How Radar Actually Works: From Pulse Timing to Synthetic Aperture

You point electromagnetic energy at something and wait for it to bounce back. That is the entire concept behind radar, reduced to one sentence. Everything else, the thousands of patents, the billions of euros in military contracts, the entire discipline of radar signal processing, exists because that simple idea runs headfirst into physics, noise, and the inconvenient behaviour of real targets in a real atmosphere.

Radar (Radio Detection and Ranging) has been operational since the late 1930s. Nearly a century later, a €3 chip on a car bumper and a €500 million airborne early warning system both rely on the same physical principle: transmit a signal, receive its echo, and extract information from what comes back. The difference between those two systems is how they deal with the complications.

The Basic Idea: Pulse and Echo

A radar transmitter emits a short burst of electromagnetic energy. That energy propagates outward at the speed of light (approximately 299,792,458 m/s in vacuum, slightly less in the atmosphere). When the wavefront strikes an object, some fraction of the energy scatters in various directions. A small portion of that scattered energy returns to the receiver. The time delay between transmission and reception tells you the range to the target.

The range equation is straightforward:

R = c × t / 2

Where R is the range to the target, c is the speed of light, and t is the round-trip time delay. The factor of 2 accounts for the fact that the signal travels to the target and back. A target at 150 kilometres returns an echo after:

t = 2 × R / c
t = 2 × 150,000 / 299,792,458
t ≈ 1.0007 ms

One millisecond for a target 150 km away. Radar operates on timescales where microseconds represent meaningful distances: 1 microsecond of timing corresponds to 150 metres of range.

The history of this idea is tangled across several countries. Heinrich Hertz demonstrated reflection of radio waves from metallic objects in 1887. Christian Hülsmeyer patented a "telemobiloscope" for ship detection in 1904. But the development that transformed radar from a curiosity into a military necessity happened in Britain during the 1930s. Robert Watson-Watt, working for the Radio Research Station, demonstrated in February 1935 that a Heyford bomber could be detected by its reflection of BBC shortwave radio transmissions, in what became known as the Daventry experiment.

By 1937, the Chain Home system was operational along Britain's eastern coast. It was crude by modern standards: the transmitters used fixed, flood-illuminating antenna arrays operating between 20 and 55 MHz (HF/VHF bands), with separate receiving stations. The antennas were strung between steel towers over 100 metres tall. Range accuracy was about 1.6 kilometres, bearing accuracy roughly 1 to 2 degrees, and height estimation was poor. But Chain Home gave the RAF something invaluable during the Battle of Britain in 1940: approximately 20 minutes of warning before Luftwaffe formations crossed the Channel. That margin, barely enough time to scramble fighters and climb to altitude, changed the outcome of the air campaign.

The same period saw parallel radar development in Germany (Freya and Würzburg systems), the Soviet Union, France, and the United States. But Britain's decisive contribution was the cavity magnetron, developed by John Randall and Harry Boot at the University of Birmingham in 1940. The cavity magnetron could generate kilowatts of power at centimetre wavelengths (microwave frequencies), enabling compact, precise radar systems. It has been called the most valuable cargo to cross the Atlantic during the war, when the Tizard Mission brought a prototype to the US in September 1940.

The Radar Range Equation

The most important equation in radar engineering describes how much power returns to the receiver from a target at range R. It connects every parameter that matters: transmitter power, antenna performance, target characteristics, range, and wavelength. Understanding it reveals why radar design involves relentless tradeoffs.

Start from first principles. The transmitter radiates power P_t through an antenna with gain G_t. The antenna concentrates energy into a beam rather than radiating uniformly in all directions. At range R, the power density (watts per square metre) at the target is:

S_target = (P_t × G_t) / (4π × R²)

This is the inverse-square law: power spreads over the surface of a sphere of radius R, with the antenna gain concentrating it relative to an isotropic radiator.

The target intercepts some of this power and scatters it. The radar cross section (σ, in square metres) characterises how effectively the target scatters energy back toward the radar. A target with RCS σ acts as if it captures energy from an area σ and re-radiates it isotropically. The power density of the scattered field back at the radar is:

S_return = (P_t × G_t × σ) / ((4π)² × R⁴)

Note the R⁴ term. The signal has made a round trip, suffering inverse-square spreading in both directions. This is the single most important fact in radar engineering: received power drops as the fourth power of range. Doubling the range reduces received power by a factor of 16. To double your detection range, you need 16 times the power (or 16 times the antenna aperture, or some equivalent combination).

The receiving antenna captures this power density over its effective aperture A_e. For an antenna with gain G_r, the effective aperture is:

A_e = (G_r × λ²) / (4π)

Where λ is the wavelength. The received power is then:

P_r = S_return × A_e
P_r = (P_t × G_t × G_r × λ² × σ) / ((4π)³ × R⁴)

This is the radar range equation in its standard form. For a monostatic radar (transmitter and receiver co-located, using the same antenna), G_t = G_r = G, and the equation simplifies to:

P_r = (P_t × G² × λ² × σ) / ((4π)³ × R⁴)

Each term tells you something about the design:

The practical question is: can you detect the target? Detection occurs when the received power exceeds the minimum detectable signal (S_min) of the receiver, which is determined by the receiver noise floor and the required signal-to-noise ratio (SNR). The noise power in the receiver is:

