L
Admin
Before you report the EVM of a transmitter or an amplifier, you need to know the EVM floor of the measurement chain itself. On high-order QAM (256-QAM, 1024-QAM and above) the analyser, the signal source and the cabling can easily dominate.
1. Units first
EVM in percent and in dB are related by EVM_dB = 20*log10(EVM_% / 100). So 1 % is -40 dB, 3.5 % is about -29 dB and 0.5 % is about -46 dB. Keep the conversion handy; specs mix both.
2. Phase noise contribution
For small values, the integrated RMS phase error (in radians) of the LO chain contributes roughly the same amount to EVM. An integrated jitter of 0.5 degree RMS is about 0.0087 rad, so about 0.87 % or about -41 dB on its own. Integrate the phase noise over the band that the demodulator's carrier tracking does not remove; the tracking loop bandwidth and the subcarrier spacing both matter for OFDM.
3. The input-level bathtub
Sweep the analyser input level (or the reference level and attenuation) and plot EVM:
- Too low: the noise floor of the analyser front end and ADC dominates.
- Too high: compression or ADC clipping, especially with high PAPR OFDM signals.
The best EVM sits at the bottom of the curve. Re-check whenever the signal bandwidth or PAPR changes.
4. Equaliser and demodulator settings
- Channel estimation from reference symbols only versus from reference plus data can change the result by several tenths of a dB.
- Equaliser on or off: a flat-response setup may not need it, but a long cable or filter does. Report which mode the result used.
- Make sure the frame and numerology settings match the waveform exactly; small mismatches look like poor EVM.
5. Establish the floor with a loopback
Connect the signal source directly to the analyser with the same cables, attenuators and level as the real test. Measure EVM. This is your system floor. Then insert the DUT. If the DUT result is within about 3 dB of the floor, the measurement mostly reflects the setup, and you should either improve the setup or state the result as an upper bound. Roughly, independent contributions add as the root sum of squares in percent.
6. Other contributors
- Source EVM (DAC resolution, I/Q imbalance, its own phase noise)
- Frequency response ripple across wide channels (100 MHz and above)
- Spurs falling in-band, and interference picked up by unshielded cables
What residual EVM floor do you get on your bench at FR1 100 MHz channels, and which setting moved it the most?
1. Units first
EVM in percent and in dB are related by EVM_dB = 20*log10(EVM_% / 100). So 1 % is -40 dB, 3.5 % is about -29 dB and 0.5 % is about -46 dB. Keep the conversion handy; specs mix both.
2. Phase noise contribution
For small values, the integrated RMS phase error (in radians) of the LO chain contributes roughly the same amount to EVM. An integrated jitter of 0.5 degree RMS is about 0.0087 rad, so about 0.87 % or about -41 dB on its own. Integrate the phase noise over the band that the demodulator's carrier tracking does not remove; the tracking loop bandwidth and the subcarrier spacing both matter for OFDM.
3. The input-level bathtub
Sweep the analyser input level (or the reference level and attenuation) and plot EVM:
- Too low: the noise floor of the analyser front end and ADC dominates.
- Too high: compression or ADC clipping, especially with high PAPR OFDM signals.
The best EVM sits at the bottom of the curve. Re-check whenever the signal bandwidth or PAPR changes.
4. Equaliser and demodulator settings
- Channel estimation from reference symbols only versus from reference plus data can change the result by several tenths of a dB.
- Equaliser on or off: a flat-response setup may not need it, but a long cable or filter does. Report which mode the result used.
- Make sure the frame and numerology settings match the waveform exactly; small mismatches look like poor EVM.
5. Establish the floor with a loopback
Connect the signal source directly to the analyser with the same cables, attenuators and level as the real test. Measure EVM. This is your system floor. Then insert the DUT. If the DUT result is within about 3 dB of the floor, the measurement mostly reflects the setup, and you should either improve the setup or state the result as an upper bound. Roughly, independent contributions add as the root sum of squares in percent.
6. Other contributors
- Source EVM (DAC resolution, I/Q imbalance, its own phase noise)
- Frequency response ripple across wide channels (100 MHz and above)
- Spurs falling in-band, and interference picked up by unshielded cables
What residual EVM floor do you get on your bench at FR1 100 MHz channels, and which setting moved it the most?