Time-Domain NMR and the Fourier Transform

Introduction

Modern pulsed NMR instruments do not directly draw a row of frequency peaks as a sample sits in a magnetic field. A radio-frequency pulse creates transverse nuclear magnetisation, and the receiver records its oscillating decay over time. Fourier transformation converts that free induction decay, or FID, into a frequency-domain spectrum. The quality of this conversion depends on acquisition duration, sampling, phase and processing choices, not only on the molecule.

Core explanation

At equilibrium in a static field B₀, nuclear spins produce net magnetisation mainly along the field direction. An RF pulse near resonance rotates some magnetisation into the transverse plane. Transverse magnetisation precesses at frequencies determined by local chemical shifts and couplings, inducing an alternating voltage in the receiver coil. The recorded voltage is the FID. It contains the sum of contributions from all observed resonances, so it may look like a complicated decaying oscillation rather than a set of obvious peaks.

For a simple single resonance with effective transverse decay time T₂ , the time signal resembles s(t)∝exp(−t/T₂ )cos(2πν₀t+φ). The asterisk indicates that field inhomogeneity as well as intrinsic spin–spin relaxation can broaden the observed decay. Different resonances contribute different oscillation frequencies. Fourier transformation decomposes the superposed time oscillations into their frequency components, producing peak positions, widths and intensities. After reference and scaling, chemical shifts are usually reported in ppm rather than raw hertz.

Sampling imposes limits. The spectral width must be chosen so signal frequencies fit within the sampled range; otherwise aliasing can fold resonances to incorrect apparent locations. Acquisition time sets the spacing of discrete frequency points and limits how finely nearby frequencies can be distinguished. Roughly, a longer observed time window permits finer frequency resolution, while a rapidly decaying FID contains intrinsically broad lines that merely collecting more empty noise cannot sharpen. These are different constraints: digital point spacing and physical linewidth.

Zero filling appends zeros after the measured FID before Fourier transformation. It can interpolate the displayed frequency grid and make peak shapes smoother, but it does not recreate information that was never acquired or separate genuinely unresolved lines. Apodisation multiplies the FID by a window function. Exponential weighting suppresses late-time noise and may improve apparent signal-to-noise, while broadening lines. A resolution-enhancing window can sharpen display at the price of amplified noise or distortion. Processing trades properties rather than offering free sensitivity and resolution simultaneously.

Fourier results are initially complex, with real absorption and imaginary dispersion components. Phasing rotates these components to give correctly shaped absorptive peaks. Incorrect phase can distort intensities or make peaks partly negative. Baseline correction, solvent suppression and reference calibration may also be needed. Every step should be reported when quantitative interpretation matters, since aggressive processing can create apparent features or hide weak signals.

Repeated scans improve signal-to-noise approximately with the square root of scan number when noise is independent and conditions are stable. Four times as many scans provide roughly twice the signal-to-noise, not four times. Long experiments may also require recycle delays sufficient for relaxation, especially for quantitative integration. A beautiful-looking spectrum can still be quantitatively unreliable if pulse and timing choices bias different nuclei unequally.

In two-dimensional NMR, the experiment repeats while an indirect evolution time t₁ is incremented. The direct FID is collected in t₂ for every t₁ increment. Fourier transformation along both axes generates a two-frequency correlation map. The single-dimensional FID therefore supplies the basic language for understanding 2D acquisition and the time cost of many t₁ increments.

Step-by-step reasoning

Start with an RF pulse creating transverse magnetisation. Explain precession and decay as the receiver's time signal. Identify the chosen spectral width, sampling interval and acquisition duration. Fourier transform to frequency space, then evaluate phase, baseline and any windowing before assigning peaks. Separate instrumental resolution from physical relaxation broadening.

Visual explanation

Imagine a wavy line whose amplitude shrinks from left to right: the FID. A Fourier transform rearranges its simultaneous oscillations into frequency peaks. A longer intact wave train allows closer frequencies to be told apart, whereas a rapidly fading wave produces broad peaks even on a fine digital grid.

Real-world analogy

Listening to several musical notes played at once gives a mixed sound wave. Frequency analysis can recover the notes, but a recording that ends very quickly makes close pitches difficult to distinguish. This resembles Fourier NMR, though NMR amplitudes additionally decay through spin relaxation and field imperfections.

Real-world example

An analyst sees two nearly overlapping proton peaks in a reaction mixture. Increasing acquisition duration and checking magnetic-field shimming may help if digital resolution and field homogeneity are limiting. Merely adding zero-filled display points cannot resolve signals whose physical linewidths already overlap strongly.

Why?

Why does a shorter FID tend to give broader lines? A rapidly decaying time signal contains a wider spread of frequencies by Fourier relationship. In NMR, fast transverse dephasing or field inhomogeneity shortens the coherent oscillation and broadens the corresponding spectral peak, even if the nominal digital frequency grid is very fine.

Common misconception

The Fourier transform does not generate new chemical information; it expresses acquired time-domain information in a more interpretable frequency form. Zero filling does not improve true resolving power, and a narrow digital peak after aggressive processing need not reflect a physically narrow resonance.

Worked example

Suppose an FID is recorded for 0.50 s, giving an approximate digital frequency spacing of 1/0.50=2 Hz before optional zero filling. Doubling acquisition to 1.00 s reduces that spacing to about 1 Hz if useful signal persists. If T₂ is only 0.05 s, however, most signal has vanished long before either endpoint. The observed linewidth from dephasing dominates, and extra acquisition mostly records noise. The calculation distinguishes digital grid spacing from intrinsic resolution.

Quick check

1. What physical signal does a pulsed NMR receiver record before Fourier transformation? Answer: It records a time-domain free induction decay caused by precessing transverse nuclear magnetisation inducing voltage in the receiver coil.

Exam focus

Distinguish T₂ from T₂ , digital resolution from physical linewidth, and zero filling from actual longer acquisition. Explain the effect of apodisation on both noise and linewidth. If claiming an intensity ratio is quantitative, account for pulse angle, recycle delay and processing.

Advanced insight

The Fourier transform of an ideal exponentially decaying complex oscillation is a Lorentzian line. Field inhomogeneity can introduce additional broadening and altered shapes. Advanced methods use spin echoes or field shimming to address different sources of dephasing; a spin echo can refocus static inhomogeneity but not reverse all true irreversible T₂ relaxation.

Summary

Pulsed NMR acquires an FID after RF excitation and uses Fourier transformation to display resonances by frequency. Sampling width, acquisition time, T₂ decay and processing all shape the spectrum. Zero filling interpolates, windows trade noise against resolution, and proper phasing and baselines are essential before structural interpretation.

Practice questions

1. If useful signal persists, what happens to approximate digital frequency spacing when acquisition time doubles? Answer: It halves, since spacing is roughly the reciprocal of acquisition time. 2. Can zero filling recover a frequency difference absent from the measured FID? Answer: No. It interpolates a finer display grid but adds no independent measured time-domain information. 3. Why can exponential windowing make a noisy spectrum easier to view yet broaden peaks? Answer: It suppresses noisy late-time data, effectively shortening the signal and widening its Fourier lines. 4. Why is a spin echo not a cure for every source of linewidth? Answer: It can refocus some static field inhomogeneity, but irreversible spin–spin relaxation and other dynamics remain.