Signal Processing & Math

Chi-Squared Distribution

/ky-skwerd dis-trih-byoo-shuhn/
A probability distribution that models the sum of k squared independent standard normal random variables, serving as the statistical foundation for RF detection theory and CFAR threshold computation. In radar receivers, the noise power envelope follows a chi-squared distribution with 2 degrees of freedom (exponential), and non-coherent integration of N pulses produces a chi-squared distribution with 2N degrees of freedom. The Swerling target models use chi-squared distributions with 2 or 4 degrees of freedom to describe fluctuating radar cross-section statistics, directly impacting detection probability calculations for operational radar systems.
Category: Signal Processing & Math
Degrees of Freedom: k = 1, 2, 4, 2N...
Key App: CFAR Detection

Understanding Chi-Squared Distribution

If Z1, Z2, ..., Zk are independent standard normal random variables, then the sum X = Z12 + Z22 + ... + Zk2 follows a chi-squared distribution with k degrees of freedom. This seemingly abstract definition maps directly onto radar signal processing: in-phase (I) and quadrature (Q) noise samples from a matched filter output are independent Gaussians, so the power detector output (I2 + Q2) follows a chi-squared distribution with k = 2, which is the exponential distribution. This is why thermal noise power in a radar receiver has an exponential probability density.

Non-coherent pulse integration extends this directly. Summing N independent noise power samples produces a chi-squared random variable with 2N degrees of freedom. The detection threshold for a desired false alarm probability Pfa is set at the point where the complementary CDF of this chi-squared distribution equals Pfa. For example, with N = 10 integrated pulses and Pfa = 10-6, the threshold is found from the inverse incomplete gamma function. In CFAR processors, the chi-squared model is used to derive the adaptive threshold multiplier analytically, enabling constant false alarm rates even when the noise floor varies across range cells due to clutter or jamming. The Swerling target models further exploit chi-squared statistics: Cases I/II use 2 degrees of freedom (many equal scatterers) while Cases III/IV use 4 degrees of freedom (one dominant plus many small scatterers).

Key Formulas

Chi-Squared PDF (k degrees of freedom):
f(x; k) = x(k/2 - 1) · e-x/2 / (2k/2 · Γ(k/2))   for x > 0

False Alarm Probability (N integrated pulses):
Pfa = Γ(N, T/2) / Γ(N)   [upper incomplete gamma ratio]

CFAR Threshold Multiplier (CA-CFAR):
T = Nref · (Pfa-1/Nref - 1)

Where k = degrees of freedom, Γ = gamma function, T = detection threshold, N = integration count, Nref = number of CFAR reference cells. The CA-CFAR multiplier ensures constant Pfa independent of noise level.

Chi-Squared in Swerling Target Models

Swerling CaseRCS DistributionChi-Squared kFluctuation RatePhysical Model
Case 0Non-fluctuatingN/A (Marcum)NoneSphere, corner reflector
Case IExponentialk = 2Scan-to-scanMany equal scatterers
Case IIExponentialk = 2Pulse-to-pulseMany equal scatterers
Case IIIChi-squared (4 dof)k = 4Scan-to-scanOne dominant + many small
Case IVChi-squared (4 dof)k = 4Pulse-to-pulseOne dominant + many small
Common Questions

Frequently Asked Questions

How does the chi-squared distribution relate to radar detection?

In a radar receiver, thermal noise is Gaussian in each I and Q channel. The envelope detector computes I2 + Q2, which follows an exponential distribution (chi-squared with 2 degrees of freedom) under the noise-only hypothesis. The threshold is set so the tail probability equals the desired false alarm probability, typically 10-6 to 10-8. With N non-coherently integrated pulses, the summed noise follows a chi-squared distribution with 2N degrees of freedom, and the threshold is adjusted using the incomplete gamma function.

What are Swerling target models and how do they use chi-squared?

The Swerling models describe fluctuating radar targets. Cases I and II assume exponential RCS distribution (chi-squared, k = 2), modeling targets with many equal-size scatterers. Cases III and IV use chi-squared with k = 4, modeling targets with one dominant scatterer plus smaller ones. Odd cases fluctuate scan-to-scan; even cases fluctuate pulse-to-pulse. Case 0 (Marcum) is non-fluctuating. The choice of model directly affects the required SNR for a given detection probability.

How is chi-squared used in CFAR detector design?

Cell-Averaging CFAR estimates local noise by averaging power from reference cells surrounding the cell under test. The sum of N independent exponential samples follows chi-squared with 2N degrees of freedom. The CFAR multiplier T is computed so that the tail probability equals the target Pfa. For 16 reference cells and Pfa = 10-6, the multiplier is approximately 4.8 times the noise estimate. This chi-squared model allows analytical calculation of CFAR loss, typically 1 to 2 dB versus an ideal fixed-threshold detector.

Radar & Detection Systems

Request a Quote

Need precision waveguide terminations or matched loads for radar receiver testing and calibration? Contact our engineering team.

Get in Touch