Comparing a "phase" at 3MHz

[Infraviolet]

Another separate trouble in my current project... Hopefully this should be the last nasty surprise of the high frequency world which I discover in the course of this project.

I've got a 3MHz sine wave signal (the output of the 3MHz amplifier I was working on, thanks for the help there) and I need to compare whether it is roughly in phase, or in anti-phase with a 3MHz square wave. That is to say I'm trying to check whether the 3MHz sine wave is in phase with the square wave, or inverted relative to it.

In an ideal world I'd use op amps, but these signals are proving too fast for any I have to hand.

I've also failed with comparators (LM393), seems the comparators I have are too slow to operate from the signal.

I tried using a 4066 analog switch IC with the square wave on the control pin and then feeding the sine wave in to one of the switched pins with the other switched pin pulled to ground and meaurements taken there. But the 4066 seems to induce huge distortions in signals this fast, capacitance somewhere I think judging by the shapes of the severaly distorted waveform.

I also considered a 4066 for sample and hold purposes at the time of a peak, but again the distortion it introduced was enormous, large than even the strongest signal.

I thought of trying to massively amplify the sine wave and then use logic gates, but the amount of amplification needed would be very large indeed, as I need to be able to know if it is in phase or antiphase from signal strengths of +/- 20mV peak to peak to 1.4V peak to peak. Logic gate ICs look like a good solution or these speeds, but I'd somehow have to get the signal up to logic levels of voltage, for any input signal size. Is there any easy way to make a crude comparator (input signal versus a carefully selected static voltage rail) from transistors?

What should I try, the trouble is that at a 3MHz frequency everything either I'm used to can't keep up (op amps...), and capacitors cease to be something to damp out fast vibrations and instead become plain short circuits.

What avenues of approach have I got for this problem? All the things I can think of only work at lower frequencies.

Thanks

P.S. to be clear the phase is only ever in phase or in antiphase, its not quite perfectly so its a few tens of degrees out, but this situation is still evry much a digital output of in or anti, not a continuous output of phase difference.

[radiolistener]

I've got a 3MHz sine wave signal (the output of the 3MHz amplifier I was working on, thanks for the help there) and I need to compare whether it is roughly in phase, or in anti-phase with a 3MHz square wave. That is to say I'm trying to check whether the 3MHz sine wave is in phase with the square wave, or inverted relative to it.

just use second channel on oscilloscope. But first check and calibrate it with your cables, because there is a phase delay in the cable, so if you use cables with different electrical length they will have different phase delay, you're needs to take it into account. Use square wave with a sharp edge for phase delay calibration in order to get more precise result.

Since you're want to compare phase between sine wave and square wave, I suggest to use trigger for sine wave, so it's zero crossing will be exactly at the center of the oscilloscope screen and you can easily find delay between screen center and square wave edge.

[tggzzz]

It would help if you mentioned the Vac and Vdc of the signals.

Comparators are slower responding with small differential signals (i.e. small Vac). Probably the simplest solution is to get faster comparators.

Alternatively, if your two input signals are V1 and V2, then create an inverted version of one of them, say V2I. Then, depending on the phase, either V1+V2 or V1+V2I will be roughly constant and the other will exhibit larger voltage swings.

[dmills]

Diode ring mixer?

Fast comparator and nand gate then lowpass will give you a voltage proportional to phase difference.
Lots of ways to do this, just remember that your 3MHz square wave will have a mess of harmonics out to several tens of MHz and you need to respect that in the layout. 

[Infraviolet]

radiolistener, sorry, should have been mroe clear, I've already done that with the scope for checking the signal, this is about doing it in a circuit such that it outputs some simple (varying only when the relative phases of the square wave and the sine wave change, not varying as fast as the square and sine actually oscillate) digital or analog output signal.

tggzz, I can make an inverted version of the square wave pretty easily, but low pass filtering the V1+V2 and V1+V2Inv would surely gvie the same longer term average voltage for both. Also, can you recommend and faster comparator types which are relatively "generic" (that is to say other types with same pinout and performance if they go out of stock)?

dmills, the diode ring mixer, can that be run with capacitors rather than transformers to provide DC blocking? Also, would it fail whenever the variable amplitude sine wave signal has amplitudes under a diode drop (yes I can get some Schottky to make it smaller than the usual 0.7V, but schottky's still have a drop) in size?

Thanks

[langwadt]

Diode ring mixer?

Fast comparator and nand gate then lowpass will give you a voltage proportional to phase difference.
Lots of ways to do this, just remember that your 3MHz square wave will have a mess of harmonics out to several tens of MHz and you need to respect that in the layout.

I think XOR would be better

[dobsonr741]

OP, are you building a PLL? When you say ‘comparing phase’ sounds like a solution, would be helpful to know the original problem you are solving.

