// Foundation
What Is Modulation?
A radio wave by itself is just a sine wave — it carries no information. Modulation is the process of stamping information onto that carrier wave by changing one of its three properties: amplitude, frequency, or phase.
Think of it like this: a carrier wave is like a blank piece of paper. Modulation is the act of writing on it. The receiver "reads" the changes to recover the original message.
AMPLITUDE FREQUENCY PHASE AM — change the height FM — change the speed PM — shift the timing Data: 1 1 0 0 1 0 1 Data: 1 1 0 0 1 0 1 Data: 1 1 0 0 1 BIG BIG small small BIG small BIG FAST FAST slow slow FAST slow FAST phase flip! 180°
These three techniques — AM (Amplitude Modulation), FM (Frequency Modulation), and PM (Phase Modulation) — are the building blocks of all radio. Modern digital radios combine amplitude AND phase changes simultaneously to carry far more bits per second. This is where the IQ plane comes in.
// The Core Concept
The IQ Plane — Seeing the Signal

A Signal is a Rotating Arrow

Any sine wave can be described as a rotating arrow (phasor) in a 2D plane. The arrow's length is the amplitude. The angle it makes is the phase. As time passes, the arrow spins. If we freeze a snapshot, we see exactly where the arrow is pointing — and that one point tells us everything about the signal at that instant.

Why "I" and "Q"?

The 2D plane that the phasor rotates in has two axes. RF engineers named them I (In-phase) and Q (Quadrature).
I Q 0 +1 −1 +1 −1 I = cos(θ) Q = sin(θ) Point (I, Q) θ What I and Q mean physically I In-phase cosine component of signal horizontal axis of IQ plane Q Quadrature sine component — 90° offset vertical axis of IQ plane Together: s(t) = I·cos(2πft) − Q·sin(2πft) Every signal is just a point (I, Q) on this plane. Changing where the point sits = different signal.
The big idea: Any sine wave — no matter its amplitude or phase — can be represented as a single point (I, Q) on this 2D plane. Modulation = moving that point around the plane to encode different symbols. The receiver measures the point and decodes the symbol. Simple!
// Analogue Modulation
AM, FM and PM — How They Look in the IQ Plane
AM Point moves along I axis I Q big amp (=1) small amp (=0) slides along I axis PM / PSK Point moves around circle edge I Q 00 (0°) 01 (90°) 11 (180°) 10 (270°) moves around QAM Point can be anywhere! I Q 16 points = 4 bits/symbol!
AM (left): The dot slides back and forth along the I axis only. The closer to the origin = smaller signal = "0". Further out = bigger signal = "1".

PM/PSK (middle): The dot stays exactly on the circle (fixed amplitude) but rotates to different angles. Each angle = a different symbol. QPSK uses 4 angles → 2 bits per symbol.

QAM (right): The dot can go ANYWHERE in the plane — different amplitudes AND different phases. 16-QAM has 16 dot positions → 4 bits per symbol. 256-QAM has 256 positions → 8 bits per symbol. More dots = more bits per second, but the dots must be further apart or the receiver makes mistakes!
// How Bits Become Signals
Digital Modulation — Bits to Symbols

BPSK — The Simplest Digital Modulation

Binary Phase Shift Keying (BPSK) is the simplest digital modulation. It uses just 2 points on the I axis — one for "0" and one for "1". 1 bit per symbol.
BPSK Constellation I Q 1 0 (+A, 0) (−A, 0) Dashed = noise region — if noise pushes past centre = error! BPSK Waveform — bits: 1, 0, 1, 1, 0 1 0 1 1 0 flip flip flip

QPSK — Double the Speed

Quadrature Phase Shift Keying (QPSK) uses 4 points instead of 2, placing them 90° apart around the circle. Each point now represents 2 bits instead of 1 — doubling the data rate for the same bandwidth. This is the modulation used in GPS, many satellite links, and CDMA cellular.
QPSK — 4 symbols, 2 bits each I Q 00 01 11 10 4 quadrants = 4 symbols → decode by which quadrant received dot falls in QPSK Symbol Mapping BITS PHASE I Q COLOR 00 45° +0.707 +0.707 01 135° −0.707 +0.707 11 225° −0.707 −0.707 10 315° +0.707 −0.707 Grey coding: adjacent symbols differ by only 1 bit → fewer errors when noise causes a slip

