After building a noise source, how can the ENR of the new born baby be measured reliably ? There is not just one answer to this question, and, especially at higher levels of accuracy (0.2-0.5dB) the whole story gets a bit tricky, too.
The approach I am using is to compare the output of a calibrated noise source to the new noise sources I have built. The comparison is done using the noise figure application software of my spectrum analyzer. For measurements at the low end of the spectrum (1 – 100MHz), G8FEK made a special noise source for me that has calibration data from 1 to 10MHz in 1MHz steps and from 10 to 100MHz in 10MHz steps. This source was used for all measurements below.
To make an easy start, I restrict myself for now to the range of 10 to 100MHz. In this range, the characteristics of noise sources, preamplifiers, the instruments, … are available and well documented. Below 10MHz, most standard noise sources have no calibration data available, the external preamps have no documented gain and noise figure values, and the instruments themselves show much worse characteristics than at higher frequencies. I will discuss measurements below 10MHz in a separate chapter.
Spectrum Analyzer Characteristics
A not too expensive spectrum analyzer (e.g. RIGOL, Keysight CXA) has a noise figure of 25dB and above, depending on frequency (I will elaborate on that later). You may approximately derive this from the difference between the (internal preamp off) DANL values and the -174dBm cold source noise level. A 25dB+ analyzer noise level is completely useless for noise figure measurements.
Top-end analyzers (Keysight PXA, Rohde & Schwarz FSW) have noise figures of ca. 20dB. Even this is way too much to be useful. The key idea to improve the situation is to add a very low noise, high gain external preamplifier at the input side of the analyzer. This works because the noise figure of a chain of amplifiers (or instruments) is defined by the Friis formula
F is the noise factor here, the noise figure is the same expressed in dB (NF=10 * log (F)).
G is the gain of the amplifier stages (in linear terms). If we assume that the CXA has a noise figure of 30dB (i.e., a noise factor of 1000), and we use an external preamplifier like the U7227A from Keysight (> 16dB gain, 5.5dB specified maximum noise figure, varying a bit over the 10MHz to 4GHz range), then the total noise factor will be
or, with our example
instead of the 30dB noise figure we had without the external preamp. The CXA can also be ordered with an internal preamplifier (20dB gain, ca. 15dB noise figure). If we add this to our signal chain, we now have
so we have
compared to 7.5dB without internal preamp, so there is a small further improvement. The typical values for the external preamp noise figure are a bit better than our computations, so we can expect the total noise figure to be a bit less, too (ca. 5dB).
All of the above computations are valid for the “sweet spots” of the spectrum analyzer, internal and external preamps only (above 10MHz, better 100MHz). I will discuss later what effects this will have.
The Noise Figure Measurement Application
This is a piece of spectrum analyzer software that measures the noise figure of devices using a calibrated noise source and the Y method already discussed.
The setup works like that:
- The noise figure application is started, and the frequency range is set.
- The data for the calibrated noise source is entered (this is a table of frequencies and ENR values) and saved under a user-defined name.
- Internal and external preamps are activated.
- The calibrated noise source is connected directly to the external preamplifier and to the noise source control output of the spectrum analyzer.
- The setup time of the noise source is entered.
- A calibration is performed.
- It make sense to enable averaging to get a more stable display
- The DUT is now inserted between the calibrated noise source and the external preamplifier. Power is applied to the DUT.
- The application now shows noise figure and gain of the DUT over frequency in a diagram or table.
The picture below shows the noise figure and gain of a “thru” piece after calibration.
The procedure above works nicely for all kinds of amplifers. To estimate its uncertainties, there is a web-based noise figure measurement uncertainty calculator from Keysight. With the setup above, and above 10MHz, the usual uncertainties for reasonably well matched DUTs are in the range of 0.25dB.
How to Measure the ENR of a Self-Built Noise Source
The idea is simple; if we remove the calibrated noise source and replace it with a homebrew noise source having the same ENR values, the data on the spectrum analyzer screen should not change. With the adjustble noise source, the procedure runs like this:
- Perform the steps above, but leave the calibrated noise source and the external preamplifier connected after the calibration. The DUT is now a “thru” with 0dB noise figure and a gain of 0dB.
