Mine's eye

A UK company has developed real-time sonar-imaging technology that uses a computer graphics card to greatly improve the image quality for side-scan sonar, as used in mine-hunting and pipeline surveys.

Synthetic aperture sonar (SAS) techniques enhance the resolution of images processed from sideways-looking sonar by up to 10 times, but existing systems require expensive technology to estimate the motion of the platform, as well as requiring it to move relatively straight. University College London (UCL) spinout Bloomsbury DSP is using the processing power of an NVIDIA graphics card to run proprietary algorithms to deliver high-resolution sonar images in real time without these restrictions.

Sideways-looking sonar, where the system scans perpendicular to the direction of movement, is used on towed systems and, increasingly, on autonomous underwater vehicles (AUVs). It uses the movement along the track to form the larger synthetic aperture. Operators need to see images from it in real time for them to be useful.

There are two main problems with current techniques. The first is estimating where the platform has been as it does not always travel along a perfectly straight path.

Dr Simon Charles, chief executive officer of Bloomsbury DSP, explained: 'Because synthetic aperture relies on coherently adding the positions of the system, if you're not within half a wavelength of those positions, the whole thing falls apart. You have to estimate the motion of the platform quite accurately, which can be done using techniques such as inertial navigation, but that's expensive and, in some instances, not accurate enough. We tried to do this using the acoustics themselves to estimate the motion of the platform, which presents a big challenge to get that accuracy.'

The motion is calculated using a correlation process between one 'ping', or set of returns, with the next, with the platform having moved in the intervening time. The difficulty is separating out the six degrees of freedom through which it moves. However, as the key factor in SAS is the difference in distance between the platform and the target, movement across the track is more important than movement vertically, so vertical motion can be largely discounted.

The other challenge is that forming the images is computationally difficult and some of the techniques that have been developed for radar to do that are mainly Fourier transform (a signal-processing operation) based. These are very fast and efficient, but they mean that the path the platform takes has to be relatively straight, which is not such a problem for radar applications but sonar platforms are seldom stable enough.

A NVIDIA card allowed the company to program the system in software, rather than firmware, facilitating the process of making changes and achieving the required performance.

'People need real-time images at that level of detail for things such as pipeline survey and mine countermeasures work,' said Charles. 'Pipeline surveys typically use towed systems and operators look at real-time results from the side-scan systems as the boat goes along.

'For mine countermeasures, the long-term aim is to do automatic classifications so the system could identify the difference between mines and rocks. You want an autonomous vehicle going off and doing the survey, then come back and say it thinks there's mines here, here and here — not to download loads of data off it and do hours of processing before they can make that decision,' he added.

Mine classification algorithms are still at a very early stage, but the work would prove easier with higher-resolution images as supplied by Bloomsbury DSP's system, especially as much of the information is in the shape of the shadow formed by the target.

'The shapes of most of the common mines are known, although makers do try to keep them secret,' said Charles. 'It's complicated by the fact that the people who make mines make them look as much like rocks as they can. There's various groups such as the NATO research centre that are looking at this and trying to make it reliable enough for military use.'

Bloomsbury DSP was founded in 2004 on the back of work started at UCL in around 1999. The company is trying to get its system to work with lower-specification sonar to extend its potential market outside high-end military applications. It is also adjusting its algorithms to work with more systems and is looking to establish a partnership with a large sonar provider in order to supply complete systems.

Berenice Baker