Opamp circuit in Tayloe mixer

[seriouscoinage]

I'm looking at making a software-defined radio using the Tayloe mixer, which there's plenty of information about on the net. It basically uses a fast CMOS switch to switch the RF input between four capacitors at the local oscillator frequency. An in-phase and quadrature output is available by taking the differential voltage between pairs of capacitors. Here's a circuit similar to what I want to do:


My question is about the way that the two op-amps U1 and U2 are used to get the differential voltage. Dan Tayloe shows how this circuit is developed from the standard op-amp differential amplifier circuit (http://www.wparc.us/presentations/SDR-2-19-2013/Tayloe_mixer_x3a.pdf - page 4-6). The circuit basically makes sense - the feedback resistor and the output impedance of the switching node set the gain of the amplifier. However, there is one thing that doesn't make sense to me. In this circuit, there is a capacitor directly at the inverting input of the op-amp. I thought this seemed odd, so I simulated a similar circuit:


The capacitance at the inverting input causes a large peak and then a 2nd order rolloff instead of the normal RC filter response I would expect.

Is the circuit as shown by Tayloe (and used in many other circuits online) incorrect? Am I misunderstanding something about the way the Tayloe mixer/circuit works? If there is something wrong with this circuit, what's a better way to get the differential voltage from the capacitors with low noise? Thanks for any light anyone could shed on this issue.

[bktemp]

You did omit C2/C3. They reduce the bandwith and compensate the effect generated by C4/C19.

[seriouscoinage]

Adding a 10nF capacitor in parallel with the feedback resistor yields this frequency response:

Which has the correct rolloff frequency as governed by the RC filters at the switching node, but then has a large peak at a higher frequency. Decreasing the parallel feedback capacitor to around 470pF gives the following:

Which has a well-behaved rolloff - at the wrong frequency (the Tayloe mixer is meant to have a -3dB point related only to the input resistance and switching capacitor values.) Other than that, I suppose it would work, but it's odd that the circuit I posted has completely wrong values for these capacitors. The paper by Tayloe doesn't even mention any compensation capacitors for the op-amps - does this mean that there's something I'm not taking into account that means the issue doesn't matter, or did Tayloe overlook this issue?

[Zad]

As an aside, Mr Tayloe took an existing idea (the switched commutator) which had been invented long ago, and attempted to give it his name. There is a history of this- It's common name is the Gilbert Cell (after Barrie Gilbert), but was actually the invention of Howard Jones in 1963!

For better performance, it is common to use a high speed mux chip or pair of fast N/P mosfets.

[bktemp]

Putting a capacitor on the inverting input of an opamp is simply wrong: The circuit becomes instable because it sets the opamp to infinite gain at high frequencies. That's peak at 100kHz.
The capacitor should be connected between the inverting input and output. But this creates another problem: The opamp is probably too slow for the rf/mixing frequency. Therefore many mixers add a seperate lowpass filter between the multiplexer and the differential amplifier to attenuate the rf signal before it enteres the opamp.
In addtion to the capacitor, the simplified differential amplifier has another problem: The feedback resistor at the inverting input forms a voltage divider with the source impedance. But at the non inverting input there is no resistor, therefore the voltage is higher at this point. The ouput of the differential amplifier is not exactly the difference. This should result in a bad image rejection.
Instead of improving the circuit, Mr Tayloe made it worse...

[Yansi]

Well... this thread will be interesting! I have already one thread here with many problems with this type of mixer. (tl,dr: does not work at all)

Some of my questions were never answered, like why  200MHz BW or so opamp is usually used on the output, when the baseband is just few kHz?

I will definitely watch this thread and try to learn something from it.

[seriouscoinage]

As an aside, Mr Tayloe took an existing idea (the switched commutator) which had been invented long ago, and attempted to give it his name. There is a history of this- It's common name is the Gilbert Cell (after Barrie Gilbert), but was actually the invention of Howard Jones in 1963!

For better performance, it is common to use a high speed mux chip or pair of fast N/P mosfets.