N = k × T_s × B_n

Where k is Boltzmann's constant (1.38 × 10⁻²³ J/K), T_s is the system noise temperature in Kelvin, and B_n is the receiver noise bandwidth in Hz. The maximum detection range is found by setting P_r equal to S_min:

R_max = ((P_t × G² × λ² × σ) / ((4π)³ × k × T_s × B_n × SNR_min))^(1/4)

A worked example: a ground-based air surveillance radar operating at S-band (3 GHz, λ = 0.1 m) with P_t = 1 MW peak power, G = 34 dB (≈ 2,512), σ = 1 m² (a small aircraft), T_s = 500 K, B_n = 1 MHz, and required SNR of 13 dB (≈ 20):

import math
 
P_t = 1e6         # 1 MW peak power
G = 10**(34/10)   # 34 dB gain ≈ 2512
lam = 0.1         # wavelength in metres (S-band, 3 GHz)
sigma = 1.0       # RCS in m²
k = 1.38e-23      # Boltzmann's constant
T_s = 500         # system noise temperature in K
B_n = 1e6         # noise bandwidth in Hz
SNR_min = 10**(13/10)  # 13 dB ≈ 20
 
numerator = P_t * G**2 * lam**2 * sigma
denominator = (4 * math.pi)**3 * k * T_s * B_n * SNR_min
 
R_max = (numerator / denominator) ** 0.25
print(f"Maximum detection range: {R_max/1000:.1f} km")
# Output: Maximum detection range: approximately 245 km

That 245 km figure is optimistic. It assumes free-space propagation, no atmospheric losses, no clutter, and a target conveniently presenting a 1 m² cross section. Real-world detection ranges are typically 60 to 80 percent of the theoretical maximum, after accounting for system losses, atmospheric attenuation, and the fact that targets rarely cooperate.

Pulse Radar Timing

A pulse radar transmits a short burst of RF energy, then switches to receive mode and listens for echoes. The timing of these pulses determines the radar's ability to measure range unambiguously, and introduces constraints that shape the entire system design.

The Pulse Repetition Interval (PRI) is the time between successive pulse transmissions. Its reciprocal is the Pulse Repetition Frequency (PRF): PRF = 1/PRI. A typical surveillance radar might use a PRF of 300 Hz (PRI = 3.33 ms), while a fire-control radar tracking a single target might use 5,000 Hz or higher.

The duty cycle is the fraction of time the radar is actually transmitting:

Duty Cycle = τ / PRI

Where τ is the pulse width. A radar transmitting 1-microsecond pulses at a PRF of 1,000 Hz has a duty cycle of 0.001, or 0.1%. The average power is:

P_avg = P_peak × Duty Cycle
P_avg = 1,000,000 W × 0.001 = 1,000 W

This is why radar can achieve megawatt peak powers from systems with modest average power consumption.

Range Ambiguity

The radar measures range from the time delay between transmitting a pulse and receiving its echo. But what if the echo from a distant target arrives after the next pulse has already been transmitted? The receiver cannot distinguish this late echo from a short-delay echo of the new pulse. The target appears to be much closer than it actually is.

The maximum unambiguous range is determined by the PRI:

R_ua = c × PRI / 2 = c / (2 × PRF)

For PRF = 300 Hz:

R_ua = 299,792,458 / (2 × 300) ≈ 500 km

For PRF = 10,000 Hz:

R_ua = 299,792,458 / (2 × 10,000) ≈ 15 km

This creates a direct conflict with Doppler velocity measurement, which (as we will see) benefits from high PRFs. Radar designers categorize systems as low-PRF (long unambiguous range, ambiguous Doppler), medium-PRF (ambiguous in both), or high-PRF (unambiguous Doppler, ambiguous range).

Timing Diagram

A simplified pulse radar timing sequence looks like this:

Time →
 ┌──┐                              ┌──┐                              ┌──┐
 │TX│                              │TX│                              │TX│
─┘  └──────────────────────────────┘  └──────────────────────────────┘  └───
     ↑                                  ↑
     │ Echo from                        │ Echo from
     │ target at R₁                     │ target at R₁
     │                                  │
     ├──── t₁ = 2R₁/c ────►            ├──── t₁ = 2R₁/c ────►
 
     ←────────── PRI ──────────────────►

The pulse width τ also determines the minimum range and the range resolution. The radar cannot receive while transmitting (in most designs), so the minimum range is:

R_min = c × τ / 2

A 1-microsecond pulse gives R_min = 150 metres. Anything closer is "eclipsed" by the transmit pulse. Range resolution, the ability to distinguish two targets at different ranges, is also set by pulse width:

ΔR = c × τ / 2

A 1-microsecond pulse gives 150-metre range resolution. A 0.1-microsecond pulse gives 15 metres. But shorter pulses contain less energy, reducing detection range. The solution is pulse compression: transmit a long pulse with internal modulation (typically a linear frequency chirp) and compress it on receive using a matched filter. This gives the energy of a long pulse with the resolution of a short one. Pulse compression ratios of 100:1 to 10,000:1 are common in modern systems.

Blind Ranges and Eclipsing

Beyond minimum range, there are periodic blind ranges at multiples of the PRI where the echo from a target arrives precisely when the next pulse is being transmitted. A target at range R is eclipsed when:

2R/c = n × PRI + t_eclipse

Where t_eclipse falls within the transmit pulse duration. In practice, radars use PRI staggering (varying the PRI from pulse to pulse) to ensure that eclipsed targets appear at different apparent ranges on successive dwells, allowing the processor to resolve them.

