A noise measurement solution for the poor (2023)

The setup I made is shown below. The bottom two boxes are my tube noise sources, and the top box and spectrum analyzer make up my Poor People's Noise Figure Measurement Solution.

Click here for a higher resolution image...

Many of my fellow ham builders have an inexpensive spectrum analyzer like the RIGOL DSA815, but don't have a noise figure meter. The noise floor of the DSA815 or other inexpensive spectrum analyzer is too high to allow useful noise measurements, even with the internal preamplifier turned on. Also, the DSA815 does not have a noise source enable port, nor does it have a noise figure measurement application. So what can be done?

If you search the internet you will find all kinds of projects PANFI (automatic precision noise figure meter) or CANFI (the same in cheap instead of precise).oYou could go for an old HP 8970B noise figure meter like this one, which goes from 10MHz up to 2GHz depending on the model and for around €1-2K:

PANFIs and CANFIs generally work with SDRs as receivers, which notoriously suffer from many wrong responses ("birdies") and low accuracy. Even a cheap spectrum analyzer with a proper preamp can beat that. The HP 8970B solution works fine, but it only measures the noise figure for one frequency at a time. What we want is a graph and/or table of noise figure versus frequency.

My idea was to add remote control hardware and software to allow accurate measurement of the noise figure, although the DSA815 cannot do this alone.

• To measure noise with reasonable accuracy, we need a very low noise preamplifier. If we limit ourselves to 200MHz, the INA02186 preamp would be perfect (NF 2.3dB, gain 30dB, perfect flatness). A preamp is a must, but it's cheap and easy to make.
• We need a noise source with a level of a few dB and good flatness where we want to measure. The 2D3B tube would be ideal, or we'll have to buy a RFD2305S from G8FEK (or other noise source if the frequency is higher). If it is not truly flat, it should be calibrated to at least 0.1-0.2 dB and the calibration points documented. Any noise source of a few dB can be used for the initial test.
• We need a computer controllable box that enables or disables a noise source and can read a table (or point value) of ENR values ​​versus frequency. This box must be SCPI controllable from a host computer via USB.
• We need some software to control the overhead box and the spectrum analyzer.

The software should work as follows:

• Using a web-based user interface, we input start and stop frequencies, bandwidth, preamp data, expected DUT gain, and noise source level (or calibration points).
• First we connect the noise source directly to the low noise amplifier.
• We set the frequency range, bandwidth, and reference level of the spectrum analyzer. The preamp gain should be used as a reference level offset.
• We send a command to the noise controller to switch to the "cold" side.
• We perform a sweep (average) and measure a reference trace (plus a screenshot). Now we see the total noise level of the analyzerPlusthe LNA.
• We then send a command to the noise controller to turn on the noise source.
  We perform another averaging sweep and measure the hot reference trace.
• Now we insert the DUT that we want to measure between the noise source and the LNA and repeat the above process. We now have two tracks for the DUT.
• On the PC side, we now need to do some math to calculate a noise figure curve. Then we show a graph with the results (DUT gain and noise figure).

The next step was prototyping. I made a "mini box" containing all the necessary RF components, minus the control computer and power supply. Does it look like this:

Click here for a higher resolution image...

On the top left we have a transistor noise source (prototype), below we see the LNA with the INA02186. On the right we have two transfer switches that lead to the device under test (on the right outside the box) and to a feed-through connection (inside the box). At the top we have connections for the power supply (15 V, approx. 200 mA) and for the transfer switches.

The circuit diagrams look like this:

We have four possible states:

• CALIBRATE (no DUT), with noise OFF
• CALIBRATE (no DUT), with noise ON
• MEASUREMENT (with DUT), with noise OFF
• MEASUREMENT (with DUT), with noise ON

In the CALIBRATE position we can determine the noise figure of our measurement setup if the ENR of the noise source is known.

In MEASURE we use the AND method to calculate the noise figure of the DUT.

