Understanding NUS Reconstruction (IST Iterations)
Non-Uniform Sampling (NUS) is a powerful technique to save spectrometer time or increase resolution. However, since the instrument only acquires a fraction of the data points, the missing gaps must be filled before the Fourier Transform. iNMR uses a robust algorithm for this purpose: Iterative Soft Thresholding (IST).
When you process a NUS spectrum, iNMR asks for the "Max IST Iterations". This isn't just a limit to prevent your computer from overworking; it directly affects the quality of the final spectrum. Here is what you need to know to choose the right value:
- Speed vs. Accuracy: A low number of iterations (e.g., 50) is very fast but might leave the baseline noisy or result in inaccurate peak intensities.
- The Sweet Spot: For most routine 2D spectra, values between 100 and 500 offer the best balance. The reconstruction will be accurate enough for both visual inspection and integration.
- High Fidelity: If you are dealing with very weak signals or require the highest possible dynamic range, you can go above 1000. Be aware that iNMR becomes increasingly "meticulous" as you increase this number.
How it works under the hood
The algorithm starts with a high threshold, identifying only the strongest signals. In each step, it slightly lowers this threshold to find smaller peaks, multiplying it by a factor called λ (lambda).
iNMR automatically adjusts λ based on your chosen iteration limit. When you set a very high limit (over 1433), λ reaches 0.99. This means the threshold drops very slowly, allowing the algorithm to reconstruct the spectrum with extreme care. While this is more accurate, it is also significantly slower because the "evolution" of the spectrum happens in tiny increments.
Practical Advice
If you are using an older computer or a virtual machine, keep the iterations around 100-200. Modern multi-core processors (thanks to OpenMP support in iNMR) can handle 500 iterations in a matter of seconds.
One final tip: if your spectrum looks "clean" but some expected small peaks are missing, try doubling the number of iterations. If they still don't appear, they probably weren't captured during acquisition!