Ferkans — Interactive Telecom Tutor

The Most General Linear STC: Hassibi-Hochwald

At this point we have a neat taxonomy of linear space-time codes: Alamouti (orthogonal, rate 1, $n_t = 2$ ), OSTBC (orthogonal, rate $\le R_{\max}(n_t)$ , $n_t \ge 1$ ), QOSTBC (quasi-orthogonal, rate 1, half-diversity, $n_t = 4$ ), V-BLAST (no structure, rate $n_t$ , diversity 1). Each of these is a specific choice of the dispersion matrices $\{\mathbf{A}_q, \mathbf{B}_q\}$ . The question Hassibi and Hochwald asked in 2002 was: can we just let the dispersion matrices be anything and optimise?

Their answer — the linear dispersion code (LDC) framework — is that every linear STBC is parameterised by a tuple $(Q, \{\mathbf{A}_q, \mathbf{B}_q\})$ with $Q$ dispersion matrices in $\mathbb{C}^{n_t \times T}$ . The OSTBC, QOSTBC, and V-BLAST constraints are all special cases — subsets of the LDC parameter space cut out by algebraic conditions. By searching the LDC parameter space with mutual-information as the objective (not orthogonality as the constraint), we can find codes that approach the MIMO ergodic capacity for any target rate, at any block length $T$ .

This is the culmination of the linear-STC story and the bridge to Chapter 12 (Zheng-Tse DMT) and Chapter 13 (CDA codes). LDCs don't solve the DMT problem by themselves — they are a design framework, a space in which to optimise — but they are the right language for every subsequent space-time code. Every theorem we prove about LDCs specialises to Alamouti, OSTBC, QOSTBC, and V-BLAST by plugging in their dispersion matrices; every code we construct lives in the LDC parameter space.

Definition:
Linear Dispersion Code (LDC)

A linear dispersion code (LDC) is a space-time block code with block length $T$ , input symbols $\{s_q\}_{q=1}^K$ drawn from a constellation $\mathcal{X}$ , and codeword matrix of the form $\mathbf{X} \;=\; \sum_{q=1}^{K} \bigl(\alpha_q \mathbf{A}_q + j\beta_q \mathbf{B}_q\bigr), \qquad s_q = \alpha_q + j\beta_q,$ parameterised by dispersion matrices $\{\mathbf{A}_q, \mathbf{B}_q\} \subset \mathbb{C}^{n_t \times T}$ . The code is specified by the $Q = 2K$ matrices $\{\mathbf{A}_q\} \cup \{\mathbf{B}_q\}$ (no further constraints — they may or may not satisfy orthogonality relations).

The rate is $R = K/T$ complex symbols per channel use. The code is capacity-compatible (a term of art from Hassibi-Hochwald) if it is designed to approach the ergodic MIMO capacity at the corresponding ergodic channel.

Every linear STBC in this chapter is an LDC:

Alamouti: $Q = 4$ , matrices as listed in the remark of Def. DDispersion Matrix Expansion of a Linear STBC.
TJC rate- $3/4$ OSTBC for $n_t = 4$ : $Q = 6$ (3 complex symbols, $2 \times 3 = 6$ real dispersion matrices).
Jafarkhani QOSTBC: $Q = 8$ (4 complex symbols).
V-BLAST: $Q = 2 n_t$ with very simple dispersion matrices (columns of $\mathbf{I}_{n_t}$ and $j\mathbf{I}_{n_t}$ ).

The LDC framework is therefore not a new family of codes — it is a reparameterisation that lets you search for better codes by optimising over the dispersion matrices.

