The second-order initial value problem y" = f(x,y), y(x(_0)) = y(_0), y'(x(_0)) = z(_0) which does not contain the first derivative explicitly and where the solution is oscillatory has been of great interest for many years. Our aim is to construct numerical methods which are tuned to act efficiently on strongly oscillating functions. The frequencies involved determine the oscillatory character of the function and as the frequencies approach zero, the classical methods are obtained. The exponential- fitting tool has become increasingly popular as it is specially tailored for oscillating functions. Many classes of methods have been used with exponential-fitting and this will be discussed in more detail in the thesis. Collocation methods are considered for which the basis functions are combinations of polynomial and trigonometric terms. The resulting methods can be regarded as Runge-Kutta-Nyström methods with steplength dependent coefficients. We show how order conditions may be obtained, investigate the stability and other properties of particular methods and present some numerical results.

Any periodic signal can be decomposed into a sum of oscillating functions. Traditionally, cosine and sine segments have been used to represent a single period of the periodic signal (Fourier Series). In more general cases, each of these functions can be represented by a set of spectral parameters such as its amplitude, frequency, phase, and the variability of its instantaneous spectral components. The accuracy of these parameters depends on several processing variables such as resolution, noise level, and bias of the algorithm used. This thesis presents some background of existing frequency estimation techniques and proposes a new technique for estimating the instantaneous frequency of signals using short sinusoid-like basis functions. Furthermore, it also shows that the proposed algorithm can be implemented in a popular embedded DSPmicroprocessor for practical use. This algorithm can also be implemented using more complex features on more resourceful processing processors in order to improve estimation accuracy