Automatic Modulation Classification for low-power IoT applications

Abstract

The Internet of Things (IoT) has swiftly become one of the most important technologies in recent years. Radio spectrum access represents a stern challenge for the IoT as a consequence of the increased use of connected devices. This is particularly true for IoT devices operating in the unlicensed band where the huge demand for wireless connectivity will require techniques that use the spectrum efficiently. Avoiding training sequences enables a more efficient spectrum usage and has the additional advantage of reducing the power consumption of IoT devices, but it requires modulation identification mechanisms. This paper presents a simple yet efficient method to classify received signals according to their modulation type. We propose the application of a single hidden layer neural network with a small number of trainable parameters for performing the classification between seven different modulation types. The designed classifier achieves a maximum accuracy of 95% when the signal-to-noise ratio (SNR) of the input data is 12 dB, and in the presence of multi-path fading, sample rate offset and carrier frequency offset.