Direct Conversion Receiver (iq mixer with f_lo=f_rf)

Thread Starter

Omri Shachak

Joined Mar 27, 2018
11
hello,
i am trying to build a system that transmit a signal (10GHz) and the signal hits a reflective object(that changes its position in a sinusoidal manner, much slower than 10GHZ, close to 1Hz), and comes back to the Rx antenna.
my goal is to identify the phase (modulo 2 pi) that the signal acquire. (the phase is related to the object's position).
in order to do that, after the Rx antenna i put an IQ mixer and feed the LO with a replica of the RF signal that i send, using a wilkinson power splitter before i send the signal.
Therefore, in the IF port i will get near DC frequencies and my wanted data is almost DC (i am thinking close to a 1Hz, is it considered DC?).
My biggest concern is that, i will get LO leakage through the RF and hence i will get DC offset that i would normally be able to get rid off, but the problem is that my desired data is also DC. So i would not be able to distinguish between the data DC and the LO leakage DC.
I would gladly accept your wisdom and help.
thank you.
 

mlv

Joined Nov 6, 2017
17
I have a few questions and basic comments and then some suggestions. These will be based on a few assumptions from my side and may be things you've already considered, but it's better to be pedantic and explicit than to miss something fundamental.

The first question I had was about your "close to 1 Hz" description. Is the Doppler shift close to 1 Hz, or does the object move back and forth at 1 Hz? These are not the same. For example, if the object moves +/- A meters in a sinusoidal motion with a period of 1 second, then the max velocity is 2*pi*A (slope of A*sin(2*pi*t)). Further, if A= 3/(2*pi) ~ 0.48 m, then the max velocity is 3 m/s. As a fraction of the speed of light, this is 10 ppb, and the resulting Doppler shift on 10 GHz carrier is 100 Hz. If the motion only 1% of this and A=5mm, then the Doppler shift is only about 1 Hz...etc.

Second question: how consistent is the sinusoidal motion frequency? Can you expect coherency over multiple oscillations, and can you take the time to do your estimation over multiple cycles (i.e. many seconds)?

The answers to the above questions have a large bearing on practicality, of course. Equally fundamental are the questions of signal power, antenna gains, reflectivity/return-loss of the object, path loss, RX noise figure, IQ mismatch (would you do compensation at the receiver or live with the modulator's native image rejection?), TX to RX isolation, LO leakage in the mixer, etc. Would you sample the demodulated I/Q signal with a complex ADC?

It may be instructive to put together a simulation of your scenario. This forces many of the questions above. If you want to continue the discussion, send what you know at this point and we can try to fill in the details.

Thanks,
M.
 

DickCappels

Joined Aug 21, 2008
10,152
The simple solution relies on an assumption that the LO leakage and the resulting DC offset is constant. If that assumption is correct then an offset adjustment should take care of the problem.
 

Thread Starter

Omri Shachak

Joined Mar 27, 2018
11
I have a few questions and basic comments and then some suggestions. These will be based on a few assumptions from my side and may be things you've already considered, but it's better to be pedantic and explicit than to miss something fundamental.

The first question I had was about your "close to 1 Hz" description. Is the Doppler shift close to 1 Hz, or does the object move back and forth at 1 Hz? These are not the same. For example, if the object moves +/- A meters in a sinusoidal motion with a period of 1 second, then the max velocity is 2*pi*A (slope of A*sin(2*pi*t)). Further, if A= 3/(2*pi) ~ 0.48 m, then the max velocity is 3 m/s. As a fraction of the speed of light, this is 10 ppb, and the resulting Doppler shift on 10 GHz carrier is 100 Hz. If the motion only 1% of this and A=5mm, then the Doppler shift is only about 1 Hz...etc.

Second question: how consistent is the sinusoidal motion frequency? Can you expect coherency over multiple oscillations, and can you take the time to do your estimation over multiple cycles (i.e. many seconds)?

The answers to the above questions have a large bearing on practicality, of course. Equally fundamental are the questions of signal power, antenna gains, reflectivity/return-loss of the object, path loss, RX noise figure, IQ mismatch (would you do compensation at the receiver or live with the modulator's native image rejection?), TX to RX isolation, LO leakage in the mixer, etc. Would you sample the demodulated I/Q signal with a complex ADC?

It may be instructive to put together a simulation of your scenario. This forces many of the questions above. If you want to continue the discussion, send what you know at this point and we can try to fill in the details.

