RF Front End LNA simulation
I have designed some 50-50 matched LNA typologies. But when I design a front end on-chip mixer+LNA chain I confront an issue which really confuses me.
I have read some thesis and materials regarding this issue so I really want to dig this and get concrete answer and I would be really thankful and grateful if anyone give me advice in this case.
__________________________________________________ __________________________________________________ __
Imagine you want to simulate LNA which is going to be connected to mixer within chip (there is no matching between LNA & Mixer).
I have seen two methods
1) Mixer has been already designed and we measure input impedance of mixer (Zin mixer). Therefore when we design LNA we use output port which has real and imaginary part equal to Zinmixer real and imaginary part. So we measure S11, S21 ... NF and P1dB and etc.
The question is we may need to know LNA's output impedance to design Mixer ?
2) Second approach is used in Professor Behzad Razavi's book. It is emphasized that we have to care about voltage quantities in integrated circuit design.
So there is no port at the output and AC simulation is done. so we check AC gain (vout/vin) not S21 and basically we do not need output port or what ?
Even if you ignore mixer's input impedance (lets say you include Cload only) you have to put a port at the output of LNA to simulate NF since simulator need input and output port to plot NF.
What are the correct simulation steps to observe the real gain (or S21) or do we even care about S21 or not ?
What are the steps to observe other specs like NF and P1dB compression point for LNA ?
This is biggest question of my life ! If I totally get a concrete answer I would be more than happy :)
Thank you for your valuable time and effort.
Mixer noise figure design is different than LNA noise figure design. In a mixer have to provide the proper noise figure matching, for both RF and image frequencies, which is pretty hard to do it.
1. Designing the matching network at the RF input of the mixer to get the best mixer noise figure (as in LNA case), is not helping much. This happen mainly because mixer gain is low (or is no gain at all, as in passive mixers case).
2. In a mixer design more important than mixer noise figure is the mixer linearity (in most cases the mixer is the second stage in a system, after the LNA). A good mixer design will provide an RF input termination to get best mixer linearity, and not best mixer noise figure.
I wouldn't care about S21 of the LNA if I know that:
- the operating frequency's wavelengths is much longer than the physical distances in the chip. Then inside the package rather the output impedance of the LNA is important and the input capacitance of the mixer, not the reflection parameters maybe.
- the input of the LNA is matched well. You can check it easily with AC simulation, doesn't require S parameter simulation.
The mixer and the LNA is actually one circuit, they have to be so close to each other, and you shouldn't worry about reflection between them. So just simulate the LNA as a simple circuit not as a transmission line, if you know the maximum frequencies. My guess.
To get P1dB of the LNA I would use swept PSS with capacitive load which represents the mixer, to get NF I would use pnoise analysis.
In fact Mixer and LNA are totally antonyms words in RF & Microwaves world.
The only ones that keep saying they are the same, are the Analog IC designers which only use voltages to characterize their circuits.
I've seen analog IC designers when designing very high frequency circuits, that they never ever use power levels, as dBm. I keep telling them that voltages in microwave circuits have well defined limitations.
to get P1dB of the LNA I would use swept PSS with capacitive load which represents the mixer, to get NF I would use pnoise analysis.
I try this way,but to print P-1dB,need to select port in shcemetic, from your word,you needn't add port, i guess.so how do you do it? and use pnoise analysis to get NF,i don't know why the simluation result is so large:with sp ananlsis,the NF=5dB，while 70dB with pnoise analysis.
I didn't say that, but input port is not necessary just useful. It is easier to plot the P1dB if you sweep the input port's power inside pss analysis, then just go to Direct Plot and you are ready. This test requires only 1 port.
I don't know why you got 70dB instead of 5dB but I am sure the pnoise analysis is good. Plot your output noise and check the NF with hand calculations.
yeah,i find the problem.when pnoise analyse,i set the reference sideband=0. in fact ,it should to be zero.