Theorem: LDCs Achieve the MIMO Ergodic Capacity

Consider the i.i.d. Rayleigh ergodic MIMO channel with $n_t$ Tx and $n_r$ Rx antennas and transmit power constraint $\mathbb{E}[\|\mathbf{X}\|_F^2]/T \le E_s$ . For any $T \ge 1$ and any rate $R$ below the ergodic capacity $C_{\mathrm{erg}} = \mathbb{E}_{\mathbf{H}}[\log_2 \det(\mathbf{I}_{n_r} + (\text{SNR}/n_t)\mathbf{H}\mathbf{H}^{H})]$ , there exists a linear dispersion code with $Q \le 2 n_t n_r T$ dispersion matrices achieving rate $R$ with vanishing error probability as the outer code block length grows.

Equivalently: any rate $R < C_{\mathrm{erg}}$ is achievable by an LDC of dimension $Q \le 2 n_t n_r T$ . LDCs are capacity-universal — they can approach the ergodic MIMO capacity for any $(\text{SNR}, n_t, n_r)$ .

The proof is a Gaussian-input argument: if the $Q$ dispersion matrices span a sufficiently high-dimensional subspace of $\mathbb{C}^{n_t \times T}$ , then the transmitted codeword $\mathbf{X}$ has essentially i.i.d. Gaussian entries (by the central limit theorem applied to the dispersion expansion with many symbols). An i.i.d. Gaussian input achieves the MIMO ergodic capacity (Telatar 1995, Foschini 1996). So an LDC with $Q$ matrices spanning enough of the space inherits this optimality.

The concrete number $Q \le 2 n_t n_r T$ is the dimension of the codeword matrix space $\mathbb{C}^{n_t \times T}$ over $\mathbb{R}$ : $2 n_t T$ for a single receive antenna, multiplied by $n_r$ for coherent combining. Beyond this, extra dispersion matrices are redundant.

Show Hint

Substitute the LDC codeword expansion into the ergodic capacity formula.

Invoke Telatar-Foschini: i.i.d. $\mathcal{CN}$ input on each $\mathbf{X}_{ij}$ achieves capacity.

Show that a sufficiently rich LDC realises i.i.d. Gaussian $\mathbf{X}$ .

Proof

Ergodic capacity with Gaussian input

The MIMO capacity-achieving input is $\mathbf{X} \sim \mathcal{CN}(0, (E_s/n_t) \mathbf{I}_{n_t T})$ — i.e., each entry of $\mathbf{X}$ is i.i.d. $\mathcal{CN}$ with the appropriate per-antenna energy. Telatar 1995 showed that this input achieves the ergodic capacity $C_{\mathrm{erg}} \;=\; \mathbb{E}_{\mathbf{H}}\!\left[\log_2 \det\!\left(\mathbf{I}_{n_r} + \frac{\text{SNR}}{n_t}\mathbf{H}\mathbf{H}^{H}\right)\right].$

LDC mutual information

For an LDC with dispersion matrices $\{\mathbf{A}_q, \mathbf{B}_q\}$ and input $\{s_q\}$ drawn i.i.d. $\mathcal{CN}$ with appropriate variance, the codeword $\mathbf{X}$ is a complex Gaussian vector with covariance $\mathbf{R}_X = \mathbb{E}[\mathbf{X}\mathbf{X}^H]$ determined by the dispersion matrices. If $Q$ is large enough that $\mathrm{span}(\{\mathbf{A}_q, j\mathbf{B}_q\}) = \mathbb{C}^{n_t \times T}$ over $\mathbb{R}$ , the covariance can be tuned to match the i.i.d. Gaussian input covariance.

Dimensional lower bound on $Q$

The real dimension of $\mathbb{C}^{n_t \times T}$ is $2 n_t T$ . To span it we need $Q \ge 2 n_t T$ dispersion matrices. For coherent combining across $n_r$ receive antennas, the effective dimension relevant for capacity drops to the minimum of transmit and receive dimensions — hence $Q \le 2 n_t n_r T$ suffices. (The exact optimisation over smaller $Q$ is the subject of Hassibi-Hochwald's constructive algorithm.)