[tggzzz]

tggzz, I can make an inverted version of the square wave pretty easily, but low pass filtering the V1+V2 and V1+V2Inv would surely gvie the same longer term average voltage for both. Also, can you recommend and faster comparator types which are relatively "generic" (that is to say other types with same pinout and performance if they go out of stock)?

Please don't interleave multiple responses in one post; deinterleaving them is an unnecessary pain.

Long term average, yes. Short term average, no. You can add voltages without (slow) opamp circuits!I

You know what's important to you. Use TAoE and/or DigiKey parametric searches to find what you want.

[Weston]

Try using one of the phase comparators on the 74HC4046 VCO/PLL chip. You don't need to use the VCO or anything, but the phase comparators are useful. The signal input can be AC coupled and the datasheet says it should work with a p-p voltage down to 11mV for a 3V supply.

The type 1 comparator is an XOR gate and will give you the phase/voltage relation you need to determine if the signal is out of phase or in phase.

This is basically the same as if you used a discrete comparator to digitize the input signal and an XOR logic gate, but lower BOM/cost.

With any analog mixer your phase reading will be proportional to the input signal amplitude, which can cause challenges.

[Infraviolet]

4046 sounds an interesting idea, CD... cmos versions say they can handle up 18MHz according to the datasheet. By having this digitisation it should be able to work independently of signal amplitude so long as the signal is large than the 20 or so mV I could expect with 5v power? I'm ordering some immediately.

In the meantime, is a comparator built from discrete transistors at all feasible? That or a DIY schmitt trigger with the high and low levels both calibrated for the same reference voltage? I've got plenty of logic gates to hand to feed the signal in to once digitised by this "1 bit ADC".

[Infraviolet]

tggzzz, you mean trying to peak detect on the waveform after adding instantaneous voltages? I had been thinking that V1+V2 would give a wave (lets say V1 is in phase with V2) of amplitude A, centred about a voltage rail. Has been thinking that as V1+V2Inv would have a smaller amplitude B but the same centreing, low pass filtering couldn't see a difference of the two. But low pass pass filtering a peak detection could though.

[radiolistener]

I can make an inverted version of the square wave pretty easily, but low pass filtering the V1+V2 and V1+V2Inv would surely gvie the same longer term average voltage for both. Also, can you recommend and faster comparator

Note, that adding inverter or a low pass filter add phase delay, so the output phase will be different than input.

So, if you compare phase with square wave passed through inverter or low pass filter or even a piece of wire, then you will get wrong result.

[Infraviolet]

My situation is such that so long as the output (constant) voltage waveform can be worked backwards from in microcontroller software at a later point to know what the phase difference was before inverting (or any other process, even the slightest filtering makes a very clear phase diference at these freqs, though lengths of wire on the scale of breadboards don't chnage it appreciably) I'm ok. I don't mind the measured phases being different to the true ones, as long as they are consistently so. If I get a "they're in anti-phase" result but know there was a 180 degree shift that's ok, I can have software calculate back to what it really was so long as the extra shift is consistent over time and over minor variations in device properties. The tough task here is about getting the amplitude of the fast 3MHz sine wave as well as information that can be worked back to calculate the sine wave's in-phase vs anti-phase status as versus the square wave 3mHz waveform in to a microcontroller in the form of a number of analog or digital DC signals.

[tggzzz]

tggzzz, you mean trying to peak detect on the waveform after adding instantaneous voltages? I had been thinking that V1+V2 would give a wave (lets say V1 is in phase with V2) of amplitude A, centred about a voltage rail. Has been thinking that as V1+V2Inv would have a smaller amplitude B but the same centreing, low pass filtering couldn't see a difference of the two. But low pass pass filtering a peak detection could though.

Peak detection on its own seems unlikely to be a sufficient discriminator.

Why don't you let us in on the secret of the input waveforms: voltages, frequencies, amplitudes etc. All it needs is a photo of your scope's screen plus description of what you think important/unimportant, fixed and variable.

That will rule in/out some of the suggestions made by myself and others.

[radiolistener]

The tough task here is about getting the amplitude of the fast 3MHz sine wave as well as information that can be worked back to calculate the sine wave's in-phase vs anti-phase status as versus the square wave 3mHz waveform in to a microcontroller in the form of a number of analog or digital DC signals.

you can use high-speed comparator in order to convert sine wave into square wave. Since you're needs just 3 MHz, this is pretty low frequency, so you can use something like TLV3501.

You can find Chinese modules with TLV3501 and SMA connectors on aliexpress, here is how my one works at 10 MHz:


You're mentioned, that there is needs to work with 20 mVpp signal, as you can see, it works ok with 20 mVpp sine and produce 2.44 Vpp square wave on the output...