Yeah - I was planning to use a bus-switch chip like Tayloe shows in his paper. I know about the Gilbert Cell, but I'm didn't know it was considered a commutating mixer.

Putting a capacitor on the inverting input of an opamp is simply wrong: The circuit becomes instable because it sets the opamp to infinite gain at high frequencies. That's peak at 100kHz.
The capacitor should be connected between the inverting input and output. But this creates another problem: The opamp is probably too slow for the rf/mixing frequency. Therefore many mixers add a seperate lowpass filter between the multiplexer and the differential amplifier to attenuate the rf signal before it enteres the opamp.
In addtion to the capacitor, the simplified differential amplifier has another problem: The feedback resistor at the inverting input forms a voltage divider with the source impedance. But at the non inverting input there is no resistor, therefore the voltage is higher at this point. The ouput of the differential amplifier is not exactly the difference. This should result in a bad image rejection.
Instead of improving the circuit, Mr Tayloe made it worse...

Thanks for your insight. One note - the mixer (consisting of the source impedance and switch capacitance) forms a lowpass filter, with a rolloff typically starting from a few kHz to a few tens of kHz, which means that the op-amp shouldn't see too much high-frequency signal at its input - especially considering the op-amp used is reasonably fast.

You are right about the problems with the simplified differential amplifier - the more I think about it, I think it is a fatally flawed idea. I think I will use a two op-amp instrumentation amplifier as shown here: http://www.linear.com/solutions/1574. This should sidestep all of the problems of the circuit I showed above, while also making it so the gain of the circuit doesn't depend on the input impedance (this kind of bothered me about the circuit...)

Well... this thread will be interesting! I have already one thread here with many problems with this type of mixer. (tl,dr: does not work at all)

Some of my questions were never answered, like why  200MHz BW or so opamp is usually used on the output, when the baseband is just few kHz?

I will definitely watch this thread and try to learn something from it.

If I use an instrumentation amplifier instead of the "simplified differential" amplifier, do you think the mixer will work well? I'm not an expert, but the theory seems sound, even if the specific implementation with the odd op-amp circuit is wrong.

The op-amp shown in the circuit in my first post has a GBW of 40MHz, and with a gain of 100 it has a bandwidth of 400kHz. This is rather higher than the baseband (a few kHz), like you said, but doesn't seem ridiculous either. The rolloff of the mixer response is only -6dB an octave, so it seems like a good idea to make sure than the op-amp deals properly with reasonably high frequency signals - but that's only my intuition about this circuit, I could be quite wrong.

I think the LT1115 in the circuit was chosen more on the merit of its low noise rather than its GBW, though.

[seriouscoinage]

This is the solution I think I'm going to use. It's about as simple as it gets, I think.


This works fine and the gain only depends on the gain setting resistors, not the source impedance. However, the mixer output bandwidth is still dependent on the source impedance, which sort of bothers me. Is there some way to make sure the source impedance stays fairly close to 50 ohms over a wide bandwidth? Would a buffer amplifier of some kind add too much noise?

[Yansi]

I wouldn't recommend using this kind of instrumentation amp. Simply because the phase characteristic is different for each input leg. The Vsw2 signal is delayed on the U2 input, which will cause trouble the higher the frequency is on the input.

Simply put,  the commonmode rejection of this circuit will drop fast when the commonmode frequency will rise. Maybe you can try simulate that to, just for curiosity..

[seriouscoinage]

It's good that you mentioned common mode rejection, because the circuit that I posted in my last post has literally no common mode rejection at all  . This is the correct circuit:

Which has common-mode rejection that looks like this:

The low-frequency common mode rejection will probably be much worse in real life due to resistor mismatch, etc. The high-frequency CMRR looks fine, but that is only because the RC circuit is attenuating the high frequencies at the inputs. Without the capacitors, the CMRR looks like this:

As you say, the CMRR degrades quickly at high frequencies, as the bandwidth of U1 is much wider of U2. I'm not sure which of these graphs is actually relevant to my problem though - it seems like any common mode noise at the amplifier input would be subject to the high-frequency attenuation of the RC filter, so the bad CMRR of the amplifier at high frequencies shouldn't matter. In addition, I will be adding more filtering after this circuit to make sure no aliasing happens in the audio ADC, which should attenuate any high frequency signals, common mode or otherwise.

