Hello Brian,

I'll see if  I can reproduce with my TwinRX. Please provide some more
information to help me reproduce...

1. Center Frequency of fixed channel 0
2. Tuning range on channel 1
3. What tuning rate have you tried and what's the threshold between a clean
   spectrum and a bad one?
4. Please share screenshots of what you are seeing
5. Please share uhd_usrp_probe output so I know your serial, revision, uhd
   version, etc
6. Can you reproduce this problem when it's two channels on different
   daughterboards? In other words, ch 0 and ch 1 on the same TwinRX, and ch 0
   and ch 1 across each DB slot.

There's several switchable routes before and after each stage LO going
across channels, so it's possible there are some isolation problems between
channels. My first thought is also to remove LO sharing cables, but as you
said, it doesn't improve.

Maybe switching the switches that are not used might help give better
isolation. From schematic, if channel 1 uses its own LO, then only switch
16 is needed, and switch 14 which routes LO 1 to the sister channel 2 isn't
used. So I'm wondering if the state of switch 14 makes a difference in
terms of isolation. I'd have to check the software to see if you can
independently flip these switches, and if it's recommended, to test this
hypothesis. I will also check internally if similar issue is reported and
get back to you.

Regards,

Wan Liu

On Tue, Oct 25, 2022 at 6:16 PM Brian Padalino bpadalino@gmail.com wrote:

On 2022-10-25 18:15, Brian Padalino wrote:

I wonder if this is a radiative coupling thing?  Like the synthesizer
for the other channel briefly sweeps across the stable tuned
  frequency on the other channel, and a small amount of energy is
bouncing around inside the box and couples to the input.
  The TwinRx has an internal heatsink/shield, and I would expect very
few instances where that shield is "waveguiding".

It is absolutely the case that with modern electronics with lots of
high-speed digital logic, it's nearly impossible to make
  all your spurs go away.  This is compounded by the fact that
front-ends in receivers have become about 10dB more
  sensitive in terms of average noise figure over the last decade or
two.   So internal "spurs" that would previously never
  be seen are now showing up well above the noise floor.

My own approach to internal spurs is to make sure the outside-world
signal is strong enough to overwhelm them.  This
  isn't always practical, of course.

I have an X310 I can do some experiments with also.  But I won't be able
to get to it until end of the week or maybe even
  next week.

Hey Wan,

Thanks for the quick response.

On Tue, Oct 25, 2022 at 7:59 PM Wan Liu wan.liu@ettus.com wrote:

I tested at 915 MHz, but also 400 MHz.  It seems to show up wherever I
happen to tune the fixed channel.

I have tried with just 2 frequencies (200 MHz, 500 MHz) or the full
frequency range.  Tuning the full frequency range seems to provide more
noise in different areas of the spectrum.

I always get bad spectrum regardless of my dwell time on any particular
frequency.  The "max hold" will always show approximately the same spectral
image.  The "average" spectrum will obviously be quieter if I dwell for
longer.

Attached is a 10 ms capture of what I see often.  It's only one particular
case, but you can see what is going on.  I have included time vs. freq,
amplitude vs. freq, and amplitude vs. time plots.

Attached as uhd_probe.txt.

I am unsure exactly what you're asking for here.  My current setup has a
UBX in DB0 and a TwinRX in DB1.  If you can be more specific about what you
want me to try, I can give it a shot?

I will note 2 more observations I made:

• When trying to read the lo_locked sensor continuously, generating SPI
  traffic I presume going to the PLLs, I get clean spectrum.  This suggests
  to me that digital noise isn't the culprit here.
• The IQ file I saved and looked at (which generated the attached
  figures) shows some analog PLL-style locking for around 10 ms of time, then
  goes away.

Let me know if there's anything else I can provide to try to help out.  I
have a very long IQ capture I took (both inputs terminated, recording fixed
channel while retuning the other channel) but it's obviously too big to
share here.

Thanks,
Brian

Hey Wan,

Thanks for the quick response.

On Tue, Oct 25, 2022 at 7:59 PM Wan Liu wan.liu@ettus.com wrote:

I tested at 915 MHz, but also 400 MHz.  It seems to show up wherever I
happen to tune the fixed channel.

I have tried with just 2 frequencies (200 MHz, 500 MHz) or the full
frequency range.  Tuning the full frequency range seems to provide more
noise in different areas of the spectrum.

I always get bad spectrum regardless of my dwell time on any particular
frequency.  The "max hold" will always show approximately the same spectral
image.  The "average" spectrum will obviously be quieter if I dwell for
longer.

This Google Drive folder has a 10 ms capture of what I see often.  It's
only one particular case, but you can see what is going on.  I have
included time vs. freq, amplitude vs. freq, and amplitude vs. time plots.

https://drive.google.com/drive/folders/1lm6WiH39XV-Hwl42Myi3DiSiN170D92i?usp=sharing

Included in the drive as uhd_probe.txt.

I am unsure exactly what you're asking for here.  My current setup has a
UBX in DB0 and a TwinRX in DB1.  If you can be more specific about what you
want me to try, I can give it a shot?

