Ferkans — Interactive Telecom Tutor

Why mmWave Demands Array-Fed RIS

Conventional RIS deployments at mmWave face a hard trilemma: (1) the product path loss $d_1^2 d_2^2$ is brutal at $28$ - $140$ GHz, (2) active BS arrays need to be huge ( $N_t \geq 64$ ) to close the link budget, and (3) digital beamforming at large $N_t$ over GHz of bandwidth is prohibitive in power and cost. The conventional answer is hybrid analog-digital beamforming (MIMO Ch. 20–21) — but even that requires tight coordination between many RF chains.

The array-fed RIS takes a radically different approach: use a small active array ( $N_t \sim 8$ - $32$ ) that is very close to a large passive RIS ( $N \sim 256$ - $4096$ ). The active array provides the baseband complexity (few RF chains, conventional digital precoding); the RIS provides the aperture gain (large physical area, passive). They combine to deliver massive-MIMO-like performance at $10$ - $20\%$ of the hardware cost. This is the CommIT Group's (Caire et al.) flagship architecture for 6G mmWave.

The golden thread: the RIS programs the propagation, but now its programmability is jointly designed with an active precoder that the RIS sees in the near field.

Array-Fed RIS

An architecture combining a small active antenna array with a large passive RIS panel placed in the array's near field. The active array provides digital baseband flexibility via few RF chains; the RIS provides aperture gain. The near-field BS-RIS channel is full-rank, enabling multi-user multiplexing through one passive surface.

RIS Eigenmode

A singular-vector pair $(\mathbf{u}_k, \mathbf{v}_k)$ of the BS-RIS channel $\mathbf{H}_1$ . The active array excites the $k$ -th eigenmode via $\mathbf{v}_k$ ; the RIS then receives an excitation $\mathbf{u}_k$ that can be focused toward user $k$ . Independent eigenmodes enable multi-user multiplexing in array-fed RIS.

Definition:
Array-Fed RIS Architecture

An array-fed RIS consists of:

A small active array of $N_t$ antennas at the BS.
A large passive RIS of $N \gg N_t$ elements.
A short array-RIS distance $d_{\text{AR}}$ such that the RIS is in the near-field of the array (i.e., $d_{\text{AR}} < d_F^{\text{AR}} = 2 D_A^2 / \lambda$ , where $D_A$ is the active-array aperture).

The BS-to-RIS channel $\mathbf{H}_1 \in \mathbb{C}^{N \times N_t}$ is a near-field channel, with rank governed by the near-field eigenmodes (not just the angular spread). Then $\mathbf{H}_2 \in \mathbb{C}^{N_r \times N}$ is the (typically far-field) RIS-to-UE channel.

The system model is identical to Chapter 3's cascaded channel: $\mathbf{y} = (\mathbf{H}_d + \mathbf{H}_2 \boldsymbol{\Phi} \mathbf{H}_1)\mathbf{W}\mathbf{s} + \mathbf{w}$ , but the near-field structure of $\mathbf{H}_1$ is the new engineering variable.

,

Theorem: Near-Field Eigenmode Rank of $\mathbf{H}_1$

For an array-fed configuration with active-array aperture $D_A$ , RIS aperture $D_R$ , array-RIS distance $d_{\text{AR}}$ , and wavelength $\lambda$ :

Far-field regime ( $d_{\text{AR}} \gg d_F^{\text{AR}} = 2\,\max(D_A, D_R)^2 / \lambda$ ): $\text{rank}(\mathbf{H}_1) \sim L$ where $L$ is the number of scattering paths (typically 1-3 in mmWave LoS).
Near-field regime ( $d_{\text{AR}} \ll d_F^{\text{AR}}$ ): $\text{rank}(\mathbf{H}_1) \approx \min(N_t, N_{\text{eff}})$ , where $N_{\text{eff}} \sim (D_A \cdot D_R) / (d_{\text{AR}} \lambda)$ is the near-field "number of degrees of freedom."

In the array-fed RIS, we deliberately place the array in the near-field of the RIS: $d_{\text{AR}} \sim$ a few $\lambda$ . The rank becomes approximately $\min(N_t, N)$ — full rank in $N_t$ .

In the far-field, a BS antenna "sees" the RIS as a single direction: $\mathbf{H}_1$ has rank $\sim 1$ per LoS path. In the near-field, each active-array element has a slightly different path length and angle to each RIS element, giving a full-rank $\mathbf{H}_1$ of rank $\min(N_t, N)$ under sufficient aperture geometry. Near-field gives us many eigenmodes; far-field gives us few.

Proof

Near-field spread function

Each active-array element sees each RIS element at slightly different range and angle. The per-element phase pattern across the RIS differs meaningfully across active-array elements (unlike far-field, where all active elements see a nearly identical plane wave).

