Error Vector Magnitude (EVM)

EVM is a comprehensive metric that measures the vector difference between the ideal reference constellation point and the actual measured point after equalization. It aggregates the total impact of I/Q imbalance, phase noise, carrier leakage, and amplifier non-linearity into a single percentage or dB value. This makes it an efficient first-pass feature for distinguishing between high-quality and low-quality transmitter hardware in a fingerprinting system.

The measured EVM value is the result of several distinct physical-layer impairments combining destructively:

• I/Q Gain and Phase Imbalance: Creates a non-circular, skewed constellation.
• Local Oscillator Phase Noise: Causes a rotational blurring of constellation points.
• Power Amplifier Compression: Warps the outer constellation points inward due to AM-AM and AM-PM distortion.
• Carrier Leakage (Origin Offset): Shifts the entire constellation away from the zero point. Each impairment leaves a unique statistical signature within the overall EVM distribution.

While EVM alone is rarely sufficient for unique identification, its statistical distribution over time provides a soft biometric for device family or model classification. A transmitter with a consistently high EVM of -15 dB is easily distinguished from one operating at -30 dB. In open set recognition scenarios, a sudden, significant change in a known device's EVM baseline can indicate a spoofing attempt or hardware failure.

EVM requirements are strictly tied to the modulation order. The IEEE 802.11 standard specifies maximum EVM limits:

• BPSK (1/2 rate): -5 dB
• QPSK (3/4 rate): -13 dB
• 16-QAM (3/4 rate): -19 dB
• 64-QAM (5/6 rate): -28 dB
• 256-QAM (5/6 rate): -35 dB A fingerprinting system must normalize EVM measurements against the detected modulation scheme to make valid comparisons across different transmission modes.

EVM is calculated after the receiver performs channel equalization to remove linear distortion. The standard formula is:

EVM_RMS = sqrt( (1/N) * Σ |S_ideal - S_measured|^2 ) / |S_ideal_max|

Where S_ideal is the reference symbol and S_measured is the received symbol. The result is typically expressed as a percentage or in dB. For fingerprinting, the per-subcarrier EVM in OFDM systems provides a richer feature vector than the aggregate RMS value.

EVM has critical limitations as a standalone fingerprinting feature:

• Channel Sensitivity: Residual equalization errors from severe multipath can inflate EVM, masking the hardware signature.
• Information Loss: Collapsing multiple independent impairments into one scalar value discards the rich, discriminative structure found in bispectrum analysis or raw I/Q constellation topology.
• Temporal Drift: EVM can fluctuate with temperature, requiring drift compensation algorithms to maintain a stable baseline. For robust SEI, EVM is best used as a pre-filtering step before applying deep learning models to the raw waveform.

A primary contributor to EVM, this impairment occurs when the in-phase (I) and quadrature (Q) branches of a modulator have mismatched gain or are not perfectly orthogonal.

• Gain Imbalance: The I and Q signal paths have different amplification levels.
• Quadrature Error: The phase difference between the I and Q local oscillators deviates from the ideal 90 degrees.
• This creates a distinctive, asymmetric distortion in the constellation diagram that is highly device-specific.

Rapid, short-term random fluctuations in the phase of a signal, originating from the transmitter's local oscillator (LO). Phase noise causes a rotation and blurring of constellation points.

• Manifests as a spectral skirt around the carrier frequency in the frequency domain.
• In the time domain, it appears as random angular jitter on the I/Q constellation.
• The unique phase noise profile of a LO is a highly stable, unclonable fingerprint.

When a transmitter's power amplifier (PA) operates near its saturation point, it introduces non-linear distortion. This is characterized by:

• AM-AM Conversion: Amplitude distortion where the output amplitude is not a linear function of the input.
• AM-PM Conversion: Phase distortion where the output phase shift varies with the instantaneous input amplitude.
• This compresses the outer constellation points and causes spectral regrowth into adjacent channels, creating a unique out-of-band signature.

The difference between the intended and actual carrier frequency of a transmitter, caused by local oscillator inaccuracies. CFO causes the entire received constellation to rotate at a constant rate.

• CFO is a stable, long-term identifier because it is determined by the physical properties of the oscillator's crystal.
• It must be estimated and compensated for before other impairments like I/Q imbalance can be accurately measured.
• Distinct from Sampling Clock Offset (SCO), which causes a drift in symbol timing.

Also known as origin offset, this impairment occurs when a portion of the unmodulated carrier signal leaks through the mixer and appears at the output.

• In the baseband I/Q constellation, this manifests as a fixed DC offset that shifts the entire constellation away from the origin.
• The magnitude and angle of this offset are unique to each transmitter's mixer isolation characteristics.
• It is a key feature for Specific Emitter Identification (SEI) systems.

A higher-order statistical (HOS) technique that transforms a signal into a frequency-frequency-amplitude representation. Unlike the power spectrum, the bispectrum preserves phase information and is immune to Gaussian noise.

• Reveals quadratic phase coupling between different frequency components, a hallmark of non-linear hardware impairments.
• Provides a rich, noise-resistant feature space for deep learning classifiers.
• Used to extract subtle signatures that are invisible in standard EVM or spectral measurements.