Hi all, Thank you for your comprehensive responses! Changing the hardware increasing the number of bits might be the long term solution.
The RFI power is 50 times (17 dB) the total power form the noise form the whole times We adjusted the total output power to make sure it does not clip the samples at the ADC resulting to an standard deviation of 19 counts. The RFI is not impulsive, it is constant. I knew that correlators could operate using only the sign bit trough the van-vleck correction and total-power denormalization, but I did not know that this also applies to spectrometers. I will look into it. Thank you again, XB On Thu, Dec 7, 2017 at 3:43 PM Dan Werthimer <[email protected]> wrote: > > > > hi xavier, > > we use 8 bit wideband digitizers at arecibo and parkes where the RFI is > fairly strong. > do you think you might be able to get away with 8 bits ? : > > perhaps you know all this adc and dsp dynamic range stuff already, but > here are some thoughts: > > an 8 bit digitizer usually works well if the total power of the RFI > doesn't exceed about 10 times the total power from the noise from the whole > band. > (eg: RFI can be incredibly strong if it's confined to a relatively narrow > band compared to the total band). > although this general rule doesn't apply if the RFI is impulsive (see > below). > > have you studied the adc histogram from an 8 bit ADC at your site? > how often does the ADC saturate? > eg: if you set the signal level so the ADC RMS is around 5 (on a scale > from -128 to +127), > how often do you get -128 or +127? > > if you have strong impulsive RFI that occasionally saturate the ADC. > there are some casper blocks that can excise these low duty cycle events. > > if the total power from the RFI is significantly stronger than the noise > power from the total band, > then you might consider using an 8 bit ADC, but only using the sign bit > from the ADC. > (use the ADC as a one bit digitzer, that outputs only +1 or -1). > you'll loose some SNR with a one bit digitizer (see > thompson/moran/swenson), > and you'll need to apply the van vleck correction to measure strong > signals accurately, > but you won't have a dynamic range problem with a one bit digitizer. > > spectrometers and correlators can have 64 bits or more of dynamic range > with 8 bit or 1 bit digitizers. > (the digital part of your design can have lots of dynamic range, > independent of the ADC range). > but if the ADC saturates, and you don't handle the saturation > appropriately, > then you'll get spectral contamination no matter how many bits you use in > the post-adc signal processing). > > best wishes, > > dan > > > > > > > On Thu, Dec 7, 2017 at 2:10 PM, Xavier Bosch <[email protected]> > wrote: > >> Hi CASPERites, >> >> >> >> We are developing a wideband spectrometer (~1.6 GHz ) in a band that has >> a lot of RFI. Currently we are using the 8-bit 5-GSPS ASIAA ADC board and >> the ROACH2, and the ADC resolution appears to be inadequate for our extreme >> RFI environment. >> >> >> >> We would like to move over to a higher resolution ADC (12 to 14 bits), >> probably with a JESD204B interface, and a Xilinx development kit. We are >> considering something like this evaluation board KCU105 >> <https://www.xilinx.com/products/boards-and-kits/kcu105.html#hardware>and >> something like FMC217 >> <https://www.vadatech.com/product.php?product=528&catid_prev=0&catid_now=0> >> as a digitizer. >> >> >> Has anyone been in a similar situation dealing with RFI? >> >> Has anyone ported CASPER library to Xilinx development kits and JESD204B >> ADCs? >> >> >> Thank you, >> >> XB >> >> -- >> You received this message because you are subscribed to the Google Groups >> "[email protected]" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> To post to this group, send email to [email protected]. >> > > -- You received this message because you are subscribed to the Google Groups "[email protected]" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected].
