The HSP43891 is a video-speed Digital Filter (DF)
designed to efficiently implement vector operations such as
FIR digital filters. It is comprised of eight filter cells
cascaded internally and a shift and add output stage, all in
a single integrated circuit. Each filter cell contains a 9x9
two’scomplementmultiplier,three decimation registers and
a 26-bit accumulator. The output stage contains an
additional 26-bit accumulator which can add the contents of
any filter cell accumulator to the output stage accumulator
shifted right by 8-bits. The HSP43891 has a maximum
sample rate of 30MHz. The effective multiply-accumulate
(mac) rate is 240MHz.
The HSP43891 DF can be configured to process expanded
coefficient and word sizes. Multiple DFs can be cascaded
for larger filter lengths without degrading the sample rate or
a single DF can process larger filter lengths at less than
30MHz with multiple passes. The architecture permits
processing filter lengths of over 1000 taps with the
guarantee of no overflows. In practice, most filter
coefficients are less than 1.0, making even larger filter
lengths possible. The DF provides for 8-bit unsigned or
9-bit two’s complement arithmetic, independently
selectable for coefficients and signal data.
Each DF filter cell contains three resampling or decimation
registers which permit output sample rate reduction at rates
1
of
/2, 1/3 or 1/4 the input sample rate. These registers also
provide the capability to perform 2-D operations such as
matrix multiplication and NxN spatial
correlations/convolutions for image processing applications.
File Number
2785.5
Features
• Eight Filter Cells
• 0MHz to 30MHz Sample Rate
• 9-Bit Coefficients and Signal Data
• 26-Bit Accumulator per Stage
• Filter Lengths Over 1000 Taps
• Expandable Coefficient Size, Data Size and Filter Length
• Decimation by 2, 3 or 4
Applications
• 1-D and 2-D FIR Filters
• Radar/Sonar
• Digital Video
• Adaptive Filters
• Echo Cancellation
• Complex Multiply-Add
- Sample Rate Converters
Ordering Information
TEMP.
PART NUMBER
HSP43891VC-200 to 70100 Lead MQFP Q100.14x20
HSP43891VC-250 to 70100 Lead MQFP Q100.14x20
HSP43891VC-300 to 70100 Lead MQFP Q100.14x20
HSP43891JC-200 to 7084 Lead PLCC N84.1.15
HSP43891JC-250 to 7084 Lead PLCC N84.1.15
HSP43891JC-300 to 7084 Lead PLCC N84.1.15
HSP43891GC-200 to 7085 Pin CPGAG85.A
HSP43891GC-250 to 7085 Pin CPGAG85.A
HSP43891GC-300 to 7085 Pin CPGAG85.A
RANGE (oC)PACKAGEPKG. NO.
Block Diagram
VCCV
DIENB
CIENB
DCM0 - 1
ERASE
CIN0 - 8
RESET
ADRO - 2
RESET
SHADD
SENBL
SENBH
CLK
CLK
5
DF
FILTER
CELL 0
5
5
3
ADR0, ADR1, ADR2
2
DF
FILTER
CELL 1
1
DIN0 - DIN8
9
9
26
DF
FILTER
CELL 2
26
CAUTION: These devices are sensitive to electrostatic discharge; follow proper IC Handling Procedures.
CLKG3IThe CLK input provides the DF system sample clock. The maximum clock frequency is 30MHz.
DIN0-8A5-8,B5-7, C6,
DIENBC5IA low on this input enables the data sample input bus (DIN0-8) to all the filter cells. A rising edge of the
CIN0-8A9,B9-11, C10,
ALIGN PINC3Used for aligning chip on socket or printed circuit board. This pin must be left as a no connect in circuit.
CIENBB8IA low on this input enables the C register of every filter cell and the D (decimation) registers of every
COUT0-8B2, B3, C1, D1,
COENBA2IAlow on the COENB input enables the COUT0-8 outputs. A high on this input places all these outputs
DCM0-1L1, G2IThese two inputs determine the use of the internal decimation registers as follows:
NUMBERTYPENAME AND FUNCTION
B1, J1, A3, K4,
L7, A10, F10,
D11
A1, F1, E2, K3,
K6, L9, A11,
F11, J11
C7
C11, D10, E9,
E10
E1, C2, D2, F2,
E3
+5 power supply input.
Power supply ground input.
IThese nine inputs are the data sample input bus. Nine-bit data samples are synchronously loaded
through these pins to the X register of each filter cell of the DF simultaneously. The DIENB signal enables loading, which is synchronous on the rising edge of the clock signal.
The data samples can be either 9-bit two’s complement or 8-bit unsigned values. For 9-bit two’s complement values, DIN8 is the sign bit. For 8-bit unsigned values, DIN8 must be held at logical zero.
CLK signal occurring while DIENB is low will load the X register of every filter cell with the 9-bit value
present on DIN0-8. A high on this input forces all the bits of the data sample input bus to zero; a rising
CLK edge when DIENB is high will load the X register of every filter cell with all zeros. This signal is
latched inside the device, delaying its effect by one clock internal to the device.Therefore it must be low
during the clock cycle immediately preceding presentation of the desired data on the DIN0-8 inputs.Detailed operation is shown in later timing diagrams.
