Blind Frame Syncword Detection Using Deep Neural Networks with Input Linear Filtering

In this paper, the problem of detecting a frame syncword (SW), which is repeated in received signal stream, is addressed. We propose a preprocessing technique for deep neural networks used for estimating the SW. As SW has a specific sequence, consecutive received signals are mapped to a symbol using a linear filter in order to make the neural network train their sequence. The performance of SW detection of multilayer perceptron (MLP) and convolutional neural network (CNN) depending on the frame length and the data preprocessing are evaluated. Simulation results show that CNN has better performance than MLP and data processing improves the performance of CNN.