QAM — Packing More and More Bits

Quadrature Amplitude Modulation (QAM) uses a grid of dot positions — both amplitude AND phase are varied. The more dots on the grid, the more bits per symbol. But — critically — the dots get closer together, so the radio needs much better signal quality (higher SNR) to reliably distinguish between them.
// Visual Gallery
Constellation Gallery — From 2 to 256 Points
Each canvas below shows a constellation diagram. Every dot = one symbol = a pattern of bits. Hover or tap to see the bit count and spacing info. Notice how the dots get more crowded as the order goes up — that's why higher QAM needs better SNR!
The crowding problem: Going from BPSK to 256-QAM increases bits per symbol from 1 to 8 — an 8× gain in spectral efficiency. But the dots are now 16× more crowded (same total power, 128× more points). You need the received SNR to be about 25 dB higher to reliably decode 256-QAM vs BPSK. This is why 5G NR uses 256-QAM only close to the base station, and BPSK at the cell edge.
// Signal Quality Metric
EVM — Error Vector Magnitude
EVM is the single most important number for measuring how "clean" a modulated signal is. It measures the distance between where the dot should be and where it actually landed — expressed as a percentage of the ideal symbol amplitude.
I Q IDEAL where it SHOULD be ACTUAL where it landed EVM error vector reference vector EVM Formula EVM% = (|error vector| / |reference vector|) × 100 Lower EVM = cleaner signal = better performance EVM Requirements by Modulation BPSK / QPSK < 17% very forgiving 16-QAM < 12.5% standard WiFi 64-QAM < 5.6% LTE, WiFi ac 256-QAM < 3.5% requires clean PA 1024-QAM < 1.5% 5G NR, WiFi 6E Noise, PA nonlinearity, IQ imbalance and phase noise all degrade EVM
EVM is degraded by: noise (random scatter), PA nonlinearity (symbols push outward from their ideal positions), phase noise from the PLL/VCO (symbols rotate randomly), and IQ imbalance (constellation becomes distorted). Good RF hardware aims for EVM better than 2% at the transmitter output for high-order modulation.
// The Power Problem
PAPR — Why High-Order QAM is Hard for PAs
PAPR stands for Peak-to-Average Power Ratio. It's a measure of how "spiky" a signal is. A simple sine wave has PAPR = 3 dB (the peak is √2 × the RMS). A complex QAM signal can have PAPR of 8–12 dB or more.
Why does this matter? Your power amplifier (PA) must be sized to handle the peak power — but it spends most of its time amplifying the much lower average power. This forces you to "back off" the PA from its maximum efficiency point to avoid clipping the peaks. Higher PAPR = more back-off = less efficient = more heat = shorter battery life.
QPSK — Low PAPR ≈ 3.5 dB PA clipping limit avg power PAPR ≈3.5dB ✓ PA can operate near full power OFDM — High PAPR ≈ 10 dB PA clipping limit avg power PAPR ≈10dB! ↓ CLIPS! ↓ CLIPS! ↓ CLIPS! ↓ CLIPS! ↓ CLIPS! ↓ CLIPS! ✗ PA must back off 10 dB → wastes power & generates heat
ModulationTypical PAPRPA Back-off NeededUsed Where
BPSK / QPSK3–4 dB2–3 dBSatellite, GPS, CDMA
16-QAM6–7 dB5–6 dBLTE, 4G downlink
64-QAM7–8 dB6–7 dBWiFi 802.11n, LTE
OFDM (WiFi 802.11ac)8–12 dB8–10 dBWiFi, LTE, 5G NR
OFDM (5G NR 256-QAM)10–13 dB10–12 dB5G base stations
This is why 5G NR base stations use Envelope Tracking (ET) or Doherty PAs: these techniques allow the PA to maintain high efficiency even when the output power is fluctuating wildly with the OFDM signal. Without ET or Doherty, a 5G base station PA would be only 10–15% efficient — burning most of the power as heat.
// Hardware Imperfection
IQ Imbalance — When the Hardware isn't Perfect
In a real radio, the I and Q channels are created by two separate hardware paths — two mixers, two filters, two amplifiers. If these paths are not perfectly matched, you get IQ imbalance. Even a tiny mismatch distorts the constellation and degrades performance.
IDEAL Perfect square grid Equal spacing I and Q AMPLITUDE IMBALANCE I wider than Q → squashed ellipse I spacing ≠ Q spacing → wrong amplitude ratio even 0.5 dB imbalance → 3 dB loss in image rejection PHASE IMBALANCE Axes not exactly 90° apart → skewed grid Q axis (actual Q axis, not 90°) I and Q axes not exactly 90° apart 2° error → only 35 dB image rejection
Image Rejection Ratio (IRR)
IRR (dB) = 10·log₁₀((1 + 2ε·cos(Δφ) + ε²) / (1 − 2ε·cos(Δφ) + ε²))
ε = amplitude imbalance ratio (linear), Δφ = phase error (radians)

