Measured noise figure vs simulated

Hi all,
I am trying to understand what is going on in my balanced FET design. The design is optimized for minimum noise figure and uses lange couplers on the input and output. Looking at the simulation of each arm alone shows a noise figure of 0.7dB. If I include the couplers and look at the balanced block the NF jumps to ~5dB, but in real life the NF is less than 3dB.

So my question is: Why such a huge difference? I substituted ideal 3dB couplers in the simulation and still that gets me from 7db to 5dB.

Thanks for the help

I will be happy with the better measured result than with your calculated estimation.
There are several points to mention with FETs and symmetrical low-noise amplifiers at RF and microwave.
FETs are nice devices in which the internal noise sources ARE correlated, a difference from bipolar transistors. Using this fact, one can couple one internal source against another, and create a COLFET- a device generating (really!) a lower than ambient noise temperature to the input. Some FET amplifiers can be modified to do this. Maybe your balanced LNA allows one side to radiate a "cool" noise into the output to the other.
Measuring low noise amplifiers often require to consider important parameters like noise bandwidth, mainly if the following receiver includes a down-converter mixer. Then if you use a RF band-pass filter before the mixer to cut off one of the RF sidebands, then the real receiver noise figure must be ~3 dB higher than that measured without rejecting one of the side bands.
In measuring protocol, we distinguish DSB (double sideband) and SSB (single-sideband) noise figures if a down converter is included. Typically, NF(ssb) = NF(dsb) +3 dB.

In a balanced LNA the signal is splitted into two path loosing 3 dB (ideal case) power then the noise figure will increase 3 dB as well. After amplification the signal is combined again. But while signal is correlated, the noise is not (by definition). This means the signal will combine as sum of voltages and noise will combine as sum of powers. Always in the ideal case you will re-gain the 3 dB you loose at the input stage in both gain and noise figure. So if the two stage have the same gain and NF, the total noise figure you have to expect should be the noise figure calculated using Friis formula on one of the two arms taking into account the input splitter losses (in excess of 3 dB) and output combiner losses (in excess of 3 dB).
Imperfect phase and gain balance at input splitter, amplifiers and output splitter will worse the total noise figure.
It could be, for some reason, your simulator always combine the output as sum of powers then accountig directly the input 3 dB losses.

You both have a lot of useful info here, thank you. The unit is not a part of a downconverter, but that info will def. be helpful down the road. I have fixed my simulation to show much more realistic NF and (as it should be) the simulation now shows a NF ~1/2 dB better than what I get in the lab. Truth be told, I messed up the 4 port coupler connections, but at least now I have a much better insight into the noise of a balanced amp.