Matching the capacity

By the Telatar-Foschini theorem, the Gaussian-LDC mutual information approaches $C_{\mathrm{erg}}$ as the dispersion-matrix family spans the full space. Random coding in the outer binary layer pushes the rate to any $R < C_{\mathrm{erg}}$ with vanishing error probability. $\blacksquare$

Example: Designing a Rate-2 LDC for $n_t = n_r = 2$ : From Alamouti to Near-Capacity

Design an LDC for $n_t = n_r = 2, T = 2$ at rate $R = 2$ symbols per channel use (twice the Alamouti rate). Compare its mutual information to the ergodic capacity and to the Alamouti MI.

Solution

Parameter count

Rate $R = 2$ at $T = 2$ means $K = RT = 4$ complex information symbols, i.e. $Q = 2K = 8$ real dispersion matrices. The codeword space $\mathbb{C}^{2\times 2}$ has real dimension $2\cdot 2\cdot 2 = 8$ , so $Q = 8$ exactly spans it — we are at the critical dimension where LDC = "arbitrary complex $2\times 2$ matrix".

Hassibi-Hochwald construction

Pick the 8 dispersion matrices as $\{\mathbf{E}_{ij}, j\mathbf{E}_{ij}\}$ for $(i,j) \in \{1,2\}^2$ , where $\mathbf{E}_{ij}$ is the matrix with a 1 in position $(i,j)$ and zeros elsewhere. With i.i.d. $\mathcal{CN}$ input $s_q$ , each codeword entry $X_{ij} = s_q$ is i.i.d. $\mathcal{CN}$ — exactly the Gaussian i.i.d. input that achieves ergodic capacity. Hence the LDC mutual information equals $C_{\mathrm{erg}}$ .

Comparison with Alamouti

Alamouti at $n_t = n_r = 2, T = 2$ has $K = 2, Q = 4$ — it spans only half of the $8$ -dimensional codeword space. Its MI saturates at $C_{\mathrm{Alamouti}} = \mathbb{E}_\mathbf{H}[\log_2(1 + (\text{SNR}/2)\|\mathbf{H}\|_F^2)]$ — the SIMO-equivalent capacity, not the MIMO ergodic capacity. At low SNR, the two are close; at high SNR, $C_{\mathrm{Alamouti}} \sim \log \text{SNR}$ (single multiplexing gain) while $C_{\mathrm{erg}} \sim \min(n_t, n_r) \log \text{SNR} = 2 \log \text{SNR}$ (double multiplexing gain). Alamouti loses 50% of the capacity asymptotically at $n_t = n_r = 2$ . This is the multiplexing cost of orthogonality that LDCs avoid.

What the LDC framework gives us

The LDC at $Q = 8$ achieves the ergodic capacity but is not full-diversity for a finite-constellation input — the dispersion matrices $\mathbf{E}_{ij}$ don't give a rank-2 minimum-distance structure, so the error matrix $\boldsymbol{\Delta}$ can be rank-1 for some error pairs. This is where the Hassibi-Hochwald optimisation comes in: they search over dispersion matrices to maximise a mutual-information-plus-error-exponent objective, finding LDCs that balance capacity (rate) and pair-wise error performance (diversity). The resulting codes approach the Zheng-Tse DMT curve of Ch. 12 — the full rate-diversity tradeoff — not a single operating point.

LDC Capacity vs Number of Dispersion Matrices $Q$

Mutual information of an LDC as a function of the number of dispersion matrices $Q$ , for a given $(n_t, n_r, T)$ . The plot shows:

The ergodic MIMO capacity $C_{\mathrm{erg}}$ (red dashed) — the Shannon limit for i.i.d. Rayleigh.
The LDC mutual information (blue) as $Q$ grows from 1 (trivial) to $2 n_t T$ (full span).

Three-knob control:

$n_t \in \{2, 3, 4\}$ (transmit antennas).
$n_r \in \{2, 3, 4\}$ (receive antennas).
$T \in [2, 8]$ (block length).

