How to design: Microwave subharmonic mixer

[Yansi]

Hi!

I have seen quite a lot that subharmonic diode mixers (also often called "russian mixesr" or the "Polyakov RA3AAE mixers") are quite popular in many amateur radio construction, from HF up to microwave bands.  I am more interested in the microwave band use of this mixer, as at HF it is quite easier to make it work as it should.

Here is what is going on:



I would like to ask, if you have any tips for literature or have a first hand experience building these. Especially, I am interested in how these things are supposed to be designed to work properly at microwave frequencies.

Supposed I want to try designing a mixer for a 9cm band (3400MHz) - which is actually what I want to try. I have already had a go with this about two years ago, copied a design from some "schematic collection book", but the performance was really crap, totally unusable. Now that I have access to some proper test gear, I would like to give it a second chance.

Those lambda/4 or lambda/2 stubs are the only things straightforward I understand there. What I am after is how to guesstimate:
1) The impedance of the transmission lines to be used around the diodes
2) How to match these so the mixer will become as close to 50ohm as possible on all three ports.

The proper mixer diode pair I have available is HSMS-2822 (datasheet), which I think should be sufficient.

Please understand this is more like an experiment, mainly out of my curiosity and to improve some knowledge. I know that there are plenty of ready made 50ohm matched passive and even active mixers these days, that are very easy to implement into a design.

[rfeecs]

The typical way to design a mixer is to use a simulator.  You need a harmonic balance simulator like in Keysight ADS or NI AWR Microwave office.  My experience is this gives really good results, since the only nonlinear elements are the diodes which are very easy to model.

Even if you don't have those simulators, it may be useful to look at their app notes.
This page gives a list of app notes, but the links don't seem to work:
http://www.awrcorp.com/download-ni-awr-design-software-rfmw-circuits-mixer-application-notes
Here is one on harmonic mixers (obtained by googling), seems like a typical academic paper:
https://awrcorp.com/download/kb.aspx?file=appnotes/Harmonic_Mixer.pdf

You might want to look at what Marki Microwave has to say about simulation:
http://www.markimicrowave.com/blog/engineering/resources/mixer-simulations/

I have heard the free simulator QUCs has harmonic balance, but I have never used it:
http://qucs.sourceforge.net/tech/node30.html

Matching is often not that important.  Many broadband mixers aren't actually matched at all and still give relatively good performance.

Steve Maas has written a lot of books about mixer design:
http://www.nonlintec.com/books.html

In those, he outlines a method similar to harmonic balance that can be used to calculate impedances.  His books are quite academic and not too practically oriented.  But just using the simulation software is the way to go.

The other possibility if you don't want to simulate is just set up the basic mixer on the bench and tune it by hand.

[Yansi]

Thank you for some valuable tips.

After reading a bit, I can kind of understand the design principles of the mixers.  What I need is at least a circuit simulator capable of simulating non linear devices, like diodes. Is there anything free (or that could be made free) suitable for this?

To sum up for others interested in such MW mixers, what I understand the design process should be as following:

1) Choose suitable mixing diode. Series resistance R_s and chip capacitance C_j0 are of the main concern, 1/2*pi*R_s*C_j0 should be high enough (i.e. higher than desired frequency of operation). I think my HSMS-2822 with its 6ohm and 0.7pF is good enough for 3.4GHz.

2) Simulate the diode mixer, just two ports with the required shunt (IF return) and open (LO termination) stubs to obtain the impedance at LO port at LO frequency and to obtain impedance at port 2 (RF, IF) at the IF and RF frequencies respectively.

3) match the RF, IF ports, so that IF port is open at RF frequency and RF is open at IF frequency.

4) Preferably match also LO that it doesn't make unnecessary termination for RF and mainly the IF frequency (conversion loss increased otherwise).

5) Convert the electrical schematic to a microstrip design, simulate and optimize it (this might be the hardest part)

6) Call it a day.


Now the question is, what circuit simulator to use for the step 2, that I could fill in the model of the diode. Currently, the only RF software tools available I have are a copy of unusable Sonnet Lite and a Ansoft Designer SV, which is a free restricted (IDK how) student version I have scored somewhere on a web. And I almost forgot RFSIM99, but this won't help, that one can't do diodes/non-linear elements.

