Custom rolling average on time series

// First of all, you should have a table,
// call it TimeAxis, that stores your moments
// in time. Let's say it's in minutes. But such
// minutes must be unique across the whole
// time span. So, there must be a unique
// TimeID field (integer, say, that goes
// up by 1 - very important), so that you're
// able to move through the minutes by doing
// simple arithmetic. Connect this table to
// your fact table on TimeID.

// Then you write:

mvavg =
VAR DeltaT = 20 // Minutes for rolling average
VAR StartMoment = MAX( TimeAxis[TimeID] )
VAR EndMoment = StartMoment - (DeltaT - 1)
var MomentsToAvgOver =
    CALCULATETABLE(
        FILTER(
            VALUES( TimeAxis[TimeID] ),
            TimeAxis[TimeID] >= StartMoment
            &&
            TimeAxis[TimeID] <= EndMoment
        ),
        ALLSELECTED( 'FactTable' )
    )
var Output =
    // This condition makes sure that there are
    // in fact DeltaT moments and not fewer.
    if( COUNTROWS( MomentsToAvgOver ) = DeltaT,
        CALCULATE(
            AVERAGE( 'FactTable'[Measurement] ),
            MomentsToAvgOver,
            ALLSELECTED( 'FactTable' )
        )
    )
RETURN
    Output

Anonymous Datetimes are just decimal numbers with the integer portion as the # of days since 12/30/1899 and the decimal portion is the fraction of a day 1/24/60/60 for example.