Data for Self-Supervised Learning with Noisy Dataset for Rydberg Microwave Sensors Denoising

# File Documentation

This document describes the structure and content of all files in the current path.

## Directory Structure

```
figure/
├── Figure2/
│   ├── 100mV.csv
│   ├── 200mV.csv
│   ├── 3mV.csv
│   └── 4mV.csv
├── Figure3/
│   └── size.csv
├── Figure4/
│   ├── a.csv
│   └── b.csv
├── Figure5/
│   ├── frequency.csv
│   └── time.csv
└── ReadMe.md
```

---

## File Details

### Figure2/ Directory

This directory contains frequency-domain denoising results under different voltage conditions.

#### 1. `Figure2/100mV.csv`
**Path**: `Figure2/100mV.csv`

**Column Descriptions**:
- `frequency`: Frequency values (unit: Hz), starting from 49000, with step size of approximately 2 Hz
- `noisy`: Noisy signal values (unit: dBm)
- `denoised`: Denoised signal values (unit: dBm)
- `average`: Average reference signal values (unit: dBm)

**Data Characteristics**: Contains approximately 1000 rows of data, frequency range approximately 49000-51000 Hz

---

#### 2. `Figure2/200mV.csv`
**Path**: `Figure2/200mV.csv`

**Column Descriptions**:
- `frequency`: Frequency values (unit: Hz), starting from 49000, with step size of approximately 2 Hz
- `noisy`: Noisy signal values (unit: dBm)
- `denoised`: Denoised signal values (unit: dBm)
- `average`: Average reference signal values (unit: dBm)

**Data Characteristics**: Contains approximately 1000 rows of data, frequency range approximately 49000-51000 Hz

---

#### 3. `Figure2/3mV.csv`
**Path**: `Figure2/3mV.csv`

**Column Descriptions**:
- `frequency`: Frequency values (unit: Hz), starting from 49000, with step size of approximately 2 Hz
- `noisy`: Noisy signal values (unit: dBm)
- `denoised`: Denoised signal values (unit: dBm)
- `average`: Average reference signal values (unit: dBm)

**Data Characteristics**: Contains approximately 1000 rows of data, frequency range approximately 49000-51000 Hz

---

#### 4. `Figure2/4mV.csv`
**Path**: `Figure2/4mV.csv`

**Column Descriptions**:
- `frequency`: Frequency values (unit: Hz), starting from 49000, with step size of approximately 2 Hz
- `noisy`: Noisy signal values (unit: dBm)
- `denoised`: Denoised signal values (unit: dBm)
- `average`: Average reference signal values (unit: dBm)

**Data Characteristics**: Contains approximately 1000 rows of data, frequency range approximately 49000-51000 Hz

---

### Figure3/ Directory

This directory contains data on the relationship between training set size and MSE (Mean Squared Error) under different voltage conditions.

#### 1. `Figure3/size.csv`
**Path**: `Figure3/size.csv`

**Column Descriptions**:
- `training_size`: Training set size (number of samples)
- `4.6mV`: MSE values under 4.6mV voltage condition
- `4.7mV`: MSE values under 4.7mV voltage condition
- `4.8mV`: MSE values under 4.8mV voltage condition
- `4.9mV`: MSE values under 4.9mV voltage condition

**Data Characteristics**: Contains MSE values corresponding to different training set sizes (140, 420, 840, 1260, 2800, 5600, 9800, 14000, 28000, etc.), used for analyzing the impact of training set scale on model performance

---

### Figure4/ Directory

This directory contains comparison data of denoising results for time-domain signals at voltage 200 mV (a.csv) and 100 mV (b.csv).

#### 1. `Figure4/a.csv`
**Path**: `Figure4/a.csv`

**Column Descriptions**:
- `time`: Time points (unit: seconds), starting from 0.0000000501, with time step of approximately 0.0000001 seconds
- `noisy`: Original noisy signal values
- `transformer`: Signal values after Transformer model denoising
- `wavelet`: Signal values after wavelet transform denoising
- `kalman`: Signal values after Kalman filter denoising
- `mean_reference`: Average reference signal values

**Data Characteristics**: Contains approximately 500 rows of time-domain data, used for comparing the effectiveness of different denoising methods

---

#### 2. `Figure4/b.csv`
**Path**: `Figure4/b.csv`

**Column Descriptions**:
- `time`: Time points (unit: seconds), starting from 0.0000000501, with time step of approximately 0.0000001 seconds
- `noisy`: Original noisy signal values
- `transformer`: Signal values after Transformer model denoising
- `wavelet`: Signal values after wavelet transform denoising
- `kalman`: Signal values after Kalman filter denoising
- `mean_reference`: Average reference signal values

**Data Characteristics**: Contains approximately 500 rows of time-domain data, used for comparing the effectiveness of different denoising methods

---

### Figure5/ Directory

This directory contains denoising result data for samples, divided into frequency-domain and time-domain parts.

#### 1. `Figure5/frequency.csv`
**Path**: `Figure5/frequency.csv`

**File Header Comments**:
- Sample 1 denoising results (latter half)
- Total length: 1000, latter half index: 500, number of samples in file: 500
- Transformer MSE: 0.000067
- U-Net MSE: 0.000245

**Column Descriptions**:
- `Time(s)`: Time points (unit: seconds), starting from 50001.0010010010
- `Raw_Signal`: Original noisy signal values (unit: dBm)
- `Transformer_Denoised`: Signal values after Transformer model denoising (unit: dBm)
- `U-Net_Denoised`: Signal values after U-Net model denoising (unit: dBm)
- `Mean_Reference`: Average reference signal values (unit: dBm)

**Data Characteristics**: Contains approximately 500 rows of data, representing the latter half of Sample 1 denoising results

---

#### 2. `Figure5/time.csv`
**Path**: `Figure5/time.csv`

**File Header Comments**:
- Sample 2 denoising results (latter half)
- Total length: 1000, latter half index: 500, number of samples in file: 500
- Transformer MSE: 0.000717
- U-Net MSE: 0.012521

**Column Descriptions**:
- `Time(s)`: Time points (unit: seconds), starting from 0.0000000501
- `Raw_Signal`: Original noisy signal values
- `Transformer_Denoised`: Signal values after Transformer model denoising
- `U-Net_Denoised`: Signal values after U-Net model denoising
- `Mean_Reference`: Average reference signal values

**Data Characteristics**: Contains approximately 500 rows of data, representing the latter half of Sample 2 denoising results

---

## Data Usage Summary

- **Figure2/**: Comparison of frequency-domain denoising performance under different voltage conditions (3mV, 4mV, 100mV, 200mV)
- **Figure3/**: Analysis of the impact of training set size on model performance under different voltage conditions
- **Figure4/**: Comparison of different denoising methods (Transformer, wavelet transform, Kalman filter) on time-domain signals
- **Figure5/**: Denoising results for two samples, including performance comparison between Transformer and U-Net deep learning models