Double Balanced Mixer - Very Linear ?

For "very linear" the diode-ring mixers (and, if you really want to work at it, FET-ring mixers) have been recognized to be superior to Gilbert cell mixers for years and years. I doubt that has changed, although knowing whether a diode ring is superior to a FET ring is beyond my pay scale at the moment.

Gilbert cell mixers are better if you need to build an inexpensive small circuit that doesn't consume much power, though.

--

Tim Wescott 
Wescott Design Services 
http://www.wescottdesign.com

How does it work?

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

How does it work?

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

You can get pretty accurate 90 degree phase shifts across a decade of frequency using a fourth-order quadrature hybrid, but I really wonder if it wouldn't be easier to generate 5.1-15MHz SSB and downconvert it to

0.1-10MHz.

I'm playing with a first-order QH for antenna phasing in a DF system, which the simulations say will cancel well from 2.5-5MHz, with a += 20 degree error (that can be compensated because it's a fixed function of frequency). A fourth-order version should work similarly from 3-30MHz using the same antennae, as long as the RF levels are matched (the current design problem).

Robert, I'm making a separate post about an oscillator simulation problem I'm having with LTSpice and the Intersil CA3046 model. If you have a better model you can send to clifford dot heath at gmail, I think that might solve my problem. otherwise I'd appreciate your comments on that thread (about to post it).

Clifford Heath.

Robert,

The 10dBm drive means that the diodes go through the partly-on (non-linear) stage quickly, so they spend most of the time either fully on or fully off.

The

Trouble with the *3046 is you don't get enough transistors for a Gilbert cell in one device, so you may have matching problems. If you already have an input transformer, you can make is centre-tapped and feed the emitters of two pairs, but then you're relying on transformer matching not silicon matching again.

Effectively the same silicon (but bonded out differently) as the LM3046 (like CA3046) is available as a pre-wired Gilbert cell in the LM3101. If you don't mind paying $5 you can also get the rocket-ship HFA3101, which uses 6GHz transistors and is designed for 900MHz mixers.

The mixer in the SA601A/SA612A would probably also suit your purposes, works to VHF and is

Transformer matching can be spectacularly good. Bifilar windings (and trans mission line transformers wound with coax are a particularly fine example o f the breed) offer matching to one part in a billion. Ratio transformers (w here you have more than two wires in the wound bundle) can be good to one p art in ten million.

es

--
Bill Sloman, Sydney

Use octopart.com, not google. Google doesn't know how to sub on/off trailing letters on chip names.

In Chrome, after having used octopart, I just get a new tab, type octCHIP into the URL box. For any chip that's still available, you get prices and almost always a datasheet link in one go.

[snip]

You might want to look at the Intersil HFA3101

Also similar (but with no complete schematic), LT5560 and AD8343.

Chris

I used to work on this sort of stuff. I put a little bit of info (perhaps about as much as I can) about stuff I worked on here:

I have simulated mixers generally, a lot, using ADICE which is by far the nicest simulator I have ever used, but which I sadly don't have access to any longer. LTSpice is now the best thing I have access to and is very useful.

I haven't simulated that scheme I discussed above though.

From what you say it sounds like you are a developer of LTSpice which would mean you work at LT, is that right?

Chris

got it thanks, and I replied from that email address, too

very well. I can't get 'inside' the LTspice machine so .tranoise tends to slow things down a bit. around 5 to 10 times longer than the 'normal' speed of .tran

We still have the problem of low frequencies. The OP wanted a band starting from DC, which is IMHO impossible. Even with a low limit of 10 Hz, the required bandwidth exceeds by a wide margin those which have been realized. I have the feeling that it is in practice impossible to get analog components that are matched well enough for the entire baseband width. Digitally, it may be possible if the band is limited at the low end, but the filters will be very long to provide the necessary delay difference at the low end of the range with the sampling rate needed for the high end.

There are other methods with e.g. matched FIR filters, polyphase RC nets and mixing in the Weaver method.

For the output band, the standard method is to generate the signal at an high IF well above the upper limit of the output band and then downconverting. There is an advantage to the final mixer and output filter to go above 3 times the high end of the output band, e.g. 30.1 to 40 MHz for 0.1 to 10 MHz output.

--

-TV

On 3.4.14 21:21, Tim Wescott wrote: .

The last advance in very linear mixers seems to be using CMOS analog switches to do the switching instead of transistors or diodes.

Have a look at .

--

-TV

is

should

Well most double balanced mixers are very similar to Gilbert cell multipliers. The very wide bandwidth makes phasing type twin mixers impossible for conversion to SSB. The remaining choice it to mix to a much higher frequency (1 GHz) and use a SAW filter or similar to split the sidebands and mix down again with heavy filtering to do the image rejection.

You need some more slack in the bandwidths to make this reasonable.

?-)

On a sunny day (Thu, 03 Apr 2014 19:47:16 -0700) it happened RobertMacy wrote in :

No reply received as of this morning. try double_balanced_mixer at p a n t e l t j e dot com obvious space removed and at and dot replaced by .

Tim, do you have any idea why that is. "In theory" I would think the Gilbert cell would be better, 'cause it doesn't have all the odd harmonics that the switchers have. In practice I can imagine that switching is much easier than multiplying.

George H.

Sounds like you want a single chip, But if not, here is the design I always think about think about when I hear high performance , wide dynamic range receiver.

In following years new quad mixer chips have come out for further improvements, that replace the SD8901. I don't know if the later mixer chips make it to 100MHz. I suspect they do. I originally saw this in QST, I think this might be the author of that article. Mikek

Boom-kssssshhhh.... I meant, "how does it work", not "how well does it work."

Doing that in any generality for strongly nonlinear circuits is an intrinsically difficult problem, or so I would have thought.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net