According to the datasheet for the ADIS16251 rate sensor, the sensor has an angle random walk of 3.6 degrees/sqrt(hr), and the datasheet for what I'm told is one of its recommended replacements, ADIS16265, states that it has a lower angle random walk of 2 degrees/sqrt(hr). However, in testing this myself, I was not able to confirm that this was the case.
To calculate the ARW, I plotted the Allan variance curves of the two sensors using data collected at 3 Hz over a period of about 72 hours. As per IEEE-STD-952-1997 and FAQ: Gyroscope Angle Random Walk, the ARW is obtained by taking the value of the Allan deviation (square root of Allan variance) when the averaging time is 1 second. Applying this method, it seems from the plot below that the ARW of the two sensors is not significantly different, with 0.03081 degrees/sqrt(s) (1.85 degrees/sqrt(hr)) for the 16265 and 0.02903 degrees/sqrt(s) (1.74 degrees/sqrt(hr)) for the 16251.
This approach always seemed a little sketchy to me, however, since that same IEEE standard states that the rate ramp can be found by reading off the value of the plot when the integration time is 3 seconds, yet in the plot I've provided, the slope at 3 seconds is not +1/2 as required.
Next I tried using a system identification approach to determining ARW, where I modeled the total Allan variance of the rate sensor as the sum of the Allan variances contributed by the various error sources in the rate gyro, including quantization noise, ARW, bias drift, rate ramp, and rate random walk. Using this model, I applied a log-least-squares fit to the same data used to make the plot above and found that the ARW of the 16265 was 0.0291 degrees/sqrt(s) (1.75 degrees/sqrt(hr)), whereas the 16251 had an ARW of 0.0228 degrees/sqrt(s) (1.368 degrees/sqrt(hr)), indicating that the older sensor in fact has lower ARW. I've included a plot of the fitted Allan variance curves below.
I've repeated this procedure for two other pairs of sensors of each type and each time, the result I've shown here is repeated. Any idea what can explain this discrepancy?