To satisfy the immense bandwidth demands of 6G and millimeter-wave (mmWave) communications, wireless systems must deploy massive antenna arrays. These arrays provide the high beamforming gains necessary to overcome severe path loss and signal blockage at high frequencies. However, a fundamental physical bottleneck stands in the way: the radio frequency (RF) chain.
In a traditional, fully digital transceiver architecture, every single antenna element requires its own dedicated RF chain, which includes power-hungry and expensive components such as digital-to-analog converters (DACs), mixers, power amplifiers, and filters. As arrays scale to hundreds or thousands of elements, the cost, power consumption, and routing complexity of a fully digital architecture become unsustainable.
To bridge this gap, researchers have developed architectures that decouple the number of active RF chains from the number of physical antennas. By shifting a portion of the signal processing from the digital baseband into the analog or electromagnetic wave domains, these techniques achieve the performance of massive arrays at a fraction of the hardware overhead.
The Hybrid Precoding Paradigm
The most prominent solution to the RF chain bottleneck is hybrid beamforming. Instead of performing all precoding in the digital domain, a hybrid architecture splits the operation into two stages: a low-dimensional digital precoder connected to a small number of RF chains, followed by a high-dimensional analog precoder implemented using an analog network.
[Digital Baseband] ---> [Few RF Chains] ---> [Analog Phase-Shifter Network] ---> [Many Antennas]
Mathematically, this can be viewed as a matrix factorization problem. The ideal, fully digital precoding matrix $F \in \mathbb{C}^{N_t \times N_s}$ (where $N_t$ is the number of antennas and $N_s$ is the number of data streams) is factorized into the product of an analog precoding matrix $F_{RF} \in \mathbb{C}^{N_t \times N_{RF}}$ and a digital precoding matrix $F_{BB} \in \mathbb{C}^{N_{RF} \times N_s}$, where $N_{RF}$ is the number of RF chains.
Why does this approximation work without destroying system capacity? The answer lies in the physics of high-frequency channels. Millimeter-wave and Terahertz channels are typically spatially sparse, meaning signals propagate along only a few dominant paths. Because the channel's intrinsic degrees of freedom are low, we do not need full digital control over every antenna element to maximize capacity.
Foad Sohrabi et al. proved in Hybrid Digital and Analog Beamforming Design for Large-Scale Antenna Arrays that a hybrid architecture can exactly replicate the performance of a fully digital system, provided that the number of RF chains is at least twice the number of data streams ($N_{RF} \geq 2N_s$). To find the optimal matrices under strict hardware constraints, Xianghao Yu et al. introduced a framework in Alternating Minimization Algorithms for Hybrid Precoding in Millimeter Wave MIMO Systems that decouples the optimization of the unconstrained digital precoder and the element-constrained analog precoder, yielding near-optimal spectral efficiency.
Minimizing Analog Network Complexity
While hybrid beamforming successfully reduces the number of RF chains, conventional fully-connected hybrid designs still require a high-precision, adaptive phase shifter to connect every RF chain to every antenna element. As the array grows, the number of phase shifters scales quadratically, introducing a secondary hardware bottleneck.
To mitigate this, researchers have explored two structural variations:
1. Sub-Connected Architectures
Instead of mapping every RF chain to every antenna, a sub-connected architecture divides the antenna array into distinct sub-arrays, where each RF chain is mapped exclusively to its own dedicated subset of antennas.
Shiwen He et al. analyzed this approach in Energy-Efficient Transceiver Design for Hybrid Sub-Array Architecture MIMO Systems, demonstrating that while sub-connected architectures experience a slight reduction in beamforming flexibility compared to fully-connected designs, they drastically lower power consumption and hardware routing complexity by eliminating the vast majority of phase shifters.
2. Fixed Phase Shifters and Switch Networks
Another approach replaces expensive, continuously adjustable phase shifters with a small pool of simple, fixed-value phase shifters and a dynamic switch network.
In Hybrid Precoding in Millimeter Wave Systems: How Many Phase Shifters Are Needed?, Xianghao Yu et al. demonstrated that an architecture using a dynamic switch network connected to as few as 10 fixed phase shifters can achieve a spectral efficiency nearly identical to an ideal fully digital system. This shifts the hardware burden from complex analog circuitry to intelligent, algorithmic switch routing.
Precoding in the Wave Domain
A more radical departure from traditional architectures bypasses discrete analog circuits entirely by performing signal processing directly in the electromagnetic wave domain using metamaterials.
Reconfigurable Holographic Surfaces (RHS)
Instead of steering beams by altering the phase of a signal before it reaches an antenna, an RHS integrates the beamforming mechanism into the radiation surface itself. An RHS utilizes a continuous or ultra-dense arrangement of metamaterial elements fed by a surface wave. By electrically controlling the radiation amplitude of individual surface elements, the surface acts as a programmable holographic mask that shapes the escaping wavefronts.
As detailed by Ruoqi Deng et al. in Reconfigurable Holographic Surfaces for Future Wireless Communications, this holographic approach allows multiple beams to be steered simultaneously without requiring any phase shifters or dense RF routing networks, offering an ultra-thin, low-cost alternative for high-frequency communication.
Stacked Intelligent Metasurfaces (SIM)
Taking wave-domain processing a step further, Stacked Intelligent Metasurfaces (SIM) act as an analog, physical computer. A SIM consists of multiple layers of programmable metasurfaces stacked in front of a small array of source RF chains. As an electromagnetic wave passes through these layers, each layer applies a spatial phase shift to the wave.
[Source RF Chains] ---> [Metasurface Layer 1] ---> [Layer 2] ---> [Layer 3] ---> [Free Space]
(Wave-Domain Processing)
Marco Di Renzo et al. introduced this concept in Stacked Intelligent Metasurfaces for Multiuser Beamforming in the Wave Domain, showing that a SIM can physically compute the spatial precoding matrix at the speed of light as the wave propagates through the structure. This architecture completely eliminates the need for digital multiuser precoding processors and high-resolution data converters.
This wave-domain computing capability extends naturally to complex spatial tasks. For instance, Xing Jia et al. demonstrated its utility for near-field beamfocusing in Stacked Intelligent Metasurface Enabled Near-Field Multiuser Beamfocusing in the Wave Domain, while Yong Liang Guan et al. utilized it to physically compute a 2D Discrete Fourier Transform for direction-of-arrival estimation in Two-Dimensional Direction-of-Arrival Estimation Using Stacked Intelligent Metasurfaces.
By substituting expensive, power-hungry electronic hardware with mathematical operations embedded directly into physical structures and analog networks, these hybrid and wave-domain architectures provide a scalable path forward for next-generation wireless networks.