Doppler: Measuring Velocity

When a target moves relative to the radar, the returned signal undergoes a frequency shift. This is the Doppler effect, and it gives radar the ability to measure radial velocity directly.

For a target moving with radial velocity v_r toward the radar, the Doppler frequency shift is:

f_d = 2 × v_r × f_0 / c

Where f_0 is the radar carrier frequency and c is the speed of light. The factor of 2 appears because the signal travels both to and from the target, so the effective relative motion is doubled.

A worked example: a radar operating at X-band (10 GHz) observing a vehicle moving at 100 km/h (27.8 m/s) directly toward the radar:

f_0 = 10e9      # 10 GHz
v_r = 27.8      # m/s (100 km/h)
c = 3e8         # speed of light
 
f_d = 2 * v_r * f_0 / c
print(f"Doppler shift: {f_d:.0f} Hz")
# Output: Doppler shift: 1853 Hz

A 1,853 Hz shift on a 10 GHz carrier. This is a tiny fractional change (about 0.2 parts per million), but it is straightforward to measure with coherent signal processing.

CW Doppler Radar

The simplest Doppler radar transmits continuously (CW, Continuous Wave) at a single frequency and measures the frequency shift of the return. This is exactly how a police speed radar works: a small CW transmitter at Ka-band (34.7 GHz typically in Europe), pointed at traffic, with a frequency discriminator measuring the Doppler shift.

CW Doppler radar measures velocity but not range. There is no pulse timing to extract a delay. You know how fast the target is moving, but not how far away it is.

Pulse-Doppler Radar

Pulse-Doppler radar combines range measurement (from pulse timing) with velocity measurement (from Doppler shift). The key insight is that a coherent pulse train (successive pulses with a known, stable phase relationship) is equivalent to sampling the Doppler signal at the PRF rate.

Each pulse return from a moving target has a slightly different phase compared to the previous one. Over a sequence of N pulses (a Coherent Processing Interval, or CPI), the processor performs a Discrete Fourier Transform (DFT) across the returns from each range bin. The result is a range-Doppler map: a two-dimensional matrix where one axis is range (from pulse timing) and the other is Doppler velocity (from the DFT across pulses).

import numpy as np
 
# Simulated pulse-Doppler processing (conceptual)
N_pulses = 64           # Number of pulses in CPI
N_range_bins = 1024     # Range bins per pulse
 
# raw_data shape: (N_pulses, N_range_bins)
# Each row is one pulse return, each column is one range bin
raw_data = np.zeros((N_pulses, N_range_bins), dtype=complex)
 
# ... fill raw_data with received samples ...
 
# Doppler processing: FFT across the pulse dimension for each range bin
range_doppler_map = np.fft.fft(raw_data, axis=0)
 
# The result has shape (N_pulses, N_range_bins)
# Axis 0: Doppler bins (velocity)
# Axis 1: Range bins (distance)

Velocity Ambiguity

Just as range is ambiguous beyond R_ua, velocity is ambiguous beyond the maximum unambiguous Doppler frequency. Since the Doppler signal is sampled at the PRF rate, the Nyquist theorem limits the unambiguous Doppler range to ±PRF/2. The corresponding maximum unambiguous velocity is:

v_ua = ± (λ × PRF) / 4

For X-band (λ = 0.03 m) at PRF = 10,000 Hz:

v_ua = ± (0.03 × 10,000) / 4 = ± 75 m/s (± 270 km/h)

For a low-PRF surveillance radar at 300 Hz:

v_ua = ± (0.03 × 300) / 4 = ± 2.25 m/s (± 8.1 km/h)

The low-PRF case is useless for airborne target velocity measurement. This is the range-Doppler ambiguity dilemma: you cannot have both long unambiguous range and wide unambiguous velocity with a single PRF. Fighter aircraft radars (like those in the Eurofighter Typhoon or Dassault Rafale) use medium-PRF waveforms with multiple PRFs, resolving ambiguities in both range and Doppler by comparing measurements across different PRF settings using techniques analogous to the Chinese Remainder Theorem.

Clutter and Signal Processing

If the only signals arriving at the radar were target echoes, radar would be straightforward. The problem is that every surface within the radar's beam also reflects energy. The ground, the sea, rain, buildings, vegetation, birds, insects: all of these produce returns that compete with the target of interest. This unwanted return is called clutter, and dealing with it consumes most of the effort in radar signal processing.

Types of Clutter

Ground clutter is the return from terrain within the radar beam. For a surface radar, the main beam may illuminate the ground at low elevation angles, and antenna sidelobes illuminate terrain in all directions. Ground clutter is strong (terrain RCS per unit area can be -5 to -20 dB/m², depending on surface type and frequency) and approximately stationary, though wind-blown trees and moving vehicles add spread.

Sea clutter is the return from ocean waves. It is highly variable, depending on sea state, wind speed, grazing angle, polarization, and frequency. Sea clutter has a non-trivial Doppler spectrum because the wave surface is moving. At high sea states, breaking waves produce strong, spiky returns ("sea spikes") that mimic small target echoes.

Weather clutter is the return from precipitation. Rain, snow, and hail all reflect radar energy. Weather clutter is actually the signal of interest for meteorological radars, but for surveillance and tracking radars, it obscures targets. At S-band, moderate rain (4 mm/hr) produces returns equivalent to an RCS of roughly 10⁻⁶ m² per cubic metre of rain volume, which sounds tiny but integrates over enormous volumes.