If we plot the four curves on a spectrum analyzer (Keysight CXA is used at the moment because it has 4 traces) we get something like this (just an example):

The yellow line is CAL OFF, the blue line is CAL ON, the pink line is MEASURE OFF, and the green line is MEASURE ON. The CAL curves are equidistant as they should be, which means that the noise source (2D3B is used here) is flat and the preamp gain is also flat. The MEASURE curves show that since the distance between ON and OFF is about the same as in CAL mode, the gain drops rapidly (-3dB at around 20 MHz) and the NF should be quite small. A slight decrease in the distance between the MEASUREMENT curves can be seen at higher frequencies, indicating a moderate increase in the noise figure of the device under test.

The rest is arithmetic and we have a noise figure curve from 100kHz to 100MHz (*)

On the RIGOL DSA815TG we only have three curves available, so we have to do this one by one (or perhaps all the CAL curves together and then all the MSR curves).

What is missing now is the USB controlled box plus some SCPI programming.

It took some time, but now it's done. I called it "Noise Measurement Box" and it contains all the necessary hardware and is SCPI controllable. It's built into a 19″ frame as usual, with the front panel looking like this:

Click here for a higher resolution image...

On the left we have four switches that control the instrument:

• LOC/USB toggles between local mode and USB computer control.
• CAL/MSR toggles between a direct connection (A ports) and a connection to the DUT (B ports). Both connections must use the same cable types for the errors to clear.
• COLD/HOT turns the noise signal on and off. The noise power is "factory set" here to 7 dB ENR (i.e. 5 mA anode current).
• INT/EXT continuously triggers the noise input or waits for an external trigger signal (such as other commercial noise sources). The BNC input for this is on the back of the device.

In the middle we see the CAL and MSR ports, inputs and outputs. An LCD display shows the current status, either local or remote.

Further to the right, we see another N-type connector leading to the spectrum analyzer, as well as some LEDs showing the current settings (useful if you're not sitting close enough to decipher the small LCD screen).

The inside looks like this (USB cable not installed yet):

Click here for a higher resolution image...

From top to bottom, left to right, we see the power switch and the power filter. Above is the power supply for the tube anode voltage (120V) and the heater controller (9V). To the right of the supply you can see the anode current regulator, as well as the tube itself, which is packed in a Schubert box for shielding. The generated noise exits on a semi-rigid SMA cable on the far right.

One row down we have a 15V power supply for the amplifier/switch box and computer board on the right, with many wires going to the front panel switches and LEDs.

Downstairs is another tin box with the HF relays and the INA02186 low noise MMIC amplifier (gain approx. 30 dB, NF approx. 2.3 at 200 MHz). Noise enters from the left, on the top cover we see the outputs to the CAL and MSR N-type connectors on the front panel, and the output to the spectrum analyzer is on the far right and also leads to a front panel N-connector .

Some basic measurements

The noise level was checked on a Keysight N9000A CXA and found to be around 7dB. First I tried a COLD source, an external INA02186 amplifier (battery operated) and got a display like this (AVG 10, preamp int. ON, BW 10 kHz, 0.2 dB/div):

That's -101.3 +/- 0.3 dB. And then HOT, with the same scale:

We have 94.9 +/- 0.3 dB I would say. That seems fine to me. I suspect that the relays consume some of the flatness that a "pure" noise source can offer with the same tube.

The adaptation at the input port of the DUT is also interesting. I tested this on a Keysight E5071C and got the following results (on COLD first):

Better than 30dB up to 70MHz, Better than 25dB up to 175MHz, Better than 20dB up to 300MHz. Ok, now the HOT match:

Not so nice, but reasonable (guess the length of the cables!). Match better than 30dB up to about 50MHz, better than 25dB up to 150MHz and better than 18dB up to 270MHz.

Measurements of the device under test

Next, I performed noise figure measurements on a Norton BFY90 amplifier using the Keysight N9000A CXA integrated noise figure application as a reference. The noise source came from my noise meter box in manual mode, the rest of the settings didn't change. The NF plot looked like this:

That's a pretty constant 4dB (plausible, see the BFY90 datasheet), with a very plausible gain of 7.6dB. Now let's look at the screenshots of the humble RIGOL DSA815TG and add some calculations. The cold comes here:

Rotonda -91.5dBm, what alittle birdaround 28 MHz. Now it's hot:

That's softer -87dBm. Now we need to extract the tracking data and do averages and bins just like the Keysight app does. If we create 10 MHz wide containers, we get a table like this:

Click here to view the full EXCEL file of the footprints...