Thanks,
M.
first of all, thank you for your answer.
before i go on i should have mentioned earlier that the so called object i am referring to is the human heart, and my purpose is measuring the heartbeat of a human being standing in front of my machine for couple of minutes. I did not say it before because i felt like cheating a bit by asking exactly what my purpose is but i guess it is relevant to your answers.

the answer for the first question: "The first question I had was about your "close to 1 Hz" description..."
the answer is that the object is moving back and forth at 1 Hz. (the object moves for a short distance in 1 hz so its velocity is very small and there is no measurable doppler shift in this case).
as to the second question: " how consistent is the sinusoidal motion frequency? " .
the sinusoidal motion frequency is not very consistent at least from my limited understanding of the human heart. but i can have the person wait for a couple of minutes for the heartbeat approximation.

DickCappels thank you i guess you are right and i can do an offset adjustment.


As you suggested i am adding a diagram of my system, and would appreciate if you can point me to parts i made a crucial mistake.
the red part is supposed to deal with the TX to RX isolation & LO leakage in the mixer by subtracting a constant signal from the received signal.
the blue part is the LO drive.

if you have questions please let me know, thank you again.
 

mlv

Joined Nov 6, 2017
17
Ok, that clarifies a lot of things, though I definitely didn't intend to have you disclose more than you wanted to.

Yes, you can try to remove a constant DC at baseband or scaled version of the LO at RF. I think that part is apparent, and you're already working to address this by subtracting at RF (this probably addresses the mixer coupling more than others, since the mixer coupling probably dominates). However, this DC removal isn't without challenges. More on that in a bit...

First, I think it would be good to talk about the signal model. This is needed before talking about removing impairments. It isn't that you have frequency modulation due to Doppler. There just isn't enough displacement, as you noted. I am guessing that you're relying on movement to get a difference in distance that then gives a slight phase modulation. An alternative might be that reflectivity changes with heart cycle that gives amplitude modulation, but that seems less likely. Either way, there is some modest modulation on the LO; I'll assume for this discussion that it's PM and not AM.

The PM due to movement of a person's chest would come from movement on the order of a mm or two at best. The next question is: what fraction of a wavelength is this? For an LO at 10GHz, the wavelength is 3e8/10e9 = 30 mm. So the phase modulation is roughly +/-(0.75mm/30mm)*2*360deg = +/-18 degrees (factor of 2 from reflection causing the variation in total distance traveled to be doubled).

Note, however, that this is only true if the entire skin surface that is reflecting the signal is moving by +/-0.75mm. I expect that areas away from the heart (or carotid if you use that area) may not move as much. If the scattering surface is then moving to varying degrees, you get a large dispersion, with many paths (multipaths if you consider them discrete or an integral if continuous) summing at the receiver with varying amounts of PM, ranging from nearly zero PM to a peak of best-case variation, and thus reducing the signal PM seen at the receiver.

Back to the "DC" portion or interferer/impairment aspects. The LO can couple directly, either from TX to RX chains/antennas or in the RX mixer and result in DC at baseband. Additionally, scattering returns that are not related to the heartbeat signal are interferers and likely live between DC and the 1 Hz PM in frequency. These can be due to unmodulated reflections, physical swaying/movement of the person's body, or breathing movement and need to be tracked, making it more difficult to remove the baseline, since it could wander. Anything causing a mm of movement could be a problem.

In the end, you need more signal to noise+interference_or_impairment ratio. This is where averaging can help, but only if the signal is coherent/consistent enough to be beneficial. If it changes slightly and you don't know it, you can average too long and cause signal annihilation in the integration.

Fundamentally: what is your PM-modulated signal power versus the powers of (1) noise; (2) LO coupling; (3) unmodulated LO reflection; (4) baseline wander. Next, analyze what performance can be achieved; simulate the signal plus impairments to test algo performance. Then analyze captured data if available to see if the model fits.

Overall, the signal identification problem here seems like a modal analysis problem for estimating the frequencies and complex amplitudes of your RX signal. I might try matrix pencil or min norm approaches for modal analysis, though these are *not* light-weight computationally. They may, however, be useful as an analysis tool for data captures.

Another impairment that will be present is an image of the PM signal. The mixer that you listed in the other thread has IR of ~15 dB. However, if you get your signal clean enough to see this image, I think you would be happy... ;)

I'm not sure that any of this rambling is anything new/useful, but it was good for me to get some insights to your challenges.
M.
 

Thread Starter

Omri Shachak

Joined Mar 27, 2018
11
thank you mlv, i appreciate your help very much. some of the things you mentioned i have not considered or knew how to deal with and will give me more stuff to think about and optimize my system.
 

BR-549

Joined Sep 22, 2013
4,928
Have patient hold head stationary, but relaxed. Successful image sweeps of the face....should show a facial color change with heartbeat.

 
Top