Problem with some Soft Rock-like mixers... and solution

[AB2EZ]

I have noticed that there is a problem with the design of certain Soft Rock-like quadrature mixers... that causes the outputs to not be in quadrature (among other things): e.g. Soft66DB2, Soft66RF.

Refer to the first attachment below:

Note that the i.f. or r.f. inputs of the digitally controlled bus switches are directly connected together.

For a number of reasons, this is not good. For example, when the top mixer (in this application) switches on, the load impedance looking into its i.f. (or r.f.) input changes from high impedance (open) to essentially zero Ohms (i.e. the impedance looking into the op-amp that the output of the mixer is connected to). This causes a modulation of the signal at the input to the mixer that is synchronized with the in-phase switching control waveform (the local oscillator). As a result, there is a strong component on the input of the mix that is at the beat frequency.

Since the inputs of the two mixers are connected, one gets a variety of very undersirable effects.

There are a number of ways to redesign this circuit to fix this problem. One way is shown in the second attachment. Note that the i.f. or r.f. input signal source has a relatively low impedance (for example 50 Ohms). By using a separate series resistor for each input, one obtains a high degree of isolation between the two mixer inputs.

I tried this change on one of my Soft Rock-like mixers (Soft66DB2), and it worked great. The outputs are now in quadrature.. as they should be... and the outputs appear to be cleaner, as well. Prior to making this change, the outputs were definitely not in quadrature... not even close. [The change involved cutting two traces traces, and adding an additional series input resistor]

Stu  

[WA1GFZ]

Stu,
Most of these tayloe mixers rely on a 50 ohm source to get the correct time constant at the input of the op amp. I prefer a resistive divider with a 50 ohm shunt of a 6 db pad on a low Z source. The problem with using them at IF is the excessive gain ahead of the mixer. My homebrew with a 6dB pad has an MDS of about -130 dBM so any gain ahead of it limits the dynamic range. I also reduce the op amp gain to around 20 dB since the Tayloe will process a couple volts of RF. Reducing op amp gain will hurt MDS but not a problem if the mixer is in the second IF.

[AB2EZ]

Frank

Hi!

I never knew that the Tayloe mixers, that you and others have referred to, are similar to these Soft Rock / Soft Rock-like mixers.

Based on your post, I read Tayloe's paper (available on the Web, of course).

One important difference (I think), in the Tayloe mixer v. the Soft66RF mixer, is that, in the Tayloe mixer, each bus switch is only on for a quarter of each cycle. There is always one, and only 1, bus switch turned on.

In the Soft66DB2, each of the switching control inputs is a square wave at the local oscillator frequency... with the two square waves offset by a quarter of a cycle.

This, I believe, makes a subtle difference... leading to the interaction effect I described in my earlier post (above). The two bus switches interact during the time when they are both on (1 quarter of each cycle). In the unmodified design... when both bus switches are on, the paralleled i.f. inputs see two low impedance loads. When only one of the bus switches is on, the paralled i.f. inputs see only one low impedance load.

Stu 

[WA1GFZ]

Stu,
My Homebrew is a double balanced Tayloe which if I remember has two on at a time. It has been a while. My concern with your high series resistor is the long time constant at the output. This could limit the input bandwidth to the sound card. I have not followed the soft66DB2 or anything after the soft rock 6.
Look at the article from QEX that used to be on the FLEX site as a download. It talks about the time constant and its effects.

[AB2EZ]

Frank

I read the four (4) QEX articles available on the FLEX Website. These are very interesting and well written articles.


It was intresting to note something that changed between the first article and the last article.

In the first article, Gerald Youngblood shows the Tayloe design... essentially as it was shown in Tayloe's own publication (with reference to Tayloe's work, of course). In that design, a pair of op-amps are used at the output of the Tayloe sampling detector. One of the four (4) Tayloe outputs goes to the + input of one of the op-amps, and the complmentary (inverted) Tayloe output goes to the - input of that same  op-amp. But: the - input of an op-amp doesn't care about the capacitor at the output of the Tayloe sampling detector that feeds it... i.e. the - input is a zero impedance input.

[Note, the Soft66RF (and many other Soft-Rock like mixers) really don't use the Tayloe design, per se. They use the - input of the op-amp that is at the output of each bus switch... and they don't employ the Tayloe integrating capacitor. This is why the bandwidth of these mixers is not sensitive to the source impedance of the signal going into the mixer. Only the gain of the mixer depends upon the source impedance of the signal going into the mixer. The bandwidth is set by a capacitor in parallel with the feedback resistor in the op-amp]

Going back Gerald Youngblood's articles....

In the 4th article, he describes an implementation that uses a pair of instrumentation amplifiers (both + and - inputs are high impedance) instead of a pair of op-amps. This makes a lot more sense to me if you want to avoid the problem I described in my first post of this thread.

Stu

[WA1QHQ]

Stu,

I wonder if you had a chance to review the July/August 2002 QEX article by Rod Green VK6KRG entitled The Dirodyne; A New Receiver Architecture. In the article Rod described the problem you were seeing and attributed it to LO leak through to the RF input port of the Tayloe mixer / digital commutating filter. The solution he implemented was similar to yours only he put the individual resistors on the output of each switch.