Moving Target Indication (MTI)

The simplest approach to clutter rejection is MTI filtering. Since ground clutter is mostly stationary (zero Doppler), and targets of interest are moving (nonzero Doppler), you can subtract successive pulse returns. Any signal that is the same from pulse to pulse (stationary clutter) cancels, while moving target signals survive.

A simple two-pulse MTI canceller computes:

y[n] = x[n] - x[n-1]

This is a high-pass filter in the Doppler domain, with a null at zero Doppler (and at multiples of the PRF). The frequency response is:

H(f) = 2 × sin(π × f × PRI)

The null at zero Doppler removes stationary clutter, but it also removes targets with Doppler shifts near zero or near multiples of the PRF (the "blind speeds"). Targets moving at blind speeds are invisible to the MTI filter.

More sophisticated MTI systems use staggered PRIs to shift the blind speeds, or multi-pulse cancellers (three-pulse, four-pulse) to deepen the clutter notch while reducing the blind speed problem.

CFAR Detection

After Doppler filtering, the radar must decide which range-Doppler cells contain targets. The classical approach is to set a threshold: if the signal in a cell exceeds the threshold, declare a detection. But the noise and residual clutter level varies across the range-Doppler map (due to terrain variations, weather, jamming, and other factors). A fixed threshold results in excessive false alarms in high-clutter regions and missed detections in quiet regions.

Constant False Alarm Rate (CFAR) processing solves this by adapting the threshold locally. The standard cell-averaging CFAR (CA-CFAR) estimates the local noise level by averaging the power in reference cells surrounding the cell under test, then sets the threshold as a multiple of this estimate:

Threshold = α × (1/N) × Σ P_reference_cells

Where α is a scaling factor chosen to achieve the desired false alarm probability. If the cell under test exceeds the threshold, a detection is declared. This maintains a constant false alarm rate regardless of variations in the background level.

Variants include greatest-of CFAR (GO-CFAR), smallest-of CFAR (SO-CFAR), and ordered-statistic CFAR (OS-CFAR), each handling different clutter edge and multi-target situations differently.

The Challenge in Practice

Clutter rejection is what separates a working radar from a screen full of noise. A modern airborne radar like the Euroradar CAPTOR-E (on the Eurofighter Typhoon) must detect small targets (RCS < 1 m²) at ranges over 100 km while looking down at the ground, where clutter returns can be 40 to 60 dB stronger than the target. This requires doppler processing with clutter notches deeper than 60 dB, combined with CFAR detection, sidelobe blanking, and adaptive beamforming. The processing load is enormous: millions of range-Doppler cells, each requiring complex arithmetic at real-time rates, every radar dwell.

Antenna Systems and Beamforming

The antenna determines where the radar looks and how precisely it can measure angle. Two parameters matter above all others: gain (how much the antenna concentrates energy in a particular direction) and beamwidth (how narrow that concentration is).

Parabolic Dish Antennas

The classical radar antenna is a parabolic reflector. A feed horn at the focal point illuminates the dish, which collimates the radiation into a narrow beam. The 3 dB beamwidth (the angular width at which power drops to half its peak) is approximately:

θ_3dB ≈ 70 × λ / D  (degrees)

Where λ is the wavelength and D is the dish diameter. An S-band radar (λ = 0.1 m) with a 5-metre dish has a beamwidth of about 1.4 degrees. A larger dish or shorter wavelength gives a narrower beam.

The gain of a circular aperture is:

G = (π × D / λ)² × η

Where η is the aperture efficiency (typically 0.5 to 0.7). For the same S-band, 5-metre dish with η = 0.6:

G = (π × 5 / 0.1)² × 0.6 ≈ 14,804  (≈ 41.7 dB)

Parabolic dishes are mechanically rotated to scan the beam across the surveillance volume. Airport surveillance radars (ASR) typically rotate at 12 to 15 RPM, completing a full 360-degree scan every 4 to 5 seconds. The radar at Athens International Airport, like most European airports, uses an S-band ASR for approach control.

Phased Array Antennas

A phased array replaces the single dish with an array of many individual radiating elements. By controlling the phase (and optionally the amplitude) of the signal at each element, the beam can be steered electronically without mechanical motion.

Consider a linear array of N elements spaced at distance d. When each element radiates with a progressive phase shift Δφ between adjacent elements, the beam steers to angle θ_s from broadside:

sin(θ_s) = (Δφ × λ) / (2π × d)

The conventional element spacing is d = λ/2. This choice is significant: it is the maximum spacing that avoids grating lobes (spurious beams at angles where the element contributions add constructively, creating ambiguous beam directions). With d = λ/2, the visible region spans ±90 degrees from broadside without grating lobes.

The array factor (the radiation pattern due to element arrangement, independent of individual element patterns) for a uniform linear array is:

AF(θ) = sin(N × π × d × sin(θ) / λ) / (N × sin(π × d × sin(θ) / λ))

The beamwidth narrows with the number of elements: θ_3dB ≈ 0.886 × λ / (N × d). For a 1,000-element linear array at λ/2 spacing, the beamwidth is about 0.1 degrees.

PESA vs. AESA

A Passive Electronically Scanned Array (PESA) uses a single transmitter and receiver, with passive phase shifters at each element. All elements share the same transmitter signal, and beam steering is accomplished by adjusting the phase shifters. PESA systems can steer the beam rapidly (microseconds, compared to seconds for mechanical rotation), but the single transmitter is a bottleneck and a single point of failure.

