Cost Equation
for the SKA
Sander Weinreb, Caltech/JPL, sweinreb@caltech.edu
Larry D’Addario, NRAO and
SETI Institute, ldaddario@astro.berkeley.edu
1.
General
form of spreadsheet
2.
Performance
and cost parameters
3.
Results:
Cost vs antenna diameter and other parameters
4.
Front-end
model
5.
Signal
processing model









Front-End
Cost and Performance Model
|
Cooler Type |
Physical Temp |
Cooler Cost, K$ |
Noise @ 10 GHz |
Noise @ 50 GHz |
Noise @ F GHz |
|
None |
300 |
1 |
40 |
200 |
4*F |
|
Peltier |
180 |
3 |
24 |
120 |
2.4*F |
|
Pulse or Klemenko |
60 |
6 |
10 |
50 |
1.0*F |
|
Gifford-McMann |
15 |
20 |
4 |
20 |
0.4*F |
Total
Front-End Cost, FE$
FE$ = Cooler$ +
3(Bands)*2(Polarizations)* LNA$
LNA$ = $0.8K*(Fmax/10)^.3
i.e. FE$ = $11.9K for Pulse Cooling with Fmax = 20 GHz
Total
Electronics Cost, EL$, per Antenna
EL$ = FE$+
3*Feed$+LO$+IF$+Fixed$
Feed$ = $1K, LO$ = $5K, IF$ = 2*(0.5 + 2*BW),
Fixed$ = $1K
i.e. EL$ = 11.9 + 3*1 + 5 +
2*4.5 +1 = $29.9K
Tsys =
Tlna + [10 + 4* (F/10)]
Signal
Processing Cost Model
·
EACH ANTENNA
1. Digitization
Cd
= [kd (B/Nch)e
+ cd] Nch
(exponent e is approximately 2)
Cost
per channel increases faster than bandwidth => more channels
·
Presently expensive --
total for array is more than the correlator!
2. Tracking
·
Primarily delay and
phase tracking of sources
·
Per antenna per beam --
multi beaming is probably needed to mitigate small FOV of station beam.
·
All-digital
implementation assumed
·
EACH STATION
1. Summation to form phased-array, per beam
·
Multi-beaming probably
needed due to small station beam
2. Transmission to central processing facility
·
available bandwidth
shared among beams.
·
CENTRAL PROCESSING (CORRELATION)
1. Per antenna section
·
filter bank for L spectral channels
·
cost grows as B log(L) if minimum-memory FFTs are used
2. Interconnection section
·
can dominate cost for
large N
·
natural architecture
grows as B N 3
·
clever architecture
grows as B N log(N)
3. Cross correlation section
·
grows as B L N 2 (note
L dependence, even for FX
architecture)