copy and paste from the designer of one of my wideband units .i own a few different units
Free air is a horrible calibration gas for applications where rich accuracy is important, basically for all performance applications.
Wideband controllers do not measure lambda directly, they measure pump current, the amount of current required to balance the sensor at stoich. The relationship between pump current and lambda is defined in the bosch datasheet, and from that you can calculate lambda.
This is taken directly from the bosch LSU 4.2 datasheet.
The curve Pump Current (Ip) Vs Lambda curve is divided into 2 sections. One section for Ip <0, rich, and another for Ip >0, lean. Each section has its own totally different electro-chemical reaction, and thus that is why the curves are very different.
It might seem that given the curve, one can use free air as a calibration gas to calculate a compensation value to apply to all points on the curve. however the curve above is a "Nominal" curve, that is the expected curve for a perfect sensor, there are no confidence/tolerance bounds on the curve above.
With the above curve, all I am trying to do is to visually illustrate that the rich and lean curve are very different and using the lean curve to come up with a compensation value for the rich curve is stretching reality.
This is taken directly from the bosch LSU 4.2 datasheet.
For a new sensor @ Lambda = 1.7, lean gas, the tolerance of the measured lambda is 0.05. For a new sensor @ Lambda = 0.8, rich, the tolerance of the measured lambda is 0.01.
Note, that the magnitude of the pump current for 0.8 lambda is roughly the same as the the magnitude of the pump current for 1.7 lambda, according to the Ip vs Lambda graph, this just means that 0.8 lambda vs 1.7 lambda is a good basis illustrating the rich vs lean tolerances of the sensor.
Why is accuracy of the lambda sensor 0.01 lambda for 0.8 lambda rich gas, and why does that accuracy fall to 0.05 for 1.7 lambda lean gas? the answer is that bosch calibrates each sensor using rich gas at the factory using a lazer cut calibration resistor. if bosch were to use lean gas for calibration, the lean accuracy would be very good and the rich accuracy would suffer. How bad would the rich accuracy be if lean gas was used for calibration? I do not know, but assuming 0.05 accuracy is reasonable IMHO, that translates to about 0.735 AFR accuracy in rich gas for a gasoline engine.
Also note that using free air for calibration is "free", bosch takes the extra step to use rich gas, if free air was as good as rich gas for calibration of the rich curve, Bosch would have saved the money and used free air instead.
Factory calibration of the bosch sensor using rich gas is included in the price you pay for the sensor. Throwing the calibration away and relying on free air calibration which reduces your rich accuracy is just foolish.
Also note that after 500 hours of bench time the accuracy of the lambda sensor drops to 0.02. Free air calibration will compensate for the aged sensor, however the end result is still worse accuracy than 0.02 lambda accuracy for rich gas.
Coles notes: Free Air calibration will ensure your lean accuracy is very good and the rich accuracy is very poor. Rich gas calibration done by bosch from the factory ensures each sensor has 0.01 lambda accuracy for rich gas, but the accuracy for lean gas is much worse. If you are tuning for performance, rich, then you do not want free air calibration. If you are running a very lean burn engine, then free air calibration is a useful feature.
