cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
norume
Adventurer
Adventurer
870 Views
Registered: ‎04-01-2015

RF-ADC Training Signal?

Jump to solution

The "Background Calibration Process" of "Zynq UltraScale+ RFSoC RF Data Converter" (PG269) says, "For applications that requires fast convergence, consider providing a training signal that can be eventually shut off to train the GCB and TSCB blocks right after the startup state machine is completed."

What should this signal look like?  A square wave, a sine wave, or something else completely?  What frequency?  Is there an optimum range of frequencies?

 

0 Kudos
1 Solution

Accepted Solutions
klumsde
Moderator
Moderator
729 Views
Registered: ‎04-18-2011

Hi @norume 

You need to put a signal in the desired region. So the training signal must be in the same nyquist zone as the "Real" signal. 

It must also match the mode1/mode2 requirement. 

You must also be careful about what signal you use. The ADC needs diversity of data for it to converge it's background calibration. For that reason you can't pick a signal that is an integer sub multiple of the sample rate and in phase with the sample clock. 

You also would have the option here to take the calibration coefficients from a "Trained ADC"and bypass them using the save/load feature. 

Hope it is clear

Keith

-------------------------------------------------------------------------
Don’t forget to reply, kudo, and accept as solution.
-------------------------------------------------------------------------

View solution in original post

0 Kudos
3 Replies
drjohnsmith
Teacher
Teacher
826 Views
Registered: ‎07-09-2009
As far as I know,
its the amplitude of the signal that's important,
it needs to be above around -30 db, and symmetrical around zero,

A system I was on used "white" noise, but that as far as I can remember , was just convenient to that system,
<== If this was helpful, please feel free to give Kudos, and close if it answers your question ==>
0 Kudos
norume
Adventurer
Adventurer
762 Views
Registered: ‎04-01-2015

Yes, the documentation does specify a lower limit (-40 dBFs -- i.e. 1% of full scale) but provides no other guidance other than that signals at DC, and at 1/16, 1/8, 3/16, ... the sampling frequency are ignored. 


Gain Calibration Block (GCB)

This block corrects for the gain difference among the interleaving sub-RF-ADCs. Any residual difference results in spurious signal at +/-fin + (k/N)*Fs, where Fs is the sample rate of the RF- ADC, N is the number of sub-RF-ADCs, and fin is the frequency of the input signal. The GCB requires the presence of an input signal with power greater than -40 dBFs for an accurate correction of the gain skews. Signal bins at f = k*Fs/(2N) are ignored and do not count towards the signal power. This block should be in the freeze mode whenever the input signal power drops below -40 dBFs for longer than 100 μs.

Time Skew Calibration Block (TSCB)

This block corrects for the time skews among the interleaving sub-RF-ADCs. Any residual difference can result in spurious signal at +/-fin + (k/N)*Fs, where Fs is the sample rate of the RF- ADC, N is the number of sub-RF-ADCs, fin is the frequency of the input signal The TSCB also relies on the presence of an input signal with input power greater than -40 dBFs to provide accurate estimate and correction of the time skews among sub RF-ADCs. This block should be in the freeze mode whenever the input signal power drops below -40 dBFs for longer than 100 μs

0 Kudos
klumsde
Moderator
Moderator
730 Views
Registered: ‎04-18-2011

Hi @norume 

You need to put a signal in the desired region. So the training signal must be in the same nyquist zone as the "Real" signal. 

It must also match the mode1/mode2 requirement. 

You must also be careful about what signal you use. The ADC needs diversity of data for it to converge it's background calibration. For that reason you can't pick a signal that is an integer sub multiple of the sample rate and in phase with the sample clock. 

You also would have the option here to take the calibration coefficients from a "Trained ADC"and bypass them using the save/load feature. 

Hope it is clear

Keith

-------------------------------------------------------------------------
Don’t forget to reply, kudo, and accept as solution.
-------------------------------------------------------------------------

View solution in original post

0 Kudos