Chapter Summary

Chapter 16 Summary

Key Points

• 1.

  The Marchenko-Pastur law describes the limiting spectral distribution of large i.i.d. Gram matrices. Its density is supported on $[(1-\sqrt{\beta})^2, (1+\sqrt{\beta})^2]$ and is the bedrock of asymptotic MIMO analysis.

• 2.

  The Stieltjes transform $m_\mu(z) = \int (x-z)^{-1}\,d\mu(x)$ summarizes a spectrum into a single analytic function. Fixed-point equations for $m$ yield the limiting density via Plemelj's inversion formula.

• 3.

  In MIMO with i.i.d. Rayleigh fading, the LMMSE output SINR concentrates around a deterministic fixed-point $\gamma^* = \text{SNR}/(1 + \beta\gamma^*/(1+\gamma^*))$. This replaces per-realization Monte Carlo with an analytic formula for system design.

• 4.

  The Donoho-Tanner phase transition is a sharp cliff $\rho^*(\delta)$ in the $(M/N, s/M)$ plane. Below the curve, $\ell_1$-recovery succeeds with probability one; above it, recovery fails. The curve is the right benchmark for any practical sparse-recovery algorithm.

• 5.

  Approximate Message Passing saturates the Donoho-Tanner curve with $O(MN)$ per-iteration complexity, making the asymptotic benchmark operationally achievable.

• 6.

  Free probability linearizes the addition and multiplication of asymptotically free random matrices: the R-transform adds under $\boxplus$, the S-transform multiplies under $\boxtimes$. Together with the Stieltjes transform, these form the core analytical machinery for composite random matrix expressions.

• 7.

  Deterministic equivalents are sharp: the gap between random and limit shrinks as $O(1/n_r)$. For $n_r \geq 16$ the prediction error is typically below 0.3 dB in massive MIMO link budget analysis.