I will note 2 more observations I made:

• When trying to read the lo_locked sensor continuously, generating SPI
  traffic I presume going to the PLLs, I get clean spectrum.  This suggests
  to me that digital noise isn't the culprit here.
• The IQ file I saved and looked at (which generated the attached
  figures) shows some analog PLL-style locking for around 10 ms of time, then
  goes away.

Let me know if there's anything else I can provide to try to help out.  I
have a very long IQ capture I took (both inputs terminated, recording fixed
channel while retuning the other channel) but it's obviously too big to
share here.

Thanks,
Brian

Hello Brian,

Thank you for the additional information.

Regarding  #6, I meant that if you have two TwinRX daughterboards, see if
you get this problem when the fixed channel is on one daughterboard, and
the tuned channel is on the other.

Regarding screenshots, are you referring to any particular frequency and
time region, or is everything above the noise floor associated with the
tuning? In other words, is the clean spectrum where there is nothing above
the noise floor in both time and frequency plots?

Also can you explain what you mean by "shows some analog PLL-style locking
for around 10 ms of time, then goes away"? Are you referring to the burst
from 3 ms to 13 ms, or something specifically at 10 ms?

Lastly, what are your spectrogram parameters to generate the waterfall?

I'll reach out again after I attempt to reproduce.

Regards,

Wan

On Tue, Oct 25, 2022 at 8:29 PM Brian Padalino bpadalino@gmail.com wrote:

Hey Wan,

On Tue, Oct 25, 2022 at 10:53 PM Wan Liu wan.liu@ettus.com wrote:

Ah, I see.  Unfortunately my setup is a mixed USB/TwinRX setup so maybe it
isn't exactly testing what you're asking for, but I did use the subdev spec
to target the UBX RX2 for hopping around, and the TwinRX Channel 0 was
fixed.  In this case, the fixed spectrum stayed nice and clean the whole
time.

The captures were taken with terminated RF inputs.  Channel 0 of the TwinRX
was fixed at some frequency (I believe 400 MHz) and Channel 1 was hopping
around.  The recording was observing Channel 0 - the fixed frequency
channel.  When no hopping happens, there is clean spectrum with just a
noise floor which is what I expected to see.  When hopping is happening,
every so often we will see these sweeping signals show up.  They last
around 10 ms or so and then the spectrum is back to being clean.

I meant the phenomenon that starts at around 3 ms and lasts until around 13
ms.  It looks like an analog PLL settling to me.  Here is a zoomed in
version:

https://drive.google.com/file/d/1NDax78i3UQh7X_R4g8SHBkBLibI1ICQZ/view?usp=sharing

I am using an FFT size of 2048 with a blackmanharris window of the same
size, and overlapping by 1024.  My MATLAB command is:

spectrogram(slice, blackmanharris(2048), 1024, 2048, 50e6, 'centered');

Sounds good.  Let me know if you need any other data or clarifications on
what I am seeing.

Thanks,
Brian

After a bunch of testing, I ended up with the following solution which
seems to have fixed the vast majority of the issue.  There's still extra
noise, but not nearly as bad as it was previously so I'd appreciate it if
Ettus still looked into a more complete solution.

For now, I just enabled the "mute till lock detect" feature of the ADF5356
and ADF4351 PLLs.  I modified gen_adf5356_regs.py and gen_adf4351_regs.py
to default it to be on, and the ld_cyc_count to be the longest possible.

Brian

On Wed, Oct 26, 2022 at 9:38 AM Brian Padalino bpadalino@gmail.com wrote:

On 2022-10-26 16:03, Brian Padalino wrote:

Hmm, interesting.  This is basically a "smoking gun" that it's the
actual synthesizer outputs, rather than some digital signalling--
  which you'd previously all-but-eliminated.

Interestingly, the switches used on the synthesizer outputs are
high-isolation, non-reflective switches, providing about
  60dB of port-to-port isolation, and terminating the
switched-away-from port to reduce tendency for lines to radiate.
  I'm kind of getting lost in the schematic, but I wonder if there's a
case where the switches are actually in the wrong state?

My guess is that if this is a "sneak path" issue, it will require a
hardware rev to fix, but if it's just a combination of
  muting the synthesizers during tuning, combined, perhaps, with a
slightly-wrong truth table for the various RF switches
  in the design, that's just software.

Hello Brian,

I set up my HW and SW, but I'm having some trouble reproducing after some
initial playing around with UHD examples and gnuradio. I assume you have
some UHD program that records RX to file and while you repeated make tune
requests to ch 1, is this correct? To make sure we are on the same page,
you provide a sample program that reproduces your steps for tuning and
recording the IQ samples.

You also mentioned " When trying to read the lo_locked sensor continuously,
generating SPI traffic I presume going to the PLLs, I get clean spectrum."
Can you also show in your sample program how to reproduce these steps

Once I can get the same spectrogram as you, I'll also look into the "mute
till lock detect" feature of the ADF5356.

Regards,

Wan

On Wed, Oct 26, 2022 at 4:03 PM Brian Padalino bpadalino@gmail.com wrote: