MPU-9150 (IMU) – what happens at fullscale range boundaries

It looks like the Gyro and the Magnetometer are slightly miscalibrated. In excel, it looks like they are about 135 degrees apart. Try changing your formula to be this instead:

=DEGREES(ATAN2(U3,T3))+Dcal

Where Dcal is some calibration value. It looks like the data matches up fairly well when Dcal is about 135 degrees.

I'm unfamiliar with the Intel Edison, but this sounds like some internal buffer overflow in the MPU-9150. It looks like the MPU-9150 has an internal FIFO buffer between the I2C interface and the actual motion engine. When you do a printf (presumably through some JTAG based debugging interface emulating a serial port) is probably much slower than the time it takes to read out a sensor value. So what is likely happening is that you are removing sensor values using I2C and printf much slower than the motion sensor is writing values. At some point the buffer overflows because of this, and puts the MPU into some error state.

Whenever you design a real-time system, you must keep in mind that debugging printf may be thousands of times slower than everything else which is happening in your main loop. One simple way to see if this is the problem is to only print out 1 out of every 1000 sample values (read all samples through I2C, but only print out one in 1000). If this fixes it, your problem is indeed in the debug printouts and not anything specific to the I2C driver.