Table 10.3 Selected correlator modes
# of Digitizers |
Bandwidth/ Digitizer |
Cross-pol Products? |
Channels/ Product 2 bit X 2 bit |
Channels/ Product 2 bit X 4 bit |
Channels/ Product 4 bit X 4 bit |
8 |
2 GHz |
Yes |
64 |
-- |
-- |
8 |
2 GHz |
No |
128 |
-- |
-- |
8 |
1 GHz |
No |
256 |
128 |
64 |
8 |
500 MHz |
Yes |
256 |
128 |
-- |
8 |
250 MHz |
No |
1024 |
512 |
256 |
4 |
2 GHz |
Yes |
128 |
64 |
-- |
4 |
1 GHz |
No |
512 |
256 |
128 |
4 |
500 MHz |
Yes |
512 |
256 |
128 |
4 |
250 MHz |
No |
2048 |
1024 |
512 |
2 |
2 GHz |
Yes |
256 |
128 |
64 |
2 |
1 GHz |
No |
1024 |
512 |
256 |
2 |
500 MHz |
Yes |
1024 |
512 |
256 |
2 |
250 MHz |
No |
4096 |
2048 |
1024 |
Normal multiplication = multiplier is 4-level X 4-level
2 X 4 multiplication = multiplier is 4-level X 16-level
4 X 4 multiplication = multiplier is 16-level X 16-level
This is accomplished in the following manner. In normal mode, the multipliers in every chip are used to multiply the two most significant bits coming out of each FIR filter (which carries 7-bit precision in its final stage). In 2 X 4 mode, half the elemental multipliers work on normal data, and half work on the two most significant bits from one antenna and the two next most significant bits from the other antenna, but at the same delay values as the normal multipliers for that particular channel. In 4 X 4 mode, 4 different cross- products are calculated for the same delay values:
"high" bits X "high" bits
"high" bits 1 X "next highest" bits 2
"next highest" bits 1 X "high" bits 1
"next highest" bits 1 X "next highest" bits 2
In either 2 X 4 or 4 X 4 modes, it is up to software to reassemble the various products from the accumulators which service them in order to get the full-precision product.
It is because some multipliers must be assigned to the extra cross-products that the frequency resolution decreases.