PDSP16488A MA
those in a single device. This compensates for the twelve
delays added to the convolver sums in the second row, plus
an additional eight delays to compensate for the partial width
of the first device in the secind row.
Four devices can also be used to give an 8x8 window,
but with a 30 MHz pixel clock. Each device is configured to
provide a 4x4 partial window, but the maximum pixel rate is
reducedfrom40to30MHzbecauseoftheresponseoftheline
delay expansion circuitry. Intermediate precision is restricted
to 16 bits, since time multiplexed data outputs cannot be used
above 20 MHz.
the normal way and has its MASTER and SINGLE pins left
open circuit.
The internal convolver sum, in the device producing
the final result, must be delayed by 4 pixels to match the
inherent delay in the expansion output from the other device.
Thisisactuallyachievedbydelayingthepixelinputstotheline
stores [ Register D bits 3:2 = 01 ]. No additional delay in the
expansion input is needed, but the pipeline delay used to
produce DELOP must be four clocks greater than that given
in Table 4 for a single device. The DELOP output is redundant
in one of the two devices.
This configuration requires no additional delay in the
expansion inputs, and the inputs to the line stores in both
devices in the second row must be delayed by 8 clock cycles
[ Register D bits 3:2 = 10 ]. The DELOP output needs twelve
additional clock delays to match the processing delay.
Figures 14 and 15 show non-interlaced and interlaced
versions of the above 8 x 8 and 4 x 4 arrangements
Figure 16 shows how four devices can also be used to
provide an 8x8 window, with 16 bit pixels and 20MHz clock
rates. The expansion data from a previous device needs no
additional delay since the partial window size in each device
is only 4x4. The internal convolver sums from each device in
the second row must be delayed by 8 Clks and the DELOP
output must have 12 additional delays. If this arrangement is
to be used in a non-interlaced application, the field store must
be replaced by four line delays.
Two devices can also be used to support systems
requiring 16 bit pixels. With this approach the 16 x 8 multipli-
cation is mechanized as two 8 x 8 operations, with the results
added together after the most significant half has been shifted
by 8 places to the most significant end. This shift operation is
controlled by Register D, Bit 1. Both convolvers are pro-
grammed to contain the same coefficients. The convolved
output can theoretically grow to 30 bits, and the appropriate
field must be selected before using the gain control.
Examples of this operating mode are shown in Figure
13. Each device must be configured in the same 8 bit pixel
operating mode, but the device producing the final result must
use the 8 place shift option on its internal sum.
The least significant 8 bits of the pixel are connected to
the MASTER device and the most significant 8 bits are
connected to the device producing the final result.. The
internal sum in this device must be delayed by four pixels to
match the delay in the expansion output from the first device.
Thisisactuallyachievedbydelayingthepixelinputstotheline
stores( Register D, bits 4:2, = 001 ]. The expansion input
needs no additional delay [ Register D bits 1:0 = 10 ].
The actual pixel precision can be any number of pixels
between 8 and 16, and may be a signed or unsigned number.
Any unused, more significant bits, must respectively be either
sign extended or be tied low.
SIX DEVICE SYSTEMS
As shown in figure 17, six devices, each in an 8Wx4D
mode using 8 bit pixels, can provide a 16W x 12D window at
20MHz clock rates. Expansion inputs from previous devices
in a row [but not the first device in each row] need an extra 4
Clks of delay since the partial window is eight pixels wide.
Internal convolver sums need a differential delay of 12 Clk
cycles from row to row [ Register D bits 3:2 = 11 ].
DELOP must have four additional pipeline delays in order
to match the total processing delay. This output can be
obtained from either device.
The DELOP output must have 32 additional delays to
match the total processing delay.
EIGHT DEVICE SYSTEMS
FOUR DEVICE SYSTEMS
Two additional chips will extend the above six device
configuration to a 16 x 16 window. Internal convolver sums
musthavedifferentialdelaysof12clockcyclesbetweenrows,
as in the six device system. The DELOP output needs 44
additional clock delays.
Four devices, each in the 8x8 mode, can be used to provide
a 16 x 16 window, with 8 bit pixel resolution and 10 MHz clock
rates. The partial sum from the first device in each row must
be delayed by eight pixel clocks before it is added to the result
from the next device. This provides the eight pixel displace-
ment to match the width of the window. The delay is actually
providedbyfouradditionaldelaysintheexpansioninputtothe
next device, plus the inherent four clock delays in outputing
results from the first device. Register D, Bit 0 controls the
additional delay.
The internal convolver sums, in the two devices in the
second row, must be delayed by 12 clocks before they are
added to the result from the first row. This twelve clock delay
is necessary because of the combination of the eight pixel
horizontal displacement delay , and the four clock delay in
outputing the result from the last device in the top row. It is
actuallyachievedbydelayingthepixelinputstothelinestores.
(Register D, bits 3:2 = 11 ].
NINE DEVICE SYSTEMS
Nine devices each in the 8 x 8 mode will provide a 24
x 24 window with 8 bit data and 10 MHz pixel clocks. This is
shown in Figure 18. Expansion data inputs from previous
devices in a row [ but not the first device in each row ] need an
extra 4 Clks of delay. The internal convolver sums need
differential delays of 20 Clk cycles between rows. Sixteen of
the latter delays can be provided internally by setting Register
B, bit3, and also Register D, bits 3:2. The four extra delays
must be provided externally.
The DELOP output needs 56 clock delays in addition
to the 29 required for the 8 x 8 single device configuration.
The DELOP output must have 20 delays additional to
19
MITEL [ MITEL NETWORKS CORPORATION ]
