DATE:
2015-05-22
UNIVERSAL IDENTIFIER: http://hdl.handle.net/11093/413
SUPERVISED BY: López Valcarce, Roberto
UNESCO SUBJECT: 1203.21 Sistemas de Navegación y Telemetría del Espacio
DOCUMENT TYPE: doctoralThesis
ABSTRACT
The incessant demand for enhanced communication services has rendered current RF spectrum allocation policies obsolete. The astronomic price of licensed channels stands in sharp contradiction with their heavy underutilization and propels the adoption of novel spectrum management paradigms where spectral resources are assigned in a dynamic fashion, possibly respecting hierarchical relations among multiple classes of users. In those scenarios, it is expected that emerging communication systems will leverage spectrum awareness information to drastically improve spectrum usage eficiency.
Spectrum sensing comprises a collection of signal processing procedures intended to characterize spectrum occupation along time, frequency and space based on the observations of the RF environment reported by one or more sensors. It is the purpose of the present thesis to contribute to this field by putting forward a number of statistical methods that capitalize on the special features of different communication scenarios to reliably obtain detailed occupancy information at low implementation costs.
We first propose a family of techniques aimed to detect, relying on the noisy observations of a sensor with one or multiple antennas, the presence of constant magnitude and/or bandlimited transmissions in a frequency band of interest, as motivated by the new regulations of the Federal Communications Commission. Next, we address the problem of detecting constant-magnitude and Gaussian-distributed waveforms in time-varying channels, relevant in this context since spectrum sensing algorithms typically require long observation windows to meet the stringent performance requirements enforced by spectrum regulations in low-SNR conditions.
We then apply sub-Nyquist acquisition techniques to characterize the occupancy state of a wide frequency band via inexpensive sensing architectures with minimal computational resources, where spectral prior information typically available in practice -- e.g. spectral masks, roll-off factors, etc. --, is exploited to enable compression. To minimize the sampling rate, we extensively analyze the general problem of recovering second-order statistical information of wide-sense stationary processes from compressed measurements. Our results in this direction are of application well beyond spectrum sensing contexts.
Finally, we look at the problem of spectrum cartography, where the goal is to construct power spectrum maps characterizing the spectrum utilization not only along frequency and time, but also across space. We propose several methods capable of learning those maps based on the highly compressed observations reported by a collection of inexpensive sensors. A presente tese de doutoramento enmárcase no contexto de sistemas de sensado espectral con especial atención a esquemas de acceso dinámico ao espectro. Baixo este último termo englóbanse aqueles sistemas que pretenden incrementar a e ciencia de uso dos recursos espectrais existentes mediante unha flexibilización das políticas empregadas para a súa asignación. Nos esquemas empregados na actualidade, a autoridade reguladora do espectro electromagnético asigna bandas de frecuencia a grandes operadores, tipicamente mediante un sistema de poxas multimillonarias. Paradoxalmente, recentes campañas de medidas puxeron de manifesto o elevado grado de infrautilización que impera nestas bandas con licenza. Na vista destes resultados, varias comunidades de expertos suxiren considerar a opción de que usuarios sen licenza exploten estes baleiros espectrais para as súas actividades de comunicación a condición de que non degraden a calidade da comunicación do usuario con licenza cando este último accede ao canal. Por esta e outras razóns, tórnase necesario investigar mecanismos que permitan aos chamados usuarios secundarios coñecer, en cada momento, o estado do canal. O problema principal radica en empregar un conxunto de observacións do canal, proporcionadas por un sensor espectral, para dar resposta a unha ou máis preguntas do tipo: hai algún usuario primario a operar na banda b? se é así, con que potencia se recibe? Para proporcionar tal resposta, un riguroso conxunto de ferramentas estatísticas ponse en funcionamento para crear funcións dos sinais observados que devolven decisións que satisfacen certos requirimentos, tamén formulados en termos estatísticos. O noso abano inclúe metodoloxías importadas da teoría da detección, como o cociente xeneralizado de verosimilitudes ou a procura de tests invariantes, da teoría da estimación, como os estimadores de máxima verosimilitude, ou da teoría de aprendizaxe estatístico, como a máquina de vectores soporte.