I/Q optical modulator
Two push-pull Mach–Zehnder modulators nested inside a third, 90° apart: one sets the real part of the optical field, the other the imaginary part.
Open in the canvas →Background
An optical field has an amplitude and a phase, or equivalently two independent components a quarter of a cycle apart: the in-phase part I and the quadrature part Q. A modulator that changes only the intensity uses one of them. Coherent optical links use both, and both polarizations, which is most of how fibre capacity grew from 10 to 100 gigabits per second and beyond on the same glass[1].
The device that writes I and Q is three interferometers. The light is split in two. Each half passes through its own Mach–Zehnder modulator, driven push-pull: a phase +φ/2 in one arm and −φ/2 in the other. The two arms then recombine to a field proportional to sin(φ/2) or cos(φ/2), depending on the port — a real number that goes smoothly from +1 through zero to −1, with no phase rotation along the way. That is what push-pull buys: driving one arm alone gives the same intensity but drags the phase with it, which is chirp[2]. One of the two branches is then delayed by a quarter of a wavelength, and the branches are recombined. The output field is I + iQ, with I and Q set independently by two voltages.
Setting each of them to ±1 gives the four points of QPSK; more drive levels give 16-QAM and denser constellations[2]. In practice the three interferometers are waveguides on one lithium niobate or indium phosphide chip a few centimetres long, and the path lengths that have to be held to a small fraction of a wavelength are held by the chip itself, with three slow bias voltages trimming what is left[3].
One feature is intrinsic rather than a flaw: combining two fields in quadrature on a 50:50 combiner puts half of their power in the port that is not used. An I/Q modulator at full drive passes at most half its input.
What this setup demonstrates
The same device in free space, at 1550 nm: a splitter, two Mach–Zehnder interferometers built from beamsplitters and mirrors, a phase modulator in each of their four arms, a fifth holding the 90° shift, and a combiner. Every path is traced and the fields are added with their phases, so the readings come from the geometry and not from a formula for the device.
Each child interferometer has two outputs. One goes on to the combiner; the other lands on a monitor detector, which is where a real modulator has its monitor photodiode. As shipped, both children are driven by a full 180° (+90° and −90° in their arms): the monitors are dark, I = Q = 1, and the output and the unused port each carry 0.5.
Things to try, in this order.
Change a sign. Swap the two drives of the Q modulator to −90° and +90°. Q is now −1: a different QPSK symbol. Nothing on the screen moves, because the power is I² + Q² and a detector does not see the phase of the light. That is the meaning of quadrature — the two components do not interfere.
Remove the quadrature. Set the 90° shifter to 0° and repeat. Now the two branches do interfere: with both at +1 the unused port takes everything and the output is dark, and flipping the sign of Q moves all of it to the output. The sign that was invisible is now the whole signal, which is also why a real device needs that bias held.
Turn one down. Reduce the I drives towards 0° and its light moves from the output to the I monitor, as sin²(φ/2). With both children undriven nothing leaves the modulator at all: it is biased at its null, as a real one is.
The drives here are static phases. Nothing is modulated in time, so there is no symbol stream, no bandwidth, no drive voltage or Vπ, and no chirp from an imperfect push-pull. The phase modulators are ideal: exactly the phase asked for, at this wavelength, with no loss.
The detectors measure power. There is no coherent receiver — no local oscillator, no 90° hybrid — so the constellation itself is never displayed; what the setup shows is the power relations that follow from it. The interferometers are exactly balanced because they are drawn on a grid: there is no drift, no bias control, no finite extinction from unequal splitting, and none of the waveguide loss of a real chip.
The interference is that of an ideal monochromatic CW source through unitary non-polarizing beamsplitters and fully reflective flat mirrors, the same model as the Mach–Zehnder example.
Related components
References
- K. Kikuchi, “Fundamentals of coherent optical fiber communications,” Journal of Lightwave Technology 34, 157–179 (2016)
- P. J. Winzer, R.-J. Essiambre, “Advanced optical modulation formats,” Proceedings of the IEEE 94, 952–985 (2006)
- C. Wang et al., “Integrated lithium niobate electro-optic modulators operating at CMOS-compatible voltages,” Nature 562, 101–104 (2018)