Enhanced Signal Area Estimation for Spectrum-Aware Systems based on Image Processing Techniques

Abstract

In many practical application scenarios, radio communication signals are very frequently represented as spectrograms, which represent the received signal strength measured at multiple discrete time instants and frequency points within a specific time interval and frequency band, respectively. Radio spectrograms have been used for time-frequency signal analysis in spectrum-aware systems for many purposes. An important aspect in the processing of radio spectrograms is the region that each individual radio transmission or signal component occupies in the time-frequency domain within the spectrogram, which in this research is referred to as Signal Area (SA). The concept of SA precisely determines the occupied bandwidth and the start/end times of each individual radio transmission. The capability to obtain this information accurately from a radio spectrogram can be useful in many practical applications, including spectrum surveillance (both for enforcement of spectrum regulations and gathering of signal intelligence in military applications), radio signal interception and identification, electronic warfare and radio environment spectral awareness (for instance, in databases for spectrum sharing systems). Consequently, the process of Signal Area Estimation (SAE), which entails determining the subsets of spectrogram points that belong to one or more SAs, is an important function is spectrum-aware wireless communication systems.

The interest of this research is on how to accurately determine the SAs present in a radio spectrogram obtained from empirical spectrum measurements based on the application of techniques from the field of image processing. Image processing techniques can be employed to this end by treating spectrograms as images, where each time-frequency point in the spectrogram is seen as an image pixel. This point of view converts the problem of SAE in a noisy spectrogram into the problem of recognising rectangular shapes in a noisy image, for which several image processing techniques can be employed. The unique characteristics of the SAE problem in radio spectrograms require a tailored research study. In this context, the main aim of this thesis is to explore the feasibility of using image processing techniques to enhance the accuracy of SAE, and to propose novel SAE methods based on image processing techniques. To this end, a broad range of relevant techniques from the field of image processing are explored, including morphologic operations, edge detection and flood fill techniques, the Hough transform along with other heuristic methods and solutions based on deep learning techniques for the processing of images. The proposed methods are evaluated both with software simulations and using an experimental hardware platform specifically built to this end. The obtained simulation and experimental results show that the methods proposed in this thesis can provide significant accuracy improvements compared to other SAE methods in the existing literature.