Replication Data for: Multi-Dimensional Wireless Signal Identification Based on Support Vector Machines

The dataset includes spectral correlation function (SCF) estimations by FFT accumulation method (FAM) for totally 4500 signals with 20000 I/Q samples (but only 16384 samples are used). The signals belong to three different cellular communication standards: GSM, WCDMA, and LTE. The signals have been received from different channels with multipath, fading, and noise. Furthermore, the dataset provides other features such as Fast Fourier Transform (FFT), Autocorrelation (ACF), and Power Spectral Density (PSD) in linear scale. The dataset can be used to validate the designed classifier model aiming to identify cellular communication signals. For each signal, the dimension of SCF estimate (alpha domain profile maximizing over spectral frequency) is 1*32769. There are four train sets which must be used together (SCF_train1.mat, SCF_train2.mat, SCF_train3.mat, and SCF_train4.mat). Four train sets for each feature have 3000 signals totally, and two test sets for each feature have 1500. The label of the cellular communication standards are given in dataset as follows: WCDMA -> 0 LTE -> 1 GSM -> 2 The compressed file includes: 1. ACF Folder 2. FFT Folder 3. PSD Folder 4. SCF Folder Each folder above consists of two folder: Test and Train. The test set is located in the Test folder as two parts and the train set is located in the Train folder as four parts. The contents of .mat files: training_class_k : denotes class labels corresponding to the training_data_k, its dimension is 750*1 double training_data_k : includes the kth quarter of the training data, its dimension is 750*32769 double test_class_k : denotes class labels corresponding to the test_data_k, its dimension is 750*1 double test_data_k : includes the kth half of the test data, its dimension is 750*32769 double The dataset has been used for the paper "Multi-Dimensional Wireless Signal Identification Based on Support Vector Machines" submitted for possible publication in IEEE Access. Please cite this paper, if you use the dataset.

本数据集包含基于快速傅里叶变换累积法（FFT Accumulation Method, FAM）计算得到的频谱相关函数（Spectral Correlation Function, SCF）估计值，共计涵盖4500个信号，每个信号原始包含20000个同相/正交（I/Q）采样点，但实际仅使用其中16384个采样点。
上述信号分属三种不同的蜂窝通信标准：全球移动通信系统（Global System for Mobile Communications, GSM）、宽带码分多址（Wideband Code Division Multiple Access, WCDMA）以及长期演进（Long Term Evolution, LTE）。所有信号均通过包含多径、衰落与噪声的不同信道接收。
此外，本数据集还提供线性尺度下的多项特征，包括快速傅里叶变换（Fast Fourier Transform, FFT）、自相关函数（Autocorrelation Function, ACF）与功率谱密度（Power Spectral Density, PSD）。该数据集可用于验证面向蜂窝通信信号识别的分类器模型。
针对每个信号，其频谱相关函数估计值（在频谱频率维度上取最大值得到的α域轮廓）的维度为1×32769。
训练集共分为四部分，需联合使用，对应文件分别为SCF_train1.mat、SCF_train2.mat、SCF_train3.mat与SCF_train4.mat。各特征对应的训练集总计包含3000个信号，而各特征对应的测试集则包含1500个信号。
数据集内的蜂窝通信标准标签对应关系如下：WCDMA → 0，LTE → 1，GSM → 2。
压缩包包含以下内容：
1. ACF文件夹
2. FFT文件夹
3. PSD文件夹
4. SCF文件夹
上述每个文件夹均包含两个子文件夹：Test（测试集）与Train（训练集）。测试集以两部分形式存储于Test文件夹内，训练集以四部分形式存储于Train文件夹内。
.mat文件的内容说明如下：
- training_class_k：表示与training_data_k对应的类别标签，维度为750×1的双精度数组
- training_data_k：包含第k个训练数据四分之一子集，维度为750×32769的双精度数组
- test_class_k：表示与test_data_k对应的类别标签，维度为750×1的双精度数组
- test_data_k：包含第k个测试数据半子集，维度为750×32769的双精度数组
本数据集曾用于论文《基于支持向量机的多维无线信号识别》（Multi-Dimensional Wireless Signal Identification Based on Support Vector Machines），该论文已提交至IEEE Access以待发表。若您使用本数据集，请引用该论文。