Watch the LDC curve saturate at the capacity when $Q$ reaches the critical value $2 n_t T$ . This is the constructive content of Thm. TLDCs Achieve the MIMO Ergodic Capacity: with enough dispersion matrices an LDC is capacity-achieving. Compare with Alamouti's $Q = 4$ (a fixed point well below the curve for $n_t > 2$ ) and OSTBC's $Q = 2 K_{\max}(n_t)$ .

Parameters

n_t

2

n_r

2

Block length

T

4

LDC Design as Convex Optimisation

Hassibi and Hochwald 2002 pose the LDC design problem as follows: minimise the pairwise error probability bound (Chernoff-type, from Ch. 10 §10.3) over the choice of dispersion matrices $\{\mathbf{A}_q, \mathbf{B}_q\}$ subject to a power constraint $\|\mathbf{A}_q\|_F^2 + \|\mathbf{B}_q\|_F^2 =$ const. The PEP bound, after Jensen's inequality, becomes an expectation of $\det(\mathbf{I} + \alpha \mathbf{Q} \boldsymbol{\Delta}\boldsymbol{\Delta}^H)^{-1}$ where $\mathbf{Q}$ is a convex function of the dispersion matrices. The objective is concave in the Gramian $\mathbf{A}_q^H\mathbf{A}_q + \mathbf{B}_q^H\mathbf{B}_q$ — a classical log-det convex-optimisation problem that can be solved by semi-definite programming.

The observation that LDC design is convex is what makes the framework so usable: you specify a target $(n_t, n_r, T, R)$ and let a numerical solver find the dispersion matrices. Hassibi and Hochwald 2002 Table I lists several such optimised LDCs for $n_t = n_r = 2, 3, 4$ at rates 1, 2, 3 — each one outperforming Alamouti (at rate 1) or V-BLAST (at rate $n_t$ ) in mutual information. The catch is that the decoder for a generic LDC is lattice-based (sphere decoder) — no longer linear. This is the price for closing the capacity gap.

⚠️Engineering Note

The LDC Decoder Cost: From Linear to Lattice

Every step up in rate from Alamouti (OSTBC → QOSTBC → LDC → CDA codes) costs decoder complexity. The specific cascade is:

Code	ML decoder	Complexity
Alamouti / OSTBC	Scalar matched filter	$O(MK)$
QOSTBC	Pair-wise ML	$O(M^2 \cdot K/2)$
General LDC	Lattice decoder (sphere)	$O(M^K)$ worst case, $O(\sqrt{M}^K)$ typical
CDA / Golden / Perfect (Ch. 13)	Lattice decoder	Same as LDC in principle

A rate-2 LDC at 16-QAM with $K = 4$ costs $O(16^4) = 65\,536$ metric evaluations worst case per codeword; in practice, sphere decoding reduces this to $\sim \sqrt{16}^4 = 256$ evaluations at reasonable SNR. By comparison, Alamouti costs $\sim M = 16$ . The rate improvement (2 symbols/cu vs 1) is bought at $\sim 16\times$ the decoder complexity — still tractable.

In 5G NR, the decoder budget for a physical downlink shared channel (PDSCH) MIMO layer is constrained by the UE's silicon area. At $n_t = n_r = 4$ with 64-QAM, a brute-force per-codeword lattice search is infeasible; practical receivers use either (a) closed-loop codebook precoding + per-layer LDPC decoding (standard 5G approach), or (b) approximate lattice decoders like K-best or MMSE-interference- cancellation. Pure-LDC codes thus remain a research benchmark rather than a standard feature.

