MPU-6050 Yaw over time

Our product currently uses an MPU-6050 to track head motion. We use DMP configured as per the example motion driver test.

We get the expected (minimal) yaw drift when the device is stationary (and with gyro auto comp enabled) and we can measure that and compensate in client code to get excellent zero motion yaw drift.

In use we find that DMP has occasional difficulty keeping track of yaw. A 30 degree yaw followed by a return to center usually returns within a degree or so but sometimes it's way off.

Are we expecting too much from the MPU-6050 and/or DMP ?

Is there an optimal MPU/DMP configuration for this kind if rotational orientation tracking ? i.e. does FSR make a different ? We could choose smallest FSR that captures the limit of human head acceleration if it makes a difference.

Does sample rate make a different, i.e. is 200hz better for DMP than 50 hz ?

Does pushing bias to MPU rather than DMP make a difference to DMP output ?

We're building a test rig to enable reproducable testing of DMP with various settings but if the Invensense Devs know of a sweet spot then we'd be grateful to know!

Regards

Rob James