Wireless Field Guide
Topic hub

Antennas and propagation

How antennas launch and collect energy, how signals lose power and spread in time and frequency on the way, and how standard channel models such as 3GPP TR 38.901 capture that for simulation.

Updated 15 Nov 2026 · antenna gain, antenna array, path loss, Friis equation, multipath fading, Doppler, delay spread, TR 38.901

Overview

Between the transmitter’s power amplifier and the receiver’s low-noise amplifier sits the part of the link you cannot redesign: the antennas and the space between them. The antennas decide how much of the transmitted power heads towards the receiver and how much of the arriving wave is captured. The propagation environment decides how much is lost on the way, how many copies of the signal arrive, and how quickly all of that changes as things move.

This hub covers both at the level a physical-layer or system engineer needs. It is enough to read an antenna datasheet, estimate path loss, pick a fading model for a simulation and understand what a channel model is and is not telling you. Detailed antenna design belongs to electromagnetics texts and solvers.

Key ideas

Antenna parameters

Antenna gain in dBi compares the power density in the strongest direction with that of an ideal isotropic radiator fed with the same power. It combines directivity, which is pattern shape, with radiation efficiency, which is loss. Gain is not free power. A high-gain antenna concentrates energy into a narrower beamwidth, so it must be pointed. Other datasheet parameters matter as much: input match (return loss or VSWR) over the band, polarisation and the isolation between polarisations, front-to-back ratio, and the frequency range over which those figures hold. Antennas are reciprocal, so the transmit and receive patterns are the same.

The figure regulators usually limit is EIRP, the transmit power plus antenna gain minus feeder loss, in dBm or dBW.

Arrays

An array places several elements, often half a wavelength apart, and feeds each with a controlled amplitude and phase. Adding the element signals in phase towards a chosen direction gives an array gain of up to 10 log10(N) dB for N elements over a single element, and the beam can be steered electronically by changing the phases. Spacing wider than half a wavelength allows grating lobes, unwanted copies of the main beam, at some steering angles. At 28 GHz half a wavelength is about 5.4 mm, which is why millimetre-wave phones and base stations carry arrays of many elements. The MIMO and beamforming hub covers how arrays are driven.

Path loss and the Friis equation

In free space, the Friis transmission equation gives received power as transmit power plus both antenna gains minus the free-space path loss, all in decibel units. The loss between isotropic antennas is 20 log10(4πd/λ). ITU-R Recommendation P.525 writes it as 32.4 + 20 log10 f + 20 log10 d dB, with f in MHz and d in km. At 3.5 GHz and 1 km that is about 103.3 dB. Loss grows by 6 dB for each doubling of distance.

Real links lose more. Buildings and terrain block and diffract the wave, foliage and clutter absorb it. At millimetre-wave frequencies rain and atmospheric gases add loss that lower bands can usually ignore. ITU-R publishes prediction methods for these effects: P.1411 for short outdoor links, P.1238 for indoor links, P.2108 for clutter loss, P.676 for gases and P.838 for rain. Empirical path loss models fit measurements with a distance exponent larger than 2 and a random shadowing term, usually log-normal.

TxRxbuildingdirect pathreflectedground reflectionvmultipath geometrypowerdelay spreadτ0τ1τ2delaypower delay profile
Multipath propagation. The receiver gets a direct path, a reflection from a building and a ground reflection, each with its own delay and attenuation. The power delay profile on the right plots the power of each arriving copy against delay, and the spread between first and last significant paths is the delay spread. A moving receiver also sees each path with its own Doppler shift.

Fading, Doppler and channel models

Multipath copies add with different phases, so received power swings by tens of dB over distances of a fraction of a wavelength. With many paths and none dominant, the envelope follows Rayleigh fading. A strong line-of-sight path gives Rician fading. The delay spread sets the coherence bandwidth, roughly its inverse, over which the channel looks flat in frequency. Motion adds a Doppler shift of up to v/λ. At 3.5 GHz and 120 km/h the maximum is about 389 Hz. Its inverse sets the order of the coherence time, over which the channel stays roughly constant. Clarke’s 1968 model of a mobile surrounded by scatterers gives the classical U-shaped Doppler spectrum used in many simulators.

For 5G NR evaluations, 3GPP TR 38.901 defines channel models for 0.5 GHz to 100 GHz. It specifies path loss, line-of-sight probability and building and car penetration loss for scenarios that include urban microcell street canyon, urban macrocell, indoor office, rural macrocell, indoor factory and suburban macrocell. Its fast-fading model generates clusters of rays with delays, angles and powers drawn from scenario statistics, with optional spatial consistency and blockage. For link-level work it also gives simpler profiles: clustered delay line models CDL-A to CDL-E and tapped delay line models TDL-A to TDL-E, where A to C represent non-line-of-sight channels and D and E line of sight. The Release 19 version adds a clause on channel modelling for integrated sensing and communication. These models describe statistics for comparing designs. They do not predict the coverage at a particular street corner, which needs measurements or ray tracing.

Where to start

Read /learn/beginner/07-fading-channels/ for multipath, Rayleigh and Rician fading and Doppler in simulation. /learn/beginner/03-decibels-and-link-budgets/ puts the Friis equation to work. Keep /reference/antennas-and-propagation-formulas/ to hand. Intermediate lessons on array processing and the TR 38.901 models are in preparation.

Tools and platforms

Tool What it is for Licence Platform
Sionna Sionna RT ray tracer, and TR 38.901 CDL, TDL, UMi, UMa and RMa models in Sionna PHY Apache-2.0 /platforms/opensource/
openEMS FDTD electromagnetic solver for antenna and structure simulation GPL-3.0 /platforms/opensource/
scikit-rf Reading Touchstone files and analysing measured antenna match BSD-3-Clause /platforms/opensource/
CommPy SISO and MIMO Rayleigh and Rician channel models in Python BSD-3-Clause /platforms/opensource/
Signal and spectrum analysers Measuring received power and spectra in the field or lab Commercial /platforms/keysight/, /platforms/rohde-schwarz/

Common mistakes

  • Adding antenna gain twice, once as dBi on the datasheet and again inside a “path loss” figure that already assumed it.
  • Using free-space loss for a non-line-of-sight urban link. It is a lower bound on loss, not an estimate.
  • Simulating a fading channel with a Doppler or delay spread that does not match the scenario, or with a sample rate too low to represent the delays.
  • Treating a TR 38.901 scenario as a prediction for a specific site.
  • Forgetting cable, connector and body losses between the radio and the antenna.

Going deeper