Multi-Frequency Vector Demodulation for Dynamic Lift off Suppression in Fast Scanning
Multi-frequency vector demodulation isolates surface flaws during fast scanning by projecting multi-channel complex impedance onto axes orthogonal to lift-off trajectories.

Locus
High-speed eddy current scanning over metallic surfaces exposes probe coils to mechanical jitter, structural deflection, and surface roughness. These physical shifts change the clearance between the sensor geometry and the conductive specimen ~ a distance known in non-destructive testing as lift-off. Fluctuations in lift-off drive large impedance shifts in the probe coil, which can easily swamp the much smaller signals caused by surface micro-cracks or material flaws.
As clearance changes, total coil impedance traces a non-linear path across the complex plane. Secondary eddy currents induced in the target material generate a counteracting magnetic field, reducing coil self-inductance and raising effective series resistance as ohmic losses dissipate energy. The spatial distribution of these currents decays exponentially with depth, bounded by skin depth and set by excitation frequency, magnetic permeability, and electrical conductivity.
At a single excitation frequency, lift-off movement and shallow defect responses share overlapping phase vectors on the impedance trajectory, leaving phase rotation as the sole way to isolate a flaw. When scanning speeds reach several meters per second, dynamic tilt and probe bounce shift these trajectories away from static calibration curves, as vertical velocity continuously alters secondary flux leakage and warps the expected phase angle.
Multi-frequency excitation shifts this geometry. Driving an inspection coil with two or more discrete frequencies at once sets up distinct skin depths in the material. Lower frequencies penetrate deeper, giving lift-off variations a distinct amplitude and phase signature, while higher frequencies concentrate flux near the surface, sharpening sensitivity to immediate probe proximity and surface conditions.
| Frequency (kHz) | Lift-Off Range (mm) | Delta Inductance (µH) | Delta Resistance (Ω) | Trajectory Phase Angle (deg) |
|---|---|---|---|---|
| 100 | 0.1 to 1.5 | -4.12 | +1.85 | 155.8 |
| 500 | 0.1 to 1.5 | -2.85 | +3.42 | 129.8 |
| 2000 | 0.1 to 1.5 | -0.94 | +5.11 | 100.4 |
Impedance shifts follow curved paths. Because the lift-off vector rotates across drive frequencies while defect phase responses stay tied to depth profiles, multi-frequency sensing expands the vector space from two dimensions to four or more. That divergence lets signal processing architectures separate clearance variations from real flaws.
A lift-off variation of 0.2 millimeters produces a coil impedance magnitude shift equivalent to a 100-micrometer EDM notch at 500 kilahertz.
Ignoring trajectory curvature during dynamic scans leads to high false-positive rates. Without dynamic suppression, a threshold will trigger spurious rejections whenever mechanical vibration exceeds tens of micrometers, forcing operators to slow the scanner and sacrifice throughput.

Synthesis
Generating dual or triple excitation waveforms requires direct digital frequency generators driving high-linearity power amplifiers. The drive signals need low intermodulation distortion to avoid phantom signals in the coil return path, while the digital-to-analog converters producing composite multi-tone signals require high dynamic range to maintain phase stability over temperature swings in the electronics.
When multiple sinusoidal currents drive one probe coil, transistor non-linearities create sum and difference frequencies. These intermodulation products leak into demodulation channels as spectral noise and low-frequency beats, causing cyclic baseline drift during fast scans. Linear class-AB or ultra-low distortion class-D power stages operating into tuned resonant matching networks help suppress driver non-linearity.
- Intermodulation harmonic leakage creates spurious carrier frequencies inside the signal band, distorting quadrature components before demodulation.
- Amplifier saturation non-linearity clips composite voltage peaks, introducing crosstalk between high- and low-frequency excitation channels.
- Phase jitter in clock synthesizers degrades vector stability during dynamic scans, causing baseline wander that translates into false defect signals.
- Temperature-induced drift in drive output impedance alters coil excitation currents, shifting the amplitude baseline across extended scan runs.
Impedance matching networks must maintain flat group delay across all active frequencies. Differential coil configurations reject common-mode environmental noise, but component tolerances between primary and secondary windings introduce baseline imbalances, which are further aggravated by thermal drift.
Onboard digital signal processing is intended to eliminate drive-line crosstalk, but unshielded analog front-end board layouts still couple high-frequency carrier energy directly into low-frequency receive paths, bypassing digital filtering entirely.

