TABLE OF CONTENTS The FFT is a record-oriented algorithm. A time record, N samples long, is the input, and the frequency spectrum, N samples long, is the output. Recall from Chapter 3 that N is often restricted to being a power of 2 in order to simplify the FFT computation. A typical record length for an FFT analyzer is 1024 sample points. The frequency spectrum produced by the FFT is symmetrical about the folding frequency. Thus, the first half of the output record is redundant with the second half and the sample points numbered 0 to N/2 are retained. This implies that the effective length of the output record is ( N/2) + 1. These are complex points (real + j imaginary) which contain both magnitude and phase information.
Practically speaking, the output of the FFT is ( N/2) + 1 points, extending from 0 Hz to f f. Not all of these points are usually displayed though, since the anti-alias filter begins to roll off before f f. A common configuration is 1024 samples in the time record, producing 513 unique complex frequency domain points, with 401 of these actually displayed.
The N/2 (or so) frequency domain points are often referred to as bins and are usually numbered from 0 to N/2 (e.g., 0 to 512 for N = 1024). These bins are equivalent to the individual filter/detector outputs in the bank-of-filters analyzer. Bin 0...
