Red Pitaya low-frequency transfer function

[Lenny]

Hello everybody,

I have been using my RedPitayas successfully for some PID applications. However, as I get more concerned about the precise shape of the transfer function, I measured it for pid11 with various proportional gains. I measured in the region between 0 and 200 kHz with an Agilent E5061B Network analyzer, and a 50Ohm resistor in parallel with input one of the RedPitaya. See the attached plot that shows the magnitude of the transfer function.
What I find looks good at higher gains; however at low gain, the response around 30kHz is not even monotonous with the proportional gain any more: the p=1 curve lies above the p=10 curve in a 50kHz-window. I measured similar curves for two different RedPitayas.
Does anyone know what I can do about this problem? Is it related to the input/output stage filters, or rather something I can work on, like the power supply buffer etc?

Thanks for any help! Lenny

[redpitaya]

Hi, Lenny

Hmm.

If I convert p-s from counts to gains - you have done measurements for this range of gains (approximately)
2 -1- 0.025 - 0.0025-0.00025-0
I think that in the case of tree last gains the signal/noise ratio is too low.
On LV jumper settings with 1Vpp input amplitude your output signal will go to few mV or less.

Regards

[lene85]

input.png

output.png

Hello everyone,

after a long pause this same issue bothered me so I did some scope measurements of the transfer function of my red pitaya.
I use the PID11 with only proportional gain. A setting p=4096 would digitally correspond to a gain of unity with my FPGA code.

The output goes to a 50Ohm oscilloscope input.
Th input has a 50Ohm resistance in parallel to avoid any reflection issues and is fed by a 10 kHz sine wave from a network analyzer (very proper signal, see picture).

When changing the proportional gain to low values, I get some very nasty results (see plot). In principle, the digitized input signal (about 0.6V amplitude) should oscillate in the redpitaya between +-4500. So a gain p=1 (normalized gain would be 1/4096) should just about create an output signal with an amplitude of +-1 bit or 0.25mV. So what I would expect is that this trace be basically impossible to distinguish from the p=0 reference curve. However, the curve with p=1 rather resembles the one for p=40.
And even stragner, a proportional gain between 1 and 40 gives an output which is smaller than the p=1 curve! In other words, the transfer function seems not to be monotonic with the proportional gain..

This behaviour makes no sense at all. Neither do I see this caused by the FPGA code (it is basically the same code that comes with the RedPitaya with some modifications which do not affect the proportional pid part), nor can I see an analog cause because when I set the proportional gain to 0 the 10 kHz signal definitely disappears..

So my question: What is going on here? Is there some additional circuitry which disables the output stage completely (automatically) if it is held at zero? Where does the noise of the p=0 stage come from? The power supply? Can one suppress it a little? And finally, but probably in vain: Can one have a look (or a more detailed description) of the analog input and output stage? In addition to above enigma, I have also often seen a dramatic difference between the negative and positive half-sines that go through the red pitaya, and some ugly jumps around 0V (such that for small signals, the negative-half-sine is shifted towards positive voltages and the positive one is shifted towards negative ones. This looks like the negative and positive half-sines are created by different analog stages?? It would be useful to know whether this happens at the input or output stage. In that case I would add an analog offset to my signal to avoid this nasty region.

Thanks for any help! Lenny


EDIT: The picture upload did not work with my browser. You can find the plots here:
The signal sent to the input:
https://drive.google.com/open?id=0B7nhU ... 3VMZU5zOWc
The signal output of the redpitaya:
https://drive.google.com/open?id=0B7nhU ... WlpaEc2ZjQ

[lene85]

plots are here..

[DF4IAH]

... for your amusement, check this hardware issue and do following :

- go into any generator application of the RedPitaya
- take 10 kHz for output, maximum voltage and select the pulse waveform.
- now take a good external oscilloscope and look for the rising and falling edges

1) for amusement you will find out, that a 15 MHz sine wave is overlayed onto the pulse, that would trigger the external scope.
The amplitude of this "additional signal" is abt. 10% of full scale. The RedPitaya signal gets better with a load of 50 Ohms,
but the 15 MHz signal is still there having a lower amplitude then.

2) for amusement you'll find out also, that the signal does not go to 1/2 of the amplitude when loaded by the 50 Ohm resistor,
as it should be on a 50 Ohms output.

Thinking process follows: Amusement 1 with Amusement 2 gives the idea, that the DAC output buffer is not built as an
electronic professional would expect ...

Would that explain somehow about the signal quality that you found out with your $$$ equipment? Still I had now chance to
get in touch of any schematics of RedPitaya - for a good reason?

Cheers
Uli

[lene85]

Hi Uli,
thanks for your reply. I also noted an anomaly in the relation between measured voltage at the RedPitaya output and measurement impedance. I guess one would usually insert a 50Ohm resistance in series with the output SMA, and here it seems the output buffer is directly connected to the SMA? Maybe then one can improve performance by putting a 50Ohm resistor in series directly at the output? Anybody tried that? Of course this would result in dividing the output amplitude by 2..

The 15 MHz signal would not bother me, but it definitely makes me more curious about other potentially hidden problems..

Recently there was a post in this forum with a reverse-engineered schematic, but the picture does not exist any more:
viewtopic.php?f=9&t=386
Does anyone have a copy of that file (in case it really exists)?

Anyway, this whole discussion based on the lack of information from the constructor is starting to get a little absurd, so lets hope they jump in with some news...

[Nils Roos]

I happened to archive it: RP_IN1.png

[lene85]

Thanks! Even though it would of course be even nicer to have the same picture of the output stage

For my part, I checked the DAC datasheet (http://www.nxp.com/documents/data_sheet/DAC1401D125.pdf) and on page 6 it details the differential nonlinearity to be maximally 1.1 LSB. Therefore my curves are not explained by the DAC alone.

Now to make sure, I swept the offset of the function generator of the output from counts -450 to 450 (remaining at each count for about 10ms), and finally put it to zero. The result is here:
https://drive.google.com/open?id=0B7nhU ... nVPVnoyeVE

At the transition -1 -> 0 and above, the response gets nonmonotonic. Can someone else reproduce this to tell me if its a problem with only my redpitaya? Any opinions on how to get rid of this?

Thanks!