The entire hybrid beamforming field has been benchmarking architectures using radiated antenna power as the comparison metric, implicitly accepting that the analog RF network between digital chains and antennas is inherently lossy. Zlatanov and Salakhov reveal that this loss was never physically necessary — it was an artifact of using contractive network topologies.
Conventional hybrid beamforming uses fully-connected splitter-phase-shifter-combiner networks to route signals from a small number of digital RF chains to a large antenna array. These networks are contractive by construction: even with ideal lossless passive components, the splitting and recombining process dissipates power. The community has treated this contraction as a fundamental cost of reducing the number of expensive digital chains.
The proposed architecture replaces the contractive network with a programmable unitary RF network — one that preserves all injected power by mathematical construction. The implementation uses an interlaced mixer-phase architecture: fixed (non-programmable) mixing layers interleaved with programmable diagonal phase-shifting layers. All power injected by the RF chains reaches the antenna ports without loss.
The closed-form digital beamformer and low-complexity analog programming method produce a hybrid precoder that closely matches fully-digital performance. In narrowband simulations, both continuous-phase and 6-bit quantized versions are nearly indistinguishable from the fully-digital benchmark, while conventional baselines suffer significant losses under fair power comparison.
The “fundamental gap” between hybrid and fully-digital beamforming was a design choice, not a physics constraint. Replacing contractive RF networks with unitary ones eliminates the gap entirely, revealing that the performance penalty was in the topology of the analog network, not in the principle of hybrid processing.