Year: 2022
Type: dissertation
Abstract: This thesis introduces the spectro-computational complexity (SC) analysis of communication signal algorithms. The SC analysis stems from a novel mathematical model that unifies indicators of performance of the information theory, such as capacity, throughput, and spectral efficiency (SE), along with indicators of the theory of computational complexity (CC), such as processing runtime, the number of computational instructions and asymptotic lower bound. The proposed analysis exploits the fact that both classes of performance indicators can be written as a function of the signal’s spectral bandwidth W. This thesis defines the SC throughput SC(W) of a communication signal algorithm as the ratio between the number of modulated bits B(W) and the CC T (W) required to turn the bits into the signal. Thus, an asymptotic analysis on W reveals the necessary CC (or SE improvement) to prevent the nullification of SC(W) as W grows. Based on SC(W), novel definitions derived from and homologous to concepts of the CC and information theory fields are defined, such as SC capacity, SC efficiency, optimal SC algorithm, and comp-limited signals (from computation-limited, as a reference to the homologous power- and band-limited signal regimes). With the assistance of the SC analysis, novel capacity-complexity scaling laws are presented for Orthogonal Frequency Division Multiplexing (OFDM) and some variants. In a case study, an optimal mapper for OFDM with Index Modulation (IM) was designed, implemented, and evaluated. Such ideal mapper reaches the maximal SE gain over OFDM but its complexity had been considered computationally intractable by the OFDM-IM literature. Based on the SC analysis, the exact asymptotic complexity required to sustain the maximal SE gain was demonstrated as linear (rather than exponential) on the number of subcarriers. In other case study, it was shown that the complexity of the FFT algorithm causes the throughput of OFDM to nullify on the number of subcarriers N. The FFT constraint that N must grow as a power of two 2^i (i > 0) translates into an exponential complexity on i as spectrum widens. In general, it is shown that the throughput of FFT-based waveforms (e.g. OFDM) nullifies on N unless the lower bound complexity of the Fourier transform problem verifies as linear on N, which implies in faster-than-FFT algorithms and remains an open question in computer science. Based on the SC analysis, an algorithm is proposed to an alternative formulation in which an N -point transform is replaced by several smaller transforms. This formulation is employed by OFDM in its vectorized form (V-OFDM) in order to mitigate the cyclic prefix overhead of OFDM. Thus, the proposed algorithm can replace FFT in V-OFDM. By relaxing the power of two constraint of N and, with the parameterization of the number of smaller transforms, it was verified that the complexity of the proposed algorithm can grow linearly on N rather than exponentially on i. The presented case studies illustrate how the SC analysis can guide waveform designers towards the optimal balance between capacity and complexity in the extremely large signals expected for future wireless networks.
Author: Queiroz, Saulo Jorge Beltrão de
Institution: University of Coimbra
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
Awards: Segurança em Redes Wireless através de Codificação e Jamming, Agregação de Espetro entre Bandas Licenciadas de Acesso Exclusivo e Partilhado em Redes Heterogéneas com Células Pequenas, G4724859719, G491858407, G5439335012, G5454278309, G5677667711, G572724779, G6671678162, G685756494, G715306151, G7295230126, G7353418206, G7852741794
OpenAlex ID: W7139543210
