File size: 2,987 Bytes
c89047a
 
 
fccfb5a
 
c89047a
 
 
 
 
 
 
 
 
 
 
 
 
fccfb5a
 
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
---
language:
- en
license: cc-by-sa-4.0

task_categories:
- time-series-forecasting
- tabular-regression
tags:
- photonics
- time-stretch-adc
- signal-processing
- nonlinear-equalization
- optical-communications
size_categories:
- 100K<n<1M
---



# Continuous-Time Photonic Time-Stretch ADC (CTS-PADC) Dataset

An open-access experimental dataset capturing real-world optoelectronic non-idealities for machine-learning-based nonlinear equalization and denoising in continuous-time photonic time-stretch ADCs.

- **Repository:** [OCEAN-Lab/CTS-PADC_Dataset](https://huggingface.co/datasets/OCEAN-Lab/CTS-PADC_Dataset)
- **Paper:** *An Open Experimental Dataset for Machine Learning-Based Nonlinear Equalization and Denoising in Continuous-Time Photonic ADCs* (OFC 2027)

---

## 1. Dataset Description

Photonic Time-Stretch Analog-to-Digital Converters (PTS-ADCs) overcome electronic clock-jitter limits by stretching wideband RF signals across dispersive fiber before electronic digitization ($S \approx 3$). 

However, practical physical frontends suffer from:
1. **Modulator Non-linearity:** Mach--Zehnder Modulator (MZM) harmonic distortions.
2. **Clock Mixing:** Intermodulation spurs with the 50-MHz Mode-Locked Laser (MLL) repetition rate.
3. **Channel Skew:** Filter mismatch across the 4-channel wavelength demultiplexing network.
4. **Polarization Fluctuation:** Time-varying amplitude variations in standard fiber.

This dataset provides paired distorted experimental captures and ideal ground-truth waveforms to allow researchers to benchmark their own digital signal restoration algorithms (e.g., Transformers, CNNs, State-Space Models, or Volterra filters).

---

## 2. Dataset Composition

Each record contains a paired vector of **400 temporal samples** digitized at an effective stretched rate of **1 GSa/s**.

| Partition | Frequency Grid | Power Levels | Replications | Vector Count |
| :--- | :--- | :--- | :--- | :--- |
| **In-Distribution (Train/Val)** | 0.6–1.4 GHz (5.1 MHz step, 143 points) | -20 to +14 dBm (2 dB step, 18 levels) | 300 per point | **772,200** |
| **Out-of-Distribution (OOD Test)**| 3 Unseen Frequencies (0.8, 0.9, 1.3 GHz) | All 18 power levels | 300 per point | **16,200** |
| **Total** | | | | **788,400** |

---

## 3. Data Schema

| Field Name | Type | Description |
| :--- | :--- | :--- |
| `S.Numbers` | `int64` | Capture index identifier |
| `Frequencies` | `float32` | Analog input RF frequency (GHz) |
| `Power` | `float32` | RF input power (dBm) |
| `arcsin_results` | `list[float]` | **Model Input ($x$):** 400 temporal samples from differential arcsine recovery |
| `test_data` | `list[float]` | **Target ($y$):** 400 temporal samples of the ideal stretched continuous wave |
| `expected_recovered_freq`| `float32` | Expected fundamental frequency after stretching ($f_{\text{RF}}/S$ in GHz) |

---

## 4. How to Access and Inspect the Data

### Step 1: Install Dependencies
```bash
pip install datasets pandas numpy matplotlib