Network analyzer measurements

Updated March 14, 2011

Click here to go to our main page on microwave measurements

Click here to go to our main page on S-parameters

Click here to learn how to extract effective dielectric from sample transmission line measurements

Click here to go to our page on reference planes

Click here to learn how (not) to trash a cal kit!

Click here to go to our page on RF probing

Click here to go to our page on "Smoothing is cheating!" It offers an example of using smoothing to pass a VSWR specification. Click here to see how smoothing can be used to improve group delay measurements (this is NOT cheating!)

This site used to have some great app notes on Agilent network analyzers. The site doesn't seem to be maintained these days, but you might find a few nuggets of information. Thanks to Ian for pointing out its slow demise!

Here's a link to Agilent's official PNA page.

Network analyzers fall into two categories. Vector analyzers are capable of measuring complex (magnitude and phase) reflection and transmission; scalar analyzers only measure magnitude.

One of the most essential pieces of TE in the lab is the network analyzer. It can be used to measure impedance, VSWR, loss, gain, isolation, and group delay of any two ports of a multi-port network (don't make us draw a potato with arrows here). The two big guys in network analyzers are Agilent, the 800 pound gorilla once known as Hewlett Packard, and Anritsu once known as Wiltron before they turned Japanese. Can anyone tell us why Anritsu didn't add a Playstation option to their analyzers yet?

Vector network analyzers

A word about acronyms concerning network analyzers... vector network analyzers (VNAs) are often called "ANAs" by old farts, including the Unknown Editor. ANA stands for "automated network analyzer". A long time ago during the Carter administration, the original network analyzer (H-P 8409) was not automated, in the sense that TE error correction was done by hand. Return loss measurements could not exceed the VSWR of the equipment, so you couldn't resolve beyond 20 dB return loss in most cases. Gain and insertion loss and phase were calculated from the subtraction of two measurements (first the through connection, then the DUT connection). It was a bad time to be alive...

Then the first automated network analyzers were introduced. A minicomputer (about equal to a 1000 watt, five dollar calculator) grabbed the vector data from the 8409, and did some fancy manipulations that resulted in automatic error correction and accurate magnitude and phase of the four S-parameters. It was considered magic. The next step was to build the error correction into the test equipment (no external computer) and display the error-corrected measurements in nearly real time (the original HP 8510, circa 1982). Today vector network analyzers are all automated (error correction is built in). And the acronym ANA has stuck.

This type of network analyzer consists of a sweep oscillator (almost always a synthesizer so that measurements will be repeatable), a test set which includes two ports, a control panel, an information display, and an RF cable or two to hook up your DUT. Each port of the test set includes dual directional couplers and a complex ratio measuring device. Other options include a means for bias voltage/current injection, and a computer controller to manipulate and store data. The "classic" vector network analyzer is the Agilent (HP) 8510, shown below. Depending on how much you spend, this analyzer can make measurements from 45 MHz to 110 GHz.

Before you jump into vector network analyzer measurements, you will have to calibrate the network analyzer. There are many types of calibration techniques, and even more types of calibration standards. A typical calibration will move the measurement reference planes to the very ends of the test cables. You will have the choice of calibrating for reflection or transmission only, using either of the two ports or both of them together. For most tasks you will probably calibrate both test ports for reflection and transmission, which will allow you to measure full two-port scattering matrices (S-parameters for your device under test (DUT). This is referred to as "twelve-term error correction". I don't know why there are twelve terms for two ports, I will get back to you on that.

Before you perform a calibration, you should do a little "preflight" check-out of the TE and DUT. Ask yourself:

• What frequency range do you need to measure?
• Does the cal kit, cables and any adapters you need operate over the desired band?
• Are the cables in good condition? (Connect them together and see what the effects of gently bending them have on uncalibrated transmission and reflection parameters).
• Will the cables reach the DUT? (This seems obvious, but I have seen people waste time calibrating only to discover that the test cables are too short to reach both ports of the DUT).

