Error vector magnitude is the single number a transmitter is judged by, and it is an aggregate. It tells you the constellation is degraded. It does not tell you which of six independent mechanisms degraded it, and a design team that treats it as one problem usually optimises the wrong thing.
Every transmitted symbol is meant to land at a defined point in the constellation. It lands somewhere near it instead. The error vector is the difference, and error vector magnitude is that difference expressed as a percentage of the reference magnitude, averaged across many symbols.
Because it is an average of a vector quantity, it aggregates everything that displaces a symbol — noise, distortion, timing error, phase error and quantisation alike. That aggregation is what makes it a useful acceptance criterion and a poor diagnostic.
A figure without its configuration means nothing. The published Faststream figure of 3.9 to 4.1 per cent is stated under a specific configuration for exactly this reason: modulation order, bandwidth, output power and whether pre-distortion is active all move it, and comparing two numbers measured differently is comparing nothing.
They add in power, so the largest dominates and improving anything else is wasted effort until it is addressed.
| Mechanism | How it appears in the constellation | Where it is fixed |
|---|---|---|
| Amplifier non-linearity | Outer points compressed inward, worse at higher power | Digital pre-distortion, or backing off output power |
| Phase noise | Points smeared tangentially, rotating about the origin | Local oscillator and synthesiser design |
| IQ imbalance | Constellation skewed or elliptical rather than square | Quadrature calibration, usually digital correction |
| Carrier leakage | Whole constellation offset from the origin | Mixer balance and DC offset correction |
| Quantisation and noise | Uniform scatter, independent of position | Converter resolution and datapath word lengths |
| Timing and sampling error | Points displaced along the symbol trajectory | Synchronisation loops and clock distribution |
Photographing the constellation is worth more than reading the number. The shape of the error usually identifies the mechanism before any measurement is decomposed.
Decompose before optimising. Measure with pre-distortion disabled, at several output powers and modulation orders, and photograph the constellation at each. The dominant mechanism is usually obvious once the data is laid out that way, and everything else is premature.
Error vector magnitude: the distance between where a symbol should land in the constellation and where it actually lands, expressed as a percentage of the reference magnitude and averaged over many symbols. It is the standard aggregate measure of transmitter quality.
Because modulation order, bandwidth, output power and whether pre-distortion is active all change it. Two numbers measured under different conditions are not comparable, which is why a published figure should always state its configuration.
Six mechanisms mainly: amplifier non-linearity, phase noise, IQ imbalance, carrier leakage, quantisation and noise, and timing error. They add in power, so the largest dominates — and improving any of the others is wasted effort until it is addressed.
Only the part caused by amplifier non-linearity and memory. It does nothing for phase noise, IQ imbalance or quantisation, so once those dominate, further pre-distortion capability buys nothing.
Sweep output power and look at the constellation shape. Flat EVM across power indicates a noise or quantisation limit; sharp degradation with power indicates compression. Tangential smearing indicates phase noise; a skewed constellation indicates IQ imbalance.
Written for engineers. Share it with one.
Where the published figure comes from.
KNOWLEDGECFR, DPD and the amplifier.
TECHNOLOGYThe chain EVM measures.
PRODUCTAdaptive pre-distortion.
TECHNOLOGYWhere quantisation is decided.
CAPABILITYPhase noise and IQ balance.
Tell us the specification, the constraint and the deadline. Programmes that cross silicon, radio, embedded and AI are where Faststream is strongest.