In Uncategorized

At Marki, as in much of the microwave industry, we tend to focus on frequency domain measurements. This includes S-parameters (insertion loss, return loss, isolation, rejection), parameters that can be calculated from S-parameters (amplitude balance, phase balance, common mode rejection ratio, directivity), and related parameters (group delay flatness, differential group delay). These are the parameters that are most interesting for applications such as radar, narrow band communications, and electronic warfare.

In the signal integrity industry, the people that design the high speed signaling hardware that powers backplanes on major computing systems powering the internet, among other applications, time domain is king. These engineers generally care about one thing: the data throughput that a system is capable of. Components are evaluated on the basis of whether they can successfully transmit the data. For this reason the primary metrics are eye diagrams, error vector magnitude plots, and bit error ratio measurements.

Each approach has its benefits and drawbacks, and it is important for a well rounded engineer to master both domains. Think of time domain as an aerial view and frequency domain as a microscope. The drawback of frequency domain measurements are that they can be too specific, and you can miss the forest for the trees. You know the 3 dB rolloff is at 7 GHz, and the return loss has a troubling hump at 3 GHz, but will that sink the ship or not?

In time domain measurements, you immediately see the big picture, but if anything is wrong it is difficult to diagnose. For example, a ringing in the eye could be from many sources, and there may be no way to isolate the source without resorting to frequency domain measurements. Here are a few eye distortions, and what the potential sources might be. First, here is what your basic eye diagram looks like, as plotted by microwave office:

basic eye


It is perfectly square and with minimal ringing because I selected many sample points. If you select fewer points, you get an eye that looks more realistic, like this:

undersampled eye


Which is the same way that it looks

Filtered Eye


after you filter it with a filter that looks like this

filter responseexcept with some ringing due to Gibb’s phenomenon, a result of the limited high frequency content in the signal. Ringing is among the most common phenomena experienced in a lab, but it is generally harmless, since it does not introduce noise near the sampling point. In fact, a signal can be filtered very aggressively without affecting the sampling point:

Aggressively filtered eye

as shown in this example with a 500 MHz filter (1/2 the datarate). On the other hand, high pass filtering will cause baseline wander, which shows up in eye diagrams as a split in the different levels of eyebrows that is dependent on the coding scheme of the PRBS or data sequence:

high pass filtered 128 bit eye

high pass filtered 1024 bit eye


In the 128 bit sequence, there are discrete levels that result from each bit sequence. If this bit sequence repeated then it would always look like that. The eyebrow is not necessarily from random noise, as it appears, but from deterministic artifacts of a given bit sequence. This is why coding like 8B/10B and 64B/66B can be used on lower bandwidth hardware: the low frequency components are not distorted because they are not present.

What we’ve covered so far is only half the story. There’s still phase/group delay and timing variations, which are usually much more important than amplitude considerations, but that will have to wait for another post.



Recent Posts

Leave a Comment