Auriga Microwave

Published literature and conference sessions are booming with techniques for maximizing efficiency, improving linearity, reducing size, etc. of amplifiers, but often these techniques come at the sacrifice of bandwidth. An ideal solution could realize these performance enhancements without reducing the frequency response. Designing for the widest bandwidth possible requires subtlety.

Start with good active models.


Although foundry transistor models have become much better in recent years, they are made to appeal to the widest audience possible. This “one model fits all” approach works for narrowband applications (i.e., add a few tunable components for the bench test and you’re golden). For wideband applications, you need transistor models that provide accuracy over the entire frequency range of interest. Creating a scalable, wideband behavior model that represents thermal effects (resistive and capacitive), process variation (i.e., supports Monte Carlo), and bias dependency is non-trivial. Before starting a wideband design, always compare the model against measured data. If the model is not accurate over the entire range of interest (with some margin), extract a custom model.

Start with good passive models.


Similarly, passive models have become much better in recent years. Component-level parasitic effects are the most limiting factor in achieving wide bandwidth. Substrate permittivity, mounting orientation, proximity to ground plane, and self-resonance are some of the parasitic effects often overlooked. If models with this level of detail are not available, mount the component in a representative fixture (i.e., use the same substrate as the final product), measure S-parameters, de-embed the fixture, and use the measured results in simulation. For a more versatile solution, have a custom model made (I use the Modelithics scalable model).

Create your own N-section transformer.


A common way of achieving wide bandwidth matching networks is with N-section transformers, where N is usually 2, 3 or 4. The three most popular algorithms are exponential, equal ripple, and maximally flat (listed from widest bandwidth with most ripple to narrowest bandwidth with least ripple). Using these three configurations as a starting point, the Smith Chart Utility in Agilent ADS is a powerful tool for tweaking the section impedances to find the right balance of bandwidth and ripple that works for the application (seek the Microwaves101.com free Excel calculator to get started). To minimize size, meander the RF traces and/or replace transmission lines with lumped components.

Know the system need.


Fact: Amplifiers have less gain, power, and efficiency at higher frequency than lower frequency. To achieve equal performance across a wide frequency band, low-end performance is worsened to match the high end. When designing for extremely wide bandwidths (i.e., multiple octaves or decades), this can amount to a significant degradation at the low end (i.e., -6 dB gain per octave adds up quickly). Is a flat response really necessary? Somewhere in the front-end is an antenna with positive gain per octave. Negative amplifier gain slope over frequency could be a benefit!

Know the limiting factor.


In every system, there is a bandwidth-limiting element. Always know what it is. Often, packaging is the culprit. Watch for practical limitations from feedback circuits and reactive matches. Passive components, like couplers and splitters, usually have well-defined band edges (especially when wideband). In general, any technique using a phase shifter is going to be bandwidth limited. When exploring the widest bandwidth possible, knowing the limiting factor will allow the designer to focus on the greatest area of constraint.

Dr. Nickolas Kingsley, Director of Engineering
Nick manages the engineering team at Auriga Microwave. His research interests include the design, miniaturization, fabrication, packaging and testing of RF MEMS multilayer front ends.

---

 

As published in Microwave Journal, Expert Tips, Tricks and Techniques; July 12, 2012.
Reprinted with permission.
The article may be viewed in its entirety here.