An Active Electronically Scanned Array (AESA) places a small transmit/receive (T/R) module at each element. Each module contains its own power amplifier, low-noise amplifier, and phase shifter. The advantages are substantial:

• Graceful degradation: if a few T/R modules fail, the array loses a small amount of gain and sidelobe performance but continues operating. A dish antenna with a failed transmitter is completely down.
• Simultaneous modes: different parts of the array can form independent beams, allowing the radar to track targets in one direction while searching in another.
• Wideband operation: T/R modules can operate over broader bandwidths than passive phase shifters.
• Low probability of intercept: the radar can manage its emissions more precisely, making it harder for adversaries to detect and characterize.

Modern AESA radars include the Euroradar CAPTOR-E, the Thales RBE2-AA (on the Rafale), and the Hensoldt TRML-4D ground-based system. A typical fighter AESA might have 1,000 to 1,500 T/R modules, each producing 5 to 10 watts, giving total radiated power of 5 to 15 kW distributed across the array face.

Synthetic Aperture Radar (SAR)

A conventional radar's angular resolution is limited by its antenna size: θ_3dB ≈ λ/D. For an airborne radar at X-band (λ = 0.03 m) with a 1-metre antenna, the beamwidth is about 1.7 degrees. At a range of 100 km, this beamwidth illuminates a ground patch 3 km wide. You can see that something is there, but not its shape.

SAR circumvents this limitation by exploiting the motion of the radar platform. As an aircraft or satellite moves, the radar transmits and receives pulses at successive positions along its flight path. By coherently combining these returns, the signal processor synthesizes the effect of an antenna much larger than the physical one, one whose length equals the distance the platform travels during the observation.

How SAR Works

As the platform moves, a target on the ground remains within the antenna beam for some distance of travel (the synthetic aperture length, L_sa). During this time, the Doppler shift of the target's return changes continuously: it is positive as the platform approaches, zero at closest approach (broadside), and negative as the platform recedes.

This Doppler history is unique to each target position. The SAR processor uses this fact by applying a matched filter in the along-track (azimuth) dimension. The matched filter correlates the received signal with the expected Doppler history for a target at each candidate position. Where correlation is high, a target exists.

The azimuth resolution of a stripmap SAR is, remarkably:

δ_az = D / 2

Where D is the physical antenna length. This is independent of range and wavelength. A 1-metre antenna yields 0.5-metre azimuth resolution regardless of whether the target is 10 km or 100 km away. The reason: at longer range, the target stays in the beam longer (the beam is wider), so the synthetic aperture grows proportionally, cancelling the range dependence.

This result is counterintuitive. Normally, resolution degrades with range. In SAR, the physical antenna's beamwidth determines the synthetic aperture length, and the processing gain from a longer synthetic aperture exactly compensates for the greater range.

Range Resolution

SAR range resolution is independent of the azimuth processing. It depends on the transmitted bandwidth:

δ_r = c / (2 × B)

Where B is the signal bandwidth. A 300 MHz bandwidth gives:

δ_r = 3 × 10⁸ / (2 × 3 × 10⁸) = 0.5 m

Modern SAR systems use bandwidths of 100 MHz to over 1 GHz, achieving range resolutions from 1.5 metres down to 15 centimetres.

SAR Modes

Several SAR imaging modes exist:

Stripmap SAR: the antenna beam points to a fixed angle relative to the flight path. As the platform moves, the beam sweeps across the ground, imaging a continuous strip. Resolution is D/2 in azimuth.

Spotlight SAR: the beam is steered to continuously illuminate a specific area as the platform passes. This increases the synthetic aperture length beyond what the beamwidth would allow, improving azimuth resolution at the cost of area coverage. Spotlight SAR can achieve resolutions below 10 centimetres.

ScanSAR: the beam is scanned across multiple sub-swaths, trading resolution for wider coverage. Each sub-swath is imaged at reduced resolution (because the target is illuminated for less time), but the total swath width can be hundreds of kilometres.

European SAR Systems

The European Space Agency's Sentinel-1 mission is probably the most widely used SAR system globally. Sentinel-1A (launched 2014) and Sentinel-1B (operated until 2022, with Sentinel-1C launched in December 2024 as replacement) carry C-band (5.405 GHz) SAR instruments in a 693-km sun-synchronous orbit. Operating in Interferometric Wide Swath (IW) mode, Sentinel-1 images a 250-km swath at roughly 5 by 20 metre resolution. The data is freely available through the Copernicus programme, and has been used for everything from monitoring Arctic sea ice to tracking ground subsidence in Athens due to groundwater extraction.

Germany's TerraSAR-X (launched 2007) and TanDEM-X (launched 2010) carry X-band SAR instruments capable of spotlight resolution below 1 metre. The TanDEM-X mission flies two satellites in close formation, separated by 200 to 500 metres, generating interferometric SAR data that produced the most accurate global digital elevation model available (12-metre posting, 2-metre relative height accuracy).

Italy's COSMO-SkyMed constellation (four X-band SAR satellites) provides similar high-resolution capability. These European SAR assets collectively provide continuous, all-weather, day-and-night observation of the Earth's surface, something optical satellites cannot guarantee due to cloud cover.

