Ferkans — Interactive Telecom Tutor

Why Complex Gaussians?

In baseband signal processing, signals and noise are complex-valued. The thermal noise in a receiver is modeled as $\mathcal{CN}(0, N_0)$ — circularly symmetric complex Gaussian. The i.i.d. Rayleigh MIMO channel has entries $h_{ij} \sim \mathcal{CN}(0, 1)$ . Understanding the complex Gaussian is therefore essential for any analysis of wireless systems. The key concept is circular symmetry (or "properness"): the distribution is invariant under multiplication by $e^{j\theta}$ for any angle $\theta$ .

Definition:
Proper Complex Gaussian Distribution

A complex random vector $\mathbf{Z} = \mathbf{X} + j\mathbf{Y} \in \mathbb{C}^n$ has the proper (circularly symmetric) complex Gaussian distribution $\mathbf{Z} \sim \mathcal{CN}(\boldsymbol{\mu}, \boldsymbol{\Sigma})$ if:

The real-valued vector $\hat{\mathbf{Z}} = \begin{pmatrix} \mathbf{X} \\ \mathbf{Y} \end{pmatrix} \in \mathbb{R}^{2n}$ is jointly Gaussian.
The pseudo-covariance (complementary covariance) vanishes: $\tilde{\boldsymbol{\Sigma}} = \mathbb{E}[(\mathbf{Z} - \boldsymbol{\mu})(\mathbf{Z} - \boldsymbol{\mu})^T] = \mathbf{0}$ .

Here $\boldsymbol{\Sigma} = \mathbb{E}[(\mathbf{Z} - \boldsymbol{\mu})(\mathbf{Z} - \boldsymbol{\mu})^H]$ is the (Hermitian) covariance matrix. The PDF (when $\boldsymbol{\Sigma} \succ 0$ ) is

$f_{\mathbf{Z}}(\mathbf{z}) = \frac{1}{\pi^n \det(\boldsymbol{\Sigma})} \exp\!\left(-(\mathbf{z} - \boldsymbol{\mu})^H \boldsymbol{\Sigma}^{-1} (\mathbf{z} - \boldsymbol{\mu})\right).$

Notice the normalization constant is $1/(\pi^n \det(\boldsymbol{\Sigma}))$ , not $1/((2\pi)^n \det(\boldsymbol{\Sigma}))^{1/2}$ — the factor of $2$ from the real/imaginary decomposition is absorbed. Circular symmetry means $e^{j\theta}\mathbf{Z}$ has the same distribution as $\mathbf{Z}$ (when $\boldsymbol{\mu} = \mathbf{0}$ ).

Circular symmetry (properness)

A complex random vector $\mathbf{Z}$ is circularly symmetric (proper) if $e^{j\theta}\mathbf{Z}$ has the same distribution as $\mathbf{Z}$ for all $\theta$ . Equivalently, the pseudo-covariance $\mathbb{E}[\mathbf{Z}\mathbf{Z}^T] = \mathbf{0}$ . This means the real and imaginary parts have equal covariance and are uncorrelated.

Definition:
Pseudo-Covariance Matrix

For a complex random vector $\mathbf{Z} \in \mathbb{C}^n$ with mean $\boldsymbol{\mu}$ , the pseudo-covariance (or complementary covariance) matrix is

$\tilde{\boldsymbol{\Sigma}} = \mathbb{E}\bigl[(\mathbf{Z} - \boldsymbol{\mu})(\mathbf{Z} - \boldsymbol{\mu})^T\bigr].$

Note the transpose (not Hermitian transpose). A complex RV is proper iff $\tilde{\boldsymbol{\Sigma}} = \mathbf{0}$ , which implies:

$\boldsymbol{\Sigma}_{\mathbf{X}} = \boldsymbol{\Sigma}_{\mathbf{Y}}$ (equal real/imaginary covariance),
$\boldsymbol{\Sigma}_{\mathbf{XY}} = -\boldsymbol{\Sigma}_{\mathbf{XY}}^T$ (skew-symmetric cross-covariance).

Example: The Scalar Complex Gaussian

Let $Z = X + jY \sim \mathcal{CN}(0, \sigma^2)$ . Describe the distributions of $X$ , $Y$ , $|Z|$ , and $|Z|^2$ .

Solution

Real and imaginary parts

Properness and the constraint $\mathbb{E}[|Z|^2] = \sigma^2$ imply $X \sim \mathcal{N}(0, \sigma^2/2)$ and $Y \sim \mathcal{N}(0, \sigma^2/2)$ , independent.

Envelope

$|Z| = \sqrt{X^2 + Y^2}$ is Rayleigh with parameter $\sigma/\sqrt{2}$ :

$f_{|Z|}(r) = \frac{2r}{\sigma^2}\exp\!\left(-\frac{r^2}{\sigma^2}\right), \quad r \geq 0.$

Power

$|Z|^2 = X^2 + Y^2 \sim \text{Exp}(1/\sigma^2)$ , i.e., $\mathbb{E}[|Z|^2] = \sigma^2$ .

Common Mistake: Complex Gaussian Variance Convention

Mistake:

Writing $Z \sim \mathcal{CN}(0, \sigma^2)$ and then computing $\text{Var}(\text{Re}(Z)) = \sigma^2$ .

Correction:

For $Z \sim \mathcal{CN}(0, \sigma^2)$ , the total power is $\mathbb{E}[|Z|^2] = \sigma^2$ , split equally between real and imaginary parts: $\text{Var}(\text{Re}(Z)) = \text{Var}(\text{Im}(Z)) = \sigma^2/2$ . The factor of 2 is a common source of errors in SNR calculations.

🎓CommIT Contribution(2021)

Complex Gaussian Models in MIMO-ISAC

G. Caire, A. Fengler, P. Jung — IEEE Transactions on Information Theory

This paper develops a framework for activity detection and channel estimation in massive random access, where users transmit pilot sequences and the base station must determine which users are active. The channel model is $\mathbf{h}_k \sim \mathcal{CN}(\mathbf{0}, \beta_k \mathbf{I}_M)$ , a proper complex Gaussian with large-scale fading coefficient $\beta_k$ . The detection and estimation algorithms rely critically on the properties of complex Gaussian vectors developed in this chapter: the Wishart distribution of the sample covariance, the conditional Gaussian formulas for LMMSE channel estimation, and the independence of uncorrelated Gaussian components for separating users.

massive-mimorandom-accesscomplex-gaussianView Paper →

Key Takeaway

The proper complex Gaussian $\mathcal{CN}(\boldsymbol{\mu}, \boldsymbol{\Sigma})$ is the standard model for baseband noise and Rayleigh fading channels. Properness (circular symmetry) means the pseudo-covariance vanishes, the real and imaginary parts have equal variance $\sigma^2/2$ each, and the distribution is invariant to phase rotation. Every property of the real Gaussian — marginals, conditionals, affine closure, uncorrelated $\Leftrightarrow$ independent — carries over to the proper complex case.

Historical Note: Goodman, Turin, and Complex Gaussian Statistics

1960s

The systematic development of complex Gaussian statistics dates to N. R. Goodman's 1963 paper "Statistical Analysis Based on a Certain Multivariate Complex Gaussian Distribution," which introduced the complex Wishart distribution and the associated hypothesis tests. The application to wireless communications was pioneered by G. L. Turin (1960), who showed that the baseband representation of narrowband noise through a scattering channel is circularly symmetric complex Gaussian. This model has been the default in wireless system analysis ever since.

The Complex Gaussian Distribution