Chapter Summary

Chapter Summary

Key Points

• 1.

  Vector spaces as the setting. $\mathbb{C}^n$ is the workhorse space of communications: every signal vector, channel vector, and beamforming weight vector lives here. Dimension, basis, and span are the organizing concepts that tell us how many degrees of freedom a system has and how to represent signals efficiently.

• 2.

  Inner products giving geometry. The inner product $\langle \mathbf{x}, \mathbf{y} \rangle = \mathbf{y}^H \mathbf{x}$ endows $\mathbb{C}^n$ with geometry — angles, distances, and projections all derive from it. The Cauchy–Schwarz inequality $|\mathbf{y}^H \mathbf{x}|^2 \le \|\mathbf{x}\|^2 \|\mathbf{y}\|^2$ directly yields the matched-filter beamformer as the projection that maximizes received SNR.

• 3.

  Matrices as linear maps. Range, null space, and rank characterize what a linear map can and cannot do — the rank of a channel matrix $\mathbf{H}$ determines the number of independent data streams a MIMO link can support. Hermitian ($\mathbf{A} = \mathbf{A}^H$), unitary ($\mathbf{U}^H\mathbf{U} = \mathbf{I}$), and positive definite ($\mathbf{x}^H\mathbf{A}\mathbf{x} > 0$) matrices are the three special classes that dominate wireless system analysis.

• 4.

  Eigendecomposition for Hermitian matrices. The spectral theorem $\mathbf{A} = \mathbf{Q}\mathbf{\Lambda}\mathbf{Q}^H$ reveals that every Hermitian matrix acts by scaling along orthogonal directions defined by its eigenvectors. The Rayleigh quotient $R(\mathbf{x}) = \mathbf{x}^H\mathbf{A}\mathbf{x} / \mathbf{x}^H\mathbf{x}$ connects eigenvalues to constrained optimization, a pattern that recurs throughout communications theory.

• 5.

  SVD for everything. Every $m \times n$ matrix decomposes as $\mathbf{U}\mathbf{\Sigma}\mathbf{V}^H$ — rotation, scaling, rotation. MIMO channel capacity, optimal beamforming, and low-rank channel approximation all trace back to the SVD; the singular values $\sigma_1 \ge \sigma_2 \ge \cdots$ quantify the strength of each spatial sub-channel.

• 6.

  Trace/det inequalities in capacity. The MIMO capacity formula $C = \log\det\!\left(\mathbf{I} + \text{SNR}\cdot\mathbf{H}\mathbf{H}^H\right)$ is built from trace and determinant. Hadamard's inequality, Fischer's inequality, and the concavity of $\log\det$ on the positive definite cone are the analytical workhorses for bounding and optimizing capacity expressions.

• 7.

  Kronecker products in channel models. The Kronecker product $\otimes$ structures the separable spatial correlation model of MIMO channels: $\mathrm{vec}(\mathbf{H}) \sim \mathcal{CN}(\mathbf{0},\, \mathbf{R}_t^T \otimes \mathbf{R}_r)$, factoring the full correlation into transmit and receive sides. The $\mathrm{vec}$ operator converts matrix equations to vector form, enabling compact manipulation via the identity $\mathrm{vec}(\mathbf{A}\mathbf{X}\mathbf{B}) = (\mathbf{B}^T \otimes \mathbf{A})\,\mathrm{vec}(\mathbf{X})$.

• 8.

  Matrix calculus for optimization. Gradients of the key building blocks — $\nabla_{\mathbf{x}}(\mathbf{x}^H\mathbf{A}\mathbf{x}) = 2\mathbf{A}\mathbf{x}$, $\nabla_{\mathbf{A}}\log\det(\mathbf{A}) = \mathbf{A}^{-T}$, and $\nabla_{\mathbf{A}}\mathrm{tr}(\mathbf{A}\mathbf{B}) = \mathbf{B}^T$ — reduce wireless optimization problems such as beamformer design and precoder optimization to eigenvalue problems with closed-form solutions.