SAR Processing

The core of SAR processing is two-dimensional matched filtering. In the range dimension, the processor compresses the chirp pulse (standard pulse compression). In the azimuth dimension, it applies a matched filter corresponding to the quadratic phase history of a point target.

The azimuth signal for a point target has a phase history approximately described by:

φ(t) = (4π / λ) × R(t)

Where R(t) is the instantaneous range to the target, which varies quadratically with time as the platform passes:

R(t) ≈ R_0 + v² × t² / (2 × R_0)

This gives a linear frequency modulation (chirp) in the azimuth direction, and the matched filter is therefore another chirp with the conjugate phase. The processing can be performed efficiently in the frequency domain using the Range-Doppler algorithm or more advanced algorithms (Chirp Scaling, Omega-K) that handle range cell migration correctly.

A single SAR image might require processing billions of complex samples, with computation times of minutes to hours depending on resolution and scene size. Modern SAR processors use GPUs extensively.

Weather Radar

Meteorological radar is the application where clutter becomes the signal. Rather than trying to eliminate returns from precipitation, weather radar measures them precisely to characterize storm intensity, type, and motion.

Reflectivity

The basic measurement of weather radar is reflectivity, which describes how intensely the precipitation scatters the radar signal. The reflectivity factor Z is defined as:

Z = Σ D⁶

Where the sum is over all raindrops in a unit volume, and D is each drop's diameter. The sixth-power dependence means that a few large drops contribute far more than many small drops. This makes Z highly sensitive to drop size distribution.

Because Z spans many orders of magnitude, it is expressed in logarithmic units, dBZ:

dBZ = 10 × log₁₀(Z / Z₀)

Where Z₀ = 1 mm⁶/m³. Typical values:

Most European weather radars operate at C-band (5.6 GHz), balancing range capability against attenuation in heavy rain. The S-band (2.8 GHz) used in many American systems suffers less rain attenuation but requires larger antennas.

Dual-Polarization

Modern weather radars transmit and receive both horizontal and vertical polarization simultaneously. This provides additional information about the shape, size, and type of hydrometeors (rain, snow, hail, ice).

Key dual-polarization variables:

Differential reflectivity (ZDR): the ratio of horizontal to vertical reflectivity. Raindrops are flattened by aerodynamic forces as they fall, so they reflect more in the horizontal plane. ZDR > 0 indicates oblate particles (rain). ZDR ≈ 0 indicates spherical particles (small drizzle or hail that tumbles). ZDR values help distinguish rain from hail, snow from ice pellets.

Correlation coefficient (ρ_hv): measures how similar the horizontal and vertical returns are pulse-to-pulse. Pure rain has ρ_hv > 0.99. Melting snow (the "bright band") has ρ_hv around 0.85 to 0.95. Non-meteorological targets (birds, insects, ground clutter) have ρ_hv < 0.8. This variable is excellent for filtering out biological echoes and ground clutter.

Specific differential phase (KDP): related to the phase difference between horizontal and vertical signals accumulated along the propagation path. KDP is proportional to liquid water content and is unaffected by attenuation or calibration errors, making it a robust rainfall estimator.

Doppler Velocity

Weather radar also measures the radial velocity of precipitation using the Doppler shift, just like any other Doppler radar. This reveals wind patterns within storms. A mesocyclone (the rotating updraft in a supercell thunderstorm) produces a characteristic velocity signature: approaching velocities (green) adjacent to receding velocities (red) in a tight couplet.

Volume Scanning

A weather radar performs a volume scan by completing a series of 360-degree sweeps at increasing elevation angles (typically 10 to 20 different elevations over 5 to 10 minutes). This builds a three-dimensional picture of the precipitation field, allowing meteorologists to identify storm structure at all altitudes.

European Weather Radar Networks

EUMETNET's OPERA programme (Operational Programme for the Exchange of Weather Radar Information) coordinates weather radar data across Europe. As of 2026, the OPERA network includes over 200 weather radars from national meteorological services across more than 30 European countries. The data feeds into the OPERA composite, a pan-European radar mosaic updated every 15 minutes.

When you open a weather app in Barcelona and see rain approaching from the west, the radar data driving that display likely comes from AEMET's network of 15 C-band radars across Spain, feeding into the OPERA composite. The app receives reflectivity data in dBZ, converts it to estimated rainfall rate using an empirical relationship (the Marshall-Palmer Z-R relationship, Z = 200 × R^1.6, where R is rainfall rate in mm/hr), and overlays it on a map.

Real-time radar composites across Europe are freely available through EUMETNET and many national services. The Finnish Meteorological Institute, the Dutch KNMI, and the German DWD all provide excellent public radar visualizations.

Modern Radar: FMCW and Automotive

Not all radar uses pulses. Frequency Modulated Continuous Wave (FMCW) radar transmits continuously while sweeping the frequency. This approach has become dominant in automotive radar, where compactness, low cost, and simultaneous range-velocity measurement are critical.

FMCW Principle

An FMCW radar transmits a signal whose frequency increases linearly with time (a chirp). The transmitted frequency at time t is:

f_tx(t) = f_0 + (B / T) × t

Where f_0 is the starting frequency, B is the total bandwidth, and T is the chirp duration. The reflected signal from a target at range R arrives with a time delay τ = 2R/c. At any instant, the transmitted and received frequencies differ by a constant amount called the beat frequency:

f_beat = (B / T) × τ = (2 × B × R) / (c × T)

The beat frequency is proportional to range. Measuring f_beat gives R directly:

R = (f_beat × c × T) / (2 × B)

The beauty of FMCW is that the beat frequency is low (kilohertz to megahertz) even when the signal bandwidth is large (gigahertz). This makes the receiver electronics simple and inexpensive compared to pulse radar, where the receiver must handle the full signal bandwidth.

