What impairments contribute to EVM?

EVM is the combined effect of all impairments in the signal chain. Thermal noise contributes uniformly across the constellation (Gaussian cloud around each point). Phase noise from the LO creates tangential smearing (arc-shaped spread). IQ gain imbalance makes the constellation elliptical. IQ phase error (deviation from 90 degrees between I and Q mixers) creates a parallelogram distortion. PA nonlinearity compresses the outer constellation points inward and rotates them (AM-AM and AM-PM distortion). DC offset shifts the entire constellation off-center. Each contribution adds in an RSS (root-sum-square) fashion: EVM_total = sqrt(EVM_noise^2 + EVM_PN^2 + EVM_IQ^2 + EVM_PA^2 + EVM_DC^2).

What are the 3GPP EVM limits for 5G NR?

3GPP TS 38.104 specifies EVM requirements for base station transmitters: QPSK requires less than 17.5% EVM. 16QAM requires less than 12.5%. 64QAM requires less than 8%. 256QAM requires less than 3.5%. These limits apply at the antenna connector and include all transmitter impairments. For user equipment (UE), the limits are slightly more relaxed. The tighter limit for 256QAM (3.5%) means the total contribution from PA nonlinearity, LO phase noise, IQ imbalance, and all other sources must fit within a 3.5% RMS error budget. Achieving this typically requires DPD and careful frequency planning to keep LO phase noise contribution below 1% EVM.

How does EVM relate to SNR and BER?

For noise-dominated impairments (Gaussian errors), EVM in percent relates to SNR as: SNR (dB) = -20 log10(EVM/100). So 3.5% EVM corresponds to 29 dB SNR, and 8% EVM corresponds to 22 dB SNR. This mapping is exact only when the error is Gaussian and uniformly distributed. PA compression creates non-Gaussian errors (concentrated on outer points), and phase noise creates correlated errors, both of which can produce different BER than the Gaussian model predicts. In practice, 3GPP EVM limits are set conservatively to ensure that the pre-FEC BER is low enough for the LDPC decoder to achieve the target block error rate of 10 percent or less.

Signal Processing

Error Vector Magnitude

EVM (%)

A 5G NR transmitter sends a 256QAM symbol intending to place it at the constellation point (0.9239, 0.3827). The receiver measures it at (0.9100, 0.4010). The error vector has magnitude 0.022, which relative to the average symbol power is 2.3% EVM. Across millions of symbols, the RMS of all these error vectors gives the EVM measurement. It captures every impairment in the transmitter: PA compression pushes outer points inward, LO phase noise smears all points tangentially, IQ imbalance stretches the constellation into an ellipse, and DC offset shifts the whole grid. A single percentage number captures the aggregate health of the entire transmit chain.

Category: Signal Processing

Unit: % RMS or dB

3GPP 256QAM Limit: <3.5%

Regulatory Limits and What Drives Them

Modulation	3GPP NR BS	3GPP NR UE	WiFi 6/6E	Equiv. SNR	Dominant Impairment
QPSK	17.5%	17.5%	N/A	15.1 dB	Noise
16QAM	12.5%	12.5%	N/A	18.1 dB	Noise, PA compression
64QAM	8%	8%	−25 dB (5.6%)	21.9 dB	PA, phase noise
256QAM	3.5%	3.5%	−30 dB (3.2%)	29.1 dB	PA, PN, IQ imbalance
1024QAM	2% (Rel-17)	N/A	−35 dB (1.8%)	34.0 dB	All impairments critical

EVM definition:
EVM(%) = √(P_error / P_reference) × 100
EVM(dB) = 10·log(P_error / P_reference)

RSS combination of impairments:
EVM_total = √(EVM_noise² + EVM_PN² + EVM_IQ² + EVM_PA² + EVM_DC²)

EVM to SNR (Gaussian):
SNR(dB) = −20·log(EVM/100)
3.5% EVM = 29.1 dB SNR, 8% EVM = 21.9 dB SNR

Common Questions

Frequently Asked Questions

What impairments contribute?

Noise (Gaussian cloud), phase noise (tangential smear), IQ gain imbalance (ellipse), IQ phase error (parallelogram), PA compression (inward push on outer points), DC offset (center shift). Add in RSS: EVM_total = √(ΣEVM_i²).

3GPP limits for 5G NR?

QPSK: 17.5%. 16QAM: 12.5%. 64QAM: 8%. 256QAM: 3.5%. Measured at antenna connector, all TX impairments included. 256QAM requires DPD and careful LO phase noise budgeting (<1% EVM from PN alone).

EVM to SNR to BER?

SNR = −20·log(EVM%). 3.5% = 29 dB SNR. Exact only for Gaussian errors. PA compression and phase noise create non-Gaussian patterns with different BER than the model predicts. 3GPP limits set conservatively for LDPC decoding margin.

Related Terms

← Envelope Tracking FBAR →

TX Testing

EVM Budget Analyzer

Enter individual impairment contributions (phase noise, IQ error, PA compression) and compute total EVM. Check compliance against 3GPP and IEEE limits for your modulation scheme.

Build EVM Budget