1 of 24

Random signals and adaptive filtering

Department of Data Communication Networks and Systems

Lecturer Shukhrat Palvanov

2 of 24

What is a Random Signal?

A random signal, also called a stochastic signal or process, is a time-varying quantity whose values cannot be predicted exactly but can be described statistically.�Unlike deterministic signals (such as sinusoids or square waves), random signals exhibit unpredictable variations in amplitude, phase, or frequency over time. Their behavior is governed by probability laws rather than explicit mathematical equations.

Nature of Randomness:�Random signals represent uncertainty or noise present in physical systems such as electrical circuits, wireless channels, or biological processes.

Examples:

    • Thermal noise generated by resistors
    • Shot noise in electronic devices
    • Atmospheric noise in radio communication
    • Fluctuations in speech or ECG signals

3 of 24

Statistical Description of random signals

To understand random signals, we use statistical parameters instead of explicit equations:

Mean value: average level of the signal.

Variance: measure of how much the signal deviates from its mean.

Autocorrelation: measure of similarity between the signal and its time-shifted version.

Power Spectral Density (PSD): frequency domain representation showing power distribution.

4 of 24

Why Random Signals Matter�

Random signals are essential in signal and image processing because:

  • They model noise and uncertainty in systems.
  • They help design filters that can reduce noise and improve signal quality.
  • They are used in applications such as communications, medical imaging, and radar systems.

5 of 24

Random Variables and Random Processes�

In signal and image processing, many signals are uncertain or unpredictable due to noise and other external factors.

�To describe and analyze such signals mathematically, we use the concepts of random variables and random processes.

�These are fundamental tools for modeling and understanding random behavior in signals.

6 of 24

Random Variable

A random variable is a numerical quantity whose value is determined by the outcome of a random experiment.�It provides a way to represent random phenomena mathematically.

Continuous random variable: takes any value within a range (e.g., temperature, voltage noise).

Discrete random variable: takes countable values (e.g., number of errors in a transmission).

A random variable X is characterized by:

Probability density function (PDF): , which shows how probabilities are distributed over possible values.

Cumulative distribution function (CDF): the probability that X does not exceed x.

Example:�If X represents thermal noise voltage, then X may follow a Gaussian (Normal) distribution with mean 0 and variance σ^2.

7 of 24

Random Process�

A random process (or stochastic process) is an extension of the random variable concept to time.�It is defined as a collection of random variables indexed by time:

Each value of t corresponds to one random variable, and the complete set over time forms the random process.

Example:�Thermal noise measured at different times — each sample is random, but together they form a random process X(t).

8 of 24

Sample Function and Ensemble

A sample function (or realization) is one possible waveform of the random process over time.

The ensemble is the set of all possible sample functions.

Each time you observe the process, you may see a different waveform, but all follow the same probability law.

9 of 24

Continuous and Discrete Random Processes

  • Continuous-time random process: X(t), where t is continuous.
  • Discrete-time random process: X[n], where n takes integer values (used in digital signal processing).

In digital image processing, we mostly deal with discrete random processes, because images and digital signals are sampled and quantized.

10 of 24

Power Spectral Density (PSD)

The Power Spectral Density (PSD) describes how the power of a signal is distributed over frequency.

It is a key tool in analyzing random and noisy signals in both time and frequency domains.

For a wide-sense stationary (WSS) random process X(t)X(t)X(t), the PSD is defined as the Fourier Transform of its autocorrelation function:

11 of 24

White Noise and Colored Noise

White Noise

12 of 24

Colored Noise

13 of 24

Concept of Filtering in Random Signal Processing

Filtering is the process of extracting useful information from a signal while reducing noise or unwanted components.

In random signal processing, filters are used to separate signal and noise, estimate unknown parameters, or enhance system performance.

A filter transforms an input signal x(t) into an output signal y(t):

y(t)=h(t)∗ x(t)

where h(t) is the filter’s impulse response, and * denotes convolution.

14 of 24

Filtering in Random Environments

When input contains random noise, filter design must consider statistical characteristics of both the signal and noise.

Performance is often measured using Mean Square Error (MSE) between the desired signal 𝑑(𝑡) and the output 𝑦(𝑡):

MSE=𝐸[(𝑑(𝑡)−𝑦(𝑡))2]

15 of 24

Adaptive Filtering

Adaptive filtering refers to a class of filtering techniques where the filter parameters change automatically to adapt to variations in the input signal or environment.

Unlike fixed filters, adaptive filters can learn and update their coefficients in real time to minimize the error between the desired and actual outputs.

Why Adaptive Filtering?

  • Many real-world systems are non-stationary, meaning their statistical properties change over time.
  • A fixed filter may not perform well under such conditions, while an adaptive filter tracks changes dynamically.

16 of 24

Adaptive Filtering

17 of 24

Structure of an Adaptive Filter

18 of 24

Adaptive Filter Algorithms

Adaptive filters rely on learning algorithms to update their coefficients in response to changes in the input data.�These algorithms determine how fast and accurately the filter adapts to varying environments.

General Weight Update Equation

w(n+1)=w(n)+Δw(n)

where the update term Δw(n) depends on the specific adaptation algorithm used.

19 of 24

Main Adaptive Algorithms

1. LMS (Least Mean Squares) Algorithm:

Simplest and most widely used.

Updates weights using instantaneous error:

w(n+1)=w(n)+μx(n)e(n)

μ= step size (controls convergence speed).

Advantages: simple, low computational cost.

Drawbacks: slower convergence for highly correlated inputs.

20 of 24

Main Adaptive Algorithms

2. NLMS (Normalized LMS):

    • Improvement of LMS; step size normalized by signal power.

    • More stable and faster convergence.

3. RLS (Recursive Least Squares):

  • Minimizes the weighted least-squares error using all past data.
  • Very fast convergence but high computational cost.
  • Suitable for rapidly changing environments.

4. Kalman Filter:

    • Optimal adaptive filtering algorithm for systems with state-space models.
    • Provides excellent estimation performance in dynamic systems.

21 of 24

Applications of Adaptive Filtering

Adaptive filters are widely used in signal, image, and communication systems where the environment or signal statistics change over time.�Their ability to self-adjust makes them ideal for many real-world applications.

1. Noise Cancellation

Removes unwanted noise from signals in real time.

Used in:

    • Speech enhancement (microphones, headsets)
    • Biomedical signal processing (ECG, EEG noise reduction)
    • Active noise control (ANC) in headphones

Principle: The adaptive filter estimates the noise and subtracts it from the noisy signal.

22 of 24

Applications of Adaptive Filtering

Echo Cancellation

  • Eliminates echo in telecommunication or audio systems.
  • Example: Hands-free phones or video conferencing.
  • Adaptive filter models the echo path and subtracts estimated echo from the received signal.

Channel Equalization

  • Compensates for signal distortion caused by communication channels.
  • Adaptive filters adjust parameters to maintain reliable data transmission under varying channel conditions.

23 of 24

Applications of Adaptive Filtering

System Identification

  • Estimates or tracks an unknown system by adapting its coefficients to match the system’s behavior.
  • Used in control systems and fault detection.

Prediction and Estimation

  • Predicts future signal values (e.g., stock prices, speech signals).
  • Used in adaptive forecasting and tracking systems.

Image Processing Applications

  • Adaptive smoothing and edge-preserving filtering.
  • Noise reduction in time-varying or non-stationary image data.

24 of 24

Tasks for understanding

Task 1 – Random Signal Generation and Analysis�Generate a random signal, analyze its mean, autocorrelation, and power spectral density (PSD).

Task 2 – Adaptive Noise Cancellation Using LMS Filter�Use an adaptive filter to remove noise from a signal.