[TimFox]

This device has been around for almost 50 years, and may still be in production.
https://www.onsemi.com/download/data-sheet/pdf/mc1496-d.pdf
The last bit of prose in the datasheet discusses phase detection, but they don't give an example circuit.

[David Hess]

Any fast multiplier will work, including the 1496, RF mixer, or synchronous demodulator.  For logic level signals, an XOR gate will work.

A cycle by cycle decision could be made by delaying one signal 90 degrees, say with an integrator, and then using that to drive a sampler (sample and hold, diode bridge sampler, whatever) to sample the output of the other signal, a form of sampling phase detector.

[Weston]

4046 sounds an interesting idea, CD... cmos versions say they can handle up 18MHz according to the datasheet. By having this digitisation it should be able to work independently of signal amplitude so long as the signal is large than the 20 or so mV I could expect with 5v power? I'm ordering some immediately.

In the meantime, is a comparator built from discrete transistors at all feasible? That or a DIY schmitt trigger with the high and low levels both calibrated for the same reference voltage? I've got plenty of logic gates to hand to feed the signal in to once digitised by this "1 bit ADC".

Having the input signal vary over such a large range is a bit of a design challenge. Any DC offset in the comparator or overdrive dispersion is going to distort the waveform and cause phase and duty cycle errors. At the lower bound of 20mV it would probably not be practical to make a comparator from discrete transistors. You might even need a relatively high performance IC based comparator.

The datasheet for the 74HC4046 says it should work. Also, it seems to have better low signal performance at low supply voltages, I assume thats related to biasing / offsets related to the supply voltage. So try running it at 3V?


It would add complexity, but if you have issues with the lower end of your signal range, is might make sense to have an amplifier with automatic gain control buffering the input signal to amplify it to a constant amplitude. It would make an analog phase comparator more precise (by normalizing the input amplitude so your output volts/degree is constant)  and make a digital phase comparator more precise by reducing issues with comparator offset and dispersion at low signal levels.

[Terry Bites]

Have a look at the veritable AD8302. Its a magnitude and phase analyser in a single ic.
Its so easy to use it makes you well up. (We love you Barry Gilbert....)
Low cost $10ish modules abound, eg www.amazon.co.uk/ad8302/s?k=ad8302 A bloody bargain.
It can do your 3MHz while buffing its nails but you may have to increase the input caps for low frequency (3MHz) working!

[Infraviolet]

4046's and TLV3502 (dual version) are on their way to me. I have an odd idea with diodes as switches which I'm going to try until the ICs turn up, will post more about it after trying. I'm going to give a try to each of those options before worrying about other ideas for getting the phase difference measurement from my signals.

"Why don't you let us in on the secret of the input waveforms"
Sorry I haven't described the system in full yet, I tend to see threads as a one problem area per thread discussion, but here goes:

The system is a PCB based resolver, I've challenged myself to make it all without relying on any special single function chips (there are resolver processing chips already manufactured) or highly specialised components (the type where supply suddenly disappears, this has been on my mind for even the smallest projects since the chip crisis propagated through as far as small volume buying is concerned). 

The underlying principle is working REALLY well so long as the rotor is close to the coils (exactly as expected and intended given how little power it is needing), but due to ratios of inductance to capacitance for getting LC tank circuits to have decent Q factors, and with my L being limited by how much coil geometry I can fit in to certain areas of a PCB, I am forced up to 3MHz to make this work. In an ideal world I'd be down at several hundred KHz where I'm more familiar and can use pretty typical op amps, but L and C didn't want to play nicely.

I transmit a 3MHz signal from one coil and another receives it, the amplitude and "phase" of the receivied signal are controlled by the angle of a metal rotor. By "phase" I mean in-phase vs anti-phase, never does the signal shift by other relative phases*. And as far as "phase" goes I don't care about phase diferences induced in pre-transmitted or post-received areas of circuits, as they stay consistent with angle. I'm getting a receievd signal which is a sine wave, a pretty clean one indeed once I eliminated some excessive length wires and made sure to probe with the pointy tip and ground ring of an o-scope probe rather than the hook-in-cover and ground-clip part. Its amplitude after amplification (with transistors) can vary, depending on rotor angle, from around 20mVpp to 1.5Vpp (I may have quoted slightly different numbers at some points as these exact values vary with design iterations and I'm working on further improvements, but given the noise I'm seeing on signals of under 5mVpp being able to measure signals down to 20mVpp seems feasible. With further amplification measuring down to 20mVpp, despite more ampliciaftion, would be desirable as it would improve angular resolution). I can, where necessary, use capacitors to couple this 3MHz sine, whatever its amplitude at the time, on to any different reference centre level of voltage I please (within the limits of Gnd to 5V I'm designing for the final system to need as its own supply rails).

