Hi Mandana >> It is important to ensure that sufficient signal is coming out after >> requanisation to ensure >> reasonable correlator efficiency while avoiding clipping. This >> requires a calibration step that >> can be done using autocorrelation products. >> > Thanks for the useful comments. > Is there any suggestions on the calibration strategy and how to monitor for > clipping. Right now, I am going to re-adjust the gain on bins that are below > 1000 auto-corr units and try to make the spectrum as flat as I can ( by > looking at the spectrum every second). Once this is settled (or maybe it > won't), I will re-adjust gains only once every so-many minutes maybe. Does > this sound reasonable?
Our calibration strategy basically involves the following steps; 1. Take the autocorrelation products for the input being calibrated. 2. Divide each product by the number of accumulations. This gets you back to the average power of one output value of the requantiser for each frequency channel of your input. 3. The real part of these autocorrelation values (XX*) contains real(x)^2+imag(x)^2 where x is average output of the requantiser for each channel. real(x) and imag(x) are 4 bit values so real(x)^2+imag(x)^2 is a 9 bit value. We try to get this 9 bit value into the range 20-30 in the band of interest. This depends on the dynamic range of your signal of interest, more head room may be needed if you don't want clipping. These steps will need to be repeated a few times as quantisation is not a linear process and should be done using an input that matches the final signal of interest as closely as possible (power levels especially). The process can be automated using a script. The amount of times you need to redo it depends on how much your signal changes, for a similar power input it should not be necessary to do it very often (our RF input level is held at a certain level so we can pretty much calibrate once). The DC channel output of the FFT often has a higher power than other channels so this may cause clipping after requantisation even though the rest of the signal is fine. It may be useful to multiply this channel with 0 to remove it as it generally is not useful. We have a snap block connected to the output of the requantiser that can be used to take occasional snapshots of the data stream to check for clipping and a bit in a status register that can be monitored. It would be more important to monitor the ADC and FFT for clipping as these affect multiple channels. Hope this helps Regards Andrew
