YYou wrote on May 30th, 2008, 5:53pm:I was considering that this noise at the LO port is stationary, so with the noise shaping by the LC tank & HPF, the noise power is only peaked at LO frequency.
I can't understand why noise spectrum is peaked at LO only when noise is stationary.
I think noise spectrum is also peaked at LO even when noise is cyclostationary.
YYou wrote on May 30th, 2008, 5:53pm:Referring to the paper 'noise in RF-cmos mixer: A simple physical model' by Darabi and Abidi, an interferer(noise) at a frequency of fin at LO port will produce harmonics at the mixer output at fin, 2f(LO)+/-fin, 4f(LO)+/-fin,..., so the big noise at LO frequency will not be observed at IF(DC for direct conversation receiver). Now I realize that this noise maybe cyclostationary, so the above statement is not applicable to this situation.
Noise at mixer output is cyclostationary. Also noise from LO buffer is cyclostationary.
But whether cyclostationary or not is not issue in your case, I think.
YYou wrote on May 30th, 2008, 5:53pm:I also tried to verified this by simulation. Here is the setup:
-Case 1: check the noise at IF using PSS/Pnoise
a. Input RF port: PAC=1 (say, RF freq band at 2G)
b. LO(divider input)= 2*2G=4G
c. PSS (fund=4G, Beat=2G)
Wrong. fund=beat=2GHz.
But you should not use term "beat" in PSS setting. Expression of "beat" is not proper.
Fundamental frequency of PSS is not beat frequency.
I can't understand why Cadence has kept to use this expression of "beat" in PSS's setting UI.
There is no "beat" in analysis statement.
But I don't mind "beat" since I don't use Cadence SpectreRF for RF circuits design.
YYou wrote on May 30th, 2008, 5:53pm:d. Pnoise(start=1K to 3M, max.sideband=20, output=IF_output, input=rf_input, reference band=-1)
e. after simulation, I checked noise summary and saw large portion of noise(1K to 100K) is from 1/f noise from LO buffer/DividerBy2.
-Case 2: check the noise at LO buffer output(using the same circuits as case 1)
a. same as case1
b. same as case1
c. same as case1
d. Pnoise(start=1K to 3.5G, max. sideband=20, output=LO_buffer_output, input=rf_input, reference band=-1)
e. I checked noise summary again after simulation, I expect to see the biggest noise contributor at 2G is from the 1/f noise from the transistors in LO buffer/dividerBy2, but the result is the opposite, most of it is white.
Is my setup right ? What is the noise spectrum I should see at the LO buffer output ?
Of course, improper. Your simulation points in pnoise are not proper, so 1/f noise is lost in case-2.
If you observe noise of IF frequency, you have to set like following.
aho pss fund=2G harms=8 maxacfreq=40G errpreset=moderate annotate=status outputtype=freq
boke (
Any_Out_Node 0) pnoise start=1K stop=3M maxsideband=20
+ sweeptype=absolute annotate=status
If you observe noise of LO frequency, you have to set like following.
aho pss fund=2G harms=8 maxacfreq=40G errpreset=moderate annotate=status outputtype=freq
boke (
Any_Out_Node 0) pnoise
start=1K stop=3M maxsideband=20
+
sweeptype=relative relharmnum=1 annotate=status
You don't have to set iprobe=rf_input and reference band=-1 when you don't see NFssb and Gain.
"reference band=-1" is not proper especially for case-2.
YYou wrote on May 30th, 2008, 5:53pm:The last question is how to solve this issue:
- Now I am using single balanced mixer, if using double balanced mixer, the noise can be largely rejected since it is like common-mode signal. Right ?
Right. But not largely. And the noise is not like common-mode signal.
YYou wrote on May 30th, 2008, 5:53pm:- If I stick to single balanced mixer, is there any way to lower the noise contribution from LO part ?
It might be improved if you change LO and RF ports.