Practical Constraints

•
Sphere decoder complexity averages $O(M^{K/2})$ at moderate SNR; grows to $O(M^K)$ worst case
•
Lattice basis reduction (LLL) pre-processing helps but adds its own $O(K^3)$ cost per channel realisation
•
In current standards, simpler structured codes + iterative LDPC+MIMO detection are preferred

Historical Note: Hassibi-Hochwald 2002: The Unifying Framework

2002

Babak Hassibi and Bertrand Hochwald published "High-rate codes that are linear in space and time" in IEEE Trans. IT, vol. 48, no. 7, July 2002, pp. 1804-1824 — a long paper (20 pages) that did three things simultaneously:

Unified the linear-STC zoo. Defined the LDC as the most general linear space-time code and showed that Alamouti, OSTBC, QOSTBC, and V-BLAST are all special cases. Every linear STBC fits in the LDC framework.
Formulated design as optimisation. Posed the LDC design problem as minimisation of the PEP bound over dispersion matrices; showed it is convex in the Gramian and solvable by SDP. Provided a numerical algorithm and tabulated optimised LDCs for common $(n_t, n_r, T)$ .
Proved capacity-achieving universality. Showed that LDCs can approach the ergodic MIMO capacity for any $(n_t, n_r)$ as $Q \to 2 n_t n_r T$ — the ergodic capacity of the Rayleigh MIMO channel is reachable by a linear dispersion code.

Hassibi — a Caltech professor working in MIMO theory since the late 1990s — and Hochwald (at Bell Labs, later at Notre Dame) were the natural authors: both had previously written landmark papers on multiple-antenna capacity (Hochwald-Marzetta 1999) and differential unitary STC. Their 2002 paper is the natural endpoint of the linear- STC thread: everything beyond LDCs is either non-linear (CDA codes, Ch. 13) or adapts the LDC framework with extra constraints (e.g., approximate universality, DMT-optimality).

The paper is technically demanding but conceptually clean. It is the reference that subsequent LDC-design papers cite in their introductions; it is also the paper that cleanly separates "code design" (the choice of dispersion matrices) from "code analysis" (the resulting rate, diversity, and decoder complexity). Every subsequent research paper on linear MIMO codes implicitly uses its vocabulary.

Alamouti vs OSTBC vs QOSTBC vs V-BLAST vs LDC

Property	Alamouti ( $n_t=2$ )	OSTBC ( $n_t > 2$ )	QOSTBC ( $n_t=4$ )	V-BLAST	LDC (general)
Rate (sym/ch. use)	1	$\le 3/4$ ( $n_t = 3,4$ )	1	$n_t$	any $R \le \min(n_t, n_r)$
Diversity	$2 n_r$ (full)	$n_t n_r$ (full)	$2 n_r$ (half of full)	$n_r - n_t + 1$ per layer (ZF)	design-dependent
Decoder	Scalar MF	Scalar MF	Pair-wise ML, $O(M^2)$ per pair	ZF-SIC or MMSE-SIC	Sphere / lattice, $O(M^{K/2})$ typ.
Orthogonality	Yes ( $\mathbf{X}\mathbf{X}^H = \\|\mathbf{s}\\|^2 \mathbf{I}$ )	Yes	Block-diagonal only	None	Optional / design choice
Ergodic-capacity gap	0 at $n_t=2$	Grows with $n_t$	Grows with $n_t$	$\to 0$ at high SNR	$\to 0$ by construction
Closed-form codeword	Yes (Alamouti matrix)	Yes (Hurwitz-Radon)	Yes (2×2 Alamouti blocks)	Identity-like (each antenna streams a symbol)	Optimised numerically
Canonical reference	Alamouti 1998	Tarokh-Jafarkhani-Calderbank 1999	Jafarkhani 2001	Wolniansky et al. 1998	Hassibi-Hochwald 2002
In 3GPP standards?	Yes (WCDMA, LTE TM2, NR fallback)	No	No	Effectively, as SM with LDPC + MMSE-IC	No (research benchmark only)

Common Mistake: LDC Is Not OSTBC: Don't Assume Linear Decoding

Mistake:

Reading "LDC is the general linear STC" as implying "every LDC has an OSTBC-style linear matched-filter decoder" or "LDCs preserve the orthogonality of Alamouti at higher rates".

Correction:

LDC is "linear" in the sense that the codeword matrix $\mathbf{X}$ is a linear function of the information symbols $\{s_q\}$ . The decoder is generally not linear. Only the special sub-classes — OSTBC and the trivial rate-1 cases — admit a scalar matched-filter decoder. A generic LDC at rate 2 or higher has a non-orthogonal codeword Gramian $\mathbf{X}\mathbf{X}^H$ that is a matrix polynomial in the symbols, and its ML decoder is a lattice / sphere decoder. The "linear" is in the encoder, not the decoder.

This is also why you cannot "just use the Alamouti decoder" on a general LDC — you have to respect the non-orthogonal structure with a lattice decoder. For practical rate-2 LDCs at $n_t = n_r = 2$ the decoder runs at $O(M^2)$ per codeword (tractable); for rate-4 LDCs at $n_t = n_r = 4$ it runs at $O(M^4)$ (needs sphere decoding to be practical).

,

Why This Matters: Forward: LDCs, DMT, and Cyclic Division Algebra Codes

LDCs can approach the ergodic MIMO capacity (Thm. TLDCs Achieve the MIMO Ergodic Capacity) but they do not automatically achieve the diversity-multiplexing tradeoff $d^*(r)$ of Chapter 12. The DMT requires a stronger property: the error matrix $\boldsymbol{\Delta}$ must have full rank for every symbol pair at any rate — an algebraic condition that LDCs don't generically satisfy. Chapter 13 introduces cyclic division algebra (CDA) codes — a non-linear-in-the-symbol but algebraic-structured construction that guarantees full-rank error matrices for all rates up to $\min(n_t, n_r)$ , thereby achieving the full DMT.

The Golden code (Belfiore-Rekaya 2003), the Perfect codes (Oggier et al. 2006), and the Elia-Kumar-Pawar-Kumar-Caire LAST constructions (2006) are all CDA codes — genuine DMT-optimal constructions that upgrade the LDC framework with extra algebraic structure (number- theoretic irrationalities, roots of unity, cyclotomic extensions). The bridge from "linear" (LDC) to "DMT-optimal" (CDA) is the subject of Chs. 12 and 13.

🔧Engineering Note

MIMO in Wi-Fi and LTE: How the STBC Story Maps to Standards

The space-time coding story of this chapter maps to deployed standards roughly as follows:

Wi-Fi 802.11n (2009): 2×2 and 4×4 MIMO with spatial multiplexing (V-BLAST-style, no STBC). Optionally, Alamouti-style STBC mode for legacy coverage. Primary coding is LDPC or BCC.
Wi-Fi 802.11ac (2013): Same architecture, extended to 8×8. No STBC; closed-loop precoding with explicit beamforming feedback.
LTE TM2 / TM3 (2008-2012): Open-loop transmit-diversity mode based on Alamouti for 2 Tx antennas, and block-Alamouti for 4 Tx. Used for cell-edge and control channels.
LTE TM4-TM9: Closed-loop codebook precoding (rank 1 to $\min(n_t, n_r)$ ), effectively V-BLAST with beamforming.
5G NR (2018): Codebook-based precoding (Type-I, Type-II) with subband granularity. No STBC in main data channels. Alamouti still appears in some broadcast/control.

The absence of OSTBC, QOSTBC, and LDC in mainstream standards is not because those codes don't work — they demonstrably work. Rather, the combination of LDPC outer codes + spatial multiplexing + closed-loop precoding dominates in the typical cellular operating regime (moderate SNR, partial CSIT, long coherence time). Alamouti at $n_t = 2$ is the exception: it is simple, open-loop, and provides a meaningful diversity gain at cell edge where CSIT is unreliable.

