Non-uniform sampling (NUS) and non-linear reconstruction methods, are increasingly gaining acceptance in the NMR community. While increasingly common in the literature, further adoption of NUS techniques are haunted by a particular challenge: determining optimal sampling strategies that 'guarantee' a suitable spectrum. The question of how much data it is necessary to acquire seems impossible to solve a priori, However, the problem can be approached using data already acquired by the spectrometer; additional samples, or additional scans, can be collected based on deficiencies found in already-measured data.
In addressing this problem, concepts of sensitivity and resolution must be revisited, where NUS methods can be employed not only to achieve high resolution by sampling long evolution times but also be utilised to achieve gains in sensitivity by non-uniform signal averaging. The relative intensity of thermal noise signals has long been an issue affecting NMR, typically addressed by collecting multiple scans of each sample point. By not requiring the same number of scans for each sample point, instrument time can be optimised to improve sensitivity at key time points ─ or free up time to collect additional samples.
Here, we will discuss new non-uniform sampling and signal averaging techniques being developed to reduce acquisition time of multi-dimensional experiments via real-time analysis. We will also share some unorthodox solutions to the challenges posed by prioritising between spectral sensitivity, resolution and completeness. Several strategies will be discussed ─ from the successful, to the promising, to those fellow investigators may wish to avoid.