We are apparently within 0.2-0.3 dB of the CXA's reference result. We do have a peak here (which is probably the peak around 28 MHz on the RIGOL cold image), the rest is very good for a cheap spectrum analyzer that costs a fraction of the software AF option only on the Keysight N9000A (RIGOL DSA815TG is around €1200, Keysight N9000A 8.5GHz with internal preamp, NF app and some other devices are over €30K even without tracking generator). The RIGOL solution/noise measurement box is definitely good enough for amateur NF measurements up to about 200 MHz.

Software Considerations

Of course, you could do all the measurements manually. Set the LOC/USB switch to local, connect your CAL and DUT wiring, connect the spectrum analyzer and record the 4 curves for MSR/CAL and COLD/HOT, and then you can use the Y factor and Calculate DUT NF. As simple as that (*)

But the box can do even more. First, you can use this as part of your spectrum analyzer's LF measurement application (if you have one, only the best analyzers have that feature). You need to connect the noise source enable wire to the rear BNC connector, set the correct noise source setup time (>100ms please, we are using mechanical relays here), enter a spot ENR of 7dB, set the input as external and go. I use this method to verify that the result of manual measurements and the homemade solution (see right below) are correct.

Second, if you don't have a noise figure application built into your spectrum analyzer (...remember, we originally started as a "noise figure measurement solution for the poor"), you can use the noise figure measurement box integrated noise. software. Controlled by a PC-SCPI software layer (I wrote it myself, not a big deal), the instrument provides the following API (including SCPI commands):

• To connect
• Identify (*IDN?)
• Separate
• Reset (*RST)
• Clear Error (*CLS)
• Turn the amplifier on and off (:INSTR:AMP:SET ON | OFF)
• Set CAL or MSR (:INSTR:MODE:SET CAL | MSR)
• Set the noise source to COLD or HOT (INSTR:TEMP:SET COLD | HOT)
• Get ENR Table (:INSTR:ENRTAB?)

I'm using another SCPI library (also written by myself) to control a standard spectrum analyzer (like a RIGOL DSA815TG if you like). Using these two pieces of software allows for a solution where everything is done automatically and what you output is a table of measured noise figures in the specified frequency range, plotted on a graph created by the famous GnuPlot package.

The user interface is a web form. Enter all the measurement parameters, press run and wait. The software does the rest.

Results

Below are some example output for a Quad J310 JFET Amplifier and a BFY90 Norton Amplifier.

On a CXA, the noise figure is very similar (note that the CXA image goes to 100 MHz instead of 50 MHz). The gain is gone a bit more. As mentioned, this amp's match is quite poor, which is probably the reason for this error.

Now a Norton BFY90 amp:

I verified these results on a Keysight N9000A CXA with the Noise Figure app using a 2D3B noise source. For the BFY90 amp it looks like this (only up to 100 MHz, sorry):

I'd say the homemade solution is only a few tenths of a dB different from what the CXA measures.

Finally, a good test of your machines math: try a 10dB attenuator! Here we expect a gain of exactly -10dB and a noise figure of exactly 10dB as well. It's clear that the signal levels are getting very low now, making the graphics stronger, but it's a good test nonetheless. Bandwidth increased to 300kHz (or could have gone for more rounds of averaging).

Noise aside, NF (estimated) is around 9.7 and gain is around -10. seems good !

Just to be safe, the software also saves all the screenshots of the CAL/MSR/COLD/HOT spectra. This is to document that no birdie invalidated the measurements. Below is a correct example (BFY90, MSR/HOT, 10-100 MHz):

Some tips for successful measurements

While the box does everything automaticallyis it soSome points that you could consider to make better measurements:

minimize drift. Warm up your device to avoid temperature changes during measurement. The Keysight device calibrates itself, the DSA815TG is automatically calibrated by my software before taking a measurement. If the DUT changes the gain or noise figure, it would be wise not to let my software control the power supply, but to apply power from the DUT half an hour before the measurement. Commercial devices further minimize drift by cycling through the CAL/MSR/COLD/HOT multiple times per measurement point, which introduces quite a bit of vibration into the relay. If your DUT is just warming up, this will compensate for inaccuracies at a specific frequency, but not at another frequency measured some time later. In the end, heating is always smarter, and then you can do it all with just 4 relay settings.