Orthogonality
Vector demodulation extracts real and imaginary components through analog mixers or digital multiply-accumulate channels locked to carrier reference clocks. Each excitation frequency yields a pair of quadrature components ~ In-phase (I) and Quadrature (Q) ~ so dual-frequency excitation produces four continuous channels: I1, Q1, I2, and Q2. Together, these channels form a time-series vector describing total electromagnetic coupling between probe and target.
Lift-off suppression relies on mapping these multi-dimensional vectors into a reduced subspace where movement vectors land on a null projection axis. Linear vector demodulation applies a transformation matrix M to raw vector V, with matrix elements calculated during setup by cycling the probe through a controlled lift-off range over a flaw-free area of the specimen.
| Demodulation Strategy | Active Frequencies | Lift-Off Range (mm) | Flaw Signal Residual (dB) | Lift-Off Suppression Ratio (dB) |
|---|---|---|---|---|
| Single-Phase Rotation | 500 kHz | 0.5 | -1.2 | 18.4 |
| Dual-Frequency Linear Subtraction | 100 kHz, 500 kHz | 0.5 | -0.4 | 34.1 |
| Dual-Frequency Vector Matrix Inversion | 100 kHz, 500 kHz | 0.5 | -0.1 | 46.8 |
| Triple-Frequency Non-Linear Projection | 100 kHz, 500 kHz, 2 MHz | 1.0 | -0.05 | 52.3 |
The coefficients of the transformation matrix balance the high-frequency vector against the low-frequency vector. Because lift-off trajectories at different frequencies are non-parallel in complex space, a linear combination exists where components generated purely by gap variation cancel to zero. Defect signals, which carry frequency-dependent phase shifts from depth attenuation, do not cancel, leaving the flaw response intact on the orthogonal channel.
Standard EN 1711 requires phase angle stability within plus or minus 0.5 degrees across the active scanning temperature band to maintain flaw detection validity.
Digitizer quantization bounds dynamic range, while field leakage alters complex mutual inductance. When transformation matrices are derived under static conditions, acceleration during fast scanning distorts probe geometry relative to the surface normal. Under contract conditions specifying compliance with ISO 15549, equipment specifications state that phase projection matrices must be updated or compensated when probe inclination angles vary by more than two degrees along high-speed scanner axes.
Bandwidth
Scan speeds over two meters per second compress spatial defect pulses into narrow time windows, requiring a wide signal path response. Filter design after quadrature mixing must balance noise rejection against dynamic phase distortion: low-pass filtering removes high-frequency carrier ripple and intermodulation products, but filtering too aggressively adds phase delay and smears localized flaw signals.
If filtering introduces unequal group delay across low- and high-frequency channels, vector subtraction occurs between components captured at slightly different probe positions along the surface. This spatial misalignment breaks suppression matrix performance during acceleration phases.

