How to determine Zin, Yout for RF transistor, from A parameters

On 05/06/2017 06:39 PM, Kevin Aylward wrote:

> S parameters are only valid as a linear approximation, and give
> incorrect results for stability because of this. They are completely
> useless for determining distortion, phase noise, or the best operating
> current for a design, or for dealing with variations. Because, now get
> this, they are only a linear approximation.

Does it matter that much? If one's designing say a chain of RF 
amplifiers, then it would seem the nonlinearity of the final is going to 
dominate everything, and there are ways of dealing analytically with 
weakly nonlinear systems.

> Hardly ever, today. You are obviously not someone that needs to design
> ASICs, that work first time, and work in their millions.
>
> Small signal analysis has absolutely zero chance of doing that.
>
> Once you know what you are doing, setting up worst case runs and
> parameter sweeps over 10,000s of transient runs, will achieve what is
> absolutely impossible to do with manual equations. The world has moved on.

idk. it sounds a lot like you're saying the only "right" way to design 
analog RF circuits is to already be an expert analog RF designer. Then 
you wouldn't need to muss with any math because you would know what to 
do, intuitively.

I don't know how helpful this advice is for mere mortals.

On 05/06/2017 07:09 PM, John Larkin wrote:

> How do you know what to simulate? Just plunking parts on a screen and
> running Spice only works for very simple circuits. Try designing a
> 5-resonator hairpin filter by fiddling.

If you were a _real_ RF design man you'd just know. jeez.

On 2017-05-08 15:14, Okkim.Atnarivik@invalid.fi wrote:
> On Monday, May 8, 2017 at 10:43:22 AM UTC+3, Jeroen Belleman wrote:
>> On 2017-05-06 05:50, dakupoto@gmail.com wrote:
>>> Could some RGF/microwave guru on this newsgroup
>>> please help ? I am looking at the datasheet for
>>> a ONSemi RF transistor, with the S parameters
>>> listed for various frequencies. What would be
>>> a quick and easy(that I can code as a simple
>>> C program) way to convert these to the
>>> corresponding Zin and Yout. Thanks in
>>> advance for your hints/suggestions.
>>>
>>
>> Since no one took the trouble to simply reply to your
>> question, here we go.
>>
>> The S11 parameter is the input reflection coefficient.
>> It's just a different way to represent input impedance.
>> S11 is in fact what you would measure across the
>> arms of a Wheatstone bridge with the relevant impedance
>> in one of the arms and Z0 = 50 Ohm resistors in all the
>> others. Converting, Zin = Z0*(1+S11)/(1-S11).
>>
>> The same conversion applies to S22 and Zout. Yout=1/Zout.
>>
>> Bear in mind that these are linearized parameters of
>> a basically non-linear device at specific bias conditions
>> and with a specific excitation level.
>>
>> All complex number math, naturally.
>>
>> Jeroen Belleman
>
>    For an unilateral transistor i.e. S12 nonzero, the formulae are
> a bit more complicated:
>
> Zin = Z0 * ((1+S11)*(1-S22)+S12*S21) / ((1-S11)*(1-S22)-S12*S21)
> Yout = 1/Z0 * ((1+S11)*(1-S22)+S12*S21) / ((1+S11)*(1+S22)-S12*S21)
>
>    One obtains those from signal graph analysis, or (as I did) by
> copying from Pozar's book.
>
>    Regards,
>             Mikko
>

I didn't try to verify in detail, but it looks like there's a
error in there, because those expressions do not reduce to mine
if one sets S12=0.

Jeroen Belleman

On Sun, 7 May 2017 19:49:42 -0700 (PDT), tabbypurr@gmail.com wrote:

>On Monday, 8 May 2017 01:52:15 UTC+1, k...@notreal.com  wrote:
>
>> Halfa megabuck is nothing.
>
>not in my universe

Well, I'll be buying damned few mask sets myself but a half-a-magabuck
isn't a lot of money when talking about ASIC or custom chip
development.  Gate arrays may be significantly less, of course.

>wrote in message news:tp61hcdr752v3ammue1gdfcp5e2ckv3mk6@4ax.com...

