Mixer noise figure

[rfguy2020]

Hi,

I'm designing RF receiver, which need to tolerate relatively high input power (~20dBm). I have been using MAX2021 IQ demodulator, which has high P1dB (30dBm). Problem is the noise. According to datasheet, NF is 19dB with 11dBm blocker. It is expected that NF will increase at 20dBm to ~30dB even though signal level is 10dB below P1dB.

I'm looking for alternative by using discrete mixers to see if I can improve the noise issue. I'm looking minicircuits appnote: https://www.minicircuits.com/appdoc/AN00-010.html?srsltid=AfmBOopvJ-Xw_gk1eQKPgeVJRe298EOmUNFCUei1CDaibiEs5TyGFjyG
It does not say anything about the noise but LO power should be 10dB higher than RF signal when using diode mixer. In case of FET mixer, I should use LO power 3dB higher than RF power. It sounds like FET mixer would be better choice, but I'm not sure what happens to noise.

Then I was looking Marki microwave site: https://markimicrowave.com/technical-resources/white-papers/mixer-basics-primer/
It says that "FET and CMOS mixers are typically used in higher volume applications where cost is the main driver and performance is less important. For the more challenging, high performance applications, Schottky diode mixers are used almost exclusively." It sounds like diode mixer is better choice.

Anyway, should I look transistor or diode based mixer for my application?

[szoftveres]

What is your application that requires linear handling of +20dBm RF power while having to operate at low noise figure? Usually these are contradicting requirements.

[ejeffrey]

Problem is the noise. According to datasheet, NF is 19dB with 11dBm blocker. It is expected that NF will increase at 20dBm to ~30dB even though signal level is 10dB below P1dB.

This is a tall order for any mixer.  That's a lot dynamic range you are looking for out out a fundamentally non-linear device.  9 dB of that noise figure is strictly the insertion loss.  Not sure exactly what you are trying to do, but I would really look into using linear filters to separate your blocker from your signal, then add some low noise gain to the small signal before mixing it.

Also, are you sure that it's even possible?  Presumably your 20 dBm blocker is going to have some wideband noise floor.  That is noise that the mixer can't even do anything about.

[KE5FX]

Traditionally you use an LNA to solve noise figure problems, and a diode or FET ring mixer to deal with dynamic range problems caused by strong signals.

[mawyatt]

There's a very interesting type of passive mixer that's based upon a multiple phased LO called PolyPhase or N-Path Mixer. We starting working with this based in early 2000, later DARPA got involved, now it utilized in some advanced commercial SDR receivers, even understand Apple is utilizing this.

https://www.eevblog.com/forum/rf-microwave/polyphase-or-n-path-mixer/

This mixer has some very unique and physics defying properties, like below 2dB NF (you can't do this, a passive mixer must have a NF > 3.92dB, yet decades ago measurements revealed sub 2dB NF), very high IIP, antenna impedance matching that follows LO without any "tunable" elements, direct down-converted I and Q outputs and so on.

If interested please spend some serious time studying the mentioned IEEE papers, and likely many more now, it's a very deep dive

We can discuss more if OP is interested.

Best 

[tggzzz]

There's a very interesting type of passive mixer that's based upon a multiple phased LO called PolyPhase or N-Path Mixer. We starting working with this based in early 2000, later DARPA got involved, now it utilized in some advanced commercial SDR receivers, even understand Apple is utilizing this.

https://www.eevblog.com/forum/rf-microwave/polyphase-or-n-path-mixer/

This mixer has some very unique and physics defying properties, like below 2dB NF (you can't do this, a passive mixer must have a NF > 3.92dB, yet decades ago measurements revealed sub 2dB NF), very high IIP, antenna impedance matching that follows LO without any "tunable" elements, direct down-converted I and Q outputs and so on.

If interested please spend some serious time studying the mentioned IEEE papers, and likely many more now, it's a very deep dive

Seconded.

The applicability will depend, to some degree, on your centre frequency and bandwidth. If the technology is at all appropriate for your use cse, then that thread is worth studying in some detail.

Another keyword is "Tayloe mixer", after a person who obtained a patent.

I'll note that it is easy to create a simple N-path filter using discrete components (4 or 8 capacitors, one resistor, and a discrete IC analogue switch mux/demux [1]). Playing around with it will give a gut feel for its operation which is, IMHO, valuable when trying to understand the theory.

[1] I did that in ~1979; components and frequencies have improved significantly When I tried to explain its operation back then, people couldn't understand the explanation. Clearly my teaching skill has improved since then 

[mawyatt]

Seconded.

