Problems adjusting the power levels with large out of band birdies.


    The 700 Mhz receiver was looking at the region W44. There was a 40 Mhz filter centered at 710 Mhz after the pre-amps. The signal was mixed to 750 then 260 Mhz. A 12.5 Mhz and .78 Mhz band were input to the correlator. The .78 Mhz band was first filtered by a 260 IF filter 5 Mhz wide before being sent to the correlator chassis. There is a 50 Mhz IF filter centered at 260 MHz in the correlator chassis for each signal path. The signal then goes to the power counters (which see all 50 Mhz) and the A to D converter (8 bits) which digitizes the data and then passes it to the digital filters, and then the correlator. The correlator was being run in 3 level stokes mode. 6 sets of 100 records (3 seconds per record) were taken. Position switching was done every 100 recs. The power levels were adjusted using the 0 lags once for the on position and then once for the off position. After the first set, the attenuators for the on or off were restored prior to the new on or off. The tv station channel 54  was turned off for these measurements. The power counter data was normalized by dividing each record of a group of 100 by the median for that group. The problem is that the 12.5 Mhz band 0 lag power decreases as the power counter increases. It is worse for the run that begins with the correlator measured/optimum power set to 1.5.  You would expect that a birdie outside the 12.5 Mhz band could cause the power counter value to increase while the 0 lag data remained constant, but the 0 lag data should not decrease. We must be saturating something in the digitizer, digital filters or conversion to 3 levels. The .78 Mhz band has no negative slope. Of course the power counters are only seeing a 5 Mhz bandwidth because of the IF filter.

    When filtering to a narrow bandwidth before the digitizer you run into the problem that all of the bits of the digitizer will not be used.  For a .78 Mhz bw there is an optimum power level. The digitizers are set up assuming that they see 50 Mhz noise. If you filter this ahead of time then the power at the digitizer will drop by 50Mhz/filter width. In our case this was 10 db. If the digitizer normally are set for full scale operation, then we had sqrt(10) fewer bits.