Borrowed a Coherent IQ Modulation Chip: A Hands-On Walkthrough

I spent two days trying out Xili Optoelectronics' coherent IQ modulation chip. A photonics module company I'm invested in needed to choose a modulator, the supplier sent over an evaluation board, and I ran through it myself. Here are the spots where beginners tend to get stuck.

First, let me clarify a few terms. Coherence looks at the phase of light — like upgrading from a black-and-white photo to being able to see shades. IQ modulation splits information into two paths: one called I (in-phase), one called Q (quadrature). The two carriers are 90 degrees out of phase, like two perpendicular ribbons woven together, so you can pack more information into the same bandwidth. The constellation diagram is those dots on the screen — one dot represents one symbol, and the tighter the dots cluster, the cleaner the signal. Bias is the operating point you set for the chip, similar to where you tune a radio station. An evaluation board is the chip soldered onto a small board with power and RF ports, so you can test it easily.

Preparation and power-up

1. Gather everything first. Evaluation board, adjustable regulated power supply, arbitrary waveform generator (to produce the I and Q baseband signals), a sampling oscilloscope or coherent receiver that can show constellation diagrams, a few RF cables, two attenuators. Read the manual once first, focusing on power-up sequence and bias voltage range.

2. Wiring. Connect power ground first, then power up. After powering up, don't rush to add signals — first check whether the current falls within the range marked in the manual. In my testing, the no-load current was slightly higher than nominal, which is normal, no need to panic.

3. Adjust bias. The IQ modulator has several operating points — one each for I, Q, and phase — all of which need to be tuned to the quadrature position. The easiest way is to use the chip's companion bias control circuit to let it sweep on its own. I swept manually once: the output power curve has two valleys and one peak. Find them, then add signals.

Adding signals, and the pitfalls I hit

Start with small baseband amplitude, then increase. My first round I opened it too much, the driver saturated directly, and the constellation diagram turned into a blur — I thought the chip was broken. Lowering it fixed it.

Connect the QPSK signal, and four dots should appear on the screen. At this point check two things: whether the dots are scattered, and whether there are extra mirror dots. Mirror dots leak out when the I and Q amplitudes or delays aren't balanced — adjusting gain and delay lines can suppress them.

Then there's bias drift.

Bias offset directly degrades SNR and EVM, with temperature and aging being the two main sources.

I confirmed this in testing — the dots I just tuned start scattering after half an hour, and it's more obvious once temperature rises. Later I switched to closed-loop bias control, using a pilot tone plus coherent detection to monitor it. I hear the convergence precision can reach the ±0.03 Vπ level, which manual tuning can't achieve.

The last pitfall is phase synchronization. Coherent reception requires the local oscillator light and signal light to be in phase — the local oscillator is the reference beam at the receiver end. Beginners get stuck here most easily. My crude method was to first not split paths, but split two paths from the same laser for self-coherence. Once that worked, I switched to an independent local oscillator with carrier recovery.

The chip itself isn't hard to use — what's hard is the surrounding ring of bias, delay, and phase. Get any one of the three wrong and the constellation diagram won't give you face. Whether this ring can be built into the solution matters far more than the chip's nominal specs looking good, and it's the first thing I ask about when looking at teams like this.

Next step: simulate IQ modulation and demodulation in Python and NumPy. Once you have a feel for the constellation diagram and image-frequency suppression, then go to hardware — it'll save you half the debugging time. After getting QPSK working, push toward 16QAM.