DRAFT · this page is still being checked; its numbers and wording may change.
A synthetic aperture sonar forms its image by adding thousands of echoes in step, so a fault in the receiving hardware rarely shows where it happened. Four common faults are injected here into one simulated survey and followed from a single recorded ping to the seabed image and a point target.
Tour
Controls
One ping as recorded
SAS image of the patch
8 m × 8 m, along-track up, range 12 to 20 m to the right; the truncated cone, the cylinder and the point target are marked.
Point target, 1.2 m
Along-track up, range to the right, in decibels against the clean point's peak. The seabed is subtracted, so only the point, its sidelobes and any interference remain.
How the images are made
The sonar is the array the rest of this site describes: 36 receivers 30 mm long at 300 kHz with 60 kHz of bandwidth, advancing 34 phase-center spacings per ping so that consecutive pings share two phase centers, on a straight track 10 m above a flat sand bed. The bed is simulated with ApertureLab's point-scatterer engine at 20,000 scatterers per square meter, with the truncated cone, the cylinder and a wedge as meshes whose acoustic shadows the simulator casts, and a point target on the bed. The survey is simulated once.
Each fault is then written into a copy of the recorded data, the complex baseband samples of every receiver and ping, with a fixed random seed. Tone and noise levels are set against the root-mean-square echo between 18 and 20 m of slant range, the middle of the 15.6 to 22.4 m that the patch spans. The tone sits at 310 kHz on every receiver; locked, it starts each ping with the same phase, and free-running, with a random one. The noise is independent on each receiver. A weak receiver has its samples scaled; a ping phase error multiplies every receiver of a ping by the same phase.
Every copy is beamformed by time-domain back-projection onto the bed with the true track, on a 1.25 cm grid with no glint suppression. The beamformer's time-varying gain is normally estimated from the data, which would partly divide a tone or noise back out of the image; here every copy uses the gain estimated from the clean data, as a sonar with a fixed gain would. The seabed image uses one brightness scale for all settings.
The point target's response is isolated by simulating the scene a second time without it and subtracting the two images, which is exact because the imaging chain is linear. The second scene receives the same weak receiver or ping phase as the first, but not the tone or the noise, so the point-target panel shows the point under the fault together with any interference, without the bed. Shadow depths compare the median brightness behind each object with open bed at the same ranges. The images were computed ahead of time for every setting, 27 in all.
For a survey flown over real deep-water bathymetry, see From bathymetry to SAS. For how the same scene focuses when objects stand above the bed, see The surface of focus; for the imaging chain, including micronavigation, see Image formation.