Practical Constraints

•
Open-loop STBCs (Alamouti) are used where CSIT is unreliable (cell edge, high mobility)
•
Closed-loop precoding dominates in the mainstream cellular regime
•
Rate- $>1$ OSTBCs / QOSTBCs / LDCs are not in mainstream 3GPP / IEEE 802.11 releases
•
Research LDCs and CDA codes (Ch. 13) remain open-source / academic-reference only

📋 Ref: IEEE 802.11n/ac, 3GPP TS 36.211, TS 38.211

Quick Check

An LDC is designed with $Q = 4$ dispersion matrices at $n_t = n_r = 2, T = 2, K = 2$ . Without additional constraints on the dispersion matrices, this LDC is not necessarily:

Orthogonal (OSTBC) — the dispersion matrices might not satisfy Hurwitz-Radon-Eckmann

Rate 1 — the rate is $K/T = 2/2 = 1$ by construction

Linear — by construction, $\mathbf{X}$ is linear in the symbols $\{s_q\}$

Carrying 2 complex symbols — $K = 2$ by construction

Correction:

Orthogonal (OSTBC) — the dispersion matrices might not satisfy Hurwitz-Radon-Eckmann

OSTBC is a strict sub-class of LDC with the additional Hurwitz-Radon-Eckmann anti-commutation constraints. Plain LDC with $Q = 4$ at $n_t = n_r = 2, K = 2$ is parameterised by 4 arbitrary $2\times 2$ dispersion matrices — the Alamouti code is one choice, but generic choices violate orthogonality and require a lattice decoder.

Linear Dispersion Code (LDC)

A space-time block code of the form $\mathbf{X} = \sum_q (\alpha_q \mathbf{A}_q + j\beta_q \mathbf{B}_q)$ for complex input symbols $s_q = \alpha_q + j\beta_q$ and fixed dispersion matrices $\mathbf{A}_q, \mathbf{B}_q \in \mathbb{C}^{n_t\times T}$ . Introduced by Hassibi and Hochwald 2002 as the most general linear STBC framework; specialises to Alamouti, OSTBC, QOSTBC, and V-BLAST by specific choices of the dispersion matrices. Can achieve the MIMO ergodic capacity as $Q \to 2 n_t n_r T$ .

Linear Space-Time Block Code

A space-time block code whose codeword matrix $\mathbf{X}$ is a linear function of the information symbols. Equivalent to an LDC. Every code discussed in this chapter — Alamouti, OSTBC, QOSTBC, V-BLAST, LDC — is linear in this sense. Non-linear constructions (cyclic division algebra codes) are the subject of Chapter 13.

Linear Dispersion Codes

The Most General Linear STC: Hassibi-Hochwald

Definition: Linear Dispersion Code (LDC)

Theorem: LDCs Achieve the MIMO Ergodic Capacity

Ergodic capacity with Gaussian input

LDC mutual information

Dimensional lower bound on $Q$

Matching the capacity

Example: Designing a Rate-2 LDC for nt=nr=2n_t = n_r = 2nt​=nr​=2: From Alamouti to Near-Capacity

Parameter count

Hassibi-Hochwald construction

Comparison with Alamouti

What the LDC framework gives us

LDC Capacity vs Number of Dispersion Matrices QQQ

Parameters

LDC Design as Convex Optimisation

The LDC Decoder Cost: From Linear to Lattice

Historical Note: Hassibi-Hochwald 2002: The Unifying Framework

Alamouti vs OSTBC vs QOSTBC vs V-BLAST vs LDC

Common Mistake: LDC Is Not OSTBC: Don't Assume Linear Decoding

Why This Matters: Forward: LDCs, DMT, and Cyclic Division Algebra Codes

MIMO in Wi-Fi and LTE: How the STBC Story Maps to Standards

Quick Check

Linear Dispersion Code (LDC)

Linear Space-Time Block Code

Definition:
Linear Dispersion Code (LDC)

Example: Designing a Rate-2 LDC for $n_t = n_r = 2$ : From Alamouti to Near-Capacity

LDC Capacity vs Number of Dispersion Matrices $Q$