>On Sun, 7 May 2017 19:49:42 -0700 (PDT), tabbypurr@gmail.com wrote:

>>On Monday, 8 May 2017 01:52:15 UTC+1, k...@notreal.com  wrote:
>
>>> Halfa megabuck is nothing.
>
>>not in my universe

>Well, I'll be buying damned few mask sets myself but a half-a-magabuck
>isn't a lot of money when talking about ASIC or custom chip
>development.  Gate arrays may be significantly less, of course.

It depends on what you are doing.

0.18u BiCmos is great for a whole bunch of stuff. Maybe $50k -$100k for an 
engineering prototype build.

22nm, another story in cost. High precison analog usually means gates > 1u. 
Low noise means > 10u, anyway so...

-- Kevin Aylward
http://www.anasoft.co.uk - SuperSpice
http://www.kevinaylward.co.uk/ee/index.html 

"Jan Panteltje"  wrote in message news:oeo0db$vsh$1@news.datemas.de...

On a sunny day (Sun, 7 May 2017 20:55:15 +0100) it happened "Kevin Aylward"
<kevinRemovAT@kevinaylward.co.uk> wrote in
<BIqdnXfG6Yo55pLEnZ2dnUU7-UPNnZ2d@giganews.com>:

>              yer good old JFET
>               |-------------- choke--- + 12
>  ------------>|          |
>|   |         |--        |
>|  === C         |       |
>L   |------------|      ===
>|  ===          [ ]      | decoupling
>|   | 2C         | 1k    |
>  ---------------------------------- GND
>
>Ho hummm...
>
>>That is not an oscillator.

>Really... I does oscillate here, been doing that for decades.

An oscillator, is one that is complete. One that can actually be plugged 
into another piece of kit.

There are standards. Where is the CMOS or Cliped sine buffer?


>>Worked for me.
>
>
>>What is its phase noise?
>>What is is stability over temperature and power supply variations?
>>How many will be in spec out of a million made?
>>What is its spec?
>>What's its frequency sensitivity to load?

>Good question.

And no answer.

>....
>
>>You never actually build any RF did you?
>
>>I don't build anything today, I haven't for many years. I modied my 
>>Marshall
>>amp a while back though. It had a || loop for the effects. Ho humm...

>Does it use vacuum tubes?

That one had one tube in it. Model is AVT150.  I have another one, all 
Transister, MG100. I gave up on the tubes many years ago. Could not be 
bothered with the weight.

>>However, I do have an OCXO asic  in production that runs up to 100MHz, has 
>>a
>>few ppb stability, and flat band phase noise approaching -165db, in a very
>>small package.

>Very impressive,. -165 dB wonder how you measure that.

Should have said dBc, but I guess you knowing all about oscillators, would 
know the standard terminology.

I should expand on this. This is the noise after the limiter/squarer. Raw 
oscillator designs can be -190dBc to -180dBc. A key point is that this is 
with an internal 2.5V supply, in a package of 7mmx5mm. Any mere mortal can 
get that sort of performance in a 1" square with 12V.

Very few applications use a sine wave, so only a final CMOS or CLIPPED Sine 
output is relevant. Clipped sine don't mean what it says. It means a 1.2V 
regulated output of various shapes, including square. Its a historic term.

>What do you use it for, QAM 65535? Or QAM 32 bits?

I don't use it for anything. My company sells them to those that do. They 
are used in various areas such as telecom, gps, adcs, modems, whatever.


-- Kevin Aylward
http://www.anasoft.co.uk - SuperSpice
http://www.kevinaylward.co.uk/ee/index.html 

"bitrex"  wrote in message news:gP_PA.74499$yk6.36581@fx06.iad...

>On 05/06/2017 06:39 PM, Kevin Aylward wrote:

>> S parameters are only valid as a linear approximation, and give
>> incorrect results for stability because of this. They are completely
>> useless for determining distortion, phase noise, or the best operating
>> current for a design, or for dealing with variations. Because, now get
>> this, they are only a linear approximation.

>Does it matter that much?

Yes.

> If one's designing say a chain of RF amplifiers, then it would seem the 
> nonlinearity of the final is going to dominate everything, and there are 
> ways of dealing analytically with weakly nonlinear systems.

Its everything. The list is endless. Once one part needs an advanced tool, 
you need to have the advanced tool. Its irrelevant whether over bits may not 
require it.

Today, all the easy jobs have been done. At one time oscillators only needed 
stability of 50 ppm. Now some systems requite 1 ppb.

