Under approximation of the altitude covariance

[SivertHavso]

Currently the altitude covariance is approximated by using the Horizontal Dilution of Precision (HDOP) times two instead of the Vertical Dilution of Precision (VDOP):

nmea_navsat_driver/src/libnmea_navsat_driver/driver.py

Lines 242 to 243 in f60d849

	current_fix.position_covariance[8] = (
	2 * hdop * self.alt_std_dev) ** 2 # FIXME

While from my very anecdotal experience this is a decent approximation, at times the VDOP will be greater than HDOP*2 leading to an under approximation of the covariance. The inverse holds true as well, but an under approximation is in most cases worse than an over approximation of the covariance.

VDOP isn't available in any of the NMEA sentences currently parsed, but it is available in GSA sentences so there is the potential of parsing these and the changing the above lines to:

current_fix.position_covariance[8] = (vdop * self.alt_std_dev) ** 2 

As a temporary fix I'm over inflating the approximated covariance before inputting into any Kalman Filters, but it would good to have this improved at some point.