Matching EVM Normalization Before Comparing RF Test Results

Two instruments can report different error vector magnitude values for the same captured signal because they normalize the error differently. Before adjusting the transmitter to explain the discrepancy, compare the measurement definitions. A relative error number is incomplete without its reference.

Identify the normalization reference

The Rohde and Schwarz note on EVM comparison explains normalization to average reference power versus peak constellation power. The choice can change the reported value for a constellation with multiple amplitude levels.

Record the selected modulation, reference constellation, normalization mode, and units. If a particular communications specification governs acceptance, use its prescribed definition and measurement procedure rather than an instrument’s generic default.

Compare like statistics

Distinguish RMS EVM from peak or percentile results. Check the number of symbols, packets, and captures included. A result averaged over a long stable payload may differ from one that includes acquisition or burst edges.

For a generic illustration, the same error amplitude divided by a larger reference amplitude produces a smaller percentage. That arithmetic does not mean the physical waveform improved. The reference has changed, not necessarily the error.

Check receiver corrections

Record frequency correction, phase tracking, equalization, timing recovery, and any excluded subcarriers or symbols. These settings determine which impairments remain in the reported error.

A powerful equalizer can remove effects that another setup leaves visible. Both measurements can be internally valid while answering different questions about the transmitted waveform or complete channel.

Use a common capture where possible

Analyze the same recorded data with matched settings to separate algorithm choices from hardware differences. If separate instruments are required, characterize input level, attenuation, bandwidth, and source connection so the measurement path does not introduce another uncontrolled variable.

Check the instruments’ residual EVM at the relevant signal level and bandwidth. A cleaner transmitter cannot be resolved indefinitely by a measurement receiver whose own error dominates.

For a disputed result, export the reference and measured symbol coordinates for a small common interval. Recomputing the error sum and reference sum separately can identify a normalization disagreement without attributing it to analog hardware.

Save configuration files or an explicit settings record with the reported result. Include normalization, statistic, observation interval, correction settings, and the applicable acceptance method. A useful comparison demonstrates that the numbers describe the same quantity before attributing a difference to the board. Reporting only a percentage beside a modulation name leaves enough ambiguity to make an unchanged transmitter appear better or worse solely through measurement configuration.