Content
Summary
Digital signals in NMR and MRI are measured at discrete time points. The Nyquist theorem says the sampling rate must be at least twice the highest frequency present, otherwise signals fold back (alias) to wrong frequencies. In NMR this appears as folded peaks; in MRI as wrap-around artifacts.
Intuition
Imagine filming a spinning wheel with a camera that takes pictures at fixed intervals. If the wheel turns slowly, you can follow it. If it turns almost a full turn between frames, it looks like it is moving slowly or even backwards. Sampling an NMR signal works the same way: the receiver records the voltage only at regular dwell times. Each sine wave needs at least two points per cycle to be identified. Faster oscillations are indistinguishable from slower ones, so they appear at false frequencies. The spectral width equals the sampling rate, so it defines a frequency window; peaks outside the window are folded back inside. In MRI, the field of view is set by the spacing of k-space samples, so anatomy outside the field of view wraps into the image.
Key equations
Sampling rate is the inverse of the dwell time between points.
Nyquist frequency: the highest frequency that can be correctly represented (for complex quadrature data, the window is ±f_N, total width f_s).
A true frequency f appears shifted by an integer n times f_s into the observed window.
In MRI, the field of view is the inverse of k-space step size; too large a step shrinks the FOV and causes wrap-around.
Derivation
- Take a signal cos(2π f t) sampled at times t_m = m Δt.
- The samples are cos(2π f m Δt).
- Add any integer multiple of f_s = 1/Δt to f: cos(2π (f + n f_s) m Δt) = cos(2π f m Δt + 2π n m).
- Since n m is an integer, the cosine values are identical, so f and f + n f_s give the same samples.
- Hence frequencies are only determined modulo f_s, so we must restrict the window to a width f_s, i.e. ±f_s/2 for quadrature detection.
- Any true frequency beyond this window is folded to f − n f_s inside it.
Worked example
A proton NMR spectrum is acquired with a spectral width (sampling rate) of 5000 Hz, centred on 0 Hz. A peak is truly at +3200 Hz offset. Where does it appear? Also, what dwell time is required?
- Window is ±2500 Hz because f_s = 5000 Hz.
- The peak at 3200 Hz is outside the window.
- Subtract f_s: 3200 − 5000 = −1800 Hz, which lies inside the window.
- So the peak appears at −1800 Hz.
- Dwell time Δt = 1/5000 Hz = 200 microseconds.
- To avoid aliasing, increase the spectral width to at least 6400 Hz (covering ±3200 Hz), so Δt ≤ 156 microseconds.
Answer The peak aliases to −1800 Hz; a spectral width of at least 6400 Hz (dwell ≤ about 156 µs) avoids it.
Common mistakes
- Thinking Nyquist means sampling at exactly the signal frequency rather than at least twice the highest frequency.
- Believing aliasing can be fixed after acquisition; folded peaks are indistinguishable from real ones.
- Forgetting that quadrature detection makes the window ±f_s/2 rather than 0 to f_s/2.
- Confusing MRI wrap-around with motion artifacts; wrap-around arises from the FOV being smaller than the object.
- Ignoring noise: filters are needed because noise outside the window also folds in.