Sampling and averaging of the gyro rate data, guided by Allan Variance analysis, is key to getting the best out of CRS39 and to allow the earth rate signal to be drawn out of the noise.
Two sampling schemes that we would recommend are:
(a) In our test chambers we use a16 bit (successive approximation register ADC) National Instruments card. We sample the CRS39-03 fully differentially (Rate and Ref), at 10KHz, without any anti-aliasing filters. We then average every set of 10 samples to produce data at 1 KHz. This data is then analysed for AV and Noise.
(b) In our IMUs, we use 24 bit sigma delta ADCs, outputting sampled data at 10 kHz. Actual sampling at the sensor end is around 192KHz. Again the CRS39-03 is sampled differentially (Rate and Ref). We average every 10 samples to produce a data set at 1KHz.
Analysis of the resulting data using Allan Variance techniques will determine the optimum averaging time. However, the optimum averaging time may be longer than the user can accept and we have typically used 15s averaging for each compass point (90deg separation) measurement.
Minimising temperature variation over measurement cycle will improve accuracy, either by thermal shielding or provision of thermal mass. Improvements may also be found by fully enclosing the gyro in a metal enclosure, minimising any 'metal detector' or field effects.
Temperature compensation is recommended - linear or third order may be required depending on the actual conditions seen by the gyro - temperature range, rate of change of temperature, as well as actual rates applied and temperature.
Averaging measurements taken at index positions of 180deg to each other can help by draw out the real bias of the gyro by removing earth rate. Temperature fitting is also possible by comparing changes between measurements at the same index position against bias and temperature.