I have got the time to calculate the launch window, using the launch time of the last MEO launch, and the fact that the MEO 2/5 plane's RAAN is spaced 120 degrees to the east of the MEO 3/4 plane (source: http://www.9ifly.cn/forum.php?mod=redirect&goto=findpost&ptid=9786&pid=86869
Compass M3/4 was launched at 20:50 UTC on April 29 (4:50 am local on the 30th). Assume that the launch date is September 20 LT, i.e. 143 days after the last launch, the movement in longitude for the orbit plane to cross the launch site (i.e. at the same longitude and mean solar time) is:
143 / 365.25 * 360 = 140.9 degrees
Now we consider the precession of the orbit RAAN. Apparently the movement of the RAAN towards the west for a prograde circular orbit can be approximated by this equation:
d-omega = 9.97 * (Re / a) ^ 3.5 * cos(i), where Re = Earth's radius (6378 km), a = semi-major axis and i = inclination
For the Compass MEO constellation, i = 55 degrees and a = 6378 + 21500 = 27878 km, so we have d-omega = 143 * 9.97 * (6378 / (6378 + 21500)) ^ 3.5 * cos(55) = 4.7 degrees
So if we are launching the satellites to the same plane as MEO 3/4, then the launch time should be earlier by (140.9 + 4.7) / 360 * 23 h 56 min = 9 h 41 min (where the Earth's rotational period is 23 hours and 56 minutes). That corresponds to 11:09 UTC.
Since the actual plane for MEO 2/5 is 120 degrees to the east, the launch time will be later than the time above by 120 / 360 * 23 h 56 min = 7 h 59 min. Therefore the required launch time is 19:08 UTC on the 19th (3:08 am LT on the 20th). The error should be within 10 minutes
(I did the same calculation for the GPS IIF-3 launch, and it came out to be 5 minutes later than the window opening time).
For every 1 day delay of the launch, the launch time will move earlier by around 3.2 minutes, and vice versa.
(Source of the calculation method: http://www.9ifly.cn/forum.php?mod=redirect&goto=findpost&ptid=4786&pid=97613