[rfeecs]

You can simulate a nonlinear circuit with SPICE.  But for a mixer, harmonic balance is better because your IF and RF are usually very different frequencies, so SPICE transient analysis requires very long simulation with small time steps to cover both low and high frequencies.

Again, for something free, I suggest Qucs:
http://qucs.sourceforge.net/

It looks like here is an example of a double balanced diode mixer simulation, refer to section 6:
http://www.gunthard-kraus.de/qucsstudio/Tutorial_part%202_HB.pdf

Like I said, I've never tried Qucs myself, so no idea how well it works, but it is something to try.

[Yansi]

Looks very interesting, certainly because in depth tutorials seem to be available. Will try the QUCS. Although I would like to primarily improve my skill with LTSpice. However if the HB is the way to go here... (I read in multiple sources that HB simulation is the proper way to do it).

This seems promising, so far. Now I just need some spare time to play!


//EDIT: Just downloaded the QUCS. Works out of the box. Surprised, surprised...

//EDIT2: Fucking hell. That thing is not intuitive. And already found a few minor bugs. Still chaven't figure out how to place anything else than a wire in the editor. Also can't display any simulation results. Ugh...  I should really read some tutorials.
In the intuitivity regard, hats off to the people who are behind RFSIM99. This one is VERY nice and simple to use.

[Yansi]

After eating through a couple of tutorials, it is not as bad as it looked.

I have acquired a SPICE model for the HSMS-2822 diode from the Broadcom website.  I need to transfer that model to the QUCS - I am working on it now.

What I do not understand is the discrepancy between the content of the SPICE model file and the table in the datasheet.

Here you can compare.  Why are some of the parameters so different?

Code: [Select]

*SPICE model for HSMS-282x
*The parameters are for a single diode (HSMS-2820). Parameters also apply
*to the individual diodes within multiple diode configurations.
*
.SUBCKT hsms 1 2
DCD1    1     2      DMOD   
.MODEL  DMOD D(IS=4.8E-8, CJO=0.649E-12, VJ=.56, BV=26.7, IBV=10E-4
+ EG=0.69, N=1.067, RS=7.8, XTI=2, M=0.5)
.ENDS

[Yansi]

So instead of trying to import the SPICE, i have used a universal diode model and modified  some of the parameters according to the SPICE model.

I then tried to run my first simulation. A volt-ampere characteristic of the HSMS-2820 diode. 

[Yansi]

So far I can do this. Used a bog standard diode model and typed in the values from the SPICE file, that I have downloaded from Broadcom. Then I can read the port impedances using S parameter simulation, which already is a big improvement over what I was able to do last time I attempted such mixer.
Sure, the diode model is junk, as it doesn't include the real component, only models the diode's die, but better than nothing. Will try to improve it further.

Now I need to switch to the HB simulation and try to find the LO power optimum. Then start matching the port impedances.

What characteristic impedance should the stubs have? I have never found any mention of that in any literature. Just see people usually use rather thick traces, so the impedance definitely is lower than 50ohm. (image below I used 50ohm for them).

[rfeecs]

The width of the line depends what you are using it for.  For example, if you want a short circuit quarter wave line to look like an open circuit, the bandwidth will increase for a higher impedance narrower line.  Similarly if you want an open circuit quarter wave line to look like a short, the bandwidth will increase for a lower impedance wider line.

But there are limits on how narrow or how wide you want to go.  Too narrow has more loss and may be harder to process, too wide takes up more space and eventually will start to "mode" where at some frequencies it is a quarter wave along the width.  So it is a compromise.

So now you are looking at small signal S-parameters.  You will see the same impedances regardless of power.  You won't see any non-linear behavior unless you do harmonic balance.

I'm not sure if you can easily look at large signal S-parameters with QUCS.  Those would be defined as the ratios of the harmonic components (A1/B1, etc) for the frequency you are interested in.  For Microwave Office it is easy.  Those are the S-parameters presented in the harmonic mixer paper from the AWR site.

You might start out as a sanity check to try something easier, like a diode detector:  source driving a series diode and a load resistor.  You should be able to see the rectifying action.  Then add a capacitor across the load and you should see DC voltage vary with RF power.

