Skip to content

Basics of Fourier Transform

This post was originally written in English.

There are several algorithms for transforming data from the time domain to the frequency domain as below.

Fourier Series

A Fourier Series（傅里叶级数） is summarized as creating a complex waveform by summing pure sine waves with different amplitudes and frequencies, and to decompose a complex signal into a sum of sinusoids of different amplitudes and frequencies.

Dirichlet Conditions

Dirichlet Conditions specify a set of conditions that must be met before a signal can be decomposed into a Fourier Series:

  • The signal is a mathematical function, i.e., one and only one y-point corresponds to each x-point.
  • The signal is periodic.
  • The area bounded by the signal over one period is finite.

Decompose into a Fourier Series

A complex signal that meets the Dirichlet Conditions can be represented by a sum of sinusoids:

\[ f(t)=a_0+A\{\sum_{n=1}^\infty[a_n cos(n \omega_1 t+\phi_n)+b_n sin(n \omega_1 t+\phi_n)]\} \]

where:

  • \(a_0\) is the DC component.
  • \(A\) is an overall scale factor for all harmonic components.
  • \(\omega_1\) is the frequency of the fundamental.
  • \(n\) is an integer multiplier of the fundamental frequency for each harmonic term.

This proves that not only can we sum a series of sine and cosine waves to create any other wave, but also that the frequencies of the sinusoids are integer multiples (harmonics) of a single fundamental frequency.

Discrete Fourier Transform (DFT)

Discrete Fourier Transform (DFT): takes amplitude versus time data, and then translates to amplitude versus frequency data.

Mathematically, the algorithm is a series summation of the product of each sample times a complex number:

\[ X(b)=\sum_{n=0}^{N-1}x[n](cos(2\pi nb/N)-jsin(2\pi nb/N)) \]

where:

  • \(n\) is one of \(N\) samples.
  • \(N\) is total number of samples.
  • \(b\) is one of \(B\) frequency bins (each bin represents a frequency range of \(F_s /N\)).
  • \(j\) is the imaginary operator.

The DFT algorithm uses each sample point in the summation from 0 to N-1 for each analyzed frequency. All N sample points contain information about all B frequencies, thus each of the B frequencies for which information is desired requires a summation of N time sample products. Because of the reasons above, processing a DFT is slow, because \(N^2\) calculations are necessary. For example, a 2000 point DFT requires 4 million calculations, often floating point calculations, which are slower than integer calculations.

Fast Fourier Transform (FFT)

Fast Fourier Transform (FFT) remedies the DFT speed problem by skipping over portions of the summations which produce redundant information. Rules for using FFT:

  • The number of sample points must be a power of 2 (\(2^n\)).
  • The number of additions and multiplications is: \(\frac{N}{2}\log_2 N\).

References & Acknowledgements

  • Fundamentals of Testing Using ATE
  • The-Fundamentals-of-Mixed-Signal-Testing_Brian-Lowe

Original: https://wiki-power.com/
This post is protected by CC BY-NC-SA 4.0 agreement, should be reproduced with attribution.