Hanwei/Winsen recommend a 24 hour "burn-in" period of use before the sensor is used for real applications. SGX Sensortech also recommend this in their documentation FAQ.
The two sensors were run alongside each other with the CPX board measuring the analogue output voltage during the first 15 hour run and second 17 hour run. The power for these runs was provided by the USB powered CPX VOUT. This means the boards were only running at 4.6V rather than the recommended 5.0V (+/- 0.1V).
The analogue output voltage measurement was direct from the sensors without additional external resistors which puts the CPX board more at risk of over (3.3V) voltage on the inputs but this was interactively monitored during the start of the runs. The graphs show the sensors tend to spike for a second or two at start-up. Additional load resistors were added for the subsequent gas tests to lower the voltage to a suitable level.
Something happens at 184 minutes which both sensors pick up, this was probably something environmental like a nearby window being closed. The sensors do appear to be very sensitive to cool draughts.
The downward spikes on MiCS-5524 plot are when the voltage is being checked with a cheap multimeter!
The MQ-3 reaches a stable value in less time on the second phase of the burn-in period. The MiCS-5524 reaches stability far quicker in both runs.
The noise becomes more visible on graphs with a narrower range on y axis. It's remarkably low at around +/- 1mV on MiCS-5524 output to CPX input.