app/redpitaya/double_iq_pid_vco only positive frequency-deviations at output?

[aewallin]

When testing the double_iq_pid_vco example (on redpitaya14), we seem to get only positive frequency deviations, and when the output of the PI-servo is a negative number the DDS output disappears (or the tuning-word is completely wrong)

This video shows the output-DDS when the input-signal is close (<1Hz) to the demodulation-DDS frequency. The positive frequency-deviations seem ok, but the signal disappears (or jumps very far) when the PI-servo output changes sign.
https://youtu.be/UNrcE8mEVow?feature=shared

some guessing on what the issue could be:

1. the 'unsigned' keyword is used here https://github.com/oscimp/app/blob/b3d592c92f03a6798bbccbb023f46f92720c75b4/redpitaya/double_iq_pid_vco/design/double_iq_pid_vco.tcl#L124
   could this cause some problems when the PI-output is negative?

2. is there documentation on how the 'sign' variable works for https://github.com/oscimp/fpga_ip/tree/master/pidv3_axi ?

[jmfriedt]

We tried reproducing your setup and document the findings at
https://github.com/oscimp/app/tree/master/redpitaya/double_iq_pid_vco/doc

At the moment our setup seems to work, but I do see strange behaviour in case of overflows. This is also a very outdated Buildroot/oscimpDigital image so I must assess if this conclusion is still valid with the current version. I highlighted in the settings the important parameters, including the "dynamic" (i.e. user settable) bit shifter to prevent overflow. The overflow is clearly visible on the data2ram output, here streamed through 0MQ and displayed with GNU Radio. Do not be concerned with the sine wave segments: the datastream is not continuous (not using the newer ping-pong data2ram) and this behaviour is expected (missing data during FIFO communication).

Please also find a video in the same directory: this is a Red Pitaya running on its internal oscillator and not on the external frequency source, hence the huge frequency offset I tried to cancel by tuning the external synthesizer.

[aewallin]

Thanks, we made some progress with this and the double_iq_pid_vco example now seems to be working. We can try to reproduce the overflow-behaviour at some point, so that we can recognize it and avoid it.
With a 10 meter polarization-maintaining fiber used at 1348nm we were now able to test fiber-noise cancellation using the example - many thanks for making oscimp available and documenting it!

next we want to try the ping-pong data2ram (hopefully gap-free data collection), and also collect both I and Q outputs of the demodulating mixer, to more easily extract the phase in radians.

[jmfriedt]

Yes replacing the data2ram with the ping-pong version written by Baptiste is a real improvement. Baptiste provided fine demonstration examples at https://github.com/oscimp/app/tree/8c5b226e68bd895f09154c20ecd89e98aa45b7a5/redpitaya/dataReal2ram_pingpong and https://github.com/oscimp/app/tree/8c5b226e68bd895f09154c20ecd89e98aa45b7a5/redpitaya/dataComplex2ram_pingpong (Real and Complex inputs respectively) which should help figure out how they work.

One open question for me is that, since the beginning we have addressed frequency transposition using an analog analogy of NCO + mixer and then struggled to decimate to reach the few tens of kS/s with large FIR filters that include as many delays as tap cycles (limiting the closed loop control bandwidth). On the other hand White Rabbit is using the DDMTD [1] with an output rate equal to the beatnote between the two driving signals: I wonder if our choice was wise and what would have been the impact of going the White Rabbit Way. I do not have the answer :(
Emilien provided a DDMTD IP for OscimpDigital but we do not have (yet) a working example integrated in the app/ directory.

[1] Moreira & al, Digital Dual Mixer Time Difference for Sub-Nanosecond Time Synchronization in Ethernet