Range and Velocity Resolution

Range resolution is determined by the bandwidth, just as in SAR:

δ_R = c / (2 × B)

A 77 GHz automotive radar with 4 GHz bandwidth achieves:

δ_R = 3 × 10⁸ / (2 × 4 × 10⁹) = 0.0375 m ≈ 3.75 cm

Velocity measurement in FMCW uses the same principle as pulse-Doppler: transmit multiple chirps and analyse the phase change between successive chirps for each range bin. A moving target introduces an additional phase shift from chirp to chirp, which the processor extracts via FFT across chirps.

The complete FMCW processing chain is therefore two FFTs:

1. Range FFT: FFT of samples within each chirp, converting beat frequency to range.
2. Doppler FFT: FFT across chirps for each range bin, extracting velocity.

import numpy as np
 
# FMCW radar processing (simplified)
N_samples = 256    # ADC samples per chirp
N_chirps = 128     # Number of chirps in a frame
 
# raw_data shape: (N_chirps, N_samples)
raw_data = np.zeros((N_chirps, N_samples), dtype=complex)
 
# ... fill with received IF samples ...
 
# Step 1: Range FFT (along fast-time axis)
range_profiles = np.fft.fft(raw_data, axis=1)
 
# Step 2: Doppler FFT (along slow-time axis)
range_doppler = np.fft.fft(range_profiles, axis=0)
 
# range_doppler[doppler_bin, range_bin] gives the complex amplitude
# for each range-velocity cell

77 GHz Automotive Radar

The 76-81 GHz band (commonly called "77 GHz") has become the standard for automotive radar. At these frequencies, the wavelength is about 3.9 mm, which allows physically tiny antennas (a few centimetres across) to achieve narrow beams. The wide bandwidth available (up to 5 GHz) enables fine range resolution.

A modern 77 GHz radar chip (such as those from Texas Instruments, Infineon, or NXP) integrates the transmitter, receiver, chirp synthesizer, ADC, and DSP onto a single silicon-germanium or CMOS die. The chip costs roughly €3 to €8 in volume production. A typical automotive radar module, including the chip, antenna PCB, housing, and connector, costs €30 to €80.

A typical car sold in Europe in 2026 carries between three and five radar sensors:

• Front long-range radar: 77 GHz, narrow beam, range up to 250 metres. Used for adaptive cruise control, automatic emergency braking, and highway driving assist.
• Front corner radars: 77 GHz, wide field of view (±75 degrees), range up to 100 metres. Used for cross-traffic alert and intersection assist.
• Rear corner radars: similar to front corners, covering blind spots and rear cross-traffic.

These radars operate in FMCW mode, producing range-Doppler-angle measurements at frame rates of 20 to 50 Hz. The angular dimension is measured using multiple receive antennas (typically 4 to 12 channels) and digital beamforming or MIMO virtual array techniques.

ADAS and Autonomous Driving

Automotive radar's key advantage over cameras and lidar is its robustness in adverse weather. Radar works in rain, fog, snow, dust, and direct sunlight, conditions that degrade or disable optical sensors. At 77 GHz, atmospheric attenuation in heavy rain (50 mm/hr) is about 10 dB/km, noticeable but manageable at the 100 to 200 metre ranges relevant for automotive applications.

The European New Car Assessment Programme (Euro NCAP) increasingly requires radar-dependent safety features for top safety ratings. As of 2026, automatic emergency braking (AEB) for vehicles and pedestrians is mandatory for a five-star rating, and AEB for cyclists and junction scenarios is heavily weighted. All of these depend on radar, either alone or fused with camera data.

The challenge for autonomous driving is not individual sensor performance but sensor fusion: combining radar, camera, and lidar measurements into a coherent perception of the environment. Radar provides range, velocity, and angle. Cameras provide texture, colour, and object classification. Lidar provides dense 3D point clouds. Fusing these in real time, resolving conflicts between sensors, handling sensor failures, and maintaining accuracy across all conditions is an ongoing engineering challenge.

Radar Cross Section and Stealth

The radar range equation contains a term, σ, the radar cross section, that describes the target. RCS is not simply the physical area of the target. It is the effective area that, if it intercepted the incident power density and re-radiated it isotropically, would produce the same reflected power density as the actual target. RCS depends on the target's shape, size, material properties, the radar frequency, and the observation angle.

What Determines RCS

A flat plate perpendicular to the radar beam produces a very large RCS (proportional to the fourth power of its size relative to wavelength) because it acts as a specular reflector, directing most energy straight back. A sphere has an RCS roughly equal to its physical cross-sectional area (πr²) in the optical regime (radius much larger than wavelength). A corner reflector (three mutually perpendicular flat surfaces, like the interior corner of a room) is one of the strongest radar reflectors for its size because it returns energy back along the incoming direction regardless of angle (within a wide angular range).

Conversely, smooth curved shapes scatter energy away from the radar, and shapes with edges and flat surfaces at oblique angles direct specular reflections away from the source.

