From https://www.robot-electronics.co.uk/htm/srf08tech.html
ANN mode (Artificial Neural Network) is designed to provide the multi echo data in a way that is easier to input to a neural network, at least I hope it is - I've not actually done it yet. ANN mode provides a 32 byte buffer (locations 4 to 35 inclusive) where each byte represents the 65536uS maximum flight time divided into 32 chunks of 2048uS each - equivalent to about 352mm of range. If an echo is received within a bytes time slot then it will be set to no-zero, otherwise it will be zero. So if an echo is received from within the first 352mm, location 4 will be non-zero. If an object is detected 3m away the location 12 will be non-zero (3000/352 = 8) (8+4=12). Arranging the data like this should be better for a neural net than the other formats. The input to your network should be 0 if the byte is zero and 1 if its non-zero. I have a SOFM (Self Organizing Feature Map) in mind for the neural net, but will hopefully be useful for any type.
| Location 4 |
Location 5 |
Location 6 |
Location 7 |
Locations 8 - 35 |
| 0 - 352mm |
353 - 705mm |
706 - 1057mm |
1058 - 1410mm |
and so on |
Locations 2,3 contain the range of the nearest object converted to inches, cm or uS and is the same as for Ranging Mode.
From https://www.robot-electronics.co.uk/htm/srf08tech.html