Effective DoF

Standard near-field DoF theorem: the number of "resolvable beams" between two apertures of sizes $D_A$ and $D_R$ at distance $d$ is approximately $D_A \cdot D_R / (d \lambda)$ . This is a generalization of the aperture-angle DoF to the near-field.

Choose $d_{\text{AR}}$ small

To maximize DoF, minimize $d_{\text{AR}}$ . At $d_{\text{AR}} \sim$ few $\lambda$ : $N_{\text{eff}} \gtrsim N_t$ , so $\text{rank}(\mathbf{H}_1) = N_t$ full rank. This is the design sweet spot for array-fed RIS. $\blacksquare$

Array-Fed RIS Architecture Diagram — A small active array of $N_t = 8$ antennas mounted a few $\lambda$ in front of a large passive RIS with $N = 512$ elements. The array illuminates the RIS with near-field beams; the RIS reflects toward distant UEs with far-field coherent gain. Result: $\sim N_t$ independent streams, each with $\sim N$ aperture gain — hybrid-beamforming-like performance at a fraction of the cost.

Array-Fed RIS vs. Conventional Alternatives

Three architectural alternatives at mmWave:

Large fully-digital array ( $N_t = 256$ , $N_{\text{RF}} = 256$ ): best performance, prohibitive cost ( $\sim 256$ high-power ADCs at $> 20$ GHz).
Hybrid analog-digital ( $N_t = 256$ , $N_{\text{RF}} = 8$ ): trades performance for cost. $\sim 3$ - $5$ dB below fully-digital.
Array-fed RIS ( $N_t = 8$ active + $N = 512$ passive): trades both for hardware simplicity. Active array is conventional digital; RIS is near-field passive. Close to hybrid performance.

The array-fed RIS is the cost-effective third option. Its advantage grows as frequency increases (mmWave → sub-THz): active arrays become exponentially expensive, while passive RIS stays cheap per element.

Array-Fed RIS vs. Massive MIMO vs. Hybrid BF

Compare the sum-rate achievable by three architectures at matched cost: large fully-digital, hybrid analog-digital, and array-fed RIS. Vary the BS-UE distance; see how array-fed RIS dominates at mmWave/sub-THz where path loss is severe.

Parameters

Active array antennas

N_t

8

Passive RIS elements

N

512

Carrier frequency (GHz)28

Max BS-UE distance (m)100

Key Takeaway

Array-fed RIS: small active, large passive, near-field coupling. The active array provides digital-processing flexibility and multi-stream capability; the passive RIS provides aperture gain and physical coverage. They combine to approach massive-MIMO performance at $10$ - $20\%$ of the hardware cost — the key architectural argument for mmWave / sub-THz 6G.

⚠️Engineering Note

Array-Fed RIS Deployment

Practical array-fed RIS deployment:

Physical form: active array and passive RIS mounted on the same or adjacent structures, with $d_{\text{AR}} \sim$ few $\lambda$ (e.g., $5$ - $30$ cm at 28 GHz, $2$ - $10$ cm at 140 GHz).
Array-RIS alignment: the array must be near-field-centered on the RIS; off-axis alignment degrades the eigenmode structure.
Integration: active array typically mounts on a rooftop; RIS panel on an adjacent wall or building facade; they are connected via a known-phase backhaul.
Control latency: active array uses conventional BBU processing (ns latency). RIS controller uses a few-microsecond latency for phase reconfiguration.
Calibration: the near-field BS-RIS link requires one-time high-accuracy calibration at deployment; subsequent drift is slow (months).

Practical Constraints

•
Typical $d_{\text{AR}}$ at 28 GHz: $10$ - $30$ cm.
•
Total BS + RIS power: active array $\sim 10$ - $50$ W; RIS $\sim 1$ - $5$ W (passive).
•
RIS panel size: $0.5$ - $2$ m² at mmWave, smaller at sub-THz.
•
System cost: $\sim 5\times$ less than an equivalent fully-digital array.

Common Mistake: Don't Place the Array in the Far-Field of the RIS

Mistake:

"Put the array far from the RIS for convenience; the geometry doesn't matter too much."

Correction:

Array-fed RIS only works in the near-field configuration. If $d_{\text{AR}} > d_F^{\text{AR}}$ , the array-RIS channel becomes rank-1 (far-field), and the RIS sees the array as a single effective source — no eigenmode multiplexing. The architecture collapses to a conventional "weak BS + passive RIS" setup, losing all the multi-user multiplexing benefit. Always design $d_{\text{AR}} < d_F^{\text{AR}} / 5$ as a rule of thumb, with a safety margin.

The Array-Fed RIS Architecture