Summary:

Internet of Things (IoT) technology has become ubiquitous in many applications and its use is expanding very rapidly. However, the expansion of the IoT faces a significant scalability challenge, i.e. the very dense deployment of communicating devices is currently limited. This thesis focuses on providing solutions to the problems associated with the scalability of LoRa-type networks in both directions of communication, uplink and downlink. In long-range networks, such as LoRa, the downlink is critical as it limits the number of acknowledgements that can be sent, and therefore reliability. It also limits the ability to update devices, which could be critical when they are deployed for decades. To overcome these problems, we propose a solution, inspired by non-orthogonal multiple access (NOMA) techniques, to increase by at least an order of magnitude the number of devices that can be addressed. While the approach differentiates devices by the power allocated to them, it differs from the vast majority of previous work on the NOMA power domain in that it does not require interference cancellation. Instead, it takes advantage of the chirp spread spectrum of the modulation scheme, where, at the end of the decoding phase, the information carried by a symbol is found at the position of a peak in the Fourier domain. In most cases, the information from different users results in different peak positions, creating no interference. In this sense, we are approaching avoidance schemes such as time or frequency hopping, but without using code. This thesis proposes a new solution based on NOMA in the power domain that does not suffer from the limitations induced by interference cancellation residuals. The proposed scheme, including preamble detection and channel estimation, is presented and evaluated by simulations. We show that our scheme increases the number of connected devices by an order of magnitude compared to the current system, while maintaining full compatibility with the LoRa physical layer standard. On the uplink side, we present a new receiver capable of demodulating multiple users transmitted simultaneously on the same frequency channel with the same spreading factor. From the point of view of non-orthogonal multiple access, it is based on the power domain and uses serial interference cancellation. Simulation results show that the receiver allows a significant increase in the number of devices connected in the network. Finally, this thesis deals with the detection of uplink signals in a LoRa network using an approach based on deep learning. Two strategies are proposed: regression for bit detection based on a deep feedforward neural network and classification for symbol detection based on a convolutional neural network. These receivers can decode the signals of a selected user when several users are transmitting simultaneously on the same frequency band with the same spreading factor. Simulation results show that both receivers outperform conventional LoRa in the presence of interference. In addition, the results show that the approach introduced is relevant for addressing the issue of scalability.

Abstract:

Internet of Things (IoT) technology has become ubiquitous in many applications and its use is growing very rapidly. However, the expansion of the IoT faces a significant scalability challenge, i.e., the very dense deployment of communicating devices is currently limited. This thesis focuses on providing solutions to the problems associated with the scalability of LoRa-like networks in both the uplink and downlink directions of communication. In long-range networks, such as LoRa, the downlink is critical because it limits the number of acknowledgements that can be sent, and therefore, the reliability. It also limits the ability to update devices, which could be critical when they are deployed for decades. To overcome these problems, we propose a solution, inspired by non-orthogonal multiple access (NOMA) techniques, to increase the number of addressable devices by at least an order of magnitude. While the approach differentiates devices by the power allocated to them, it differs from the vast majority of previous work on the NOMA power domain because it does not require interference cancellation. Instead, it takes advantage of the spectral spread of the modulation scheme (chirp spread spectrum), where, at the end of the decoding phase, the information carried by a symbol is found at the position of a peak in the Fourier domain. In most cases, information from different users results in different peak positions, creating no interference. In this sense, we come close to avoidance schemes such as time or frequency hopping, but without using any code. This thesis proposes a new solution based on NOMA in the power domain that does not suffer from the limitations induced by interference cancellation residues. The proposed scheme, including preamble detection and channel estimation, is presented and evaluated by simulations. We show that our scheme increases the number of connected devices by an order of magnitude over the current system, while maintaining full compatibility with the LoRa physical layer standard. Regarding the uplink, we present a new receiver capable of demodulating multiple users transmitted simultaneously on the same frequency channel with the same spreading factor. From the non-orthogonal multiple access point of view, it is based on the power domain and uses serial interference cancellation. Simulation results show that the receiver allows a significant increase in the number of devices connected in the network. Finally, this thesis deals with the detection of signals, in uplink, of a LoRa network through an approach based on deep learning. Two strategies are proposed: regression for bit detection based on a deep feedforward neural network and classification for symbol detection based on a convolutional neural network. These receivers can decode the signals of a selected user when multiple users transmit simultaneously on the same frequency band with the same spreading factor. Simulation results show that both receivers outperform conventional LoRa in the presence of interference. Furthermore, the results show that the introduced approach is relevant to address the scalability issue.