_inspection.png"):
    """
    Plots the distorted input waveform versus the ground truth continuous-wave target
    in both the time and frequency domains.
    """
    record = df.iloc[sample_index]

    # Extract metadata
    rf_freq = record["Frequencies"]
    rf_power = record["Power"]
    expected_freq = record.get("expected_recovered_freq", rf_freq / STRETCH_FACTOR)

    # Parse temporal vectors
    raw_distorted = record["arcsin_results"]
    if isinstance(raw_distorted, str):
        x_distorted = np.array(json.loads(raw_distorted)[:VECTOR_LEN])
        y_ground_truth = np.array(json.loads(record["test_data"])[:VECTOR_LEN])
    else:
        x_distorted = np.array(raw_distorted[:VECTOR_LEN])
        y_ground_truth = np.array(record["test_data"][:VECTOR_LEN])

    # Time and frequency axes
    time_ns = np.arange(VECTOR_LEN) / SAMPLE_RATE_GSPS  # Time in nanoseconds
    freqs_mhz = np.fft.rfftfreq(VECTOR_LEN, d=1.0 / (SAMPLE_RATE_GSPS * 1e9)) / 1e6  # Frequency in MHz

    # FFT Spectra (in dB)
    fft_distorted = 20 * np.log10(np.abs(np.fft.rfft(x_distorted)) + 1e-6)
    fft_ground_truth = 20 * np.log10(np.abs(np.fft.rfft(y_ground_truth)) + 1e-6)

    # Plot
    fig, axes = plt.subplots(1, 2, figsize=(14, 4.5))

    # Time-Domain
    axes[0].plot(time_ns, y_ground_truth, label="Ground Truth (Ideal CW)", color="black", linestyle="--", linewidth=1.2)
    axes[0].plot(time_ns, x_distorted, label="Distorted Input (arcsin_results)", color="crimson", alpha=0.75, linewidth=1.0)
    axes[0].set_title(f"Time-Domain Waveform\n($f_{{RF}} = {rf_freq}$ GHz, $P_{{in}} = {rf_power}$ dBm)", fontsize=11)
    axes[0].set_xlabel("Time (ns)", fontsize=10)
    axes[0].set_ylabel("Amplitude (V)", fontsize=10)
    axes[0].grid(True, linestyle=":", alpha=0.6)
    axes[0].legend(loc="upper right", frameon=True)

    # Frequency-Domain
    axes[1].plot(freqs_mhz, fft_ground_truth, label="Ground Truth Target", color="black", linestyle="--", linewidth=1.2)
    axes[1].plot(freqs_mhz, fft_distorted, label="Distorted Spectrum", color="crimson", alpha=0.75, linewidth=1.0)
    axes[1].axvline(expected_freq * 1e3, color="blue", linestyle=":", label=f"Stretched Carrier ({expected_freq * 1e3:.1f} MHz)")
    axes[1].set_title(f"Power Spectrum (FFT)\nExpected Peak: ~{expected_freq * 1e3:.1f} MHz", fontsize=11)
    axes[1].set_xlabel("Frequency (MHz)", fontsize=10)
    axes[1].set_ylabel("Magnitude (dB)", fontsize=10)
    axes[1].grid(True, linestyle=":", alpha=0.6)
    axes[1].legend(loc="upper right", frameon=True)

    plt.tight_layout()
    plt.savefig(save_path, dpi=300)
    print(f"Saved diagnostic plot to: {save_path}")
    plt.show()


# =====================================================================
# 4. Main Entry Point
# =====================================================================
if __name__ == "__main__":
    # Set to True if reading from downloaded CSV files locally
    USE_LOCAL_CSV = False

    if USE_LOCAL_CSV:
        df = load_via_local_csv("training_data.csv")
    else:
        # Stream / download directly from Hugging Face
        df = load_via_huggingface(split="train")

    print("\nDataset Summary:")
    print(f"Total Rows: {len(df):,}")
    print(f"Columns: {list(df.columns)}")

    # Plot sample 0
    plot_sample_diagnostics(df, sample_index=0, save_path="padc_sample_plot.png")