If this circuit isn't appropriate though, what would you recommend? The ordinary three op-amp instrumentation amplifier? Something like this: http://www.analog.com/media/en/technical-documentation/data-sheets/AD8429.pdf might be good... I definitely need to do more research.

[Yansi]

I dunno, personally I'd be glad if my Tayloe mixer worked at all, with the single opamp circuit. 

What about this one? http://www.analog.com/media/en/technical-documentation/data-sheets/AD620.pdf
120kHz BW at 40dB gain. (not bad!) And also have seen them on ebay/Ali for reasonable price.

[seriouscoinage]

I think I will use this one: http://www.analog.com/media/en/technical-documentation/data-sheets/AD8421.pdf. Runs from a 5V supply, good bandwidth, (2MHz at 40dB,) fairly low noise. I think a discrete instrumentation amp would work fine, but using an integrated one gives me less opportunities to screw up, and doesn't seem to be more expensive than the equivalent op-amps I'd need. Probably best not to overthink it - people have gotten the (apparently flawed) circuit I posted in the first post to work, so I can't do much worse than that (famous last words?)

[Yansi]

I'd rather think how to make it work correctly with jellybean parts, rather than choosing fancy inamps. But just my opinion, as none of my two experiments with this type of mixer worked properly. To be exact, one didn't work at all, the second was barely usable.

It would be nice to solve a correct circuit topology without flaws first and then start to exchange parts for better one. Or do we have the circuit solved now completely?

[seriouscoinage]

I had a look at your own thread on SDR, and my project is not so different from yours. I want to build an SDR that does the HF band from 3-30MHz. First I will build a board with the LO (Si570 from 12-120MHz) and mixer on board, which is usable on its own using a PC soundcard, and then look at building a standalone microcontroller-based demodulator. Only difference is I plan to use a 11kHz IF instead of doing a direct-conversion design.

I don't really have any direct experience with this circuit yet, so I can't say for sure why your circuit didn't work properly, or if what I'm changing will fix it. I understand the concern about using fancy high-performance parts instead of more commonly available ones - I guess I'm working on the assumption that the actual mixer circuit is correct, and it was just the op-amp arrangement that needed to be fixed, so I don't see the problem in putting some nice parts after the mixer... Maybe I will pay for my overconfidence.

[Yansi]

Oh noezzzz  Si570.   What about Si5351A?  Compare the price! And also if I remember right, it is possible to generate quadrature signal directly. I have bought few pieces of this interesting chip, but still laying in the drawer, no spare time to even test it.
The only advantage of the 570 being it's very low jitter. Dunno, how much issues could those <70ps from the 5351 bring in the design. (like raising the noise floor of the receiver... but I'm no expert on radio)

I am usually quite suspicious about these simple circuits with almost miraculous properties. I wish you the best to sort the circuit out! I might get some spare time soon (in Summer) to try some more progress with my SDR.

[seriouscoinage]

I think you're right - I believe that phase noise increases the noise at the output of your mixer because the frequency content in the LO that is "spread out" from the centre frequency mixes with frequencies to the side of the signal you want to receive, mixing noise and interfering channels at close frequencies down into your baseband. I'm no radio expert either, but that's my understanding of it.

The Si570 does have better performance than the Si5351, but to be honest the main reason I chose the Si570 is because there was more information about it being used in successful SDR project - I haven't had any experience with SDR yet so I thought it was better to start with a proven solution. To be honest I don't know if the better phase noise of the Si570 even matters in this case - I think experimenting with the Si5351 is a good idea, considering it's about 1/10th the cost.

I will definitely post my results when I put this circuit together - I have ordered the boards and parts, which should all be here in 2-3 weeks or so.