Example — 0.5 dB amp + 2° phase error:
ε = 10^(0.5/20) = 1.059, Δφ = 2° = 0.0349 rad
IRR ≈ 32 dB — good enough for 64-QAM but not 256-QAM (needs >40 dB)

Solution: Digital IQ calibration (applied at baseband, corrects both errors simultaneously)
// Modern Wideband Modulation
OFDM — Why Every Modern Radio Uses It
Orthogonal Frequency Division Multiplexing (OFDM) is the modulation technique behind WiFi (802.11a/g/n/ac/ax), 4G LTE, 5G NR, DAB radio, and digital TV. Instead of modulating one carrier, OFDM splits the data across hundreds or thousands of closely-spaced, independent sub-carriers.
The problem OFDM solves: In a multipath environment (indoors, urban), the signal bounces off walls and arrives at the receiver via multiple paths with different delays. This causes some frequencies to be boosted and others to be cancelled — "frequency-selective fading". A single wide carrier gets destroyed. But if you use thousands of narrow sub-carriers, only a few get affected at any moment — the rest are fine.
OFDM: Many Narrow Sub-carriers Share the Bandwidth f |S| fading dip only 2 carriers lost! Δf = 1/T Sub-carriers are ORTHOGONAL — peak of each lines up with nulls of all others No inter-carrier interference even though they overlap in frequency! Each sub-carrier carries its own QAM symbol independently
The "orthogonal" part is the genius: The sub-carriers are spaced exactly so that when you look at sub-carrier #5, all other sub-carriers are at their zero crossing. They overlap in the frequency domain but don't interfere with each other. This is achieved using the Fast Fourier Transform (FFT) — the entire OFDM symbol is generated as one IFFT operation at the transmitter and decoded with one FFT at the receiver.
SystemSub-carriersSub-carrier spacingSymbol durationModulation per SC
WiFi 802.11ac (20 MHz)64 (52 data)312.5 kHz3.2 μsup to 256-QAM
LTE (10 MHz)600 data15 kHz66.7 μsup to 64-QAM
5G NR (100 MHz)3300 data30 kHz33.3 μsup to 256-QAM
DVB-T2 (8 MHz)32K279 Hz3.59 msup to 256-QAM
// Decision Guide
How to Choose a Modulation
ModulationBits/SymbolReq. SNR (BER 10⁻⁶)PAPRBest Used When
BPSK110.5 dB3.0 dBGPS, deep space, very low SNR links
QPSK210.5 dB3.5 dBSatellite, CDMA, long range IoT
8-PSK314 dB3.5 dBDVB-S2 satellite TV
16-QAM417 dB6.5 dBLTE near-edge, standard WiFi
64-QAM623 dB7.5 dBLTE inner cell, WiFi 802.11n
256-QAM829 dB8.5 dBWiFi 802.11ac/ax close-range, 5G NR
1024-QAM1035 dB9 dBWiFi 6E, 5G NR mmWave close range
Adaptive modulation: Modern radios don't pick one modulation — they constantly adapt based on SNR. Your phone automatically uses 256-QAM when close to a 5G tower (high SNR) and falls back to QPSK near the cell edge (low SNR). LTE calls this "Adaptive Modulation and Coding (AMC)" and it's why your download speed varies as you move around.
// Play With It
Interactive Constellation Viewer
Adjust the noise level and watch what happens to the constellation. Notice how BPSK survives far more noise than 64-QAM. This is exactly what a real receiver sees.
EVM:  |  Estimated BER:  |  Bits/symbol: