EVM or Error Vector Magnitude is a measure of how the measured constellation points deviate from the ideal locations.
EVM(%) = sqrt(P_error / P_reference) * 100
P_error = RMS power of error vector
P_reference = power of outermost constellation point
Mathematically, we can write:
(P_reference will depend on the scaling of the 16QAM constellation)
As an example, if P_error = 0.1, then EVM(%) = sqrt(0.1 / 2)*100 = 22.4%
The larger the EVM, the further the constellation point deviates from the exact position, and the higher the Bit Error Rate (BER). This deviation can be caused by noise (AWGN), phase noise, channel impairments, group delay variations, AMAM and AMPM non-linearity for example.
There is usually an EVM requirement in the Standards for the transmitter, but it is also a good measure for the receiver where an EVM bathtub curve is often observed.