In most respects the Flexible Filter Bank (FFB) needs the tied-array to be set up in the same way as for VLBI with the exception of
This was contributed by William Deich
(wavelength/92 cm) x 55 arc sec
The filterbank receives the tied-array output of the array, and its field of view is thus limited to the synthesized beam at the phase center. At 92 cm, this is 55 arc sec (if the full 2.7km of the array is used).
Any standard WSRT frequency can be used. Usually long wavelengths are chosen because pulsars have steep spectra.
32 x 2 channels
There are 64 channels in the filterbank. The channel frequencies are set by mixing the 8 IF's from the DCB (continuum backend) with 8 LO's that are internal to the filterbank.
The usual configuration is to divide the 8 LO's into 4 pairs. Both members in each pair are set to the same frequency. One of each pair is mixed with the X polarization from an IF and the other with the Y polarization. Thus there are effectively 32 channels in each of two polarizations.
0 - 3 MHz per channel.
The channels can be filtered by either narrowband or wideband filters. The narrowband filter can be set to b_narrow = 100/(2^n) kHz, n >= 0.
The wide-band filter can be adjusted to any of 124 bandwidths that uniformly span b_wide = 354 kHz -- 2.8 MHz.
The signal is AC coupled in the filterbank, so that the DC level is removed from the signal. At this writing, the effective time constant for following the DC level is ~ 0.15 sec. Thus one cannot detect pulses that vary on timescales much longer than 0.15 sec. However, this time constant is to be replaced soon with a programmable time constant that can be adjusted between ~0.15 and ~100 sec. The programmable time constant will be available within two months -- in time for use with proposals submitted for the 15 Mar 1995 deadline.
10 -- inf microsec (100 -- 0 kHz)
The output of the filtered channels are smoothed by an "audio filter" that has a rectangular bandpass. The bandwidth is programmable: b = 100/(2^n) kHz, n >= 0.
This corresponds to smoothing times tau = 1/b = (2^n) * 10 microsec, for n >= 0
51.2 microsec -- inf.
The smoothed signal is sampled at intervals that can be dt = (2^n) * 51.2 microsec, for n >= 0.
(The hardware can actually support sampling as fast as every 1.6 microsec, but this ability is moot because the present control computer cannot transfer the sampled data sufficiently quickly to disk or tape.)
absolute time keeping to approximately 0.1 -- 1.0 microsec
If desired, scans can be triggered to begin on a multiple of a 10-second tick from the WSRT maser-controlled clock. (At this writing (21 August 1996), there is a design flaw in the FFB so that the actual data-taking start time starts at any time up to one sampling interval dt after the 10-second tick. This interval is effectively random and thus introduces an uncertainty of dt/2. However it will be corrected in the near future.)
Absolute start times require knowing the difference between the maser clock and UT. This can be calibrated to about 0.1 microsec by comparing the maser clock with the time from a GPS receiver.
If you do not start taking data on a 10-second tick, then the start time is only as accurate as the FFB computer clock -- typically set by hand with 1-second accuracy.
The sampled data are two-bit digitized, then transferred to disk or tape.
If all 14 antennas are used, the nominal gain of the WSRT is G = 1 K/Jy.
However, there are additional losses in SNR due to quantization when using the filterbank. The signal is 3-level digitized before summing the 14 signals (per IF) in the "adding box", followed by 4-level digitization in the filterbank. The combined increase loss in SNR (compared to the unquantized case) reduces the effective gain of the overall system to
G_eff = 0.71 G
The time to detect a pulsar is t = (m T_sys / eta G)^2 (2 N b)^-1 (W/(P-W)) S^-2
where
E.g. Assuming the above parameters and W/P = 0.2 [broad pulse!], t = 75 (10 mJy / S)^2 (500 kHz / b) sec
16.03125 bytes per sample interval.
The 2-bit data occupy 128 bits (16 bytes) per 64 channels. The additional space is required for header information.
Recording to tape: 35 hr x (dt / 409.6 microsec)
Recording to disk: 2.7 hr x (dt / 102.4 microsec)
If the sample interval is 409.6 microsec or slower, data can be written directly to high-density Exabyte tape, for a maximum recording space of ~310 x 10^6 samples per channel (35 hr at 409.6 microsec/sample!).
However, faster sampling requires writing the data to disk, and then transferring the data to tape when not acquiring data. The maximum disk file that can be used at present is about 1.5 GB, or about 2.7 hr of data at a sampling interval of 102.4 microsec.