【Editorial Discussion】

A chain that looks clean on a CW tone, or on a narrow single-carrier test signal, can still fail EVM once the same ports carry a real 5G NR waveform. The residual EVM floor of the instrument is a useful check, but it is not the same question as “what changed when the waveform changed.” Before you retune the DUT, it is worth writing down which property of the wideband signal actually moved the number.

1) Compare the two results at the same reference plane, the same average power, and the same instrument input range. A CW tone has almost no crest factor; a 5G waveform does. If the analyser, a driver amplifier, or the DUT itself is set up for the CW average and then clips or compresses on the peaks, EVM will fail even though the CW tone looked excellent. Record the input level, attenuation or range, and whether the level was set for average or for peak.

2) Frequency response across the occupied bandwidth is invisible to a single tone at centre frequency. In-band ripple, a filter skirt, a cable or fixture that is only moderately flat, or an equaliser setting that does not match the numerology can all raise EVM on the real waveform while leaving a CW or mid-band tone untouched. A quick through or loopback of the same 5G waveform, at the same level, tells you how much of the error belongs to the setup.

3) Demodulator and waveform settings must be part of the note. The wrong SCS, bandwidth, or frame configuration, or a channel estimate that does not match the test model, will look like a DUT EVM failure. If the CW test did not exercise those settings at all, it cannot clear them.

4) Do not treat the CW result as a substitute for a residual-floor check with the real waveform. A loopback or a known-good source with the same 5G configuration answers a different question from a CW sanity check. If the two methods disagree, write down which term you think moved — crest and range, in-band response, or demodulator setup — before changing the DUT.

Open question: the last time EVM was fine on a CW or narrow test signal and failed on the real 5G waveform, what actually closed the gap — peak-to-average and instrument range, in-band frequency response, or demodulator / waveform settings? Please share the band and which term you think was dominant; no customer names or unverified numbers needed.