July 18, 2016
With many designs moving their way up to higher microwave and even millimeter-wave frequencies, it is important for signal integrity (SI) engineers to remember there is a different uncertainty distribution than at lower frequencies. For subsystem modeling, accurate Electrical-to-Optical (E/O) and Optical-to-Electrical (O/E) characterizations are critical at higher data rates. Careful transfer/calibration processes and a high-stability broadband VNA allow transmission uncertainties < 0.5 dB (1550nm/110GHz or 1310nm/70GHz) and model eye height improvements > 15%. For SI engineers this means shorter design cycles and greater confidence in their products.
A basic component measurement setup for these scenarios, based on a broadband network analyzer, is shown in Figure 1. In an example using this configuration, the optical power used was 0 to 6 dBm and RF power was –10 dBm. The network analyzer had a bandwidth of 70 kHz to 145 GHz, although only subsets of that frequency range were necessary. Data rates of up to 80 Gbps (NRZ and PAM-4) were used in the subsystem models.
Understanding the Measurement Process
The measurement process begins with a reference photodiode characterization using an electro-optic sampling process that introduces specific uncertainty components, including those related to a network analyzer calibration, photodiode mismatch, drift, optical system limitations and bandwidth limitations. Initial uncertainties were combined using covariance data on the Type B elements; subsequent steps used a simpler RSS combination for stochastic terms and a worst-case summation for deterministic terms.
The characterization is used to calibrate a reference modulator via de-embedding and then user photodiodes. Finally, a user modulator can be measured by de-embedding the user photodiode. In each follow-on step, an additional network analyzer measurement is involved that introduces calibration/mismatch and measurement mechanics (including connector repeatability and cable flex), as well as noise floor contributions. Optical power enters into the process via its influence on net noise floor and nonlinear distortion.
A summary of the contributions at two frequencies is shown in Figure 2 although, in absolute terms, all contributions have increased at 110 GHz compared to those at 20 GHz. Measurement mechanics and noise floor contributions have relatively increased. The former suggests the increasing importance of connector and cable care. The latter is mainly from typically increasing conversion loss and points to the value of higher broadband RF power and reduced IF bandwidth. The user calibration and mismatch relative contribution has not changed but have increased in absolute terms, which emphasize the importance of the user calibration and the general trend of increasing device mismatch with frequency.
For RF calibrations, many choices, including coaxial and on-wafer, are available over this bandwidth. Higher quality variations can lead to a direct transmission uncertainty of ~0.3 to 0.5 dB at 110 GHz, while more modest choices may double that. A non-optimal calibration choice could triple the first category in Figure 2 while a sub-optimal measurement hardware choice could increase the second and third category absolute contributions by an order of magnitude.
Possible Characterization/calibration Issues
To see the effect of characterization/calibration issues on modeled bit stream parameters, consider the calculation process. In the analog-like portions of the system, the frequency domain data sets are often cascaded and an impulse response is generated from the net frequency domain data with a Chirp-Z or similar transform. That broadly integrates the data over the frequency sweep, prior to convolution with the data stream. Thus, any highly oscillatory defects are likely to have small impact while a frequency-sloped defect (from a calibration tracking issue, drift, or sometimes repeatability) may have an amplified effect.
In the frequency domain, these two classes of effects are shown in Figure 3. The oscillatory defect may occur from a calibration issue on the network analyzer or a reference plane problem from a mischaracterized adapter. The slope problem could occur from drift on the measurement hardware, repeatability, or a reference plane problem from an un-characterized adapter.
As an example, a 110 GHz system was characterized using the procedure discussed earlier with a highly stable VNA and quasi-optimized calibration steps and run using a 50 Gbps NRZ signal. The resulting eye diagram is shown in Figure 4. If an oscillatory defect of 1 dB peak at high frequencies is introduced, the eye diagram was indistinguishable from that in the figure. If a slope error was introduced of 1.5 dB at the high end, the eye height was reduced by ~10% and the width by ~4%.
For a more physical experiment, the same 110 GHz system was characterized with a simplified process where a heterodyne characterization and phase-modeled process was used, resulting in an increased characterization error on the order of 0.5 dB at 110 GHz. The final user calibration was only a transmission normalization and a 1.85-1mm coaxial adapter used in the measurement was not de-embedded (resulting in both magnitude and phase deviations approaching 0.5 dB and 5°).
The resulting eye diagram is shown in Figure 5. Here the eye height has reduced by ~15% and there are additional baseline distortions partially from the phase distortions that were not linear with frequency. This amount of eye reduction may or may not be critical depending on other parts of the system, but the point is the model result was noticeably distorted by common characterization/calibration issues on the converter components.
To learn more, you can download a new Anritsu white paper entitled Wideband Optical Modulator and Detector Characterization: Uncertainties and the Impact on Eye Diagrams/Time Domain Modeling.