The channel estimation algorithms are running at the receiver part, which is not standardized. This means, it is up to implementation and it’s vendor specific.
However the typical approach is that channel estimation is done based on subframe / TTI basis (which is 1ms). This is because a single allocation for a UE is done in this timeframe.
Speaking of the estimation itself, there are different approaches as well. You may e.g. do a joint 2D estimation or separated Time and Frequency estimations for a UE allocated PRBs only. Another approach is that the estimation is done on larger portion of PRBs than the actual allocation, because it provides the UE with more points (based on pilots / RSs) for the interpolation, which may be in that case with better resolution.
