Efficient Beamspace Processing for Large-Scale Antenna Arrays

Modular tiled architectures can simplify RF packaging in large-scale antenna arrays and also offer practical advantages for digital signal processing (DSP) in Multi-User MIMO (MU-MIMO) systems. By applying beamspace processing techniques that exploit the sparsity of mm-wave channels, our proposed architecture shifts most DSP tasks to occur locally within each tile. This design enables dimensionality reduction and parallelization, which helps lower the computational workload for both inference and training compared to traditional full-array setups. Additionally, these methods help reduce the length of training periods needed. This work is a collaboration with Professor Upamanyu Madhow’s group at UCSB, as part of the SRC/DARPA JUMP 2.0 CUbiCCenter.