The transducer embedded processing theme was introduced to this DTC in recognition that almost all sensors in operation today rely heavily on signal processing methods to overcome limitations caused by the environment of operation or the hardware used to implement them. This theme concerns itself with embedded processing, considered separate and distinct from general signal processing methods or high-level data processing. In other words it is processing needed at the transducer in order to allow the maximum performance to be realised from the sensor. Examples of embedded processing would include electronic image stabilisation of an EO sensor or the cancellation of jammers within a radar sensor. It would not include data fusion of a number of sensors.
In order to provide a holistic view of processing technology we have included both advancement of processing itself and research into the tools needed to exploit the advances in processing. This is in recognition of the fact that there is an explosion in signal processing options available to a sensor designer but many programmes underestimate the time and costs associated with turning algorithm ideas into low latency real-time processing systems. As a result the full flexibility of emerging technology is not fully exploited owing to the high cost of development and proving of the resultant system. Part of this programme will assess the possibility of overcoming this limitation.
Another important factor which is crucial to the effective use of processing within sensor systems is to have a good understanding of processing requirements and the likely availability of processing capability. The well-known Moore’s Law, expressing the growth in available computing capacity, currently shows no signs of reaching its limits, but it is less well understood that whilst the growth in processing power per unit volume continues unabated, the growth in processing power per unit of power consumption is modest, indeed, very modest. This fact has major implications for many military sensor applications, as most military platforms are severely constrained in availability of power and cooling. This means that processing cannot be regarded as an inexhaustible resource for future systems, and this places a premium on developing processing algorithms and software implementations which are highly efficient.
This is perhaps best illustrated by the failure so far to capitalise on the richest, most extensive source of free data that exists most of the time in most places – the data that most informs human and animal systems. Computer vision is the key to machine intelligence. It is very difficult and remains an infant technology, but there have been significant advances in both 3D generic vision for structure analysis, and scene interpretation in terms of 3D model-based processes, which are promising real capabilities in autonomy. As military thinking turns increasingly towards the use of autonomous platforms, the development of computer vision is becoming a matter of urgency.
Overall transducer embedded processing sits on the interface between the analogue sensor input and the digital output to the command and control system.
In selecting the programmes for this theme there has been careful consideration of the balance between “quick win” year one projects and longer term incremental projects, which aspire to make a difference within what is recognised as a big topic area. A major underlying theme of the first year work is to baseline the current situation in order to allow greater focus on priority issues in latter phases of the DTC. The work has been grouped into a number of sub-themes, which are summarised below, together with the planned contributors to the research programme.