[Yansi]

I have experimented with RF power already, the S-parameter simulation seems to pretty much ignore the RF power - so the sim result is nice, but more like wrong (it is wrong already, as I have not a correct component model reflecting the packaging of the die). Exactly as you have said.

I have briefly tried the harmonic balance, but the qucs always fails with some unknown error, telling me I should contact the SW author.   

I will try once again, but, not sure how.

Regarding the stub width - I should have probably known that. Thanks for commenting!

[Yansi]

The QUCS is starting to slowly piss me off.

One fucking hour I am trying to display an  FFT.  It still keeps telling me that variable RF.Vt does not exist, yet it offers me it to be plotted to a graph. Guess what? The variable exists, however the fucking time2freq function does not exist! I did copy that straight from a tutorial.  Meanwhile someone have decided they change it to Time2Freq (yes, case sensitive). But it still tells me the variable does not exist instead. 

The Harmonic Balance simulation seems broken beyond usability. Doesn't work with anything. Is even ignoring my iteration count settings. Gets broken right away (won't do anything) with placing any kind of microstrip line into the circuit.

Also, the HB simulation does not work with power source - broken too. It took me some time to figure out I need AC voltage source, add a series resistor externally and calculate the voltage from dBm manully. Aaaaargh!!! 

//EDIT. Crap!  I can't even display a spectrum of a waveform. Because it tells me the waveform I am just displaying in a graph does not exist. I am quite disappointed with the QUCS being not usable at all, as nothing works as it is supposed.  So I will try to get my hands on something better. Like the proper tools. ADS or AWR MO. This way I can't do anything with it.

[Yansi]

I have finally found what I have been doing wrong. There is a BIG difference between QUCS and QUCS Studio. These are two very different programs, although the GUI look almost identical to the point I could not figure the difference out for a week. What a joke!! 

Now I have the correct fork of the original "QUCS" named "QUCS Studio" which is exactly what was needed the whole time.

The harmonic balance simulation is now working without any signs of issues, I have also already produced a few interesting graphs, like for example the frequency spectrum at the RF port and the conversion loss figure as a function of LO level.

Please mind you the simulation was done with a simplified model of the diode, I still need to work on creating a more realistic model of the diode.

I also need to figure out how to get a port impedance at a specific frequency when using the harmonic balance simulation. Currently I have no idea.

[biastee]

After eating through a couple of tutorials, it is not as bad as it looked.

I have acquired a SPICE model for the HSMS-2822 diode from the Broadcom website.  I need to transfer that model to the QUCS - I am working on it now.

What I do not understand is the discrepancy between the content of the SPICE model file and the table in the datasheet.

Here you can compare.  Why are some of the parameters so different?

Code: [Select]

*SPICE model for HSMS-282x
*The parameters are for a single diode (HSMS-2820). Parameters also apply
*to the individual diodes within multiple diode configurations.
*
.SUBCKT hsms 1 2
DCD1    1     2      DMOD   
.MODEL  DMOD D(IS=4.8E-8, CJO=0.649E-12, VJ=.56, BV=26.7, IBV=10E-4
+ EG=0.69, N=1.067, RS=7.8, XTI=2, M=0.5)
.ENDS

Some of the parameters are different because they were measured at different times; i.e. different wafers.

So far I can do this. Used a bog standard diode model and typed in the values from the SPICE file, that I have downloaded from Broadcom. Then I can read the port impedances using S parameter simulation, which already is a big improvement over what I was able to do last time I attempted such mixer.
Sure, the diode model is junk, as it doesn't include the real component, only models the diode's die, but better than nothing. Will try to improve it further.

You will need to put the die model inside the package model (the HSMS-282x chip is available in different packages). See AN1156, specifically fig. 3, for idea on how to do this.  AN1124 provides models for different packages.

How did the project go?

[Yansi]

Hi and thank you for commenting. Project was scrapped, due to: a) not having access to any measurement equipment to make any verification and b) not having any functional simulator to make the harmonic-balanced (non-linear) simulation of the diode mixer.  The QUCS Studio seem to just crap out when two ports are used instead of just one.

I can't say it can't be done; diode mixers are pretty common thing, however proper tools needs to be used and knowledge available. Unfortunately it seems I do not have neither of those. So I better stick with HMC213 for projects and hope it will do something.