Mark WA1QHQ

[AB2EZ]

Mark

Yes... putting a 5100 ohm resistor on the output side of each switch would be almost equivalent, and easier to do. One could just cut two traces, and put each resistor in a convenient pair of existing plated through holes on each side of each cut. This assumption is that the source resistance of the i.f / r.f. input (that drives all of the switches in parallel) is a small fraction of 5100 Ohms... e.g. 50 Ohms.

The only additional thing, using this approach, is to remove the 2.5 volt DC bias on the input of each switch. This is easy enough to do. It's actually better to not have this bias present on the inputs of the switches.

At the risk of fixing what is no longer broken... I'm going to try the output side fix... to verify that it also works... in case anyone who has one of these Soft Rock-like mixers wants to use that fix.

Thanks
Stu
[P.S. my QEX collection only goes back to 2006... and ARRL doesn't make the QEX archive available on-line, even for members. I can't find the article anywhere else either]

[KF1Z]

Interesting reading if you follow the threads, and names around...

From 8 years and more ago.......

http://groups.google.com/group/rec.radio.amateur.homebrew/browse_thread/thread/b4cabf2c30124eb0?fwc=1

[AB2EZ]

Mark

Well... I tried the approach of putting the series resistors on the output side of the switching device. The result was disappointing. An image appeared, centered at 0 Hz i.f. frequency (I was using an 11kHz i.f. frequency). I tried various embellishments... that I was hoping would make this new image disappear... but none of them worked. [After I went back to the input side configuration, I was thinking about what was wrong... and it occurred to me that this new image would be a symptom of non-linearity occurring in the bus switches. Perhaps I needed to play around a bit with the biasing]

So... I went back to the approach of placing the series resistors on the input side... and its working again (no images).

This is one of those examples where: the difference between theory and practice is larger in practice than it is in theory.

Anyway... although its simple and inexpensive, in terms of component costs... I don't think the Soft66DB2 (or similar designs) is what I would use if I were designing/building a quadrature mixer, for r.f. or i.f. applications, from scratch.

I think I would use the approach recommended by Frank et al. I would use a Tayloe quadrature sampling detector... and I would run all of the outputs into high impedance amplifier inputs (as per the Flex Radio approach described by Gerald Youngblood in the 4th of his series of articles).

Stu

[WA1QHQ]

Stu,

I believe that Rod's idea was the resistors were providing a high impedance to develope noise voltage across and that was why he had the resistors on the output. He did report that the resistors on the output which were 1/4 of the value of the single input resistor cleared up the noise problem. Rod never clearly defined what was the source of the noise unfortunately. It seems that you have made great progress on this front and once I build up my softrock board I will try it out and see if I can see the same noise and try your fix.

[WA1GFZ]

stu,
I think you need the 2.5 volt bias so the switches don't conduct through substrate diodes to 5 volts and ground. Also a single supply op amp wants to work above ground. My homebrew had plus and minus 15 volts on the inst amp so it would process inputs that produced 20 volts peak to peak outputs. The switches work well up to about 4 volts peak to peak. All of thai was well above the saturation point of the sound card.
There was a guy somewhere who used integrators after the switches with good luck. I never tried it but sounded like a good idea since everything was differential; after the switches.

[AB2EZ]

Frank
Mark
et al.


[Please read down to the "punch line" at the end]

In the process of "debugging" my Soft66RF, I learned an enormous amount about the subtleties of this type of quadrature mixer.

I learned a lot about the effect of having a switching device between the signal source and the - input of an op-amp (e.g. it is critical that the signal source be biased at the same DC level as the + terminal of the op-amp, or you will introduce a variety of problems).

I learned a lot about the behavior of FET-based bus switches, and the importance of biasing them around half way between the on and off  levels of the switching/control waveform (e.g. if you want the switching/control waveform to do a good job of opening and closing the path from source-to-drain, you want the control voltage to be at least 1 volt higher than the signal passing through when the switch is in the on state... and you want the control voltage to be at least 1 volt lower than the signal being blocked when the switch is in the off state.

I learned a lot about Tayloe mixers...  and I discovered that the principle behind the Tayloe mixer is directly applicable to a research project I am working on here at the university: i.e. a mixer that shifts a 1THz (1000 GHz) signal down to the microwave range, using a switch that is controlled by an optical local oscillator (whose power level is modulated at the THz rate). The modulated optical signal modulates the carrier density in a material with an extremely short recombination lifetime... thus modulating the resistance of the path that the THz input signal takes from input to output.

I admit that one has to be somewhat compulsive to spend (at least) 10 hours thinking about a technical problem... but I also believe that every time you work on a technical problem, and get engaged in understanding what you don't understand, you learn a tremendous amount that is useful in other technical situations you will encounter. This reconfirms my view that students who only think about the technical stuff they are being presented with when they are preparing for an exam get essentially nothing out of the courses they are taking. Students that feel compelled (and who have the time available to them) to think about things they don't understand, until they understand them, get value out of the courses they take. Unfortunately, many of today's students are either unmotivated to think about the stuff they are being presented with... or they are working full time while they are taking a full course load... and they don't have the time to think about stuff that they don't understand.

Stu

[WA1GFZ]

Then there are us weirdo hams

The guy who owns this lab has a full sized 3 element 80 meter beam.