CAEECentre for Automotive Embedded Engineering · HSR Layout, Bengaluru
Module 16 of 20 · Advanced

Random Processes, Noise and Correlation

A random process is a time-indexed family of random variables. It models electrical noise, fading, jitter and unpredictable sensor disturbances.

Advanced80 min study timePrerequisite: Random variables and Fourier analysis
Module learning progress0 of 3 stages

1 · Why this mathematics matters

Characterise electrical noise, correlation, stationarity and power spectral density.

2 · Prerequisites and learning approach

Prerequisites: Random variables and Fourier analysis. Begin with the physical meaning, identify every symbol and unit, study the fixed-scale graph, then solve and verify the examples.

3 · Symbols and units

x(t)random process
Rxx(τ)autocorrelation
Sxx(f)power spectral density
τtime lag

4 · Concepts and textbook formulas

1. Mean and stationarity

A wide-sense stationary process has constant mean and autocorrelation that depends only on time separation.

Mean and stationarity equation

How to read it: identify the output quantity first, then the variables, parameters, units and assumptions. Test zero, unity and limiting values before using the equation in a design.

2. Autocorrelation

Autocorrelation measures similarity with a delayed copy and reveals periodicity and memory.

Autocorrelation equation

How to read it: identify the output quantity first, then the variables, parameters, units and assumptions. Test zero, unity and limiting values before using the equation in a design.

3. Power spectral density

The Wiener-Khinchin relation connects autocorrelation and power spectrum.

Power spectral density equation

How to read it: identify the output quantity first, then the variables, parameters, units and assumptions. Test zero, unity and limiting values before using the equation in a design.

4. Noise models

White noise has flat PSD; thermal-noise voltage grows with resistance, bandwidth and temperature.

Noise models equation

How to read it: identify the output quantity first, then the variables, parameters, units and assumptions. Test zero, unity and limiting values before using the equation in a design.

5 · Step-by-step worked examples

Example 1: Foundation

Doubling measurement bandwidth doubles noise power and increases RMS noise by sqrt(2).

Verification: check the unit, sign, order of magnitude and physical trend.

Example 2: Thermal noise

Given: R=10 kohm, T=300 K and B=100 kHz.

Thermal noise solution equation

Engineering meaning: Noise voltage grows with the square root of resistance, temperature and bandwidth.

Example 3: Averaging uncorrelated noise

Given: Average N independent measurements.

Averaging uncorrelated noise solution equation

Engineering meaning: Averaging sixteen independent samples reduces RMS noise by a factor of four.

6 · EE/ECE applications

  • receiver noise analysis — use the governing equation to predict parameter sensitivity, confirm a simulation and interpret measured behaviour.
  • clock jitter — use the governing equation to predict parameter sensitivity, confirm a simulation and interpret measured behaviour.
  • sensor filtering — use the governing equation to predict parameter sensitivity, confirm a simulation and interpret measured behaviour.
  • communication channels — use the governing equation to predict parameter sensitivity, confirm a simulation and interpret measured behaviour.

7 · Common mistakes

  • Assuming one noise waveform represents the complete process.
  • Confusing amplitude spectrum with power spectral density.

8 · Practice problems

  1. Interpret the autocorrelation width of a noise process.
  2. Relate autocorrelation and PSD using Wiener–Khinchin.
Verification habit: predict the direction of change before calculating, then check units, sign, limiting cases and engineering reasonableness.