RCS of Common Objects

RCS values vary enormously:

The stealth fighter figures are remarkable. An aircraft with a physical size of 15 metres presents an RCS comparable to a marble. This does not mean stealth aircraft are invisible, it means they are detectable only at much shorter ranges (recall R_max goes as σ^(1/4), so reducing RCS by a factor of 10,000 reduces detection range by a factor of 10).

Stealth Design Principles

Stealth aircraft design focuses on three approaches:

Shaping: avoid flat surfaces perpendicular to likely radar directions. The faceted design of the F-117 (the first operational stealth aircraft) directed specular reflections away from the illuminating radar. Modern stealth aircraft (B-2, F-22, F-35, the European FCAS/Tempest future concepts) use continuous curves computed to minimize backscatter across a range of angles, with all edges aligned to a few specific angles. This concentrates the scattered energy into narrow angular sectors ("flashes") at known directions, ensuring that the RCS is very low at most observation angles.

Radar-absorbing materials (RAM): coatings and structural materials that absorb radar energy and convert it to heat. RAM works by matching the impedance of the surface to free space (minimizing reflection) and dissipating the energy internally. Carbon-loaded foam, ferrite-loaded paints, and honeycomb structures with resistive cards are common RAM technologies. RAM is most effective at higher frequencies (X-band and above) and becomes impractical at lower frequencies (VHF/UHF) where the required material thickness becomes excessive.

Internal weapons carriage and engine inlet treatment: engines are strong radar reflectors because the spinning compressor blades act as efficient scatterers. Stealth aircraft use S-shaped inlet ducts that prevent a direct line of sight from the radar to the engine face. Weapons are carried internally to avoid the large RCS contributions of externally mounted missiles and bombs.

Counter-Stealth

Stealth is not absolute. Several approaches can defeat or degrade stealth effectiveness:

Lower-frequency radars (VHF/UHF, operating at wavelengths of 1 to 10 metres) achieve resonance scattering from aircraft-sized features, making shaping less effective. Russian VHF radar systems (like the Nebo-M series) are specifically designed for this purpose. The tradeoff is that VHF radars have poor angular resolution (beamwidth of many degrees for practical antenna sizes), so they can detect but not precisely track stealth targets.

Bistatic and multistatic radar (where the transmitter and receiver are at different locations) can exploit the "flashes" in a stealth aircraft's RCS pattern, since the shaping is optimised for monostatic (co-located transmitter/receiver) geometries. A stealth aircraft designed to have minimal RCS toward the transmitter may scatter significant energy toward a differently-located receiver.

Passive radar systems (exploiting transmissions of opportunity such as FM radio, DVB-T television, or cellular base stations) are inherently multistatic and do not emit, making them difficult for the target to detect and respond to. European passive radar systems from Hensoldt (Twinvis) and Thales are in operational service.

Radar Bands: A Reference

Radar systems span a wide range of frequencies, each with distinct propagation characteristics, antenna sizes, and applications:

The choice of band involves fundamental tradeoffs. Lower frequencies propagate farther, penetrate rain and foliage better, and require simpler transmitters, but need large antennas for narrow beams. Higher frequencies enable smaller antennas, wider bandwidths (better range resolution), and more precise angle measurement, but suffer greater atmospheric attenuation.

The 77 GHz automotive radar sits at W-band, where the tiny wavelength (3.9 mm) allows a complete multi-channel antenna and transceiver to fit in a package the size of a credit card. A VHF early warning radar, doing the same job at 150 MHz (2-metre wavelength), needs antenna arrays tens of metres across.

Tying It Together

Radar has been operational for nearly 90 years, and its underlying physics has not changed. Electromagnetic waves still travel at the speed of light, received power still drops as R⁴, and clutter still dominates the received signal in most practical scenarios. What has changed is the processing.

The Chain Home stations of 1940 displayed raw returns on cathode ray tubes, interpreted by trained operators who could distinguish aircraft from clutter through experience and skill. A modern AESA radar like the Euroradar CAPTOR-E processes billions of operations per second, forming multiple simultaneous beams, applying adaptive clutter suppression, tracking hundreds of targets, and classifying them by their micro-Doppler signatures (the modulation caused by rotating helicopter blades, jet engine turbine faces, or a person's walking gait).

SAR has transformed earth observation. The Sentinel-1 constellation delivers freely available, all-weather radar imagery of the entire planet every six days, enabling applications from deforestation monitoring in the Amazon to ground displacement measurement in Berlin with millimetre precision.

FMCW radar has become a commodity. A 77 GHz radar chip costs less than a restaurant meal in Amsterdam, and there are hundreds of millions of them on the road, preventing collisions every day.

The mathematics underlying all of these systems is the same: the radar range equation, the Doppler shift, matched filtering, and Fourier transforms. The physics is clear, well-understood, and has been for decades. The engineering challenge lies in the details: managing clutter, maintaining coherence, processing data in real time, and building reliable hardware that works in the rain, the desert, at 60,000 feet, or on a satellite in low Earth orbit.

If you want to go further, the standard reference is Merrill Skolnik's "Introduction to Radar Systems" (now in its fourth edition). For SAR specifically, John Curlander and Robert McDonough's "Synthetic Aperture Radar: Systems and Signal Processing" remains excellent. For phased arrays, Robert Mailloux's "Phased Array Antenna Handbook" covers both theory and practice. All three are dense reading, but radar is a dense subject. The simplicity of "transmit a pulse, measure the echo" conceals layers of engineering that have kept thousands of researchers and engineers occupied for the better part of a century.