Hi,
the input frequncy amplitude might be attenuated due to an input filtering network in front of the ADC, thus leading to an SNR reduction. How does your setup look like? Theoretical or practical?
Might also be caused by the sampling itself. Imagine the (sinusodial) input waveform is never/barely sampled at its maximum, thus leading to a reduced signal amplitude in the digital domain. Here the question is how the SNR is determined (FFT ?) and how large/long is the data set.
BR