We had sort of a debate regarding the speed of the ADCs we are using. Since we're using 3.5MHz transducers, we would have been aiming at getting 7Msps minimum ADCs. However, we're using non-perfect transducers, so pulses are often quite large (2 to 5 periods), and we're looking at the envelope of the echoes. So, the echoes envelope have a width that spans around 1ms - a 1Msps ADC could be able to capture this.
Moreover, we discussed about introducing a peak-hold-reset to play the role of the envelope detector.
Since we wanted to see the impact of the ADC's speed, we simulated a decimation from 5Msps (the original data) down to 2.5Msps and 1Msps. For this decimation, we had three choices: averaging the points, getting a random one, or getting the highest point (emulating the peak-hold).
No more to say, but this, on data before the scan conversion:

Unsurprisingly, the random is less smooth than averaged, and the Max has quite an interesting answer, as it highlights the features of the image. Rest of the description (including scan converted images) is here.

All code, images and the scripts are available at : https://github.com/kelu124/murgen-dev-kit/tree/master/worklog/Images/Session_8/source_files