- Save the display for later use (we want to compare). Switch to another trace.
- Remove the calibrated noise source and attach the external preamplifier directly to the new noise source to be tested.
- Connect the new noise source to the spectrum analyzers noise source activation output.
- Adjust the noise level so it is exactly what your calibrate noise source has (of course, you have to pick a frequency where they should be exactly equal).
- Show the old and new trace in an overlay to see the deviations of the calibrated and new noise sources.
- If neccessary, fine-tune the noise level of your new noise source for best matching.
The screen below shows the calibrated reference (thin yellow line) and then, as an example, the LED noise source in blue in a range from 10 to 100MHz. Up to 50MHz, differences are in the range of 0.1dB, and up to 100MHz we stay within 0.3dB, so that should be perfectly acceptable. If your DUT noise source and your calibrated noise source have different noise levels over frequency, the two curves normally do not coincide, obviously. If the difference is explained by your calibrated noise sources ENR table, everything is still fine. Remember that the ENR of calibrated commercial diode noise sources is not very flat. If your DUT noise source is flat, the graphs shows the difference.
Here we have used the RFD2305S, with an ENR of 5.7dB at 50MHz. At 10MHz, the RFD2305S has an ENR of 5.9dB. The graph shows that the DUT is 0.15dB below that, meaning that its ENR is now ca. 5.75dB. At the 100MHz point the RFD2305S has an ENR of 5.7dB again, but now we have a real deviation of ca. 0.3dB. We may conclude that our DUT noise source is close to the calibrated noise source to 0.1dB below 50MHz, but has a higher ENR above that, up to +0.3dB.
After this exercise, you have successfully calibrated your new noise source against a calibrated source. Done !!
Noise figure measurements are full of traps and pitfalls. The more obvious ones are:
- Stray coupling of unwanted signals. This includes your mobile phone, power line communications, computer cables like USB, LAN, keyboard or mouse, WLAN and bluetooth, cordless phones, … Put all of this as far off as possible from your measurement setup. In stubborn cases, put everything in a metal box and use coax feedthrus for all signals (recommended by Rohde and Schwarz).
- Lousy cabling. Use double-shielded, high quality coax only. RG58 ? Never for noise figure measurements. SUCOFLEX 100 is OK. Remember that coax losses add to the noise figure of your DUT.
- Adapters. Especially at higher frequencies, these might add unwanted attenuation to your signals. Use as few as possible, and if you must, use high-quality parts only.
- Connectors. Keep them clean, dont use worn-out stuff. Use a torque wrench.
- Observe warmup periods. A minimum of 30 minutes is a must for your spectrum analyzer and external preamplifier (see manual).
- Tolerances in published data. For example, the U7227A external preamplifier has noise figures specified to be below 5dB. Measured values show values around 3.5dB, only going up to 5dB around 10MHz (1/f noise seems to become stronger there).
Measurement Bandwidths and Signal Levels
Especially when using very wideband noise sources, using too high a bandwidth or no low-pass filtering can overdrive the amplifier chain of your spectrum analyzer.
An example: We use a 0-4GHz 7dB ENR noise source to measure a 20dB amplifier DUT with 4GHz bandwidth at 10MHz with 10kHz of RBW. We also use an external preamp with 20dB gain and the internal preamp of the spectrum analyzer with 15dB gain. If the DUT has no noise of its own, then the total signal level for the 4GHz will be
at the pre-mixer attenuator input. For our analysis bandwidth of 10kHz, we would see a level at the screen of
Just to remember, the SFDR of the spectrum analyzer I’m using is ca. 60dB, and the difference between the total signal level and the displayed level is 56dB. This is getting a bit close, I would say. In more extreme cases, with even higher bandwidths, a higher ENR noise source or more DUT gain, you risk overdriving your mixer. Some better spectrum analyzer have preselector filters that can be used to avoid this.
Another idea is to insert a lowpass filter before the external preamplifier. If we assume a 200MHz filter here, the level would be down at
This is a safe level, I would say.