The applicability will depend, to some degree, on your centre frequency and bandwidth. If the technology is at all appropriate for your use cse, then that thread is worth studying in some detail.

Another keyword s "Tayloe mixer", after a person who obtained a patent.

I'll note that it is easy to create a simple N-path filter using discrete components (4 or 8 capacitors, one resistor, and a discrete IC analogue switch mux/demux [1]). Playing around with it will give a gut feel for its operation which is, IMHO, valuable when trying to understand the theory.

[1] I did that in ~1979; components and frequencies have improved significantly When I tried to explain its operation back then, people couldn't understand the explanation. Clearly my teaching skill has improved since then 

Actually Tayloe's patent was on a demodulator, it's even referred to as a "Detector" and didn't address the unique properties of the PolyPhase or N-Path "Mixer" whichever one prefers to call it.

It's interesting as you mention the N-Path Filter which dates back to the 60s I believe and was called a Commutating Filter back then. We utilized this filter to detect narrowband "hidden" signals within Microwave signals around 80 and never realized the Bi-Directional nature of the technique was basically a down and up (or up and down) frequency conversion with a passive Low Pass Filter in-between. So cutting it in half with some weighted recombination and you have the origins of the PolyPhase or N-Path Mixer!!

Drs Molnar and Andrews two IEEE papers created an academic whirlwind that even DARPA got involved, and never recall DARPA having an entire workshop dedicated to a circuit before!!

Likely all this academic interest was caused by the physics defying sub 2dB NF, and all the other unique properties certainly helped

Best

[tggzzz]

Yes, this was the article I found, somehow.... https://archive.org/details/bstj39-5-1321

I used the narrowband filter with very sharp skirts (can be >>10000dB/decade) to reduce noise. Q of ~4000 with 10% capacitors  what's not to like!

I too missed the mixer properties, but I knew it was a circuit that had "more in it". I couldn't find a professional use for it, and after I retired I found Tayloe's paper. Blast!

The commutation technique appears to have been triumphantly rediscovered once a decade or so. ISTR seeing an article from the late 40s; I suspect it was referenced in one of the papers you noted in you thread https://www.eevblog.com/forum/rf-microwave/polyphase-or-n-path-mixer/

[mawyatt]

The commutation technique appears to have been triumphantly rediscovered once a decade or so. ISTR seeing an article from the late 40s; I suspect it was referenced in one of the papers you noted in you thread https://www.eevblog.com/forum/rf-microwave/polyphase-or-n-path-mixer/

Vaguely remember an old paper that used a pair of car distributors driven by a synchronous motor from the mains. One distributor was the Input and the other the Output and shunt capacitors to ground from the spark plug wire terminals, and a resistor from the coil terminals of each distributor serving as the Input and Output. This is probably one of the earliest uses of Discrete Time Continuous Amplitude (DTCA) signal processing.

Best

[mtwieg]

As much as I love N-path mixers, I highly doubt that's going to help the OP. The MAX2021 they're trying to replace is specced for 650-1200MHz. An N-path mixer working for such a high frequency would require an ASIC, and most of us don't have $50K in the couch cushions.

What is your application that requires linear handling of +20dBm RF power while having to operate at low noise figure? Usually these are contradicting requirements.

I was going to ask the same. Such a high RF power usually means:
1. The RF has already been amplified quite a lot, making the NF of the mixer irrelevant
2. The signal source actually has an enormous dynamic range, which means this is a very novel application, need more info to help.

[mawyatt]

As much as I love N-path mixers, I highly doubt that's going to help the OP. The MAX2021 they're trying to replace is specced for 650-1200MHz. An N-path mixer working for such a high frequency would require an ASIC, and most of us don't have $50K in the couch cushions.

Yes, for discrete use one is limited by the logic speed for the multiple phased LO.

However, one thing we never did try was utilizing the harmonic response of the N-Path Mixer (PPM). Might be worthwhile for the OP (or other interested party) to investigate such, but note the N-Path doesn't follow the usual harmonic response of typical passive mixers, as best to review the mentioned IEEE papers (maybe this has been tried but we've been out of "loop" almost a couple of decades). So sort of using the N-Path sampling as an RF/MW Sub-Sampling process in addition to the multitude of unique features the PPM offers!!

Best

[coppercone2]

I think someone needs to try to make a n path mixer.

It makes me think of those potato semiconductor picosecond logic elements

[tggzzz]

I think someone needs to try to make a n path mixer.

It makes me think of those potato semiconductor picosecond logic elements

I think you'll find some of us have built n-path filters/mixers.

