Signal Detection Effects on Deep Neural Networks Utilizing Raw IQ for Modulation Classification

Recently, automatic modulation classification techniques using convolutional neural networks on raw IQ samples have been investigated and show promise when compared to more traditional likelihood-based or feature-based techniques. While likelihood-based and feature-based techniques are effective, making classification decisions directly on the raw IQ samples allows for reduced system complexity and removes the need for expertly crafted transformations and feature extractions. In practice, RF environments are typically very dense, and a receiver must first detect and isolate each signal of interest before classification can be performed. The errors introduced by this detection and isolation process will affect the accuracy of convolutional neural networks making automatic modulation classification decisions using only raw IQ samples. To quantify this impact, a representative convolutional neural network designed to distinguish between 8 modulation classes (2FSK, 4FSK, 8FSK, BPSK, QPSK, 8PSK, 16QAM, and 64QAM), over a generalized parameter set, is analyzed. The classification accuracy of this neural network is investigated as a function of errors in carrier frequency estimation and errors in sample rate estimation. The importance of defining upper limits on frequency and sample rate estimation errors in a detector is highlighted, and the negative effects of over-estimating or under-estimating these limits is explored.