FOURIER ANALYSIS
Overview
Fourier Analysis is the process of transforming time domain data into the frequency domain and the result is similar to that obtained by connecting up a spectrum analyser instead of an oscilloscope. It is especially useful in analysing the harmonic content of signals, perhaps to look for particular types of distortion, but it has many other uses also.
Method of Calculation
Fourier analyses are created by first executing a Transient Analysis and then performing a Fast Fourier Transform on the resulting data. This process involves discrete time sampling of the time domain data, with the result that the Nyquist sampling criterion applies. Put simply, this means that the highest frequency that can be observed is half that of the sampling frequency. However, other misleading effects can occur due to aliasing of the sampling frequency with harmonics of the input signal that are higher than half the sampling frequency. To minimise these effects, various types of window can be applied to the input data prior to the FFT.
Using Fourier Graphs
In the first instance, you need to set up your circuit as for a Transient Analysis except that you should use a Fourier graph instead of (or as well as) an Analogue graph.
Setting Up the Transform
Four fields on the Edit Fourier Analysis Graph dialogue form control the transform.
Start time and Stop time define the window of transient data that is transformed, exactly as they do for an analogue graph.
Max Frequency sets the top of the displayed spectrum, and Resolution sets the spacing between adjacent points within it. The FFT is sized from the ratio of these two numbers, rounded up to a power of two and never smaller than 256 points, and the sampling rate for the transient data follows from that size. The sampling rate is therefore always high enough for the Nyquist criterion to hold across the displayed range. A small ratio of Max Frequency to Resolution will be sampled faster than those two numbers alone imply, because of the 256 point floor.
If possible, choose a resolution that divides exactly into the fundamental frequency of the signal being analysed. Doing so places the fundamental and its harmonics on exact FFT points and avoids the spreading that otherwise occurs.
If the simulation window is shorter than one period of the chosen resolution, Proteus reports that the time data will be zero padded for the FFT. The transform will still run, but the result is being computed largely from padding rather than from circuit data, so increase the stop time or coarsen the resolution.
Windowing
The Window field selects the window function applied to the time domain data before the transform, and offers four settings:
- None, which applies no window at all. This is the default, and is the right choice when the simulation window is an exact number of cycles of the signal.
- Bartlett, a triangular window.
- Hanning, a raised cosine window and the usual general purpose choice.
- Welch, a parabolic window.
Windowing tapers the ends of the captured data so that the transform does not see the abrupt discontinuity that occurs where the capture starts and stops part way through a cycle. That discontinuity is what produces spurious sidebands around each real harmonic, so a window will usually give a cleaner spectrum at the cost of slightly broadening the peaks.
The left y-axis may be displayed either in dBs or normal units, using the Y Scale in dBs option.
To analyse the frequency content of a signal:
- Set up the circuit as for a Transient Analysis.
- Add a Fourier graph to the design and drag probes connected to the points of interest onto it.
- Choose start and stop times, and maximum frequency and resolution values, to suit the signal you are analysing.
- Choose a window appropriate to the signal.
- Press the space bar to invoke PROSPICE.
See Manual Y-Axis Configuration for information on defining y-axis limits.
See Also: