Radar Detection: Swerling Models & Monte Carlo

What the Radar Equation Doesn't Tell You

The classic radar range equation spits out a single number: the range where your received SNR hits the detection threshold. Clean, deterministic, and completely misleading if you think that's the whole story. The equation assumes your target has a fixed RCS, the atmosphere is perfectly transparent, and every component in your radar performs exactly to spec. Real targets don't cooperate.

Aircraft bank and yaw, changing their presented cross-section by 20 dB or more from one pulse to the next. Ships pitch and roll in heavy seas. Rain doesn't just sit there — it scatters your signal, adding anywhere from 0.01 dB/km in light drizzle to 20 dB/km in a tropical downpour, and that's two-way loss. Your klystron or solid-state amplifier? It's putting out ±1 dB different power from unit to unit off the production line, and another ±2 dB swing as the temperature changes from winter to summer. The radar range equation gives you a point estimate. What you actually need is a probability distribution wrapped around that estimate, accounting for everything that varies in the real world.

This walkthrough uses the Radar Detection Simulator to analyze a ground-based surveillance radar operating at 3 GHz. We'll work through target fluctuation models, add weather, and run Monte Carlo trials to see how much your detection range actually varies when you account for manufacturing tolerances and environmental uncertainty.

Target Models: Choosing the Right Swerling Case

Before you run anything, you need to pick a target fluctuation model. The five Swerling cases were developed in the 1950s and they're still the standard because they map cleanly to physical scattering mechanisms. Here's what each one represents:

Swerling 0 is the optimistic case — the target RCS is constant, so you get the full benefit of non-coherent integration. Use it for calibration spheres or trihedral corner reflectors where the geometry really is stable. For a fighter-sized aircraft at 3 GHz with pulse-to-pulse integration, Swerling 2 is the standard choice. The "many scatterers" assumption holds because you're seeing returns from the fuselage, wings, tail, engine inlets, and control surfaces all interfering with each other. The RCS follows a Rayleigh distribution from pulse to pulse.

Swerling 1 is more conservative — it assumes the target RCS changes slowly, so you're effectively integrating the same RCS value across all your pulses. That makes integration less effective and produces lower detection probability at the same SNR. Most engineers skip Swerling 1, but if you're designing a system where you need margin against worst-case target aspect angles, it's worth running both Swerling 1 and Swerling 2 to bracket your performance.

Swerling 3 and 4 apply when you have one dominant scatterer — think of a small aircraft where the engine inlet or a corner reflector on the tail dominates the return. The RCS distribution shifts from Rayleigh to a chi-squared with four degrees of freedom, which has a longer tail. You get occasional strong returns that help detection, but the median performance is similar to Swerling 2.

Setting Up the Nominal Case

Let's configure a typical 3 GHz ground surveillance radar. These parameters represent a medium-range system like an L-band air traffic control radar or a ground-based air defense sensor:

The simulator computes SNR at each range bin using the Friis radar equation — the two-way path loss version where you get R⁴ in the denominator instead of R². Then it maps SNR to detection probability using the Marcum Q-function for Swerling 0 or the appropriate noncentral chi-squared CDF for Swerling 1 through 4. Non-coherent integration of N pulses improves your SNR, but not by the full factor of N. For Swerling fluctuating targets, the integration efficiency is closer to N^0.8 because the target fading decorrelates your pulses.

With these inputs, the nominal detection range at Pd = 0.5 comes out around 180 km. That's the median detection point — half the time you'll detect at this range, half the time you won't. The 90% detection range is closer to 120 km, which is the range where nine out of ten scan opportunities will produce a detection. That 60 km difference between Pd = 0.5 and Pd = 0.9 is entirely due to target RCS fluctuation. If you were running Swerling 0 (constant target), those two ranges would be much closer together.

Adding Rain: ITU-R P.838 Attenuation

Now let's see what happens when it rains. Enable rain attenuation and set the rain rate to 16 mm/hr, which corresponds to moderate rain in ITU-R climate zone K. The simulator applies the P.838 specific attenuation model:

where k and α are frequency-dependent coefficients that vary with polarization. At 3 GHz with horizontal polarization, k ≈ 0.00155 and α ≈ 1.265. Plug in R = 16 mm/hr and you get γ_R ≈ 0.044 dB/km. That doesn't sound like much, but remember this is a two-way path. Over a 180 km path to the target and back, you're losing 16 dB. That's enough to cut your detection range from 180 km down to about 120 km for the nominal case.

The rain region is limited to the first 4 km in altitude — the so-called bright band where rain is most intense before it starts evaporating or turning to snow at higher altitudes. The simulator handles this by calculating an effective path length through the rain region rather than assuming rain all the way to the target. For a ground-based radar looking at a target at 10 km altitude, most of your path is above the rain layer.

