• State TI Thinks Resolved
  • Locked Locked
  • Replies 14 replies
  • Answers 2 answers
  • Subscribers 11 subscribers
  • Views 1137 views
  • Users 0 members are here
Support feedback
Options
Options

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TRF3722EVM: I/Q input to LO feedthrough issue

Martin Held
Prodigy 10 points
Part Number: TRF3722EVM
Other Parts Discussed in Thread: TRF3722, TRF372017, TRF3705

Hello,

I am trying to design a system using the TRF3722.  I have a PCB in hand with our design, and have also evaluated the TRF3722 and TRF372017 using the EVM moduels.  I have a 50kHz I/Q single frequency input and am using a 2.0-3.0GHz PLL frequency (though many of my tests are done at 2.686 GHz, for no particular reason).  I am seeing -13dBc of the IQ signal on the LO output.  The LO output should, in theory, just be a single spur at the PLL frequency.  I feel like I must be doing something silly, because I would expect I/Q feedthrough to the LO signal to be less than -40 to -50 dBc.  For this test I have been using an AFG, with transformers to go from single ended to balanced with the ability to adjust the bias point to the required value.  (Pardon my crappy SpecAn with a dim screen).  I have tested this with both my design and the eval boards.

I am seeing this same behavior on the TRF372017 evaluation board.

As a result of seeing baseband feedthrough to the LO, my RF output is spikey as all can be.  One would expect (with a 50kHz I/Q signal, 90 degree phasing) that the resultant output would be a single spur at the PLL frequency plus (or minus depending on phasing) the 50kHz frequency of the baseband signal.

Any ideas on what I'm missing? Thanks!

On my board all power supplies are bypassed to heck and back, ferrite isolated, and similar to the application circuitry in the datasheet.  The application is using the IF frequency to analyze the return signal in DSP for accurate phase, and having the spurs really messes things up.

  • 0
    TI__Guru 73459 points

    Hi Martin,

    Could you please advise what is an AFG? Please elaborate your exact connection. Pictures and block diagram of the setup will be helpful as it is hard to follow your exact connections.

    I agree this 13dBc of isolation should not have happened. Will you try to shift the IF to higher frequency and see the impact of the isolation? Please also see if you can reduce the IF frequency amplitude to see if the isolation drop in dB/dB fashion. 

    -Kang

  • 0
    TI__Mastermind 18095 points

    Martin:

    I think that you have two problems, both related to the BB inputs.  If all is set up correctly you should see on the output a suppressed LO carrier, a desired upper sideband, and a suppressed lower sideband.  You will also see some harmonics and LO +/- n*BB, where the 3rd harmonic on the lower sideband side will be the worst.

    Looking at picture 2 of the TRF3722 EVM, the carrier is not suppressed and you are not getting any suppression of the lower sideband.  The carrier feedthrough suppression is dependent on DC balance between all 4 ports.  Microvolts make a difference here.  I suspect that your DC offset balance is poor with your set-up.  You can measure this directly with a voltmeter on all four BB lines.  In the case of the TRF3722, the BB common mode voltage should be 0.25V.  The set-up that you describe uses a function generator and a transformer.  The transformer does not pass DC, so you cannot use the function generator to create the common mode voltage.  If the transformer has a center tap option, you could inject the 0.25V at that point for the modulator.

    The sideband suppression is determined by phase and amplitude balance between the I/Q ports.  When you see the two sidebands of equal amplitude, it indicates that one of the I/Q signals is not configured properly or is not present. You would need to set the function generator to inject quadrature BB signals to I/Q.  If the phases are the same, you will not see the suppression.  You can also probe with a scope to ensure the BB signal at the modulator pins are equal amplitude and at the appropriate phases.  I suspect that the source phase is not set up properly or that one of the paths through the transformer is not at the right level.

    I believe the -13 dBc is roughly equivalent to one BB channel missing.  If that is the case, then it would explain both the sideband issue and the carrier feedthrough issue.  Again, a probe of the BB signal at the modulator pin should reveal the situation  

    In general, I am not an advocate of the function generator-to-transformer approach as there are lots of imbalances within the BB structure that will impact modulator performance.  Ideally, a specific generator capable of quadrature BB signals like the Agilent ESG is a better approach.  If that type of instrument is not available, you could acquire a TI DAC EVM, like the DAC3484EVM, that includes the TRF3722 and TRF3705 modulator devices.  The DAC is an excellent BB generating source and the EVM is set-up for the proper common mode voltages.

