Measuring Modulator Vπ and Electro-Optic Bandwidth

Scope

This article describes bench procedures for measuring the half-wave voltage $V_\pi$ of an electro-optic modulator and its electro-optic bandwidth. Three $V_\pi$ methods are covered, in increasing order of equipment demands: fitting the DC transfer curve of a Mach-Zehnder modulator, the Bessel carrier-null method for bare phase modulators, and the small-signal quadrature-slope method. Swept-frequency $S_{21}$ measurement of $V_\pi(f)$ and the 3 dB EO bandwidth follows. Ring modulators and electro-absorption modulators need different procedures and are out of scope.

Method 1: DC transfer-curve fit (Mach-Zehnder devices)

Setup. Laser at the target wavelength, polarization aligned to the modulator's axis, into the MZM, into a power meter or slow photodiode. Sweep the bias electrode with a source-measure unit or DAC, from at least $-V_{\pi,\text{expected}}$ to $+2V_{\pi,\text{expected}}$ so the sweep contains one full maximum and one full minimum with margin. Record optical power at each step.

Fit, don't peak-pick. The transfer function is

$$P(V) \;=\; C + B\cos\!\left(\frac{\pi V}{V_\pi} + \phi_b\right),$$

and the honest procedure is to fit all four parameters to the whole curve. Reading $V_\pi$ as the voltage distance between the apparent maximum and minimum uses exactly two points of the data, both of them the flattest, worst-conditioned points on the curve, and it fails entirely when the sweep window clips a peak. A fit uses every point, works on partial fringes, and returns the extinction ratio $\;(C+B)/(C-B)\;$ and bias phase $\phi_b$ for free. The fit is easier than it looks: for any trial $V_\pi$ the model is linear in the remaining parameters ($P = C + p\cos(\pi V/V_\pi) + q\sin(\pi V/V_\pi)$), so a one-dimensional scan over $V_\pi$ with a linear least-squares solve at each step is robust and has no starting-guess problem. The Vπ Fitter runs exactly this algorithm on a pasted two-column sweep.

Sweep speed matters, in both directions. Sweep too fast and the recorded curve lags the true one; too slowly and slow drifts walk the fringe during the measurement. In lithium niobate the notorious DC bias drift shifts $\phi_b$ over seconds to minutes, which distorts a slow sweep into a chirped cosine that fits poorly; repeat the sweep both directions and compare. In silicon depletion modulators the phase is a genuinely nonlinear function of voltage (the plasma-dispersion effect goes roughly as the square root of reverse bias), so a single $V_\pi$ is only defined about a stated bias; fit locally around the intended operating point and quote it as such, and watch for photocurrent-induced self-heating at high optical power, which adds a thermal phase on top of the electronic one.

Method 2: Bessel carrier null (phase modulators)

A bare phase modulator has no transfer curve to sweep: its output power is constant. Drive it instead with an RF tone $V(t) = V_m \sin\Omega t$ and look at the optical spectrum. The field splits into sidebands with amplitudes $J_n(\beta)$, where $\beta = \pi V_m / V_\pi$, and the carrier is proportional to $J_0(\beta)$, which has its first zero at

$$\beta \;=\; 2.40483.$$

Increase the RF drive until the carrier disappears on an optical spectrum analyzer, record the drive amplitude $V_{m,\text{null}}$ at the device, and

$$V_\pi \;=\; \frac{\pi \, V_{m,\text{null}}}{2.40483}.$$

The null is sharp and the method needs no calibrated photodiode, only a spectrum view with enough resolution to separate carrier from sidebands (a high-resolution OSA at 10 GHz drive, or a heterodyne trick below that). The dominant error is knowing $V_m$ at the device: cable loss, connector reflections, and impedance mismatch all stand between the generator setting and the electrode. Measure the delivered RF power at the end of the actual cable with a power meter and convert ($V_m = \sqrt{2 Z P}$ for a matched load), rather than trusting the dial. This is also the standard way to get $V_\pi$ at a specific RF frequency rather than at DC.

Method 3: small-signal quadrature slope

Bias the MZM at quadrature, apply a small tone $V_m \ll V_\pi$, and measure the resulting optical excursion. At quadrature the transfer function's slope is $\pi B/V_\pi$, so the tone produces a peak-to-peak optical swing $\Delta P_\text{pp} = 2\pi B V_m / V_\pi$. Measure the full DC fringe once (maximum minus minimum, which is $2B$), and then

$$V_\pi \;=\; \pi V_m \, \frac{P_\text{max} - P_\text{min}}{\Delta P_\text{pp}} .$$

This is fast and lock-in friendly, but it inherits every uncertainty of holding quadrature and of the smallness assumption; treat it as a monitor during alignment, not the number for the datasheet.

EO bandwidth: Vπ(f) and S21

At high frequency the modulator responds less per volt: electrode microwave loss, velocity mismatch between the RF and optical waves, and impedance mismatch all dilute the interaction, so $V_\pi(f)$ rises with frequency. The standard measurement is electro-optic $S_{21}$: a vector network analyzer drives the modulator, a photodiode of known, calibrated frequency response detects the light, and the VNA reads the round trip. Divide out the photodiode's own response (and cables and any amplifier, from their own calibrations) and what remains is the modulator's EO response.

One convention trap accounts for half the disagreements between labs: the VNA reports electrical power, which goes as the square of the optical modulation depth. A 3 dB drop in optical response therefore appears as a 6 dB drop on the VNA. Datasheet "3 dB EO bandwidth" means the optical convention, i.e. the frequency where the VNA trace has fallen 6 dB (with detector response removed). State which convention you used; better, state both numbers.

The DC-to-RF connection is worth closing: $V_\pi(f) = V_{\pi,\text{DC}} / |m(f)|$ where $m(f)$ is the normalized EO response. A modulator with $V_{\pi,\text{DC}} = 1.7\,\text{V}$ and a 3 dB (optical) bandwidth of 60 GHz needs $\sqrt{2} \times 1.7 = 2.4\,\text{V}$ per $\pi$ at 60 GHz.

Worked example

A 15 mm x-cut thin-film lithium niobate MZM, expected $V_\pi L \approx 2.6\,\text{V·cm}$, so $V_\pi \approx 1.7\,\text{V}$. Sweep $-4$ to $+4\,\text{V}$ in 100 mV steps at one point per 100 ms; the fit returns $V_\pi = 1.73\,\text{V}$, extinction 27 dB, $\phi_b = 0.4\,\text{rad}$. Cross-check with the Bessel method at 1 GHz: carrier null at 12.4 dBm delivered into 50 Ω, so $V_m = \sqrt{2 \times 50 \times 0.0174} = 1.32\,\text{V}$ and $V_\pi = \pi \times 1.32/2.405 = 1.73\,\text{V}$. Agreement between an all-optical-power method and an all-spectrum method is the sanity check that the RF calibration is right.

Common errors

Handoff

Paste any two-column bias sweep (V, transmission in dB or linear) into the Vπ Fitter and it will fit the transfer function, report $V_\pi$, extinction ratio, and bias point, and show the residuals. The physics background lives in the half-wave voltage glossary entry.

References: Wooten et al., "A review of lithium niobate modulators for fiber-optic communications systems," IEEE J. Sel. Top. Quantum Electron. 6, 69 (2000); Chrostowski & Hochberg, Silicon Photonics Design (2015), modulator characterization chapter; Wang et al., Nature 562, 101 (2018).