The "phase" of the recieved sine signal vs the phase of the generated (pre-outgoing) square signal tells me which half of the angular region the rotor is in, 0 to 180 or 180 to 360.

I can upload o-scope traces tomorrow perhaps, but all they show is a sine wave, amplitude variable with angle, which goes in or out of phase with the other signal (square wave, 0v to 5V, but can easily make lower voltage or inverted copies) depending on which half of the angular range it is in.

I have been working with reasonable success on peak detecting methods to track the amplitude of the signal and convert that in toa fairly constant DC level output, that level only varyng when the angle of the rotor changes. I am looking to get a similar DC level (analogue or, given I only care about in-phase vs anti-phase, digital) output to tell me whether the phase matches or opposes. I would do this by processing, separately from the peak detecting which ofcourse destroys phase information, the same amplified signal as is fed in to the peak detector, and comparing it with the signal going to the transmitting coil.

My initial hope, as I mentioend early in this thread was using a 4066 analogue switch to quickly let signal received through when the transmitted waveform was high and block them when it was low, but to my unfortunate surprise this added huge distortions to the signal with spikes and other curves far out sizing the desired signal even near maximum amplitudes. Then I tried the comparators (overly slow ones in hindsight)which I had lying around from previous (slower speed) projects. Being short of ideas, and not knowing what sort of search terms would let one find things about comparing if fast signals (everything is so much easier when slower) phases, at that point I started this thread, however with the ideas you've listed I've quite a few more things to try now, thanks for all the tips.

Thanks

*probing various signals in other places the physics is as expected, the transmiting (driving) coil's signal is 90degrees (phase) ahead of the received (driven) as is always true for resonance, but there are delay elsewhere in the circuit which make the phase at the easiest to measure pre-transmitted point and the most practical post-recieved-post-amplified point close to 180 degrees (phase) apart. I see no plausible reason for, and have found no examples in testing of, these other delays to phases varying, so they can be ignored as constant when using phase diferences to know which half of angular positions the rotor is in.

P.S. the AD8302, very good value on those adapter boards, cheaper than the chip itself. Will give that a go if these solutions don't work.

[tggzzz]

4046's and TLV3502 (dual version) are on their way to me. I have an odd idea with diodes as switches which I'm going to try until the ICs turn up, will post more about it after trying. I'm going to give a try to each of those options before worrying about other ideas for getting the phase difference measurement from my signals.

I presume you realise the 4046 phase comparators (plural) require full-swing CMOS inputs.

Those comparators have 1.5ns output transistion times, which imples significant energy at 250MHz. You will need to use good quality construction techniques.

The quickest way to try an idea is often to simulate it using, e.g. LTSpice.

"Why don't you let us in on the secret of the input waveforms"
Sorry I haven't described the system in full yet, I tend to see threads as a one problem area per thread discussion, but

That often leads to the XY problem, which (for good reason) pisses a lot of people off. Avoiding the XY problem requires clear thinking and good taste.

My initial hope, as I mentioend early in this thread was using a 4066 analogue switch to quickly let signal received through when the transmitted waveform was high and block them when it was low, but to my unfortunate surprise this added huge distortions to the signal with spikes and other curves far out sizing the desired signal even near maximum amplitudes.

The design details matter: schematic showing all relevant components is necessary. A beginner is less likely to understand what constitutes "relevant".

The construction details matter: photographs are necessary. Solderless breadboards can, even with a sound design, cause problems.

[dobsonr741]

I badly wanted to understand what Infraviolet is aiming to achieve. After much deliberation, consulting with ChapGPT, I think it's a rotary angle to DC voltage converter, coils fabricated out of PCB. DC output is to be proportional to anlge, should include the 0-360 degree range.

Is that right?

[Infraviolet]

"I think it's a rotary angle to DC voltage converter, coils fabricated out of PCB. DC output is to be proportional to anlge, should include the 0-360 degree range."

Yes, Though I'm happy enough to have the DC voltage on one output vary between 0 and 180 and go through the same range again between 180 and 360, with another output to indicate which range the first is meauring from, and I expect this multiple output situation is simpler to build.

P.S. out of interest, how many very bizarre responses did ChatGPT try to suggest first?

[dobsonr741]

I've asked chatGPT to summarize, then I asked what are the inputs and outputs. Nothing off mark, it really helped. No offense, but in your writeup the intent and low level details/design detours are heavily intertwined.

[Infraviolet]

"No offense, but in your writeup the intent and low level details/design detours are heavily intertwined. "

None taken, will try to remember to avoid that for the next long reply I write.