HPC project to build AI-based phased arrays with GPUs
Julia Computing is based around the Julia programming language – offered as the fastest high-performance open source computing language for data, analytics, algorithmic trading, machine learning, artificial intelligence, and other scientific and numeric computing applications. Its flagship product is JuliaHub, a secure, software-as-a-service platform for developing Julia programs, deploying them, and scaling to thousands of nodes.
The DARPA funding was awarded as part of the agency’s Tensors for Reprogrammable Intelligent Array Demonstrations (TRIAD) program, a research area under the agency’s Artificial Intelligence Exploration (AIE) program. TRIAD addresses electromagnetic (EM) phased arrays, which contain many antenna elements that transmit and receive EM signals that are combined in such a way as to manipulate the spatial directions of the signals into beams.
Such beams can be used to transmit a signal in a specific direction or receive only those signals of interest from a desired direction, while ignoring signals emanating from other directions. The fundamental math operation behind phased arrays involves a complex multiplication of rows of coefficients, one per beam, by every array element. The beam forming operation is then followed by information extraction and processing.
“In current digital phased arrays,” says the agency in its TRIAD program notice, “every element is digitized, leading to an explosion of data requiring billions to trillions of complex operations per second. Currently, racks of high-powered processors are used in many stages of processing to handle the data processing challenge. TRIAD will create a streamlined processing approach to manage the beam forming and information processing directly within the array to significantly cut down on processing time and cost.”
This project, says the company, is a further step in its extensive machine learning research program.
“In recent years,” says project Principal Investigator and Julia Computing CTO Keno Fischer, “phased array radio frequency (RF) systems have found increasing popularity, from the SpaceX Starlink antenna to modern consumer communications standards like 5G and Institute of Electrical and Electronics Engineers (IEEE) 802.11 WiFi. With the increasing availability of low-cost radio integrated circuits (ICs) with excellent performance characteristics, the further applicability of phased array systems is now highly constrained by the availability of high performance signal processing systems capable of handling the high data rates produced by all digital phased arrays. In this research program, we’re excited to bring Julia’s industry-leading graphics processing unit (GPU) compute capabilities to this rapidly growing domain.”
Julia Computing’s Dr. Elliot Saba, who is serving as co-PI on the project adds, “We are particularly excited by the possibility of integrating these capabilities into the larger Julia machine learning ecosystem. Because Julia provides compositionality by default, as well as language-level differentiable programming, we will be able to create a fully integrated system that performs both traditional signal processing, as well as novel ML-based inference simultaneously and in near-real time. This opens up a significant opportunity space to further enhance the performance of digital phased arrays, as well as extend this work to novel research areas such as bio-sensing radar.”
The company says that it is partnering with RF expert Professor Miguel Bazdresch at the Rochester Institute of Technology to demonstrate these capabilities on purpose-built phased array testbeds, constructed from low-cost massive multiple input and multiple output (MIMO) software defined radios (SDRs) and NVIDIA GPUs. Last month the company raised $24 million in Series A funding to scale scale its production of Julia solutions and new product development for pharmaceuticals, energy, finance, and other sectors.
Related articles:
MIT programming language aims to ‘change the technical world’
Julia programming startup raises funds to advance scientific computing
Scientific ML promises ‘near interactive’ design optimization speeds
New phased array radar architectures
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
