Using Echo State Networks to Characterise Wireless Channels

We propose the use of echo state networks for the task of wireless channel characterisation to select the most similar channel to the current observed channel from a pre-defined set, based entirely on received signal information. This allows the system to select the optimal resource allocation scheme and transmission parameters from pre-computed solutions. Using suitable training data, the neural network was able to learn to characterise a signal correctly 68% of the time, which can be further improved to 72% by adding some simple location data to the signals being examined. Our system out- performs a comparable statistical method by a factor of two, demonstrating echo state networks' ability to infer information from their training data which other systems can not.