ben ebsworth
ΩElectrical Engineering

Constellation Plot

QPSK / 16-QAM / 64-QAM scatter through an AWGN channel — adjustable SNR and EVM.

knobs
knobs

QPSK constellation at 20 dB SNR. Four tight clusters, each representing a 2-bit symbol.

The Core Idea

Before a radio signal reaches your phone, someone had to map data bits onto a physical waveform. Digital modulation does this by choosing points on a 2D plane — the constellation diagram — where each point represents a unique group of bits. QPSK uses 4 points (2 bits/symbol), 16-QAM uses 16 (4 bits/symbol), 64-QAM uses 64 (6 bits/symbol). More points = faster data rate, but points are closer together = more vulnerable to noise.

The Mathematics

A modulated symbol is:

s(t) = I·cos(2πfct) − Q·sin(2πfct)

where (I, Q) is the constellation point. Through an AWGN channel:

(I', Q') = (I + nI, Q + nQ)     where nI, nQ ~ N(0, σ²)

and σ² = 1/(2·SNR_linear).

knobs

16-QAM at 20 dB SNR. Sixteen clusters, 4 bits each. Points are packed tighter than QPSK.

Switch between modulation orders and watch the clusters multiply. Each step up in order doubles the bits per symbol — but also halves the noise margin between adjacent points.

Bit error rate

The probability of decoding incorrectly depends on the minimum distance between points and noise level:

  • QPSK at 20 dB SNR: BER ≈ 10⁻⁵ (one error per 100,000 bits).
  • 64-QAM at 20 dB SNR: BER ≈ 10⁻³ — two orders of magnitude worse.

This is the fundamental trade-off: spectral efficiency vs robustness. Wi-Fi and LTE switch between modulation orders in real time — high-order QAM when signal is strong, falling back to QPSK when conditions degrade.

knobs

64-QAM at 10 dB SNR. The noise clouds overlap — many symbols will be decoded incorrectly.

Error Vector Magnitude (EVM)

EVM measures the average distance between received and ideal constellation points. A good transmitter has EVM < −30 dB (received points within 3% of ideal). EVM degrades with phase noise, amplifier distortion, and channel fading.

Further reading