Phase noise is crucial for the vast number of products out there. Its about 
squashing in 100MBs data for everyone within a 1km radius of each other, and 
so forth.

If one company is using all the right tools, how do you expect to compete 
using bear skins and knives?

> Hardly ever, today. You are obviously not someone that needs to design
> ASICs, that work first time, and work in their millions.
>
> Small signal analysis has absolutely zero chance of doing that.
>
> Once you know what you are doing, setting up worst case runs and
> parameter sweeps over 10,000s of transient runs, will achieve what is
> absolutely impossible to do with manual equations. The world has moved on.

>idk. it sounds a lot like you're saying the only "right" way to design 
>analog RF circuits is to already be an expert analog RF designer. Then you 
>wouldn't need to muss with any math because you would know what to do, 
>intuitively.

It is impossible to do modern competitive design using just intuition. Its 
too complicated. You need to have an understanding as to what to randomly 
fiddle...

-- Kevin Aylward
http://www.anasoft.co.uk - SuperSpice
http://www.kevinaylward.co.uk/ee/index.html 

wrote in message news:10gvgcdc7nt8r70tcdn5iibeepqfoor1c7@4ax.com...

On Sun, 7 May 2017 20:52:06 +0100, "Kevin Aylward"
<kevinRemovAT@kevinaylward.co.uk> wrote:

>"Jan Panteltje"  wrote in message news:oekvo1$jak$1@news.datemas.de...
>
>On a sunny day (Sat, 6 May 2017 11:24:14 -0500) it happened "Tim Williams"


>There is a big difference between chip and board level design.

My basic point through all of this was to dismiss this general idea of 
"spice isn't about the real world.... isn't good enough etc...etc..."

Pretty much every modern product today, exists because of the ics in them. 
All ic design is done in the virtual world. Only a tiny minority of ic 
designs have any dependence on unknown board level issues.

Sure, I am not an RF specialist.  However, I have done a few bits and bobs 
at the several GHz region. I have also spoken to RF specialists. Most have 
missed the boat. They think in ways passed on from their fore fathers, that 
were passed down from their forefathers.

The days of graphical techniques and manual equation solving are gone. 
Whether its electronics, designing the same bloody energy efficient car body 
shape, trading stocks and shares, its all done by running millions of 
simulations. It what TFlop computers are very good at. You just need to 
learn how to drive them correctly. Unfortunatly, as far a board level spice 
users go, most don't know how.

-- Kevin Aylward
http://www.anasoft.co.uk - SuperSpice
http://www.kevinaylward.co.uk/ee/index.html

On Sunday, May 7, 2017 at 12:58:21 PM UTC-7, Kevin Aylward wrote:
> "Jim Thompson"  wrote in message 

> >That's a remarkably ignorant statement.  During my youth slide rules
> >and Smith Charts were taught in high school and college.
> 
> Indeed, and we no longer use slide rules.

Hey, I still use slide rules!  Best way to set up a resistor ratio
while rummaging through the stock drawers for a suitable pair
to build the divider.

And a good log-log chart or nomograph is still useful (just hard to explain
to a millenial).

The digital age has all the advantages of a digital watch, and none of
the advantages of a clock dial.   

On 05/08/2017 03:45 PM, Kevin Aylward wrote:

> If one company is using all the right tools, how do you expect to
> compete using bear skins and knives?

I guess if your gig is doing designs that are intended to be sold in 
their bazillions, then yeah, it's extremely important to know everything 
about everything. Because, well, that's the gig you're in. If you know 
that your way is the only way that works in that gig then I can hardly 
argue because hey, that's your gig, not mine. ;-)

Not doing it whatever way that is is probably OK if you're in a 
different gig. Sometimes all you need is an RF amp that's good enough 
and it doesn't matter if anyone wants to buy them in quantities of 
millions. Unless every structure that could possibly ever be needed by 
anyone has already been produced in IC form by someone who can do it 
better. Maybe that's the case. Seems that way sometimes.

> It is impossible to do modern competitive design using just intuition.
> Its too complicated. You need to have an understanding as to what to
> randomly fiddle...
>
> -- Kevin Aylward
> http://www.anasoft.co.uk - SuperSpice
> http://www.kevinaylward.co.uk/ee/index.html

"Can't get a job without experience, can't get experience without a job."