If you mean the Potato Semiconductor analogue bus switches, they don't specify parameters such as switching time and charge injection. Their clock drivers do specify transition times (of 1ns), which is well controlled rather than fast.

[mawyatt]

I think someone needs to try to make a n path mixer.

It makes me think of those potato semiconductor picosecond logic elements

Many folks have built PPM or N-Path Mixers long ago (see image of Cornell chip in 65nM CMOS), see the related link earlier. Apple acquired Passiff (small company that Dr Andrews was with) over a decade ago, rumor has it this is in many Apple products today. Believe Dr Nada posted a video on the PPM some time ago, and many IEEE articles on such as well

Our comment was has anyone tried utilizing the Harmonic Content of the PPM as a sub-sampling RF/MW down converter along with all the unique PPM features. To understand this one needs to visit the mentioned IEEE articles as this is not like using a conventional passive mixer as a sub-sampler 

Best

[mawyatt]

As much as I love N-path mixers, I highly doubt that's going to help the OP. The MAX2021 they're trying to replace is specced for 650-1200MHz. An N-path mixer working for such a high frequency would require an ASIC, and most of us don't have $50K in the couch cushions.

What is your application that requires linear handling of +20dBm RF power while having to operate at low noise figure? Usually these are contradicting requirements.

I was going to ask the same. Such a high RF power usually means:
1. The RF has already been amplified quite a lot, making the NF of the mixer irrelevant
2. The signal source actually has an enormous dynamic range, which means this is a very novel application, need more info to help.

You'll need a lot more than $50K!! The last chip development we were involved with over a decade ago was >$50M total, however this was a ~Billion Transistor chip

Today's SOTA chip developments are likely pushing $1B with the nanometer feature CMOS processes

Best

[iMo]

..,,
..Today's SOTA chip developments are likely pushing $1B with the nanometer feature CMOS processes

That apply perhaps for the US DoD SOTA projects, but not for "real life" scenarios imho. Unless you mess with some new exotic R&D for normal commercial customers the today's chip developments get way much more cheaper..

[mawyatt]

..,,
..Today's SOTA chip developments are likely pushing $1B with the nanometer feature CMOS processes

That apply perhaps for the US DoD SOTA projects, but not for "real life" scenarios imho. Unless you mess with some new exotic R&D for normal commercial customers the today's chip developments get way much more cheaper..

Actually this applies to commercial SOTA chips like the ones Apple and Nividia have developed (and are developing), it's much too expensive even for USG DARPA efforts

General use/purpose chips (not massive processors) of course are much cheaper, altho if in a SOTA process the masks alone can cost $30 to 40M, even older processes the masks are very costly.

Modern chip development is not a place where mistakes are tolerated, it's far too costly and career limiting!!

BTW the PPM chip shown above was developed in 65nm TSMC CMOS in 2007~8 and far from a SOTA process. That development was't about the SOTA process but the SOTA new circuit, and probably cost ~$5M all toll. The other chip mentioned which was ~$50M development cost was in a 20 or 28nm CMOS process (can't remember), and the 2nd most complex chip DARPA had developed at the time and a very large chip with almost 1B devices. That chip implemented a specialize algorithm developed at MIT Lincoln Labs, and worked 1st time too

Here's a couple images of that chip.

Best

[rfguy2020]

The application is kind of radar. I have up to 33dBm transmit power in monostatic configuration. If I have the antenna with poor impedance matching, my received power can be level of 20dBm, but I'm interesting only of signal which has level  around -70dBm.   

[mtwieg]

The application is kind of radar. I have up to 33dBm transmit power in monostatic configuration. If I have the antenna with poor impedance matching, my received power can be level of 20dBm, but I'm interesting only of signal which has level  around -70dBm.

Ok so if you're only interested in low level signals then you shouldn't need a receiver with a high P1dB. You just need the receiver to withstand the excess power without being damaged, and which recovers from the overload quickly. Hard to give specific advice without knowing how your transmitter and receiver interface to the antenna, but there are likely ways to reduce the overload seen by the receiver (T/R switches, circulators, limiters, etc).

[szoftveres]

The application is kind of radar. I have up to 33dBm transmit power in monostatic configuration. If I have the antenna with poor impedance matching, my received power can be level of 20dBm, but I'm interesting only of signal which has level  around -70dBm.

Search for 'Monostatic FMCW radar leakage cancellation' - the signal reflecting from the antenna is -not surprisingly- fully coherent with the Tx signal (duh..) so it can be cancelled out with opposite phase same amplitude Tx signal, in place on the board.

