Signal-to-noise ratio

The ratio of target signal strength to the level of measurement noise.

A small peak in a low-concentration sample does not mean that it has been reliably detected. Peak height, concentration or difference assessments have a stable basis only when the target signal is sufficiently distinct from random system fluctuations.

Define signal and noise first

IUPAC defines signal-to-noise ratio as the ratio of signal power to noise power. Under equal-impedance conditions, the ratio of root-mean-square signal and noise amplitudes is also commonly used. Different expressions cannot be compared directly without explanation.

A spectroscopy experiment must also specify whether signal means peak height, peak area or a reading at a particular wavelength, and identify the baseline region, number of samples and statistic used for noise. Changes in bandwidth, integration time and processing also change the signal-to-noise ratio.

Microvolume does not necessarily mean low signal-to-noise ratio

Small sample volume does not inevitably reduce signal-to-noise ratio. What matters is how much useful light at the target wavelength reaches the detector, the absorption change produced by the sample, and how noise from the Blank and sample measurements propagates together.

Strongly absorbing samples transmit very little target light and may fall in a low-signal region. In dilute samples, the absorbance difference may approach baseline fluctuations. Both are difficult to distinguish, but need different responses: shortening the path or diluting may help the former, while the latter depends more on a stable Blank, a suitable path length and repeated measurements.

A smoother curve does not mean more information

Longer integration or more averaging can reduce some random noise at the cost of speed. Wider spectral bandwidth increases light throughput but may blur narrow features. Each improvement in signal-to-noise ratio comes with changed measurement conditions.

Software smoothing can make a curve look cleaner, but it cannot create information absent from the raw data and may alter peak height and width. Signal-to-noise ratio is also not equivalent to a detection limit, which additionally requires the blank distribution, calibration slope and method specifications.

Report the definitions of signal and noise together with bandwidth, sampling and processing methods to make a signal-to-noise ratio reproducible.