ULTRA LOW FREQUENCY ESTIMATION A NEGLECTED APPORTION.pdf

ULTRA LOW FREQUENCY ESTIMATION: A NEGLECTED APPORTION C.H.Tan, A. Choudhury, Y.S. Ong S.M. Veres Faculty of Engineering and Applied Sciences University of Southampton Highfield, Southampton, SO17 1BJ, UK ABSTRACT Most frequency estimation techniques exhibit larger estima- tion errors in the ultra low frequency (ULF) region (0–10 Hz asymptotically) as compared to higher frequency ranges. This paper examines this situation and proposes to model the estimation errors through function approximation there- by making a priori error estimates available for new sig- nals frequencies. The technique is shown to perform well on simulation data under varying noise conditions. Results show that it is possible to estimate frequencies in the ULF region with accuracies similar to those encountered in high- er frequency ranges. 1. INTRODUCTION Frequencies between 0.01Hz to 10Hz fall in the ultra low frequency (ULF) range. Such frequencies occur regularly in earthquakes measurements, onshore wave ripples, electro- magnetic waves [2] and structural building vibrations [9]. I- naccuracies in frequency measurements can generally be at- tributed to: (i) numerical errors in the computer, called finite wordlength effect, (ii) precision constraints on the frequen- cy content in the Discrete Fourier Transform (DFT) and (iii) noisy measurements of the signals. Most analog measure- ment instruments do not have these shortcomings as they are built upon transistors and banks of parallel filters, perform- ing calculations to display the spectrum in the frequency do- main. Most modern digital instruments are packaged with software programs to facilitate the building of spectrum an- alyzers for frequency analysis. Being solely mathematical in operation, they are more stable, not prone to temperature and humidity drift, demonstrate good performance-to-cost ratio and do not suffer from manufacturing variations or ag- ing. Therefore DSP-based techniques have quickly replaced electronics-based spectru