LBA_DAS :  Some Passband Shapes					RHF		21/7/97
				===============================

1)  Setup: a 4-bit (AT) sampler, the (partially debugged) Tied Array box with
4 identical inputs from the sampler, an N3 module with direct connection to an
AT correlator, all fed from a band limited noise source with a ~3dB rising
spectrum across 64MHz.

2)  The spectrum was effectively flattened by increasing the gain in the TA box
until its adder saturated.  Note that since the 4 "antennas" were adding
coherently, the TA data was still 4-bit before saturation, and therefore only
3-bit afterwards.  Some residual slope across the band is still present but
this is only noticable in the 32MHz bandpass.

3)  Five passbands were recorded: B1*32, B1*16, B1*16S, B2*16 LSB & USB.  The 
spectrum of B2*16L was flipped so that input frequency runs left to right 
across all plots.  Strictly speaking these are power spectra of the input data 
streams.  They differ from the actual filter frequency response in the 
transition zones at both edges, where energy from the adjacent stop bands is 
aliased in.  There is negligable distortion of the bandpass between the 1dB 
points.

3.1) A pair of B2*4 passbands (Lower flipped) from the first DAS observation 
has been included to extend the set.

4)  Analysis.

  Band edges 1_Lower & 1_Upper are measured at the 1dB points (allowing for
input noise slope where necessary).  At this point the spectrum is at 80%
(power) of the mean passband, and the adjacent alias level is <= 1.5%
(-18.3dB).

  Band edges R_Lower & R_Upper denote the equiripple bandpass.

  The frequency axis is normalised to sample frequency in the usual way.

  Bandpass 'efficiency' is calculated as 100*(Upper-Lower)/0.5.

  Ripple is 100*(peak-peak)/mean, power scale.

  For B1*32 & B1*16S, Ripple is unresolved and the expected value is shown.

  The last digit in each measured datum is subject to measurement noise.


Filter	1_Lower	1_Upper	1_B-Effy	Ripple	R_Lower	R_Upper	R_B-Effy	Edge
------	-----	-----	--------	------	-------	-------	--------	----
B1*32	0.018	.482	92.7%		~0.1%	0.046	0.454	81.7%		0.52

B1*16	0.022	0.478	91.3%		2.6%	0.037	0.463	85.4%		0.51

B1*16S	0.030	0.470	87.8%		~0.6%	0.054	0.445	78.3%		0.36

B2*16Lf	0.035	0.458	84.5%		2.9%	0.062	0.429	73.5%		0.50

B2*16U	0.041	0.464	84.6%		2.4%	0.070	0.438	73.6%		0.50

B2*4Lf	0.042	0.458	83.3%		4.9%	0.064	0.437	74.6%		0.50

B2*4U	0.044	0.461	83.3%		5.2%	0.065	0.438	74.5%		0.49

5)  Comments on 4)

  The spectra plotted always differ from the actual filter bandshape in the
transition regions on each side, due to energy in the adjacent alias bands
being folded on top of the principal response.  The distortion varies from 3dB
at band edge to less than 0.01% (ie. < -40dB) in the equiripple zone.

  B1*<n> and B1*<n>S responses for n=8,4 & 2 scale in frequency from the
B1*16(S) responses.  The 1MHz bandshapes are a little different.

  B2*<n>, n=8,4,2 & 1, responses do not scale from the B2*16.  They have ~5%
ripple through the passband.  The L and U sideband responses are mirror images
of each other.  Small differences tabulated in 3) represent experimental error.

  F1 and F2 bandshapes all scale directly from their corresponding parent B1
and B2 responses.  Given a particular (Band Splitter,Fine Tuner) configuration
the FT filter's parent response can be found from the following table.


	  F  \  BS 	*1		*2		*4		*8		*16
	  T   \
		   \_________________________________________
	*0.0625	|	B1*1	-		-		-		-
	*0.125	|	B1*2	B1*1	-		-		-
	*0.25	|	B1*4	B1*2	B1*1	-		-
	*0.5	|	B1*8	B1*4	B1*2	B1*1	-

	*1		|	B1*16	B1*8	B1*2	B1*2	B1*1
	*2		|	-		B1*16	B1*8	B1*4	B1*2
	*4		|	-		-		B1*16	B1*8	B1*4
	*8		|	-		-		-		B1*16	B1*8
	*16		|	-		-		-		-		B1*16


  The B1*<n>S bandshapes are 7.5dB down at the Nyquist band edges.  All other
responses are 6dB down.

				-----------------------------------------