Self Organizing Feature Maps Data Sets

Self Organzing Feature Maps [SOFM] Data Sets for LoRa transmitters, generated by the Batch SOFM Competitive Learning algorihtm. In the algorithm, initially, a Kohonen layer of artificial neurons (of dimensions 10x10) trains upon the data set of raw LoRa I/Qs through randomly initialized set of weights. The original ANN then 'self-organize' or cluster into a batch of six 'offspring' ANNs at every epoch. Except in the 1st epoch, the algorithm trains on the set of offspring ANNs and not the raw I/Qs. By the 200th epoch, the extent of cluster is sufficient to produce distinct SOFM patterns corresponding specifically to a particular LoRa I/Q in the raw I/Qs. The raw LoRa I/Q data, comprising of samples from six sources [5 from LoRa modules and 1 from an ARB], collected from a customized RF penetration test-bed, are also provided.用于LoRa发射机的自组织特征映射（Self Organizing Feature Maps，SOFM）数据集，由批量自组织特征映射竞争学习算法（Batch SOFM Competitive Learning Algorithm）生成。该算法初始阶段，一个尺寸为10×10的科赫农人工神经元层（Kohonen layer）将基于随机初始化的权重集，在原始LoRa同相正交（In-phase and Quadrature，I/Q）数据集上完成训练。初始人工神经网络随后将进行‘自组织’过程，或在每一轮训练（epoch）中聚类为六个‘子代’人工神经网络。除第一轮训练外，该算法均基于子代人工神经网络集而非原始I/Q数据开展训练。至第200轮训练时，聚类的充分程度已可生成与原始I/Q数据中特定LoRa I/Q一一对应的独特SOFM模式。本数据集同时提供了从定制化射频穿透测试平台采集的原始LoRa I/Q数据，该数据包含六个来源的样本——其中5个来自LoRa模块，1个来自任意波形发生器（Arbitrary Waveform Generator，ARB）。