Harmonic signals are produced by many natural sources as well as man-made sources. The detection of the harmonic signal can reveal a lot about the physical system. Various Fourier harmonic analysis based numerical methods are commonly used to detect such signals. However, recently there has been considerable interest in other non-Fourier-based methods as well, to determine the harmonics. In the present work we have studied the application of artificial neural network based machine learning in frequency identification of sinusoidal signals. We considered training sets comprising harmonic signals with randomised phase or randomised amplitude or combination of the both. Based on such training sets the trained network is then applied to detect the frequency of unknown harmonics. Here we performed aninvestigation on the relative advantages of the network trained using sets of harmonic signals with different features.
Keywords
Artificial neural network (ANN); Harmonic signals; Detection accuracy 1 Introduction Many signals associated with real life problems exhibit periodic components that repeat at fixed frequencyor at an integral multiple of the main frequency, called the fundamental, throughout the signal duration. On the other hand, a function of time, say παΊπ‘α», corresponding to any arbitrary signal data, with a given periodic variation, can in general be expressed as a weighted sum of periodic sinusoidal components written as παΊπ‘α»ΰ΅π଴ΰ΅
β π΄ΰ―sinαΊπΰ―π‘ΰ΅
πΰ― α» ΰ― ΰ―ΰଡ β¦ (1) This is also known as the Fourier series expansion of the signal 1 . Here π଴is the mean co-efficient, πis the total number of constituents, π΄ΰ― , ο·ΰ― and πΰ― are the amplitude, angular frequency and phase, respectively, of the i-th constituent. Each of these Nsinusoidals can b
User
Font Size
Information