I have 8pcs of ADT7410 sensors connected to a PIC18F47J53 microcontroller via 2 individual I2C interfaces running from 2x AA batteries. Each sensor has a 4.7µF ceramic decoupling capacitor soldered as close to the sensor as possible. The sensors are configured to "normal mode" at 16-bit resolution. The sensors are placed close to each other (the distance between them is not more than 20mm) - the temperature-wise steady environment is ensured, verified by a laboratory reference thermometer.
With this setup, 1024 consecutive measurement points are taken per each sensor and analyzed - see attached file with the results. The top chart shows all the measurands [temperature vs. time] (the blue lines indicate the maximum and the minimum values), the bottom-left chart is the output of the Fourier-Transform (which shows no periodicity in the signal), the bottom-right chart is the histogram of the readings.
The histograms are showing a bimodal and/or skewed distribution – it’s not white noise. Also, the difference between the maximum and minimum values of 1024 readings is around/above 0.1°C. These findings lead me to a conclusion, that the advertised 16-bit resolution (0.0078 °C) is not achieved (i.e. effectively, the resolution is 13-bit, even if the sensor is configured to 16-bit!). Is this the best signal-to-noise ratio, that this sensor can do - or am I missing something important?
Insight of experts would be appreciated.