Blocker Noise Figure / Reciprocal Mixing Noise test using SpectreRF (MMSIM14)

Dear Andrew,

I have so far NOT used HB analysis as I use sharp square clock and my mixer is a switching passive mixer.

So far PSS shooting method  has been fine for PSS-PNOISE analysis.

But with close-in blocker,  pss-shooting method becomes slow.

I have some queries about your suggestions:-

1:- Reasonable number of harmonics :- will it be OK if I take 50 ( this is typically I take for Pnoise-PSS-shooting analysis).

2:-  I couldn't understand your comment :-

"This way you are not trying to combine all the frequencies into a single beat frequency, but can trade off the number of harmonics for each of the two fundamentals"

Could you please elaborate in more detail ?

Kind Regards,

This is potentially quite a big thing to explain in a forum reply, as I may have to explain the whole of SpectreRF ;-) . I'll try to keep it simple though...

First of all, the difference between shooting and harmonic balance.

Shooting represents the solution in the time domain by finding a sequence of time points (the spacing of which is influenced by the highest harmonic you ask for, but it follows the signals in the same way that transient does), and ensuring that the period that you're solving for is a complete period. The number of harmonics you request doesn't really affect accuracy (it does to some extent, but it's a secondary effect), and is primarily an output parameter for the analysis. Having found the time domain solution, it can also convert that solution into a frequency domain representation.

Harmonic Balance represents the solution in the frequency domain as a sum of sinusoids (or cosines), where the length of the series is determined by the number of harmonics you ask for. The number of harmonics you ask for is therefore an input parameter which does affect the accuracy - too few harmonics and the power in the higher harmonics are squeezed (a sort of aliasing effect) into the harmonics you requested - so this impacts the accuracy. Having found the solution, it can convert this back into the time domain.

I won't go into the details of how these algorithms work. However, PSS analysis treats the solution in terms of a single common frequency - so if you have multiple tones, you have to simulate over one period of the common frequency (the greatest common divisor of the frequencies). If the ratio of this common frequency (the PSS fundamental) to the highest of the tones, then it is expensive as you need to simulate lots of periods. With harmonic balance, you really shouldn't simulate multi-tone circuits with PSS - instead you should use a multi-tone analysis.

For multi-tone analyses, you have either the hb analysis or qpss analysis. The hb  analysis is either a single-tone harmonic balance, or a multi-tone harmonic balance - depending on how many tones you enter. The idea is that each tone can specify a number of harmonics for that individual tone, and the analysis computes the harmonic combinations of each tone. So if you ask for n harmonics of f1 and m harmonics of f2, you'll get the harmonic combinations of (-n...0...n)*f1+(-m...0...m)*f2 output. This is good for the case where f1 and f2 are not harmonically related at all, or are very closely spaced. It also allows you to have more harmonics for one (more nonlinear) tone than another (more linear) tone.

The qpss analysis is similar, except it can work in either shooting or harmonic balance mode. In harmonic balance mode, it's essentially the same as a multi-tone hb analysis. In shooting mode, it's a "mixed-frequency-time" algorithm, where one signal is simulated in the time domain (the most non-linear), and other signals are represented in the frequency domain (as a harmonic series). In practice however, except in a few applications with a highly non-linear clock signal plus a single weakly non-linear signal input, shooting QPSS is usually less efficient than harmonic balance (it's a bit more complicated than that, but I can't give a 4 day SpectreRF course here...)

So, you could run an hb analysis with 50 (or an appropriate number of harmonics - i.e one where you have enough to adequately represent the nonlinearity) for the square-ish tone, and some smaller number for the other. You don't want to ask for too many harmonics though, because that will slow things down - it's a tradeoff. The good news is that in recent IC/MMSIM combinations (it's been there for a few MMSIM releases), there is an "auto" mode which will pick the harmonics for the first (typically the most nonlinear) tone automatically by doing a short time-domain simulation and picking an appropriate number of harmonics. That avoids you having to spend too long thinking about it!

There are tutorials in <MMSIMinstDir>/tools/spectre/examples/SpectreRF_workshop (rfworkshop.tar.gz or something like that - I'm off the network at the moment) to talk through this in more detail (doesn't explain the theory, but gives some typical applications). There's also the Spectre RF Theory manual too.

Hope this helps. It was longer than I intended!

Kind Regards,

Andrew..

Dear Andrew,

I tried to verify the HB analysis.

To this end, I simulated a RF passive mixer based receiver front end with both PSS(shooting) and HB analysis to measure the blocker noise figure.

The LO is set at 200MHz with squarish shape of 100ps rise/fall time. The RF input is a single tone-blocker frequency at 280 MHz and with power of  -40dBm.

For PSS:-

The beat frequency is 40MHz. I set the maxacfreq 200GHz.  To find the NFdsb of the downconverted signal , I set the ref-sideband 5.

I got NFdsb=2.78dB

pss  pss  fund=40M  harms=5  errpreset=moderate  tstab=460n
+    saveinit=yes  method=gear2only  tstabmethod=gear2only  maxacfreq=200G
+    annotate=status

pnoiseOut1  (  BB_I_OUT  0  )  pnoise  start=0.5M  stop=2.0M
+       step=0.5M  pnoisemethod=fullspectrum  maxsideband=1000
+       iprobe=PORT1  refsideband=5  annotate=status

The I did HB analysis:-

I kept LO tone and blocker tone harms as [ x y ]  and oversample factor = [2 2]

With HB = [30 17] the "BB_I_OUT" gives NFdsb at 1MHz = 2.20 dB  

with increase harms, the simulataion time increases quite significantly and even become worrisome for this simple circuit.

But it is well below the PSS 2.78dB NFdsb value.    What is the issue ? why it is not  giving a value close to the PSS value.

Also it is taking long-time got large number of harmonics . Then what is the advantage of using HB instead of PSS ?

 We have filed the case:-#45978757 in APRIL-7 but still no solution even after 2 months except getting advice of increasing the number of harmonics.

Could you please help us in this regard ?

hb5   hb  saveinit=yes  autoharms=no  autotstab=yes  oversample=[2  2]
+   fundfreqs=[(0.2G)  (0.2G+40M)]  maxharms=[30  17]
+   errpreset=conservative  annotate=status
hbnoise_mtOut1  (  BB_I_OUT  0  )  hbnoise  start=100  stop=2M
+        step=1K  iprobe=PORT1  refsideband=[1  0]  annotate=status

Please follow this up through the case you have (I see there has been some recent interaction) - I can't duplicate the work of a colleague who is already working on this and is likely to have more detail than just the information above (it's very hard to give answers to your questions without being able to see more than you've shown).

In general, passive mixers (particularly if you are using bsim3v3 or bsim4 models) are tricky to simulate accurately - this is a well know phenomenon with various papers in the literature (including one from IBM) which describe the problems with the models that lead to these inaccuracies. It's not the simulator's fault - it's the model not being good in the region that passive mixers operate.

Regards,

Andrew.

See also http://community.cadence.com/cadence_technology_forums/f/33/t/30588 and the article http://www.nsti.org/publications/Nanotech/2007/pdf/1109.pdf referenced there (the associated presentation is http://www.nsti.org/Nanotech2007/WCM2007/WCM2007-LWagner.pdf). This is probably the article that Andrew had in mind.