Spectrum-Aligned DFT for THz Massive MIMO: A Fully Digital Transmit Precoding Solution to the Bandwidth Disaster

Abstract

The transition to terahertz (THz) and millimeter-wave (mmWave) frequencies is essential for meeting the extreme data-rate demands of 6G wireless networks. However, the associated ultra-wide bandwidth induces severe beam squint—a frequency-dependent beam misalignment that drastically reduces array gain and spectral efficiency. Conventional approaches face an irreconcilable trade-off: fully digital processing offers high accuracy but with prohibitive computational complexity, while hardware-based true-time-delay (TTD) networks incur excessive power consumption and insertion loss. In this paper, we propose a novel fully-digital architecture termed Spectrum-Aligned DFT (SADFT), which eliminates beam squint purely in the digital domain via mathematical reconstruction, designed for a fully-digital transmit architecture. The SADFT organizes the computation as a single-path pruned radix-2 recursive signal flow graph—a computational structure that enables regular in-place memory access, pipeline-friendly stage-wise multiplier reuse, and direct mapping to existing FFT hardware accelerators by reconfiguring only the twiddle factor generation logic—that directly targets the frequency-variant spatial frequencies, achieving mathematically exact alignment (gain loss bounded by <inline-formula> <tex-math notation="LaTeX">$10^{-4}$ </tex-math></inline-formula> dB for the simulated configurations) through virtual dense-grid mapping and path pruning. We rigorously extend the framework to two-dimensional uniform planar arrays (UPAs), proving mathematically exact separability into cascaded 1D operations that achieve the theoretical lower bound of <inline-formula> <tex-math notation="LaTeX">$\mathcal {O}(M)$ </tex-math></inline-formula> complexity for an <inline-formula> <tex-math notation="LaTeX">$M$ </tex-math></inline-formula>-element array. Extensive simulations using THz-specific Saleh–Valenzuela multipath models and the 3GPP CDL-A clustered channel model at 300 GHz demonstrate that SADFT closely approaches the Shannon capacity, delivering over <inline-formula> <tex-math notation="LaTeX">$3\times $ </tex-math></inline-formula> capacity improvement over standard FFT-based beamforming in ultra-wideband scenarios and effectively removing the baseband computational barrier that has historically prevented fully-digital wideband beamforming from being computationally feasible. The proposed architecture offers a scalable, low-complexity solution for future extremely large-scale antenna arrays (ELAAs) in 6G systems.