Enhancing User Detection

Abstract

The internet is all around us. From our cell phones to our doorbells, almost everything is connected to the internet. With the number of devices accessing the internet surpassing the number of people on the earth, it has become critical to build wireless infrastructures that can provide a high quality experience to a large number of simultaneous users. An important task performed by cellular systems is the detection of active users at any given time based on a preamble sequence transmitted by each active user. In this thesis, we study two aspects related to user detection in cellular systems - (i) the performance of neural network based Learned Iterative Soft Thresholding Algorithm (ISTA) compared to traditional baselines using approximate message passing and Iterative Soft Thresholding Algorithm, and (ii) the impact of utilizing two receivers that work together to recover a signal from a user, compared to a single receiver approach. At its core, this research is focused on compressed sensing and sparse signal recovery. With a focus on user detection, performance of the algorithms was measured using mean squared error (MSE) as well as calculating the probability of misdetection and false alarms. Our findings show that learned ISTA outperforms ISTA and Approximate Message Passing (AMP) in terms of detection performance. For each of the algorithms examined, our findings show that using two receivers instead of one also improved the performance.