AnsweredAssumed Answered

Testing Insertion Loss

Question asked by Synergy2010 on Nov 9, 2010
Latest reply on Nov 12, 2010 by andyowen

Like • Show 0 Likes0
Comment • 4

My company has just obtained the 8753D network analyzer, and I am still new at it.
I am attempting to test for insertion loss.
Test Requirements:
Insertion loss per 100ft
N-type to N-type
Frequency Range: 1GHz
Max Loss: 14dB

Network Analyzer Set up:
Measure: S21 Transmit
Format: Log Mag
Scale Ref: Auto Scale
Center: 1GHz
Span: 500MHz
Marker: 1GHz
AVG: Averaging ON Smoothing ON

Problem: Most cables I test result in a -dB value which I am assuming is LOSS. However, when testing a shorter cable, e.g. 18inches,the result is a positive value(e.g. .2378dB). Why is this ? am I interpreting this incorrectly or am I setting something up wrong?

No one else has this question

Outcomes

4 Replies

am95405 Nov 9, 2010 9:01 AM
Mark Correct
Correct Answer
Hi,

You want to be sure to perform a calibration, a normalization or thru cal at a minimum. A two-port cal would be even better, especially if your cable has poor match.

Also, reduce the IFBW to reduce the trace noise, or use averaging. A positive value could simply be noise, or measurement error, but 0.23 is not that small.

Afsi
Like • Show 0 Likes0
Actions
jvall Nov 9, 2010 9:18 AM
Mark Correct
Correct Answer
Hmmm.... where to begin?
First: in most cases, smoothing is "cheating"...be very careful.
2nd: Slow the sweep down until you see no more changes., See: http://na.tm.agilent.com/pna/files/eld.html
3rd: As was mentioned, make sure you perform a cal first.
Like • Show 0 Likes0
Actions
am95405 Nov 9, 2010 9:31 AM
Mark Correct
Correct Answer
I was going to mention the electrical delay possibility, but then the issue seemed to be with the 18" cable, and I didn't think it was necessary to reduce measurement speed for an 18" length. But it can't hurt to test it. I missed the smoothing/averaging part.
Like • Show 0 Likes0
Actions
andyowen Nov 12, 2010 10:46 PM
Mark Correct
Correct Answer
Hi Synergy2010,

As was mentioned by my colleagues, a full two-port calibration is a great option because it corrects for two major sources of error that a simple "thru response" calibration does not: source and load match. These two terms indicate just how close to 50 ohms the source (Port 1, in your case) and load (Port 2) of the test system is. On most network analyzers, these error terms are small but not negligible and they show up most when measuring a device that is low-loss, such as 18 inches of good cable. (Please note that "calibration" here refers to a sequence of steps prior to the measurement that removes systematic errors. This is different from the yearly calibration that your instrument should undergo.)

Lets say for simplicity that your cable has no loss and has a perfect characteristic impedance of 50 ohms but the source and load match of the network analyzer are 45 ohms. The signal from the source out of Port 1 goes through the cable, unattenuated, and then sees the 45-ohm load match at Port 2. This causes most of the signal's energy to be absorbed at the load (since it's close to 50 ohms) BUT some of it comes back through the 50-ohm cable, sees the 45-ohm source, and some of it re-reflects back in the forward direction. This phenomenon is called a "standing wave" and it causes the measured ratio of incident power wave at Port 1 to the scattered power wave at Port 2 to be different than the true S21 (which assumes no load reflection) depending on the phase of these reflected signals. If you measure the same cable over a broader frequency range (or measure a longer cable while keeping the loss low) you will see "ripple" over frequency because the re-reflections add in or out of phase with the desired signal. For a very low-loss device, this ripple can add in phase and actually show a response greater than unity (0 dB)! Sorry if this kills your idea for a patent.

Besides performing a full two-port calibration, one recommendation is to connect a high-quality attenuator to Port 1 and/or Port 2. This will improve the raw source and load match thus giving you less ripple. The downside is that your dynamic range is decreased: the source power will be decreased on Port 1 and the noise floor of Port 2 will be increased. For a maximum loss of 14 dB, this is no cause for alarm. If the resulting measurement is noisy (changes from sweep to sweep), please lower the IF bandwidth or increase the number of averages.

For further reading on these sources of error, please refer to the following app note. This is the "bible" for those new to vector-error correction (calibration).
http://cp.literature.agilent.com/litweb ... -7709E.pdf
Like • Show 0 Likes0
Actions

Testing Insertion Loss

Outcomes

Related Content

Recommended Content