Prerequisites & Notation

Before You Begin

This chapter addresses a physical reality that the previous seventeen chapters systematically ignored: when the base-station aperture becomes comparable to the propagation range, the channel is no longer spatially stationary. A user no longer illuminates every antenna of an extra-large array (XL-MIMO); instead it reaches only a subset we will call a visibility region (VR). Estimating the channel now requires first detecting where the user lives on the array and then estimating the channel restricted to that region. We assume familiarity with the following prior material.

  • Near-field propagation: Fraunhofer distance, spherical wavefront, beam focusing(Review ch17)

    Self-check: Can you compute dF=2D2/Ξ»d_F = 2D^2/\lambda for a given aperture DD and wavelength Ξ»\lambda, and explain when a user at range rr is in the near or far field?

  • LS and MMSE channel estimation; pilot contamination(Review ch03)

    Self-check: Can you write the LS and MMSE estimators for a MIMO pilot observation model Yp=HSi,k+w\mathbf{Y}_p = \mathbf{H}\mathbf{S}_{i,k} + \mathbf{w} and state the MSE of each?

  • Spatial correlation matrices and the one-ring model(Review ch02)

    Self-check: Can you explain how Rk\mathbf{R}_k encodes the angular support of the user's multipath cluster?

  • Sparse recovery: LASSO, OMP, compressed sensing guarantees(Review ch12)

    Self-check: Can you state when β„“1\ell_1 minimization recovers a ss-sparse vector from mm noisy linear measurements?

  • Markov random fields, Ising model, belief propagation / loopy BP

    Self-check: Can you write the joint distribution of a 2D Ising model with nearest-neighbor coupling JJ and external field hh?

  • EM algorithm for latent-variable estimation

    Self-check: Can you write the E- and M-steps of EM for a Gaussian observation model with a discrete latent state?

Notation for This Chapter

Symbols introduced or specialized in this chapter. Throughout the chapter we use Rk\mathbf{R}_k for the per-user spatial covariance matrix (local to this chapter, not a global \ntn{} symbol) and RkR_k (plain italic) for achievable rate. See NGlobal Notation Table for the master table.

SymbolMeaningIntroduced
NtN_tNumber of BS antennas (XL-MIMO regime, typically Ntβ‰₯256N_t \geq 256)s01
N1,N2N_1, N_2Horizontal and vertical antenna counts for a UPA, Nt=N1N2N_t = N_1 N_2s01
mk∈{0,1}Nt\mathbf{m}_k \in \{0,1\}^{N_t}Binary visibility-region mask for user kk: mk,n=1m_{k,n} = 1 iff antenna nn is illuminateds01
VkβŠ†{1,…,Nt}\mathcal{V}_k \subseteq \{1,\ldots,N_t\}Visibility-region support: Vk={n:mk,n=1}\mathcal{V}_k = \{n : m_{k,n} = 1\}s01
∣Vk∣|\mathcal{V}_k|Visibility-region cardinality (number of illuminated antennas)s01
JJCoupling strength of the 2D Markov (Ising) prior on mk\mathbf{m}_ks02
hhExternal field of the 2D Markov prior (biases toward active / inactive)s02
Rk∈CNtΓ—Nt\mathbf{R}_k \in \mathbb{C}^{N_t \times N_t}Spatial covariance of the unmasked channel of user kk (chapter-local)s01
SSNumber of subarrays partitioning the XL-MIMO arrays03
Ms=Nt/SM_s = N_t / SAntennas per subarrays03
dF=2D2/Ξ»d_F = 2D^2/\lambdaFraunhofer (far-field) distance of an aperture DDs04
Apolar\mathbf{A}_{\text{polar}}Polar-domain (angle Γ—\times range) dictionary for near-field sparse recoverys04
(ψq,rq)(\psi_q, r_q)qq-th polar grid point: azimuth cosine and ranges04
L(mk,Hk)\mathcal{L}(\mathbf{m}_k, \mathbf{H}_{k})Joint log-posterior of VR mask and channel given pilot observationss05
qnq_nVariational marginal qn=Pr⁑[mk,n=1∣Yp]q_n = \Pr[m_{k,n} = 1 \mid \mathbf{Y}_p]s05
Si,k\mathbf{S}_{i,k}Pilot matrix (uplink training), Si,k∈CKΓ—Ο„p\mathbf{S}_{i,k} \in \mathbb{C}^{K \times \tau_p}s01
Ο„p\tau_pPilot length in symbolss01