Photodetector Array Power Dissipation

Radobs 22

For quantum-noise limited heterodyne detection we need a certain minimum
optical local-oscillator power at each photodiode.  Since we would like a
10 GHz bandwidth from each pixel, RC time-constant considerations dictate
that each photodetector load must be of low impedance.  There are two basic
ways to minimize the amount of local-oscillator power required.  Use
transimpedance front-end amplifiers with a PIN photodetector array, so that
while relatively high impedance loads may be employed, the transimpedance
amplifier configuration produces small RC time-constants, or we can use
Avalanche photodetectors (APDs).  It might be possible to construct an array
based on APDs at modest gain with 50-ohm loads, to avoid using trans-
impedance amplifiers, and/or to reduce the local-oscillator laser power
requirements, but there would be problems with gain stability, uniformity of
gain, and tracking of gain across the array.

For the purposes of this analysis we will consider only PIN photodetector
arrays with transimpedance amplifiers.  At this time, transimpedance
amplifiers are not available for this bandwidth, though hopefully they will
be in a few years.  Signetics has a 280 MHz 7-kohm transimpedance amplifier
called the NE5210, but its bandwidth is too small for our purposes. 
However, state-of-the-art transimpedance amplifiers with 1 GHz bandwidths
are becoming available.

A bandwidth at least as large as 10 GHz is preferred for the Multi-Channel
Spectrum Analyzer (MCSA) so that the optical spectrum can be sampled more
efficiently (rapidly) in 10 GHz chunks.  Such an MCSA is substantially wider
in bandwidth than proposed for the Microwave Observing Project (MOP), which
is typically 300 MHz, with about 10 million 30 Hz top-level bins per
polarization (left and right circular).  However, because the bin bandwidth
I have proposed for the optical search is about 100 kHz, the number of bins
per polarization for a MCSA with an instantaneous bandwidth of 10 GHz, is
only 100,000 (not accounting for any frequency overlap required).  For a
128 X 128 array there are 16,384 pixels.  If we require 128 MCSAs to sample
an entire row of pixels at a time, the number of bins is slightly larger
than proposed for MOP.  Even if we are quite extravagant and stipulate one
10 GHz MCSA for each pixel, the number of top-level bins required is only
164 times that required for a two polarization-state MOP.  On-going
developments in MCSA technology should make this easier and less expensive
to achieve in the future.

As previously mentioned, unlike with (incoherent) photon-counting arrays,
(coherent) heterodyning arrays require a sufficient amount of local-
oscillator (L.O.) laser power at each pixel to ensure shot noise limited
detection.  Ideally, we would like to use a simple 50-ohm load, but this
would require a very high local oscillator power level to maintain the
quantum noise limited sensitivity.  At visible wavelengths, about 20 mW per
pixel would be required to be quantum noise limited, so that a 128 X 128
array (16,384 pixels) would have to dissipate 328 W!  This doesn't even
include the electrical power dissipated in each photodetector, which is
several times larger.  Clearly, if we are going to illuminate all the pixels
simultaneously with the L.O., this is not going to be possible with 50-ohm
loads.  If each load could be increased to 1-kohm, then the magnitude of the
power dissipation problem could be substantially reduced.

Assuming that it is possible to produce a 10 GHz bandwidth optical front-end
with an effective load or transimpedance of 1-kohm, a local-oscillator power
of about 1 mW per pixel will be required for quantum noise limited
detection.  A transimpedance amplifier with a gain of about 20 dB would be
suitable, but it is presently a technological challenge to produce a
suitable very compact super-wideband amplifier which would be stable. 
Phase-gain margin considerations alone dictate that the amplifier must be
physically very small, in order to prevent negative feedback becoming
positive feedback at very high frequencies.  If we assume a unity gain
bandwidth of 30 GHz, the wavelength in air is only 1 cm.

For a L.O. power of 1 mW on each pixel (photodiode), the typical
photocurrent produced by the L.O. laser in the red part of the spectrum will
be about 0.45 mA per pixel.  If the reverse bias applied to the photodiodes
is 5 volts, then the electrical power dissipated per photodetector is
2.25 mW, and the total power dissipated is 3.25 mW.  This is not a problem
for a single photodiode, but in an array of 16,384 photodiodes, the total
power would be 53 W!  It is fortunate that we don't require a 2048 X 2048
array!  This is another occasion when the expression "holy smokes" is again
applicable.

