Question: When working with a sensor in the lab, I can see as little as 1 PPB of nitrogen dioxide or ozone. When I try to exploit this sensitivity in routine measurements, the signal drifts all over the place on a time scale of days. How can I get the full performance from the sensor?
Signal drift is common in most gas sensors. Otherwise, there'd be no need for calibration. Drift is generally slow, on a time scale of hours or days. In electrochemical sensors, most of the drift is in the baseline; drift in the span coefficient tends to be low. This gradual drift actually limits the sensitivity for long-term measurements. One way of improving the effective sensitivity of a sensor is to periodically measure its baseline output. This baseline can be subtracted from the routine measurements to yield a net signal which more faithfully represents the sample.
A 1995 report from our laboratory  dramatically illustrates the increase in sensitivity that can be achieved by the use of a zero filter. The TSI, Inc., nitrogen dioxide sensor will respond to rapid changes in ozone concentration of as little as 1 PPB (part per billion). Unfortunately, the sensor baseline drifts over a range of more than 200 PPB over a four day period. In most applications, this would place a practical limit on its sensitivity of 400 - 1000 PPB -- it would usually be impossible to distinguish changes in sample concentration from random baseline drift. Our study showed that by measuring the sensor baseline every 30 minutes, effective sensitivity of better than 10 PPB could be achieved.
Figure 1 shows the wildly swinging baseline drift of one of a group of five TSI Series 106 oxidant sensors exposed to ambient air over a long summer weekend. Below this curve are the upper and lower limits of the entire group of five sensors after correction for baseline drift. The five sensors agree with one another within a range of 20 PPB, or +/- 10 PPB.
This experiment was implemented by using an electric solenoid valve to switch the zero filter in and out of the sample path for five minutes of every 30 (Figure 2). The sensor reading at the end of the five minute 'autozero' period was used to correct the sensor readings for the following 25 minute measurement period.
 W.R. Penrose, L. Pan, J.R. Stetter, and W.M. Ollison, 1995. Sensitive measurement of ozone using amperometric gas sensors. Anal. Chim. Acta 313, 209-219.