Modulation classification is the process of classifying a received signal as a particular type of modulation. Wireless data demand is driving a need for improved radio efficiency and one way to achieve this is by automating classification such that no additional parameters are needed in order to demodulate. Traditional modulation recognition methods usually require prior knowledge of signal and channel parameters, which can be inaccurate under mild circumstances and need to be delivered through a separate control channel. Hence, the need for autonomous modulation recognition arises in wireless systems, where modulation schemes are expected to change frequently as the environment changes. This leads to considering new modulation recognition methods using deep neural networks.
Deep learning is a powerful classification technique that is used to classify a given input layer by dissecting its features. It is unique from the broader area of machine learning, because deep learning uses training data to determine the parameters it is exploring, as opposed to the parameters being pre-defined. The input data is fed through several layers that assign weights to determine what group a given input might belong to. Convolution neural networks (CNN) are a type of deep learning known for their performance in feature extraction. CNNs classify by finding correlations using what is known as a “universal approximator” that can learn to approximate an unknown function and classify inputs to outputs.
A friend and I decided to use MATLAB r2020B to generate a CNN for modulation classification. Thousands of signals representing 11 different modulation types (both FM and AM) were generated and used to train our network! Our results indicate that our neural network has an overall validation accuracy of 95.06% with standard channel impairment. We also investigated how our CNN would perform when we increased channel impairment parameters by a factor of 10 which resulted in less validation accuracy for specific types of modulation.
You can find our full report here!
Opmerkingen