Preston Bohm
signal processing fourier analysis quantum

Envelopes and the Fourier Transform

Packet shape → its spectrum Gaussian

Envelope shape

0.22 s
Time domain: carrier (40 Hz) under the chosen envelope
Frequency domain: |X(f)| near the carrier; shaded band marks the −3 dB width
−3 dB bandwidth
4.0 Hz
Peak sidelobe
none
Δt · Δf (RMS)
0.08
vs. Gaussian floor
1.0×
Gaussian. Minimum possible time–bandwidth product and no sidelobes; its transform is another Gaussian.
Pick a shape, then watch the spectrum. Every packet of the same duration carries the same carrier, but the envelope decides the spectrum. Hard edges (rectangular) ring with sinc sidelobes that smear energy far across the band; smooth, soft-shouldered envelopes fall off fast. The Gaussian is the unique shape that minimizes the time–bandwidth product and transforms into itself, so it puts the most energy into the narrowest clean band, which is exactly why it is the default packet in ultrafast optics, radar, and quantum mechanics.

You cannot have it both ways: a signal sharply confined in time is necessarily spread out in frequency, and the reverse. This is not a limit of any instrument but a theorem about the Fourier transform, and every wave packet obeys it. What you get to choose is the packet's shape, and that choice alone decides how much frequency a pulse must occupy and how messily that occupation falls off.

Localization costs bandwidth

A wave packet is a carrier oscillation multiplied by an envelope that switches it on and off:

\[ x(t) \;=\; g(t)\,\cos(2\pi f_0 t). \]

Its spectrum is the envelope's Fourier transform, shifted to sit around \( \pm f_0 \), so the shape of the band is entirely the envelope's transform. Shorten the envelope in time and its transform widens, ballooning the band. Measuring the spreads in time and frequency by their RMS widths \( \Delta t \) and \( \Delta f \), they obey the uncertainty inequality

\[ \Delta t \,\Delta f \;\ge\; \frac{1}{4\pi}. \]

This is Heisenberg's \( \Delta x\,\Delta p \ge \hbar/2 \) in disguise; position and momentum are a Fourier pair just as time and frequency are. The product \( \Delta t\,\Delta f \), the time–bandwidth product, has a hard floor that no cleverness beats.

The Gaussian saturates the bound

Which envelope actually reaches the floor? The Gaussian, and only the Gaussian. For \( g(t) = e^{-t^2/2\tau^2} \) the transform is again a Gaussian,

\[ g(t) = e^{-t^2/2\tau^2} \;\;\xrightarrow{\ \mathcal{F}\ }\;\; G(f) \propto e^{-2\pi^2 \tau^2 f^2}, \]

and the product is exactly \( \Delta t\,\Delta f = 1/4\pi \). Two things make this special: no other shape gets the band that narrow for a given duration, and the Gaussian transform is smooth and positive, with no sidelobes. It rolls off monotonically and never rings.

Why hard edges are expensive

Contrast a rectangular packet, the carrier gated on for a fixed time. Its transform is a sinc:

\[ \mathrm{rect}(t/T) \;\;\xrightarrow{\ \mathcal{F}\ }\;\; T\,\mathrm{sinc}(fT) = T\,\frac{\sin(\pi f T)}{\pi f T}. \]

The sinc has a forest of sidelobes decaying only as \( 1/f \), the first just 13 dB below the peak. Those tails leak energy into neighboring channels, raise a spectrum analyzer's noise floor, and wreck dynamic range. The rectangular pulse's RMS bandwidth is formally infinite, because \( \int f^2\,|{\rm sinc}|^2\,df \) diverges: a discontinuity in time always costs slowly-decaying tails in frequency. Round off the corners (triangular, raised-cosine, Gaussian, in increasing smoothness) and the sidelobes plunge.

The interactive above makes the trade concrete: every envelope is normalized to the same duration, and you can read off the −3 dB bandwidth, worst sidelobe, and time–bandwidth product. The Gaussian always sits at the bottom of the bandwidth column with no sidelobes; the rectangle has the worst of both.

Where this gets used

The payoff shows up wherever a clean, compact spectrum matters. Ultrafast lasers call pulses "transform-limited" when the time–bandwidth product reaches the Gaussian (or sech²) floor: as short as the bandwidth allows, with no wasted spectrum. Radar and communications shape pulses to suppress sidelobes so a strong return does not bury a weak neighbor. Quantum optics and NMR drive transitions with Gaussian pulses because the narrow, sidelobe-free spectrum addresses one transition without spilling onto its neighbors. And window functions in spectral estimation run this same search: a smooth envelope buying low sidelobes for a tolerable widening of the main lobe. The Gaussian is the shape the mathematics keeps returning to.

References

  1. D. Gabor, "Theory of communication," J. Inst. Electr. Eng. III, vol. 93, no. 26, pp. 429–457, Nov. 1946, doi: 10.1049/ji-3-2.1946.0074.
  2. R. N. Bracewell, The Fourier Transform and Its Applications, 3rd ed. New York, NY, USA: McGraw-Hill, 2000.
  3. L. Cohen, Time-Frequency Analysis. Englewood Cliffs, NJ, USA: Prentice Hall, 1995.
  4. F. J. Harris, "On the use of windows for harmonic analysis with the discrete Fourier transform," Proc. IEEE, vol. 66, no. 1, pp. 51–83, Jan. 1978, doi: 10.1109/PROC.1978.10837.