I am trying to measure Noise figure with Spectrum Analyzer 9 to be specific MXA N9020A) using Gain method.

I used the following formula for Noise Figure calculation:

NF= N1 - (G1+G2) - ktb + 1.7dB

N1 - measured Noise power with DUT and driver amp (-119.44dB/Hz)

G1 - DUT Gain (21.82dB)

G2 - Driver amp gain (29.4dB with noise figure around 0.7dB)

KTB - for room temperature -174dBm/Hz

1.7dB - sum of correction factor for bandwidth, log amp and detector correction noise power.

I use **Noise Marker** to get 1Hz power bandwidth. At 1GHz Span 4 MHz, 4 MHz RBW, 1Hz VBW.

So my calculation is as follow:

NF= -119.44 dBm/Hz - (21.82dB + 29.4dB) - 174 dBm/Hz +1.7 dB = 5.04dB

The the DUT measured with Agilent Noise Figure Analyzer N8975A show Noise Figure of 2.6dB.

So why I am getting such big error?

From What I deduct is the the KTB is not the correct value because the Spectrum Analyzer have some offset power in the measurement that is unknown.

I find some solution.

It look like I do not need KTB. I just need to know what the spectrum analyzer read when you apply kTb power, like thermal noise of termination.

1. I measure thermal power of the termination on spectrum analyzer when it is boosted by driver amp. (N2 = -142.16 dBm/Hz)

3. I connected my DUT with terminated input and measure output power thru boost amp on spectrum analyzer (N1 same as above = -119.44 dBm/Hz)

4. I calculate Noise Figure as follow:

NF= N1 - N2 - G1+1.7 dB =

-119.44 dBm/Hz -(-142.16 dBm/Hz) -21.82dB + 1.7 dB = 2.6 dB

Look like my second approach give me same reading as I got from NFA.

So what exactly I am missing in the first formula? Does my second formula is correct?

The only indication that I find is the the Spectrum Analyzer receiver mixer adding own noise power (in old HP application note 150-9 it around 6dB) , that probably can be anywhere from 5 to 15 dB base on mixer conversionloss.

I see that in Y-factor measurement you measure the Spectrum Analyzer noise power of setup with Noise Source off state ( same as attaching termination) during calibration.

When you see the description of Gain method nothing is mention about that.

Any comment welcome.

mirek

You might be a little surprised, but noise figure measurement can be a little complicated. What you call the "Gain Method" I would call the cold source method.

You first measure the gain of the system, then you measure the noise power out of the system. This comes from the definition of noise figure:

and putting it all together you get

So converting to dB (all the formulas above are linear, in watts) you have the Noise out (dBm/Hz)-gain(dB)-kTB. 30 or 40 years ago there might have been a factor for BW and detector (basically that is the offset for different between the average of a log and log of an average) but anything SA with a processor doesn't need it.

The trouble is you are counting on the SA to have accurate noise power measurements. It can be off by as much as 1-2 dB unless you do a noise power calibration. The error is greater above 3.6 GHz as now you are also going through a yig-tuned filter which has more error than the low frequency path.

If you use a hot/cold source, the error due to SA gain is removed by using the noise source to calibrate the noise power measurement error of the SA. Now the limit becomes the noise source ENR uncertainty.

Finally, if your input load is not ideal, you will get additional errors. By the way , how do you get the gain of your amplifiers? And you must know that mismatch between the two amplifiers will cause the combined S21 to not be the same as the sum (in dB) of the S21 of each (you have that nasty 1/(1-S22A*S11B) error ).