Summary

Chapter 28 Summary: Reconfigurable Intelligent Surfaces

Key Points

• 1.

  The RIS signal model describes a passive reflective array that applies element-wise phase shifts $\boldsymbol{\Theta} = \mathrm{diag}(\beta_1 e^{j\theta_1}, \ldots, \beta_N e^{j\theta_N})$ to the incident electromagnetic wave. The received signal $y = (\mathbf{h}_d^H + \mathbf{h}_r^H \boldsymbol{\Theta} \mathbf{G})\mathbf{w} x + n$ combines a direct BS-user path with a cascaded BS-RIS-user reflected path. The cascaded channel structure (product of two link gains) implies a "double path loss" that must be overcome by the coherent combining gain of $N$ elements.

• 2.

  Channel estimation for passive RIS is fundamentally challenging because the RIS cannot transmit or receive pilots. The cascaded channel has $O(NM)$ unknowns, requiring at least $N + 1$ pilot slots with $M$ orthogonal pilots per slot. Structured approaches including element grouping (reducing overhead to $O(N/G)$), DFT-based codebooks, compressed sensing (exploiting angular sparsity, $O(S\log N)$ overhead), and two-timescale estimation (separating quasi-static and mobile channels) can significantly reduce the training burden.

• 3.

  Joint active-passive beamforming optimises the BS precoder $\mathbf{w}$ and RIS phases $\boldsymbol{\phi}$ simultaneously. The problem is non-convex due to unit-modulus constraints $|\phi_n| = 1$. Alternating optimisation decomposes the problem into a convex active subproblem (MRT for given phases) and a unit-modulus passive subproblem, converging monotonically to a stationary point. Semidefinite relaxation (SDR) provides an approximation ratio of $\pi/4$ but has $O(N^{3.5})$ complexity. Manifold optimisation on the product of complex circles offers the best quality-complexity trade-off for large $N$.

• 4.

  The $N^2$ SNR scaling law states that with optimal phase alignment, the received SNR scales as $(\sum_{n=1}^N |h_{r,n}||g_n|)^2 \approx N^2 \mu_h^2 \mu_g^2$ — quadratically in the number of RIS elements. This contrasts with the linear ($N$) scaling of relays and massive MIMO. The $N^2$ gain arises from coherent combining across both hops of the cascaded channel. A crossover analysis shows that RIS outperforms a relay when $N > \sqrt{\alpha_r N_r / \alpha_s}$, where $\alpha_r$ and $\alpha_s$ capture the respective path loss parameters.

• 5.

  Phase quantisation with $b$ bits per element causes an average gain loss of $\mathrm{sinc}^2(\pi/2^b)$: 3.9 dB for 1-bit, 0.91 dB for 2-bit, and 0.22 dB for 3-bit. The $N^2$ scaling is preserved regardless of quantisation. Wideband beam squint, hardware imperfections (mutual coupling, amplitude variations), and the active-vs-passive RIS trade-off are additional practical considerations. Active RIS with per-element amplification addresses double path loss but introduces noise amplification and higher power consumption.

• 6.

  RIS versus competing technologies: RIS is most beneficial when the direct link is severely blocked, the RIS can be placed close to the user or the BS (reducing one hop of the double path loss), and a large surface area is available. For short-range line-of-sight scenarios, a small relay or additional BS antenna may be more cost-effective. Open problems include standardisation, physically accurate channel modelling, multi-RIS coordination, and integration with sensing and localisation.