    --RJH

  • 0 in reply to RJ Hopper
    Prodigy 10 points

    Hi RJ,

    I'm still working on a block diagram, however the problem isn't the RF outout, its the LO output... in theory, none of the baseband should be getting into the LO at all... there's no forward mechanism for that at all.  That 2nd picture you're referencing isn't the RF output, it's the LO output.  The RF output does what it should and has a suppressed sideband (even if not perfect).  The problem is I can't start adjusting my baseband balance/etc because there is such a noisy LO that when you take a perfect 90 degree phased I/Q signal, in theory, you should get a single spur output.  But, even if that were the case, I would be seeing the baseband spurs mixed into the LO then re-mixed with the 90 degree I/Q.   This is the problem I'm dealing with.

  • 0 in reply to Kang Hsia
    Prodigy 10 points

    Hello Kang,

    Arbitrary Function Generator, specifically an AFG1022 set to 50kHz on both channels.  Each channel goes to a transformer, with the center tap exposed to bias voltage so that the common mode voltage for both can be easily set and adjusted.  They can be locked to the same supply or independently adjusted depending on how I clip them.  The outputs of the transformers go to SMA connectors so that I can drive the evaluation board.  The only other things hooked up are the LO_N SMA connector to my spectrum analyzer and the USB to control and power the eval board (or +5V/3.3V for the 372017 board).

    Altering the frequency of my IF isn't really doable on my board, but on the AFG I can set it to whatever I want, I just need to set the appropriate magnitude due to the limitations of the transformers that probably start pooping out at 25kHz, so to test higher frequencies I'll probably need to grab and wire up some different magnetics (It takes 6V at the input to get 500mV at the output at that frequency).

  • 0 in reply to Martin Held
    TI__Mastermind 18095 points

    Ah, OK, this is at the LO output from the TRF3722.  I agree, there should be no BB tones on this output.  I have not tested this output using such a low BB frequency.  Usually the BB is on the order of several MHz.  You could try as Kang suggested to increase the BB frequency to see if there is any impact to coupling.  It is hard to tell, but the harmonic on the RF output seems a bit too high which may mean the ports are being over-driven.  You can try to back off the signal level and see if the coupling tones on the output drop dB-for-dB.  There are a few bias control options through programming that impact LO synthesizer, but I don't think these would impact BB coupling.  We will need to try this out on our end.  --RJH

  • 0 in reply to RJ Hopper
    Prodigy 10 points

    I did a simple test.  The setup is as follows-  Instead of using transformers to balance the baseband input (where the output magnitude is frequency dependent), I used only one baseband input as it shouldn't matter if both baseband inputs are driven or not in this particular case.  The unused baseband input was terminated 50 ohms, as were other unused inputs on the eval board.  In verifying the termination requirements, I did notice that terminating the RF output yielded a 10 dB drop in the LO spur magnitude, but the LO output was still at a rather high level.  For the test below, the RF output was turned on but not terminated, so I may repeat the experiment with the RF terminated.  (This is a manual process and not automated, unfortunately).  However, this may or may not be relevant, here is another datapoint.  It makes the chip useless, however the LO spurs go away completely if the output RF modulator is turned off completely.  (It definitely seems like something is feeding through from the RF output to the LO circuitry internally).  With the termination, there is still a definite peak at 35kHz, it's just slightly lower.

    The function generator was set to 200mV peak-peak (likely considerably underdriven for regular use), locked frequency, and 180 degree offset.  Each channel drove the corresponding BB_I_N or BB_I_P inputs.  The BB_Q_P/N inputs were terminated 50 ohms.  Steps were 5kHz up to 100KHz, 20kHz steps to 200kHz, and 100kHz steps to 1MHz.

  • 0 in reply to RJ Hopper
    Prodigy 10 points

    When I terminate the RF output, the orange line in the updated graph is what I get.  It's ~10dB below the previous, but still enough to be an issue I think.  I only bothered to repeat it to 200kHz.

