Tomas,
This problem is now resolved
When I read data from the Gyro, I save the bytes to an array called Gyro_Receive[]. I use FreeMaster to debug my hardware and software. I implemented a recorder in the tool and in the I2C read function to capture a sequence of data in real time. That is a wonderful tool!
Gyro_Rate_X is Gyro_Receive[1] << 8 | Gyro_Receive[2]. I first plotted Gyro_Rate_X and found the data dropouts reported originally. I then plotted the status byte with Gyro_Rate_X to see if there was a problem with data in the gyro. The status byte was always 0x0f. I then plotted Gyro_Receive[1] with Gyro_Rate_X and that byte was always zero. That was as expected because I have the recorder set to trigger on a positive angle rate. I then plotted Gyro_Receive[2] with Gyro_Rate_X and found that the signal was smooth, but when there was a dropout, the two variables were overlaid and negative. I then realized that the problem was with the declaration of Gyro_Receive[].
I had declared that array as char. When I changed it to unsigned char, then the data was properly assembled into Gyro_Rate_X and the problem was resolved.
Kind of dumb, I know. Especially because it took me so long to discover the cause. I can now explore some of the filtering features of the gyro, to see what will work best in my system.
I am using an ODR of 400 Hz and full scale of 500 deg/sec. I had tested two parts which told me that my problem was either software or setup. The layout is as described in the data sheet for I2C operation.