Heavier rain makes this much worse. At 50 mm/hr — a tropical thunderstorm — you get γ_R ≈ 0.21 dB/km. That's nearly 80 dB of two-way loss over a 180 km path, which reduces your nominal detection range below 90 km. At X-band (10 GHz) or higher, rain attenuation becomes the dominant loss mechanism in moderate to heavy precipitation. This is why long-range air surveillance radars operate at L-band or S-band — the rain loss is manageable.

Monte Carlo: Quantifying System Uncertainty

The nominal detection range is just the median — half of all manufactured radar systems will perform worse than that number. To see the full spread, enable Monte Carlo simulation with 50,000 trials and the following tolerances:

These tolerances represent typical manufacturing variance and operational uncertainty. Peak power varies from unit to unit due to amplifier component tolerances and drifts with temperature. Antenna gain depends on feed alignment, radome transmission loss, and pattern measurement accuracy. System losses include waveguide mismatch, filter insertion loss, and receiver front-end mismatch — all of which vary from unit to unit. Target RCS varies ±3 dB (or more) depending on aspect angle, even for the same aircraft.

The Monte Carlo result shows that the 10th-percentile detection range — the worst 10% of system-plus-environment combinations — is 95 km. That's 25% shorter than the nominal 180 km. The 90th-percentile (best 10%) reaches 155 km. This spread represents the real-world variability you'll see across a fleet of radars operating in different conditions.

The most influential parameter is target RCS, which drives nearly 60% of the detection range variance in the sensitivity breakdown. This makes sense for Swerling 2 targets: the RCS fluctuates pulse-to-pulse with a Rayleigh distribution, and the tails of that distribution dominate your detection probability at moderate SNR. The practical implication is that investing in higher transmit power or a bigger antenna has diminishing returns if you haven't properly accounted for target aspect angle variation. You can't fix a 10 dB RCS fade with 3 dB more transmit power — the math doesn't work in your favor.

The second most influential parameter is usually peak power, followed by system losses. Antenna gain is surprisingly stable if you've done your mechanical design properly. Noise figure matters more at long range where you're already SNR-limited, but at shorter ranges where you have SNR margin, noise figure uncertainty contributes less to the overall detection range variance.

Reading the ROC Curve

The Receiver Operating Characteristic (ROC) curve plots detection probability against false alarm probability for a fixed range. It answers the question: "if I relax my false alarm rate from 10⁻⁶ to 10⁻⁴, how much do I gain in detection probability at 150 km?"

At 150 km with the nominal parameters and no rain, the ROC shows Pd rising from 0.41 at Pfa = 10⁻⁶ to 0.68 at Pfa = 10⁻⁴. That's a 27 percentage point gain in detection probability for two orders of magnitude more false alarms. Whether that tradeoff makes sense depends entirely on your operational context.

For air traffic control, Pfa = 10⁻⁶ is effectively mandatory — you can't have the operator screening hundreds of false contacts per scan. For a maritime search radar with a human operator who's already looking at sea clutter and weather returns, Pfa = 10⁻⁴ may be perfectly acceptable. The operator is going to correlate contacts across multiple scans anyway, so a few extra false alarms per scan don't significantly increase workload.

The ROC curve also shows you the detection threshold knee — the point where increasing Pfa further doesn't buy you much more Pd. For most Swerling cases, that knee occurs around Pfa = 10⁻³ to 10⁻⁴. Below that, you're trading false alarms very efficiently for detection probability. Above that, you're into the noise floor and the tradeoff becomes unfavorable.

What This Simulation Won't Tell You

The simulator models thermal noise detection, range-Doppler processing gain through non-coherent integration, rain attenuation via ITU-R P.838, and target RCS fluctuation using the Swerling models. It gives you a solid foundation for link budget validation and detection range sensitivity analysis. But it doesn't model everything.

Clutter — ground, sea, weather, or chaff — isn't included. For a ground-based radar looking down at low-altitude targets, ground clutter can dominate thermal noise by 30 dB or more. You need a separate clutter model and Doppler processing to handle that. ECM and jamming aren't modeled either — if someone is actively denying your radar, your detection range collapses in ways that thermal noise statistics don't predict. Multipath from ground or sea reflections can cause deep nulls in your coverage at specific elevation angles. Antenna scanning loss — the fact that your gain drops off-boresight — reduces detection range at the edges of your scan volume.

For a full radar system analysis, those effects need their own models. But for understanding how your radar performs against a point target in free space, accounting for realistic system tolerances and atmospheric effects, this simulation gives you the essential probability framework. And most engineers skip this step entirely, relying on the deterministic radar range equation and wondering later why their field measurements don't match predictions.

Radar Detection Simulator