Calibrating a vector network analyzer

The reflection calibration for each port requires three standards, typically: an open circuit, a short circuit, and a matched 50-ohm load (for waveguide calibration, a pair of offset shorts and a load are used. An open in waveguide usually acs closer to a load due to radiation). The matched load can be a "broadband load", meaning that it has very low reflection coefficient over a lot of bandwidth, or a sliding load. Sliding loads are expensive and fragile standard which should only be used if your measurement requires great accuracy (perhaps you want to be able to tell the difference between a 1.01:1 VSWR and a 1.02:1 VSWR). The sliding load recognizes that a "perfectly matched" 50 ohm calibration standard can never exist, but a series of loads with equal mismatch but varying phase can be used to draw a circle around the center of the Smith chart, thereby solving for the perfect load. My advice to you: unless someone takes the time to show you how to use the sliding load properly, NEVER TAKE IT OUT OF THE BOX.

The particular set of cal standards (and test cables) that you use will depend on what frequency band you need to cover. Coaxial calibration kits come in type N, 7 mm, 3.5 mm, 2.92 mm, 2.4 mm, 1.87 mm and 1.0 mm. There are waveguide calibration kits for every waveguide band. Be sure not to exceed the frequency capability of the test set, cables, adapters and calibration kit (see our section on connectors for the frequency limits of different connector types).

Sage Advice: remember that cal kits are expensive, and pieces of the cal kit should NEVER be used as adapters loads in any test set. And always put the little plastic covers onto the calibration pieces, you want to prevent dirt, skin, grease, etc. from degrading the accuracy of future calibrations. We have a separate page on how (not) to trash a cal kit!

To check the validity of your calibration, as well as the general health of the test equipment, you need to look at a few things after you calibrate. If you are doing transmission measurements, check the residual error in a "through" connection (connect the test cables to each other). You should see 0 dB plus or minus 0.05 dB or better. The phase should be very close to 0.0 degrees as well. The return loss of both ports should be at least 40 dB but can be better than 60 dB if you are using good equipment. The transmission and reflection parameters should not vary significantly when you gently bend the test cables, or you have a bad connection. If you see a issue with the calibration you just did, figure out the problem before you perform another calibration, or you will be wasting your time and adding needles wear and tear to the cal kit and test cables.

Some of the options you will be asked by the network analyzer need to be discussed so you will know how to answer them.

Omit isolation: during the calibration, you can "omit isolation". If you are measuring the loss of some test cables and don't expect to see transmission data under -20 dB, go ahead and omit the isolation cal standard. But if you want to see the steep skirts of a filter or the reverse isolation of a multi-stage amplifier, you should perform the isolation step.

Averaging: performing averaging will improve the accuracy of your data, so long as you do it during the calibration as well as the actual measurement. But it will slow down the measurement process noticeably, a consideration if you have a lot of data to collect in limited time.

IF bandwidth: an option on most new network analyzers, reducing IF bandwidth also improves measurement accuracy. Try reducing from 35 kHz down to 500 Hertz.

Smoothing: smoothing is cheating, no ifs ands or buts. Smoothing reduces the "bumpiness" of a frequency response by averaging data across a couple of frequency points and using the result at one frequency. But if you need to cheat to get some data for the boss who is standing behind you, go for it. The only time that smoothing may actually improve measurement error is in group delay mode (note: this is referred to as the "aperture" setting when you are using Anritsu (Wiltron) equipment. The group delay is actually calculated from the slope of the phase angle versus frequency, and the "aperture" allows the user to define how much frequency band top take the slope over.

Auto scale: go ahead and use the auto scale to quickly display visual information on the parameter you are investigating. But when you actually plot the data on a pen plotter or printer or using an Excel spreadsheet, use a scale that makes sense. Like 2 or 5 or 10 dB per division. NOT 3 or 6 dB per division. This is because we live in a base-ten world, not a base-three world. Also, when you are plotting the same type of data for units of the same type that you are measuring, use the SAME scale for all of them so they are easy to compare.

Response calibration: this type of cal offers a shortcut to data, but offers a reduced amount of error-corrected data. A transmission response cal will merely measure the magnitude and phase of the through connection, which will be subtracted from all subsequently measured data. But you won't know anything about the magnitude and phase of the DUTs reflection coefficients. A reflection response cal will provide magnitude-only information for reflection coefficients. If you find yourself using the response cal feature often, you should consider using a scalar analyzer and stop tying up the expensive equipment.

Scalar network analyzers

Scalar network analyzers are so lowly that they don't have an acronym. So you have to say at least "scalar analyzer" or no one will know what you are talking about.

The obvious difference between a vector network analyzer and a scalar analyzer is that the scalar analyzer doesn't give you any phase information. It only detects power.

We are still working on this section, so check back later!!!