Prerequisites & Notation

Before You Begin

This chapter is the information-theoretic culmination of Part III. It takes the MIMO outage analysis of Chapter 10 and the pairwise-error analysis of space-time codes from Chapter 11 and binds them into a single, tight asymptotic identity β€” the Zheng-Tse diversity-multiplexing tradeoff. To follow the proof the reader needs the outage-probability / ergodic-capacity distinction, the distribution of the eigenvalues of HHH\mathbf{H}\mathbf{H}^{H} (Wishart), the rank and determinant criteria for space-time codes, and the Gallager-style random-coding argument in its high-SNR / error-exponent flavour. A working familiarity with exponential-equality asymptotics (f(SNR)≐SNRaf(\text{SNR}) \doteq \text{SNR}^{a}) as used in the large-deviations literature is strongly recommended.

  • MIMO outage probability and ergodic capacity(Review ch10)

    Self-check: Can you write the outage probability Pout(R)P_{\rm out}(R) for an ntΓ—nrn_t \times n_r i.i.d. Rayleigh channel at target rate RR, and distinguish it operationally from the ergodic capacity CΛ‰=EHlog⁑det⁑(I+SNRntHHH)\bar C = \mathbb{E}_{\mathbf{H}} \log\det(\mathbf{I} + \tfrac{\text{SNR}}{n_t}\mathbf{H}\mathbf{H}^{H})?

  • Wishart-eigenvalue distribution of HHH\mathbf{H}\mathbf{H}^{H}(Review ch10)

    Self-check: Can you state the joint density of the nonzero eigenvalues of HHH\mathbf{H}\mathbf{H}^{H} when H\mathbf{H} is nrΓ—ntn_r \times n_t with i.i.d. CN(0,1)\mathcal{CN}(0, 1) entries? And identify the Vandermonde factor ∏i<j(Ξ»iβˆ’Ξ»j)2\prod_{i<j} (\lambda_i - \lambda_j)^2 that drives the DMT exponent?

  • Rank and determinant criteria for space-time codes(Review ch11)

    Self-check: Can you state the rank criterion (rank(Ξ”)β‰₯r\mathrm{rank}(\boldsymbol{\Delta}) \ge r implies diversity rnrr n_r) and the determinant criterion (det⁑(ΔΔH)\det(\boldsymbol{\Delta}\boldsymbol{\Delta}^H) controls coding gain) for the codeword error matrix Ξ”=\ntnXβˆ’\ntnX^\boldsymbol{\Delta} = \ntn{X} - \hat{\ntn{X}}?

  • Alamouti scheme and V-BLAST (ZF and ML) receivers(Review ch11)

    Self-check: Can you compute the diversity order of Alamouti on an nrn_r-receive channel (2nr2 n_r), and of zero-forcing V-BLAST on an ntΓ—nrn_t \times n_r channel with nrβ‰₯ntn_r \ge n_t (nrβˆ’nt+1n_r - n_t + 1)? And explain why the latter is so much smaller than ntnrn_t n_r?

  • Gallager random-coding exponent and exponential equality(Review ch13)

    Self-check: Can you state f(SNR)≐SNRaf(\text{SNR}) \doteq \text{SNR}^{a} iff lim⁑log⁑f/log⁑SNR=a\lim \log f / \log \text{SNR} = a, and describe why additive o(log⁑SNR)o(\log\text{SNR}) terms β€” constants, polylog prefactors β€” are invisible to this notation?

  • Outage exponent of a scalar Rayleigh channel(Review ch11)

    Self-check: Can you show that for a scalar Rayleigh channel at fixed rate RR, Pout(R)≐SNRβˆ’1P_{\rm out}(R) \doteq \text{SNR}^{-1} (diversity order 1)? And extend this to Alamouti on an nrn_r-receive channel (≐SNRβˆ’2nr\doteq \text{SNR}^{-2 n_r})?

Notation for This Chapter

Symbols specific to the DMT analysis. The Chapter 10–11 MIMO notation (channel matrix H\mathbf{H}, codeword matrix \ntnX\ntn{X}, SNR SNR\text{SNR}, noise w\mathbf{w}) continues to apply and is not repeated here.

SymbolMeaningIntroduced
rrMultiplexing gain, r=lim⁑SNRβ†’βˆžR(SNR)/log⁑SNRr = \lim_{\text{SNR}\to\infty} R(\text{SNR}) / \log \text{SNR}s01
dβˆ—d^*Diversity gain, dβˆ—=βˆ’lim⁑SNRβ†’βˆžlog⁑Pout(R(SNR))/log⁑SNRd^* = -\lim_{\text{SNR}\to\infty} \log P_{\rm out}(R(\text{SNR})) / \log \text{SNR}s01
dβˆ—(r)d^*(r)The diversity-multiplexing tradeoff (DMT) curve; maximum achievable diversity at multiplexing gain rrs02
≐\doteqExponential equality: f(SNR)≐SNRaf(\text{SNR}) \doteq \text{SNR}^{a} iff lim⁑log⁑f/log⁑SNR=a\lim \log f / \log \text{SNR} = as01
Ξ”\boldsymbol{\Delta}Codeword difference (error) matrix, Ξ”=\ntnXβˆ’\ntnX^\boldsymbol{\Delta} = \ntn{X} - \hat{\ntn{X}}, size ntΓ—Ln_t \times Ls04
nt,nrn_t, n_rNumber of transmit and receive antennass01
LLSpace-time codeword block length (number of channel uses per block). Also written TT in some texts.s05
Rt\mathbf{R}_tTx spatial correlation matrix (Hermitian positive-semidefinite, size ntΓ—ntn_t \times n_t)s05
R(SNR)R(\text{SNR})Target rate at SNR SNR\text{SNR}, scaling as R(SNR)=rlog⁑2SNR+o(log⁑SNR)R(\text{SNR}) = r \log_2 \text{SNR} + o(\log \text{SNR})s01
Ξ»i(A)\lambda_i(\mathbf{A})ii-th eigenvalue of Hermitian A\mathbf{A}, ordered Ξ»1β‰₯Ξ»2β‰₯β‹―\lambda_1 \ge \lambda_2 \ge \cdotss02
min⁑(nt,nr)\min(n_t, n_r)Spatial degrees of freedom; the maximum achievable multiplexing gains01