We get a lot of questions about phase noise and noise figure, and it is a topic of interest for Ferenc Marki. The question comes up often, what kind of noise does the mixer add during the conversion process?

Our experience is that the actual noise addition from a passive silicon Schottky mixer is negligible. This is why the noise figure is very nearly the conversion loss: the mixer does not add noise, but it does attenuate the signal by an amount equal to the conversion loss. Frequently when customers have a noise problem in their conversion it is coming from the leaking of LO noise into the IF or RF port.

We can demonstrate this by performing an experiment. We take a 10 MHz low phase noise reference oscillator and measure the phase noise. Then we convert the signal to 5.01 GHz with a noisy 5 GHz LO and measure the phase noise of this output. As you can see the phase noise jumps from -154 dBc/Hz to -117 dBc/Hz at a 10 kHz offset. Finally we downconvert the signal with the same LO (with a trick I’ll explain below) and see that the phase noise has dropped back to -154 dBc/Hz, as though it had never been converted at all.

*The Trick:* Why would the mixer actually remove phase noise from the upconverted signal? Because the LO cancels it’s own phase noise out on the downconversion. As long as the LO is in phase with itself on the up and downconversion, the phase noise will cancel. Here is the same experiment done with a non-coherent LO for the downconversion step. Note that the phase noise actually increases on the downconversion as a result of the addition of the non-coherent LO phase noise.

The other important part of this is in the RF. The RF needs to *remain* coherent with the LO during it’s travels between the two mixers. If there’s nothing but a cable in between then we’re fine. If there’s something with a non-linear phase delay (a group delay) such as a filter or narrowband amplifier then we will see the phase noise on the output, as the LO doesn’t have a chance to cancel out the phase noise on the input.

So this suggests to us that the mixer doesn’t add a significant amount of phase noise. We’ll keep looking deeper into this problem and happily let you know what we find!

I really like this page. I would like to use the data plots in a tutuorial I am creating. Would that be acceptable?

Bruce,

I would be happy to allow you to use the plots for your tutorial. All I ask is that you credit this article as the original source.

Thanks for the interest.

Very good experiment and thank you for posting it.

Looking at your results, i’m wondering what would happen if repeat the experiment using, for example, the second harmonic of the LO in the upconversion and the fundamental in the downconversion. In theory, at the output there will be a signal (lets call it “Ups2”) at 5.01 GHz equal to the first upconversion in your initial experiment (lets call it “Ups1”). The question is: Should the phase noise be the same in both signals: “Ups2” and “Ups1”. What do you think?

When a signal is doubled, the phase noise increases by 6 dB (20 Log(n), where n is the multiplication factor). This means that the phase noise would be 6 dB worse for the second harmonic upconversion than it was for the fundamental.

However, to coherently downconvert you would need to double the LO, so it would still work.

And a second question: In your experiment do you have an idea (a more elaborated reasoning or even an equaton) of where the increment on phase noise in the upconverted signal comes from? Or in other words, how to relate the phase noise of the input signal with the upconverted signal. Thanks 🙂

I have a mathematical treatment for how the IF is contaminated by white noise from the LO through leakage and common mode downconversion, but not for exactly how the IF inherits the phase noise of the LO. I would imagine it to be some sort of convolution, but I’m not sure.

Hey Doug,

could you provide that mathematical treatment? Or link it?

Thanks!

Here is what I wrote 6 years ago. I offer this totally unedited and without confirmation of any sort:

Noise Factor – (S_i/N_i )/(S_o/N_o ) assuming input noise = kTB

Input Signal – PRF

Input Noise – kTB

Output Signal –P_RF∙CL_(f_RF ) (CLRF indicates the conversion loss at the RF frequency)

Output Noise – kTB+N_([email protected]_RF )∙CM_(f_RF )+N_([email protected]_(RF+IF) )∙CM_(f_(RF+IF) )+N_([email protected]_IF )∙〖LK〗_(f_IF )

Where N_([email protected]_x ) indicates the noise on the LO at frequency x, CM_(f_x ) indicates the LO common mode rejection at a frequency x, and 〖LK〗_(f_x ) indicates the small signal leakage from the LO at frequency x. Assuming the mixer is totally controlled by the LO (i.e. the RF cannot create LO mixing products), the noise contribution of the diodes is low (typically the case), there are no nonlinear effects (i.e. spurious noise products), and the noise in the RF sidebands are all kTB limited,

Noise factor = (P_RF/kTB)/((P_RF∙CL_(f_RF ))/(kTB+N_([email protected]_RF )∙CL_(f_RF )+N_([email protected]_(RF+IF) )∙CL_(f_(RF+IF) )+N_([email protected]_IF )∙〖LK〗_(f_IF ) ))

= (kTB+N_([email protected]_RF )∙CL_(f_RF )+N_([email protected]_(RF+IF) )∙CL_(f_(RF+IF) )+N_([email protected]_IF )∙〖LK〗_(f_IF ))/(kTB∙CL_(f_RF ) )

= (1+(N_([email protected]_RF )∙CM_(f_RF )+N_([email protected]_(RF+IF) )∙CM_(f_(RF+IF) )+N_([email protected]_IF )∙〖LK〗_(f_IF ))/kTB)/(CL_(f_RF ) )

= 1/(CL_(f_RF ) )+(∆N_IF)/(kTB∙CL_(f_RF ) ) where ∆N_IF= N_([email protected]_RF )∙CM_(f_RF )+N_([email protected]_(RF+IF) )∙〖CM〗_(f_(RF+IF) )+N_([email protected]_IF )∙〖LK〗_(f_IF )

Thanks a lot Doug. It was really helpful. I have a question please. You meant all passive mixers will not effect the phase noise significantly, didn’t you?

Ali,

We think so, but we have not performed this experiment with FET, non-Si, or non-semiconductor mixers. We suspect that all passive mixers will not affect the phase noise, but have not experimentally confirmed it. There is also the big caveat that some mixers are more sensitive to LO-IF noise leak through, which can affect the phase noise performance much more significantly than the devices inside the mixer.

As always though, the real phase noise experts are at

Assuming there is a variation in the group delay between the two mixers. What is the output phase noise at a frequency offset (fm) from carrier as function of the Group Delay Variation and The phase noise of the LO?

I’m not sure if I’m totally clear on the question. The group delay of the mixers shouldn’t matter in this setup, as long as it is roughly constant in the region of consideration and the LO is matched between the two mixers. If not, then the additive phase noise would show up as a function of correlation between the phase noise of the LO at different times and the group delay variation over that frequency range. I’m not familiar enough with the noise statistics to know what the functional form would be, though.

Interesting approach.

What happen with the phase noise if in between the mixers, amplifier is palace in. Does the output phase noise will include the residual phase noise from added amplifier? If that so, that will be revolutionary.

I wonder how much the splitter isolation on LO path have impact on the results. Does the LO path required some phase correction?

The phase noise added by an amplifier is mixed, not multiplied, so I don’t see a reason that you couldn’t theoretically add the phase noise at a high frequency and then mix down and have the phase noise be reduced.

The phase has to be matched such that the signal is downconverted with the exact same LO that it was upconverted with. There must be a phase delay equal to the RF transmission delay applied to the LO. As the phase walks off the phase noise will decorrelate, not immediately but depending on the frequency of the phase noise, so you might be able to get away with a lot of phase differential depending on what you are trying to do.