A block diagram of STAR is shown above.
After emerging from the instrument flange the beam first passes through
an LO diplexer which efficiently injects the local oscillator beams
into the signal path. The diplexer consists of a single, 3.162 mm thick
silicon wafer inclined 45 
to the incident beam (Mueller 1994).
The silicon wafer acts as a Fabry-Perot interferometer presenting
only 
% loss along the signal path, while reflecting
% of the LO power into the focal plane array.
The instantaneous bandwidth of the diplexer is 
GHz,
an excellent match to the IF bandwidth of the receiver.
In order to reduce optical losses, the silicon diplexer will also
serve as the vacuum window for the signal path.
An LO beam for each mixer will be produced by passing the FIR laser
output through a quasi-optical phase grating. Optical phase gratings
have been in use for over two decades. The application of phase gratings
to the generation of multiple LO beams at mm/submm wavelengths
is much more recent (Delgada 1995; Klein et al. 1997). Figure
4 is a numerical simulation of a 
array of LO beams
generated from a single LO source by a four level grating. The
beam to beam intensity variation is 
5% over a 13% bandwidth
(Klein et al. 1997). The phase grating has a transmission efficiency
of 83%. Laser micromachining (see section C.4.2) is ideally suited
for the manufacture of these gratings and will be used
to make four gratings, each offset in frequency by 10%. With these
exchangeable gratings, multiple LO beams can be generated across the full
operating range of the receiver. The high output power (several mW)
and large beamwaist ( 
mm) of the FIR laser
make it an excellent match to the requirements of the array LO.
The combined signal and LO beams then enter a passively cooled
cryostat. The cryostat (designed by Infrared Laboratory of Tucson, AZ)
has a hold time greater than 24 hours with the system fully
energized. With the receiver shutdown, the hold time will be 
hours. After passing through an infrared blocking filter, the
beams illuminate the mixer subarray. The subarray is composed of 16
mixers stacked in a 4 
matrix. Each mixer has a dielectric
lens which lies in the telescope's focal plane. The lens diameter
determines both the distance between mixers in the focal plane and the
angular spacing of the beams on the sky. By cutting away flat sections
from the sides of the lenses, the size of an array can be
significantly reduced. We have conducted beam measurements on an
80 GHz scale model of the horn/lens combination to determine the
optimum amount of truncation. We propose to truncate the lenses at
the 19 dB level ( 
), which will set the beam
spacing within each mixer subarray to 
and produce a
negligible amount of loss ( 
) and cross-talk between array
pixels. The optical system has been optimized using commercial
optical design software. Over the array aperture no significant
off-axis aberations are observed. The resulting subarray is quite
compact (31.2 
mm). At the center frequency of the array
(1900 GHz) the diffraction limited beamsize is 
. The spacing between adjacent beams on the sky will
be 
.
The 1 to 3 GHz IF output of each mixer is amplified by a low-noise HEMT amplifier mounted on a 15 K, He vapor-cooled, heat exchanger. (The IF frequency and bandwidth are set by the hot-electron bolometer response time and the availability of laser LO lines.) After further amplification at room temperature, the IF output signals are fed into an IF processor. The IF processor consists of 4 modules. Each module takes 4 receiver outputs, offsets them relative to each other by 250 MHz, and feeds them into 1 of the 4 available AOS channels. In this way, all 16 channels of STAR can be accomodated within sixteen 250 MHz sub-bands. (The IF Processor will be discussed further in section C.6.)
The receiver frontend and the four array AOS's are controlled by networked PC's. An UltraSparc is used to process the 4000 channels of spectral data during flight and will serve as an interface between the instrument and the observatory`s computer system.