[rfguy2020]

The application is kind of radar. I have up to 33dBm transmit power in monostatic configuration. If I have the antenna with poor impedance matching, my received power can be level of 20dBm, but I'm interesting only of signal which has level  around -70dBm.

Search for 'Monostatic FMCW radar leakage cancellation' - the signal reflecting from the antenna is -not surprisingly- fully coherent with the Tx signal (duh..) so it can be cancelled out with opposite phase same amplitude Tx signal, in place on the board.

Yes. I'm familiar with the carrier cancellation techniques. Carrier cancellation is not possible in this case and I need to make receiver as good as possible without the cancellation. I was wondering if there are difference in terms of noise in passive diode mixer vs. FET mixer?

[mawyatt]

Yes. I'm familiar with the carrier cancellation techniques. Carrier cancellation is not possible in this case and I need to make receiver as good as possible without the cancellation. I was wondering if there are difference in terms of noise in passive diode mixer vs. FET mixer?

If the switching device in the "mixer" also switches a bias current like in the Gilbert Mixers this creates additional shot noise, so passive mixers are preferred in low noise applications, and why the "diode mixers" are so popular, while utilizing a FET as a passive switch to replace the diode is an option. The lowest noise passive mixers have a theoretical Noise Figure of 3.92dB and thus the limit of noise performance for any type mixer, except for the somewhat recently discovered PolyPhase or N-Path Mixers. In very low noise receivers the mixer noise is hidden by the lower noise LNA, however this brings another set of issues that the designer must deal with.

We encourage you to study the passive PPM mentioned, it has some unique and remarkable features including "Mixer First" (much later title of DARPA workshop on PPM) receiver designs (see Cornell chip image). This passive mixer pushes a large shunt capacitor right up on the input (antenna) port and out-of-band signals "see" a very low shunt impedance looking into the antenna port and are reflected. From a pure signal processing standpoint this is exactly what you want to do, push the band-limiting filter as close to the input as possible, however this usually isn't possible as the filter may be too narrow to realize with a conventional preselector filter, which doesn't track the desired input frequency in frequency agile receivers.

The passive PPM solution solves these issues by using passive low Z capacitance shunt filtering imposed right at the input that tracks the LO over multiple frequency decades while achieving input impedance matching tracking without inductive or tunable elements. Thus usually no preselect filter is required and produces sub 2dB noise figures without an LNA, and provides a direct-down conversion (Zero IF) to I and Q at baseband!!

Best

[ejeffrey]

Yes. I'm familiar with the carrier cancellation techniques. Carrier cancellation is not possible in this case and I need to make receiver as good as possible without the cancellation. I was wondering if there are difference in terms of noise in passive diode mixer vs. FET mixer?

It sounds like this a Doppler or chirped system where the drive signal can overlap the signal in time but at a different frequency?  Can you prefilter the transmit with a notch filter?

It's really hard to get that much dynamic range out of a mixer.  The blocker signal you have is easily enough to partially turn on or off the switches  -- whether they are diodes or fets.  This creates high order distortion products between the LO and the blocker. Distortion is responsible for the noise in a passive mixer.  What happens is that for every distortion product mixing input signals to unwanted output spurs, there is a conjugate term that mixes out of band input noise into your signal band.

Direct digitization might help.  ADCs are more inherently linear than a mixer.  But that's a lot more complicated than just keeping your blocking signal out of the mixer.

[mawyatt]

Here's a spin-off of the PPM that yielded a fully integrated CMOS Duplexer at Cornell by Dr Apsel, OP and some folks might find this interesting.

Yuksel, H., D. Yang, Z. Boynton, E. Enroth, T. Tapen, A. Molnar, Alyssa B. Apsel. 2016."Broadly Tunable Frequency Division Duplex Transceiver: Theory and Operation." Paper presented at IEEE ICECS, Monte Carlo, December,

Yang, D., H. Yuksel, C. Newman, C. Lee, Z. Boynton, N. Paya, M. Pedrone, Alyssa B. Apsel, A. Molnar. 2016."A Fully Integrated Software-Defined FDD Transceiver Tunable from 0.3-to-1.6 GHz." Paper presented at IEEE RFIC Symposium. Nominated for best paper award, June.

https://ieeexplore.ieee.org/document/10160119

Also here's the most consolidated source for PPM (N-Path) we've found, not surprised it's Dr Molnar's group at Cornell and still quite a bit of activity

https://molnargroup.ece.cornell.edu/publications/

Best