Military Technology 06/2021

An Airbus proposal to equip the A330 MRTT with a communications relay. In addition to this role, such aircraft could also become flying proximity supercomputers for large task forces, to relieve the work of onboard servers on fighter jets and adjuncts. (Photo: Airbus) 38 · MT 6/2021 Feature Design choices and compromises will determine the overall architecture of the system. For this reason, the AI architecture is likely to be distributed over several levels, including an intermediate one represented by a tanker or another large aircraft charged with collecting and disseminating to and from the aggregating infrastructure, perhaps with an on-board supercom- puter. Of course, some capabilities will also be installed on the fighter for basic triage and possibly on the adjuncts. Yet, the most valuable informa- tion will have to be flown back to the aggregation centre on the ground, so as not to lose useful information if the aircraft is shot down. In all this, AI applications must be constantly learning, in real time. Consequently, everything must be enabled to be continuously reprogrammed through constant, consistent communication with the aggregating infrastructure, either autonomously or through an intermediate flying node. Once again, the choice of the type and amount of information to be processed on board will depend on power considerations. Thanks to its engines, the main fighter will have a few dozen megawatts available for computing functions, but UAS and adjuncts will hardly have more than few hundred kilowatts available. Between now and 2035-2040, therefore, there will be a major engineering effort to reduce energy consumption, while miniaturising onboard components. Somehow, it is a bet on cur- rent trends in storage and computing capacity. According to tech com- panies’ estimates, the physical space used by a storage unit halves every three months, so the enormous growth in data should not pose a major problem for its storage. Their exploitation, on the other hand, remains a sensitive issue, and no one is publicly determining a clear evolutionary process in this sense. Based on today’s forecasts and evaluations, made by engineers and technicians from technologically advanced companies, it seems likely the calculation capacity of a supercomputer of the class of DAVINCI-1, currently contained in numerous racks, could be condensed into a single rack within a decade. Within twenty years, this might be further reduced to a router-sized object for airborne applications. Of course, computing power and energy also means heat, which must be dissipated to avoid damage or compromise to onboard systems such as radar, or degrade the platform’s IR signatures. The choice between air or fluid cooling will be another major design choice that will affect the role to be given to each in-flight element. Building the Know-How Companies currently working on major fighter programmes intend to develop proprietary technologies that will enable them to master the triad of key technologies, thus triggering major repercussions in all high-tech sectors. In Europe, Leonardo, BAE Systems, and Thales are leading the way, while in the US, Lockheed-Martin is one of the most active, but Northrop Grumman, L3Harris, Boeing, Raytheon, and other big players are also pledging immense effort. They kicked off a process of ‘em­ powerment’ for engineers to raise their competence in the creation of unique, proprietary architectures. interactions, thanks to a few gigantic data centres with continental capac- ity. This model may be challenged by the overwhelming volume of data to be processed. One or more intermediate working levels will therefore be needed. Thinking of a military campaign, this is already evident. Data to be processed includes radar traces, radio communications, images from EO sensors, multi-spectral satellite images and much more – far more traffic than normal internet activity. Moreover, for resilience reasons, re- dundancy will be required, to avoid concentrating everything on a single infrastructure. So, if it is true that the most advanced countries think today in terms of the ‘combat cloud’ at joint and national levels, the architecture of these clouds will be far more complex than a large data centre, and will be based both on redundancy and on different circles of authorisation, sorted by user and theme/topic. For example, each military service will have its own system, interconnected with the others and from which all will be able to draw the information considered to be of common utility. Another method is dictated by multinational programmes. For example, the SCAF/FCAS or TEMPEST programmes will have common databases that can be accessed by all partner countries. However, a good com- promise will have to be found between free access by all partners and single countries’ exclusive interests. In short, great clouds will not be gigantic linear hard disks, but rather a multi-layered network of nodes with stratified, scalable sharing and security rules, defined by the end-users. Alliances such as NATO will represent a further subgroup, to which each national cloud contributes, to establish a core of shared information for use in common campaigns. AI and the Multi-Layered Cloud AI will have to support multiple and multi-layered storage and comput- ing capacities. Indeed, even for AI applications, there will be no single, monolithic capability. AI elements will be present both at the storage in- frastructure (decision-making support) and onboard the aircraft, with dif- ferent degrees of autonomy and different capabilities. For instance, at the operational level, AI will often work by inference – it will not have to learn anew, but will simply put into practice what it has learned elsewhere. A single sensor, or a single family of data, will feature elementary applica- tions for cognitive management only. But complex applications, capa- ble of merging numerous heterogeneous data before returning them to the aggregation infrastructure, will be needed at some point. The level at which this will be possible will depend on managing the computing power to be delivered in-flight. In effect, AI and supercomputing are energy-in- tensive functions that are stimulating research into both energy efficiency and the search for new supply sources and storage methods. For air- borne uses, computing power (and therefore energy) will depend on the engine, and the balance between the aircraft’s propulsion and electronic equipment. f