Can you try with gyro_bw setting 0x01 which means bandwidth is 230Hz and ODR is still 2000 Hz?
With this setting, you should able to get the noise within the spec.
Using 230Hz LPF does provide a lower noise level. However, as we talked before, the data from 230Hz has spikes in the low frequency part in the PSD plot while data from 523Hz does not. So I would like to use 523Hz LPF and aply my own filter.
However, the 523Hz data has a much higher noise level. According to the datasheet, the noise density is 0.014dps/sqrt(Hz). For 230Hz bandwidth, the RMS should be 0.014*sqrt(1.22 * 230) = 0.23dps, where 1.22 is from the equation FAE_CA1 provided. For 523Hz bandwith, the RMS should be 0.014*sqrt(1.22*523) = 0.35dps. Actually, the RMS of 523Hz data is much larger than that. What I can see is that the 230Hz LPF not only reduce energy beyond 230Hz but also attenuates frequency from 0 to 230Hz, which is weird. Please refer to the red and yellow lines in the figure below. I zoomed in the plot to show only the low frequency.
Thank you for the detail plotter.
as my colleague point out to you previously, 523Hz is almost raw data from MEMS unit, no filter applyed at all. We called it unfilter data.
This means the noise density calculation is not fit for those datas. the noise density shows the signal after filter result.
So the unfilter data with high noise is expected from sensor output.
In order to bring down the noise level, we add low pass filter with cut off frequency 230Hz. but it also introduce digitnal noise on PSD plotter. it is coming from digital front end design and confirmed by our engineer team.