5G Baseband and Edge Computing
In a centralized or cloud radio access network (C-RAN), baseband cabinets move from the bottom of mobile operator masts to edge-computing facilities in the fronthaul, where centralized baseband can be dynamically shared between radioheads. Massive co-located artificial intelligence (AI) is used to optimize the baseband performance over widely varying radiohead installations and traffic, and provide the AI heavy lifting for robots, drones, and other user equipment.
5G baseband and high-performance computing (HPC) meet in mobile edge-computing centers, and like HPC, the latest digital, network, and machine learning (ML) technologies are driving the market. Fundamental to this are digital design and signoff tools with class-leading power, performance, and area (PPA) outcomes on advanced nodes, where SoC designs incorporate increasing numbers of CPU and AI processor cores.
To enable this, Cadence provides very high capacity Cadence Advanced-Node Solutions for digital and analog, and Arm-based solutions. But as these multicore HPC SoCs reach the limits of reticle size, there is also a move to disaggregated SoCs that combine heterogeneous chiplets on a substrate for yield and mixed-node benefits, and here Cadence provides 3D-IC Design Solutions with the UltraLink D2D PHY IP for chiplet interconnection. Finally, providing dense HPC in small mobile edge-compute centers demands very high-speed board designs via Allegro PCB Designer High-Speed Option, and electromagnetic impact analysis of multi-fabric gigabit-speed signal pathways with Cadence Electromagnetic Solutions and system thermal analysis with Cadence Thermal Solutions.
Neural networks are now being developed and deployed in a wide range of markets, from communications to surveillance. The computational, power, and memory requirements to process this data are continuously increasing, with new networks and new ways to approach deep learning every day. The Cadence Tensilica DNA processor family offers a much-needed breakthrough in terms of energy efficiency and performance to meet the requirements of on-device artificial intelligence (AI), including AI software development and deployment. Cadence PCIe and CXL IP and DDR IP further complement the Cadence on-device AI.
5G software stacks will be some of the largest and most complex in the industry. Early software development is critical for successful baseband and edge computing SoCs, and Cadence hardware verification platforms such as the Palladium Z1 Enterprise Emulation System and the Protium S1 FPGA Prototyping Platform enable early software bring-up and development on work-in-progress SoC designs incorporating CPU and AI processor cores. TCAM is natively supported in the Palladium Z1 platform for SoC switch design.
For the highest-possible communication performance, Cadence provides the Cadence 112G SerDes IP. Its unique firmware-controlled adaptive power design provides optimal power and performance tradeoffs and more efficient system designs based on platform requirements. This DSP-based architecture provides superior data recovery for lossy and noisy channels.
5G infrastructure, in common with handset design, needs high-performance and highly algorithmic digital blocks for beamforming, channelizing, and signal conditioning. With AWR VSS software, wireless design teams apply behavioral, impedance-aware RF models, phased array antenna systems, and baseband signal processing blocks to develop and optimize system architectures and determine component specifications for the best overall performance. Users can simulate system metrics, such as adjacent channel power ratio (ACPR), bit-error rate (BER), and error vector magnitude (EVM), with pre-configured and user-defined virtual testbenches, as well as identify the source of spurious products and other system impairments The phased array generated supports the synthesis of user-configured antenna arrays, beam steering signal conditioning blocks, and feed network. The VSS model development kit (MDK) can be used to create custom blocks in C++.
Blocks modeled in C++ or SystemC can be retargeted and synthesized by Stratus High-Level Synthesis (HLS) for the highly demanding requirements of high-performance baseband infrastructure and creating highly efficient, low-power RTL for implementation with the Cadence Low-Power Solution.
Finally for the baseband itself, Cadence provides the Tensilica ConnX B20 DSP IP in multi-core configurations for precoding/combining, beam measurement, and tracking.