Rule out external RFI sources. Keep LED lighting, WiFi hubs, cell phones, wall warts,... away from your facility. My software photographs each spectrum analyzer measurement condition (CAL/MSR, COLD/HOT) so you can verify the absence of birdies (see the Keysight CXA image above with 4 of the traces, this is what it should look like) .

Pay attention to inaccuracies and error limits. Method Y has a known problem with devices with low gain, low noise figure, and/or highly uneven devices. You can find a link to an uncertainty calculator here:

Click here for a Keysight article explaining the uncertainties of the Y-factor method...

refinements

A few things could be improved based on the current solution:

We could include pre-DUT and post-DUT loss in our calculation. This would mean measuring the loss from the noise source to the DUT via the switch box plus the wiring from the DUT and the loss from the output of the DUT plus the wire to the input of the LNA. By using high quality cabling (semi-rigid, RG213, ...) and given the fairly low frequencies of only up to 200 MHz, I expect the attenuation values ​​to stay in the low tenths of a dB range, so the effect is unlikely to be significant. At least parts of it are also calibrated.

We could introduce a precise thermometer (based on Pt)for cold termination and/or the DUT to nullify the effects of ambient temperature changes. The good news is that there is definitely no need to compensate for the noise source of the tube, since it runs at about 100°C and relies on Schottky rather than thermal noise. Just to estimate the effect of a 10°C temperature change, it would be 0.16 dB, probably less than other sources of error.

Uncertainty Calculator Utility. Like the greats, we could create a web-based uncertainty calculator that calculates the error bounds for our measured NF values. This means that all the S-parameters of the noise source, the DUT, ... must be measured beforehand and then entered into the calculator.

Remote Noise Heads and Preamps. To eliminate wiring and switching losses before and after the DUT, we could make a "noise measurement box 2" where the noise source and LNA are external and the box only contains the power supply for them. If we use a 2D2S as a tube with 7dB ENR (@ 5mA anode current), this would be quite a bulky block on the DUT side. The LNA unit would be quite small. I expect an effect of the order of 0.1 - 0.2dB maximum.

Conclution

I am very satisfied with the results obtained and the good agreement with commercial solutions such as the Keysight N9000A CXA NF measurement software. Given the target audience (hams and builders), I'll stop at this point, wait for feedback and enjoy what I have.

(*) I was warned that every formula in a newspaper divides the number of potential readers by a factor of two. Having said that, I think a brief description of the formulas used is in order. So my solution proceeds in the following steps:

First we read the spectrum analyzer's trace tables, which give us the detected power for a range of frequencies (for example, a RIGOL has 601 points from start to finish).

We then create a series of bins (measuring points) and average the power levels (not dB) across them. My examples here use 40 containers. Smaller bins result in higher frequency resolution, but obviously show more jitter. Between 10 and 40 containers make sense. We now have average power levels for CAL/MSR/COLD and HOT for all the center frequencies of the bins we defined.

Next we calculate theSystem noise mappingwithout the DUT (which is the spectrum analyzer plus the preamplifier). All of the following calculations use factors and not dB.

ThatSystemThe Y factor is calculated by

YSystem= norteCandidate/ norteCalOff

give, using the defined ENR of our noise source (here a factor of 5)

FSystem=ENR/(YSystem– 1)

We are done in a few dB. Now let's do the same with the DUT instead. We obtain

YTotal= nortelady in/ norteMsrOff

but

FTotal=ENR/(YTotal– 1)

Now we can also calculate the gain of the DUT using

GRAMSI've got= (Nlady in- norteMsrOff) / (NCandidate- norteCalOff)

If we apply the Friis formula to add noise in a concatenated system, we have

FTotal= FI've got+ (FSystem– 1) / GI've got

if we solve for FDut we get

FI've got= FTotal- (FSystem– 1) / GI've got

and we're done. Taking the logarithm gives

NFI've got= 10*log10(FI've got)

and this is plotted on the chart along with the gain values.