Revolutionizing Radio Networks: SoftBank and NVIDIA’s Triumph in AI-Driven Massive MIMO

Press play to listen to this content Revolutionizing Radio Networks: SoftBank and NVIDIA’s Triumph in AI-Driven Massive MIMO. a successful demonstration recently utilized an ai-ran (artificial intelligence radio access network) that performed radio signal processing entirely on nvidia’s graphics processing unit (gpu). this is a major shift from traditional systems that heavily relied on dedicated hardware, specifically field-programmable gate arrays (fpgas) or application-specific integrated circuits (asics), for processing at the physical layer (phy). performance under real-world conditions. in the outdoor trial, softbank exhibited top-tier performance under real-world circumstances. they used a cutting-edge, virtualized ran (vran) structure that performs this processing by using large-scale cuda-accelerated parallel computation on nvidia’s gpu. this architecture also incorporates ai processing. throughout the trial, aitras exploited the gpu’s efficient parallel processing skills to carry out extensive matrix computations crucial for massive mimo and phy-layer signal processing. this was achieved entirely in software within the distributed unit (du). consequently, softbank affirmed the stable function of a 16-layer mu-mimo downlink in outdoor settings. in comparison to the standard four-layer setup, both spectral efficiency and throughput improved by about three times. significant validation. this success serves as a substantial validation of the phy-layer execution on a gpu. it enables stable massive mimo operations within the specified processing time of the ran. this is an important technical milestone that leads the commercialization of ai-ran. in the aitras model, the connection between the du and o-ran-compliant radio unit adheres to o-ran split option 7.2x. while some massive mimo implementations transfer parts of the uplink channel estimation and equalization processing from the du to the o-ru, aitras uses a fully software-based approach within the du. this uses the gpu computing power and aids in creating a massive mimo ecosystem using general-purpose o-rus without needing extra functions. uplink radio signals and ai processing. by combining uplink radio signals from various o-rus into a single du, equipped with an integrated ai processing unit, ai-based coordinated control for radio signal quality can be effectively executed across multiple o-rus. this improves radio signal quality through uplink channel interpolation and boosts capacity through advanced beamforming techniques. the aim with the development of aitras is to speed up the practical implementation of ai-ran, contributing to a more sustainable and flexible next-generation network infrastructure. the plan is to continue field trials and introduce aitras into commercial networks by 2026. questions & answers. what is the significance of the recent demonstration by softbank? the demonstration is significant as it showed that radio signal processing can be performed entirely on nvidia’s gpu using ai-ran, shifting away from the traditional reliance on dedicated hardware. what is the anticipated impact of ai-ran in real-world conditions? ai-ran demonstrated high performance under real-world conditions, improving both spectral efficiency and throughput by about three times. this could significantly enhance communication quality and overall capacity of base stations. what is the long-term goal for the development of aitras? the goal is to accelerate the practical implementation of ai-ran, thereby contributing to the creation of a more sustainable and flexible next-generation network infrastructure. softbank plans to introduce aitras into its commercial network by 2026.