If the bandwidth required was only 1 GHz, and if it was possible to produce
a 10-kohm transimpedance, then the L.O. power could be reduced to 0.1 mW,
with a reduction in photodiode current to 0.045 mA.  The total dissipated
power would then fall to 0.325 mW and the array power would be 5.3 W!  This
is still somewhat high, and would require a substantial amount of cooling. 
Note that this magnitude of continuous wave (C.W.) power level is available
from dye, gas and Nd:YAG lasers.

There are a number of ways to get over the array heat dissipation problem:

1.   Reduce the number of pixels that have to be simultaneously illuminated
     by the local-oscillator laser, by using a line-focused local-oscillator
     beam and a linear array, and scanning the image over the array with a
     mirror.  Total power dissipated in the linear array for a 10 GHz system
     would be about 0.42 W.

2.   Use a line-focused local-oscillator beam which scans across the two
     dimensional array in synchronization with the field scan rate.  Total
     power dissipated in the two-dimensional array for a 10 GHz system would
     also be about 0.42 W.

3.   Use a focused local-oscillator beam, and steer it so that it addresses
     each pixel sequentially in synchronization with the two-dimensional
     array output sampling, assuming sequential pixel sampling.  Acousto-
     optic deflectors might be used for this purpose, though there are some
     problems here caused by the inherent varying frequency-shift of the
     scanned L.O. beam.  Total power dissipated in the array for a 10 GHz
     system would be about 3.25 mW.  We would probably have to use a bank of
     electronic mixers to remove the differential frequency offsets produced
     by the acousto-optic modulator(s).  An added advantage of this approach
     is that a very low-power L.O. laser may be employed.

It might be possible to multiplex each element or row of elements with a
single transimpedance amplifier.  Recent years have seen major developments
in MIMIC (Microwave/Millimeter-wave Monolithic Integrated Circuit)
technology based on GaAs devices, so much may be possible here.  What ever
is done, we are basically talking about a state-of-the-art custom PIN
photodetector array that has an integral transimpedance amplifier for each
element or one that is shared between many pixels.  We should look very
closely at developments occurring in the fiber-optics industry, particularly
relating to coherent fiber-optic communications and parallel signal
processing arrays.  I even wonder if it might be possible to tie all the
photodiodes together to one transimpedance amplifier via extremely low
capacitance optical switches (a bit like photodiodes), and actually
sequentially switch each pixel into the on-state with a scanned L.O. beam? 
It would appear that an Optical SETI program will require the services of a
photodetector manufacturer for developing this specialized product.

For the purposes of future discussions, we shall assume that the array, and
mirror scanning system if employed, is equivalent to a 128 X 128 pixel
device.

Coming Attractions:
RADOBS.23  THE OPTICAL SEARCH STRATEGY
RADOBS.24  TARGETED SKY SURVEY
RADOBS.25  COPYRIGHT NOTICE
RADOBS.26  LIST OF MAIN OPTICAL SETI FILES UPLOADED TO RADOBS BULLETIN BOARD

Last week, I happened to spot a book in the Bexley library by Walter
Sullivan entitled "We are not alone - the search for intelligent life on
other worlds".  Chapter 15 in this book has the most detailed account of
Optical SETI, i.e., using laser communications, that I have seen to date in
a general interest publication, save for science fiction stories.  The
explanation for this is probably due to the book's publication date.  It was
published in 1964, only a few years after the laser was invented, and only a
few years after the start of the microwave SETI era.  At that time, there
was more interest in new SETI ideas and the ideas of Schwartz and Townes,
but eventually the microwave approach won out.

January 27, 1991
RADOBS.22
BBOARD No. 335

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* Dr. Stuart A. Kingsley                       Copyright (c), 1991        *
* AMIEE, SMIEEE                                                           *
* Consultant                            "Where No Photon Has Gone Before" *
*                                                   __________            *
* FIBERDYNE OPTOELECTRONICS                        /          \           *
* 545 Northview Drive                          ---   hf >> kT   ---       *
* Columbus, Ohio 43209                             \__________/           *
* United States                            ..    ..    ..    ..    ..     *
* Tel. (614) 258-7402                     .  .  .  .  .  .  .  .  .  .  . *
* skingsle@magnus.ircc.ohio-state.edu         ..    ..    ..    ..    ..  *
* CompuServe: 72376,3545                                                  *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *