Real-time remote processing enabled by high speed Ethernet

Abstract: A growing trend within the technologies acting as enablers for 6g, such as Massive MIMO and Large Intelligence Surfaces, is benefiting from both the communication and the positioning aspects that they can provide. As these kinds of systems are employing a large number of arrays which provide high amounts of data, a distributed hardware approach having near-antenna processing is explored in this work. The emulated set-up consists of two FPGA boards, where one is mimicking the local processing which would take place near the LIS panels, while the other aggregates all the data gathered from the panels to have joint processing. The channel state information is gathered through four LIS panels and is processed locally with the help of deep neural networks in order to achieve the position of a user. The next step is sending this data onwards to the board which acts as the central processing unit in order to fuse it, thus achieving a better estimation of the position. This thesis is proposing a hardware implementation focusing on different optimization techniques that could be used in order to achieve a low-latency and high-throughput system. For real-time applications where latency is critical, such as, for example, autonomous vehicles – a good approach are hardware accelerators tailored exclusively to the function which needs to be implemented. Finally, the communication between hardware units is also managed, with the help of an Ethernet link.