Power Consumption in DFTs for OFDM Systems
Authors
Kristensen, Jes Toft ; Simonsen, Peter August
Term
10. term
Education
Publication year
2008
Pages
106
Abstract
Specialet undersøger FFT-algoritmer (Fast Fourier Transform) til OFDM-modtagere (Orthogonal Frequency Division Multiplexing) og deres effektforbrug på en konfigurerbar hardwareplatform. Fokus er på mobile applikationer og kooperativ radio, hvor det ofte kun er nødvendigt at beregne en del af det modtagne frekvensspektrum. Det kan udnyttes af specialiserede FFT-varianter, der beregner et delmængdespektrum (SFFT), hvilket reducerer antallet af operationer og burde give lavere effektforbrug. Men en ren optælling af operationer afspejler ikke den ekstra energi, som styrende hardware og software bruger. Specialet kortlægger derfor muligheder og kompromiser, når man beregner et udsnit af spektrummet i stedet for hele. Først introduceres kooperativ radio og et OFDM-signalmodel. Derefter implementeres og analyseres to Fourier-algoritmer: en fuld Split-Radix FFT og en SFFT, der kun beregner udvalgte frekvenskomponenter, begge på en Cyclone III FPGA (Field-Programmable Gate Array). Efterfølgende måles og vurderes effektforbruget for hver implementering, og mulige forbedringer identificeres. Afslutningsvis sammenlignes algoritmerne med den teoretiske målestok for beregningskompleksitet, som normalt bruges til at evaluere FFT-algoritmer. Testene viser, at SFFT ikke er strømmæssigt fordelagtig uden yderligere forbedringer. Forbedringerne omfatter blandt andet bedre sluk-/dvale-mekanismer for delsystemer, der ikke er i brug. Hvis sådanne strømbesparende tilstande indføres, vurderes det, at SFFT kan blive praktisk, og at sammenhængen mellem beregningskompleksitet og effektforbrug i højere grad vil holde for denne implementering.
This thesis investigates FFT algorithms (Fast Fourier Transform) for OFDM receivers (Orthogonal Frequency Division Multiplexing) and their power usage on a configurable hardware platform. It focuses on mobile applications and cooperative radio, where only part of the received frequency spectrum may be needed. Specialized FFT variants that compute a subset of the spectrum (SFFT) can cut the number of operations and, in principle, reduce power. However, simply counting operations does not capture the extra energy used by control hardware and software. The thesis explores the options and trade-offs of computing a subset versus the full spectrum. It first introduces cooperative radio and an OFDM signal model. Then, two Fourier transform algorithms are implemented and examined: a full Split-Radix FFT and an SFFT that computes only selected frequency components, both mapped to a Cyclone III FPGA (Field-Programmable Gate Array). Next, their power performance is measured and potential improvements are identified. Finally, the algorithms are compared against the theoretical metric of computational complexity commonly used to evaluate FFTs. Tests show that SFFT is not power-feasible without further improvements. These include, among others, improved power-off or sleep mechanisms for subsystems when idle. With such power-saving states, SFFT is predicted to become feasible, and its computational complexity would more closely reflect actual power use.
[This abstract was generated with the help of AI]