  • 0 in reply to Martin Held
    TI__Guru 73459 points

    Hi

    could you please double check if you have properly set the baseband input to 0.25V to 0.5V common mode?

  • 0 in reply to Kang Hsia
    Prodigy 10 points

    Yes, the baseband common mode is at 0.25V, with magnitude 200mV (So full signal swing from 0.15V to 0.35V)

  • 0 in reply to Martin Held
    TI__Guru 73459 points

    Hi Martin,

    I see. I am not sure what's the next step. will have to brainstorm

  • 0 in reply to Kang Hsia
    TI__Guru 73459 points

    Hi Martin,

    To verify if it is due to modulation from the power supply coupling to the LO output, we can do the following:

    1. increase the amount of decoupling capacitance. With such low frequency coupling, the capacitance can be increased to 100uF to start out

    2. please see if you can replace the 50ohm with a higher impedance RF choke to filter out power supply modulator to the LO output port.

    please give it a try and advise the results.

    -Kang

  • 0 in reply to Kang Hsia
    Prodigy 10 points

    I will give it a shot.  I'm not super hopeful though, as I recall trying adding 100uF caps and additional bypass on several of the supplies on my version of the board (not the eval board) with no effect.  I have been busy the last few days, so pardon the delay, should have it done today or tomorrow.

  • 0 in reply to Martin Held
    TI__Mastermind 18095 points

    Martin:

    I tested the TRF3722 EVM on the bench in a configuration similar to your set-up.  I chose a reference frequency of 40 MHz and tested in integer and fractional modes with a nominal BB signal at 50 kHz.  This link has the report: https://txn.box.com/s/gn3h16xyai8gxpmu56vzns1t50ryorh1.  Note, the link is active for 1 week.

    I do see BB signal leaking into LO output, but not quite as high a level as you are seeing.  I see 50 kHz BB signal at about 36 dBc down.  This isolation performance moves dB-for-dB with BB drive level.  I found the isolation level was the worst when operating in fractional mode and the PFD frequency was set too low.  Isolation degraded to around 22 dBc when PFD set to 2.56 MHz.  Note, this is not really a valid mode of operation for fractional mode; I was just trying to see how to make the isolation worse.  In fractional mode, the PFD should be somewhere in the 20 -35 MHz range typically.

    I was not able to find any other programming/bias knobs that significantly impact the isolation performance.  Disabling the modulator does indeed kill the spur, but that is not a viable mode for operation.  Engaging gain control bit did make the performance a bit worse.  The isolation does change with BB frequency.  Frequencies within the loop filter bandwidth are worse.  Performance gets better as BB frequency increases past the loop filter cut-off.  We typically test at 5 MHz offset where the spur improves to 60 dBc.

    In short, there is some BB to LO output leakage.  Without overdriving the BB and with proper synthesizer settings the leakage should be limited to around 35 dBc or better with low BB frequency inputs.

    --RJH

  • 0 in reply to RJ Hopper
    Prodigy 10 points

    Hello RJ & Kang -

    So the PFD frequency when using the evaluation board for comparison was indeed incorrect, but I went back over my code for running the TRF3722 on my PCB and that code is reasonable.  The eval board was using the default frequency of 2.56 Mhz, and when using that, it would get -10dBc spurs on the LO.  When I increased the PFD frequency to 30.72 (using the 61.44 reference on the eval board), the spurs went to -30dBc.  I looked at my setup and code, and it is set with a 100Mhz reference clock, and FPFD is set based on the requested resolution step size.  In my case, I was always requesting 1Hz step size, so RDIV = FREF / Stepsize / (2^25) +1, in this case approximately 3 (Maybe 4 depending on where the integer rounds up at in the code, doesn't really matter here).  As a result, my FPFD is either 25MHz or 33MHz.  For reference on the eval board, I am only driving one half of the baseband, the other end is floating, the RF output is open (I cannot guarantee a matched load in my application), and one LO output is open terminated.

    Kang, I tried a 100uF cap across C42 on the eval board with no performance gain. 

    I am still trying to prove/disprove with my PCB if the LO leakage even matters for our application, and it is proving a lot more difficult than I thought.  I will verify my FPFD (via proving RDIV) but I'm pretty sure it's either 3 or 4.