Post Go back to editing

ADL5375 noise floor


I'm looking into the noise density of the ADL5375 IQ modulator and I noticed that it degrades when the LO frequency goes above approximately 4 GHz. Up to 4 GHz noise density is around -155 dBm/Hz as expected but with LO frequency of 5 GHz it degrades by 10 dB as it can be seen in the attached figure. Moreover, the noise floor degrades not only around the LO frequency but also at the low frequencies. 

Is this something I have to live with or is there a way to improve the performance? According to the datasheet noise density shouldn't degrade that much with the frequency so I wouldn't expect this to happen.

In the figure white shows noise floor with LO at 3 GHz and red shows noise floor with LO at 5 GHz. LO power is 0 dBm in both cases. 

I also nulled the LO leakage in the 5 GHz case but it doesn't make any difference.  Also. I checked the LO source and the noise density is the same at both 3 GHz and 5 GHz.

  • Hello Gec,

    I tried duplicating your results and unfortunately have not been successful. Specifically I zoomed into the 100MHz to 1GHz region that is shown elevated in the plot and did not see any appreciable difference with varied LO frequency. Although, given my spectrum analyzer floor is on the order of -110dBm, so not sure if that's making a difference or not.

    When you collect the two traces are you making any changes to the spectrum analyzer settings?  Can you zoom into the same frequency region for both sweeps, such as 100MHz to 1GHz?  How about measuring the source by itself to rule that out?

    Best Regards,


  • Hello David,

    Thank you very much for your reply.

    - There was no change in the spectrum analyser settings between the two measurement.

    - There was also no baseband input signal so that wasn't the cause. Only LO signal was present its spectrum was fine.

    - Your spectrum analyser floor seems to be the same as mine. In the attached figure the left hand axis is in dB and right hand side axes is spectral noise density in dB/Hz so that's why is -155.

    Do you think it may be the problem with the device or the evaluation board?

    I will try to do some more testing as ASAP with different spectrum analyser jut to rule it's not coming from there for some reason.

    Kind regards,

    Dragan Gecan

  • Hi Dragan,

    During my testing I applied 0.5V common mode voltage to all baseband inputs with slight offsets to cause the LO signal to pop up and pass through (opposite of nulling). Also, I had the on board amplifier bypassed, so looking at the ADL5375 output directly.

    Did you apply the common mode voltage? The mixer inputs will need this common mode voltage for proper biasing.

    I'm not sure if it's the device or board; haven't seen an issue like this before.

    Usually RF sources are the first place I look when seeing elevated noise levels as it can be quickly ruled out.

    If you do more testing and it appears to be from the device or board, the next steps would be: 1) get another board to see if it persists, 2) feedback your test setup and I'll try again to duplicate/identify the problem.

    Best Regards,


  • Hi David,

    Thank you for your help.

    I think I found the source of the noise. It is the DAC I'm using for common mode voltage and LO nulling. By using different DC supply the noise at the high frequency end goes away but at the low end it's still there I'll try to use some inline filter or capacitor to ground to see if that's also coming from the DC source.

    Since I can't use DC supply in the final product, can you recommend me a circuit for the common mode voltage supply and LO nulling?

    At the moment I'm resistively feeding DAC ports to the modulator inputs. Baseband signal is fed by differential OPAMPs. I made rough drawing of the circuit.  

  • Hi Dragan,

    You're welcome. Good to hear you were able to troubleshoot it to the baseband common mode voltages.

    In terms of baseband interface, the following are possibilities you might consider:

    1. Use differential output transmit DAC with common mode voltage as your primary driver (transmit signal and common mode).
    2. Use the single ended to differential amplifiers you have in the schematic above, with common mode coming from a DAC, such as the AD5624. You would need to be sure the AC signal isn't leaking through the DAC path, which could be done with a inductor or the resistors on each leg. It would be good to put placeholders for filtering capacitors to ground in parallel with the existing resistors to ground.
    3. Use the ADRF6521 DVGA with single-ended to differential input baluns. This would also require a DAC to apply offset voltages to do LO nulling. Although, it's two voltages instead of four. The single common mode pin could be provided as static voltage or you could use a DAC for flexibility; same goes for the gain control pin. If you go with this approach, pay close attention to the amplitude and phase imbalance of the baluns, so that you can achieve good unwanted sideband rejection.

    Best Regards,


  • Hi David,

    Thank you for the suggestions.

    Unfortunately, I realised that most of the problem is coming from from the LO source. The common mode DC is not participating that much.

    The problem seems to be phase noise of the LO source. Can you tell me what should be the phase noise spec of the LO source to achieve noise density at 5.8 GHz specified in the datasheet at (-153 dBm/Hz)?

    The LO source I'm using has phase noise of -110 dBm @ 1kHz, -125 dBm @ 1 MHz and -140 dBm @ 10 MHz (the 10 MHz value may be already hitting the noise floor of the analyser so may not be correct). When I used a source with lower phase noise the performance improved, but that source is laboratory signal generator so can't be used in a product. 

    The other observation I made is that using the HMC525ALC4 IQ mixer noise floor doesn't get worse under the same conditions using the same LO source. You can see in the attached screenshot (black ADL5375 blue HMC525). I wonder what mechanism is creating this noise degradation in ADL5375 and is not present in HMC525 part and is there a way to reduce the effect?

    Also, I did the test with another ADL5375 evaluation board and the same happened.

    Kind regards,


  • Hi Dragan,

    The LO source phase noise is often the culprit in these situations. Note that until now we have been discussing noise floor in terms of dBm/Hz, whereas phase noise is usually relative to a carrier signal in units of dBc/Hz.

    It's difficult to say specifically what the phase noise would need to be for the LO source on the ADL5375, because the composite RF output phase noise of the modulator, and resulting noise floor, may be a result of 1) the LO phase noise, 2) any noise on the baseband inputs that gets upconverted, and 3) the inherent noise of the ADL5375. It's likely those combined is producing the undesired result. I think if you have the LO phase noise and the baseband noise measured independently, then you could convert these to real power quantities (account for carrier level if in dBc/Hz), then add them in real terms, convert back to dBm/Hz units, and see if they line up with what you are measuring. Ideally you would do that for each offset frequency from a particular carrier. Although, I do realize this is laborious and you may find a quicker result by piecing together all of the evaluation boards you planned to use in your signal chain and then measuring the performance.

    If when you tested the HMC525ALC4 there was no DC component to the IF inputs, then that might explain why this solution is providing better performance, assuming that noise at baseband is coming from the DC supply. If you need to perform LO nulling, then you will need to provide a DC component to the IF pins. Another possibility is that the inherent noise of the HMC525ALC4 is better than the ADL5375. The HMC525ALC4 datasheet doesn't mention noise floor, but does give noise figure. Conversely, the ADL5375 doesn't mention noise figure, but we might imply that it's worse since the noise floor is -153dBm/Hz for 5.8GHz LO. Apologies if this isn't a straight forward answer.

    Best Regards,

  • Hi David,

    Thank you for your feedback.

    I managed to make it work by the specs at 5.8 GHz finally. To achieve that I had to high-pass filter the LO source and I had to increase the LO power to around 7 dBm (earlier the LO power was  dBm as in datasheet tests).

    Thank you for supporting me.



  • Hi Dragan,

    You're welcome. Glad to hear that it's working now.

    Best Regards,


Reply Children
No Data