What Limits Demodulation Speed in Rotary Scanners?
Rotary probe scanners operating at several thousand revolutions per minute present severe bandwidth demands. Defects as narrow as 0.2 millimeters sweep past the coil in microseconds, requiring filters that can pass these brief transients without phase lag while still suppressing carrier crosstalk.
- Scan velocity matching determines the minimum cutoff frequency needed to pass high-frequency flaw signals without spatial pulse widening.
- Linear phase FIR selection avoids phase distortion across carrier frequencies, maintaining geometric defect registration along the scan path.
- ADC oversampling ratios must exceed eight times the highest carrier frequency to prevent aliasing during digital vector mixing.
- Decimation filter delay compensation aligns time-domain signals across low- and high-frequency demodulation branches prior to vector subtraction.
| Filter Type | Cutoff Frequency (kHz) | Group Delay (µs) | Flaw Amplitude Loss (%) | Dynamic Lift-Off Noise (mV) |
|---|---|---|---|---|
| 4-Pole Butterworth | 5 | 112.5 | 32.4 | 48.1 |
| 8-Pole Chebyshev | 10 | 84.0 | 18.1 | 31.6 |
| 127-Tap Linear Phase FIR | 20 | 25.0 | 2.1 | 4.2 |
| Equiripple Digital FIR | 50 | 10.0 | 0.3 | 1.8 |
Standard low-pass filters exhibit non-linear phase behavior near their corner frequency, causing unequal phase shifts across spectral components during fast scans. Linear phase Finite Impulse Response (FIR) filters eliminate this distortion by maintaining constant group delay, though they demand significant processing resources.
Demodulation filter cutoff must exceed the spatial frequency of the narrowest expected defect multiplied by the maximum mechanical scanning speed.
What remains unresolved in current processing hardware is whether real-time adaptive filtering can dynamically recalculate transformation matrices when scanner acceleration creates sudden mechanical vibration spikes.

Coupling
Electromagnetic interaction between field coils and conductive targets depends on geometric proximity and magnetic field spread, which is dictated by coil geometry. Absolute coils generate single-ended signals that track overall proximity, leaving them vulnerable to thermal drift. Differential coil structures use opposing windings to cancel far-field noise and gradual mechanical drift, focusing sensitivity on sharp spatial gradients.
Shielding also alters mutual inductance. Placing conductive collars or high-permeability ferrite cups around inspection coils restricts flux leakage, tightening the sensing footprint on the specimen surface. A narrower footprint isolates defect signals from structural edges, but makes the measurement more sensitive to probe tilt relative to the surface plane.
Dynamic shifts in probe tilt alter the coupling footprint and produce asymmetric changes in lift-off trajectories. Because single-frequency algorithms cannot separate probe tilt from vertical displacement, multi-frequency vector demodulation relies on tracking secondary vector components ~ provided the calibration matrix was trained with tilt vector data.
Ferrite shielding narrows magnetic field spread, reducing sensitivity to edge effects while increasing localized lift-off signal magnitude.
As a practical rule, coil diameter should match the expected range of lift-off variation to maintain a linear vector response across the full scan path.
Foil
Verifying lift-off suppression algorithms requires precision reference standards made with calibrated dielectric spacers and electric discharge machining slots. Placing non-conductive polyimide shims over alloy calibration blocks simulates discrete lift-off steps without altering the underlying material properties.
Evaluating multi-frequency instrumentation requires testing across the entire operating speed range. Static calibration yields higher suppression ratios than real-world dynamic runs, where mechanical unbalance, cable flexing, and motor EMI degrade performance. Qualification procedures must test both analog coil hardware and processing firmware under active motion.
- Mount non-conductive calibration shims of known thickness ranging from 0.05 to 1.0 millimeters over the reference alloy block.
- Record multi-frequency complex voltage vectors at five discrete lift-off heights using the automated gantry at target scan speed.
- Compute the linear transformation matrix coefficients that minimize residual output amplitude across the measured lift-off span.
- Execute high-speed verification runs across calibration notches with active lift-off variation to confirm defect signal-to-noise ratio exceeds twenty decibels.
Sourcing multi-frequency demodulation hardware involves distinct commercial trade-offs. Proprietary ASICs save board space and lower power consumption, but lock designs into single-source supply chains. FPGA-based architectures using discrete analog front-ends cost more per component and draw more power, but allow second-sourcing of digitizers and amplifiers across multiple global foundries.
Lead times for high-speed, high-resolution dual-channel ADCs remain a key vulnerability, requiring buffer inventory to protect assembly schedules.