IThese nine inputs are used to input the 9-bit coefficients. The coefficients are synchronously loaded
into the C register of filter CELL0 if a rising edge of CLK occurs while CIENB is low. The CIENB signal
is delayed by one clock as discussed below.
The coefficients can be either 9-bit two’s complement or 8-bit unsigned values. For 9-bit two’s complement values, CIN8 is the sign bit. For 8-bit unsigned values, CIN8 must be held at logical zero.
filter cell according to the state of the DCM0-1 inputs. A rising edge of the CLK signal occurring while
CIENB is low will load the C register and appropriate D registers with the coefficient data present at
their inputs. This provides the mechanism for shifting coefficients from cell to cell through the device. A
high on this input freezes the contents of the C register and the D registers, ignoring the CLK signal.
This signal is latched and delayed by one clock internal to the DF. Therefore it must be low during the
clockcycle immediately preceding presentation of the desired coefficient on the CIN0-8 inputs. Detailed
operation is shown in later timing diagrams.
OThesenine three-state outputs are used to output the 9-bit coefficients from filter CELL7. These outputs
are enabled by the COENB signal low. These outputs may be tied to the CIN0-8 inputs of the same DF
to recirculate to coefficients, or they may be tied to the CIN0-8 inputs of another DF to cascade DFs for
longer filter lengths.
in their high impedance state.
DCM1DCM0DECIMATION FUNCTION
00Decimation registers not used
01One decimation register is used
10Two decimation registers are used
11Three decimation registers are used
The coefficients pass from cell to cell at a rate determined by the number of decimation registers used.
When no decimation registers are used, coefficients move from cell to cell on each clock. When one
decimation register is used, coefficients move from cell to cell on every other clock, etc. These signals
are latched and delayed by one clock internal to the device.
4
HSP43891
Pin Description
SYMBOL
SUM0-25F9, G9-G11,
H10, H11, J2,
J5-J7, J10, K2,
K5, K7-K11,
L2-L6, L8, L10,
SENBHK1IA low on this input enables result bits SUM16-25. A high on this input places these bits in their high
SENBLE11IA low on this input enables result bits SUM0-15. A high on this input places these bits in their high im-
ADR0-2G1, H1, H2IThese three inputs select the one cell whose accumulator will be read through the output bus (SUM0-
SHADDF3IThe SHADD input controls the activation of the shift and add operation in the output stage. This signal
RESETA4IA low on this input synchronously clears all the internal registers, except the cell accumulators It can
ERASEB4IA low on this input synchronously clears the cell accumulator selected by the ADR0-2 signals. If RESET
(Continued)
PIN
NUMBERTYPENAME AND FUNCTION
OThese26 three-state outputs are used to output the results of the internal filter cell computations. Indi-
vidual filter cell results or the result of the shift and add output stage can be output. If an individual filter
cell result is to be output, the ADR0-2 signals select the filter cell result. The SHADD signal determines
whether the selected filter cell result or the output stage adder result is output. The signals SENBH and
SENBL enable the most significant and least significant bits of the SUM0-25 result respectively. Both
L11
SENBH and SENBL may be enabled simultaneously if the system has a 26-bit or larger bus. However
individual enables are provided to facilitate use with a 16-bit bus.
impedance state.
pedance state.
25) or added to the output stage accumulator. They also determine which accumulator will be cleared
when ERASE is low.These inputs are latched in the DF and delayed by one clock internal to the device.
If ADR0-2 remains at the same address for more than one clock, the output at SUM0-25 will not change
to reflect any subsequent accumulator updates in the addressed cell. Only the result available during
the first clock, when ADR0-2 selects the cell, will be output. This does not hinder normal operation since
the ADR0-2 lines are changed sequentially. This feature facilitates the interface with slow memories
where the output is required to be fixed for more than one clock.
is latched on chip and delayed by one clock internal to the device. Detailed explanation is given in the
DF Output Stage section.
be used with ERASE to also clear all the accumulators simultaneously. This signal is latched in the DF
and delayed by one clock internal to the device.
is also low simultaneously, all cell accumulators are cleared.
Functional Description
The Digital Filter Processor (DF) is composed of eight filter
cells cascaded together and an output stage for combining
or selecting filter cell outputs (See Block Diagram). Each
filter cell contains a multiplier-accumulator and several
registers (Figure 1). Each 9-bit coefficient is multiplied by a
9-bit data sample, with the result added to the 26-bit
accumulator contents. The coefficient output of each cell is
cascaded to the coefficient input of the next cell to its right.
DF Filter Cell
A 9-bit coefficient (CIN0-8) enters each cell through the C
register on the left and exits the cell on the right as signals
COUT0-8. With no decimation, the coefficient moves directly
from the C register to the output, and is valid on the clock
following its entrance. When decimation is selected the
coefficient exit is delayed by 1, 2 or 3 clocks by passing
through one or more decimation registers (D1, D2 or D3).
The combination of D registers through which the coefficient
passes is determined by the state of DCM0 and DCM1. The
output signals (COUT0-8) are connected to the CIN0-8
inputs of the next cell to its right. The
enables the COUT0-8 outputs of the right most cell to the
COUT0-8 pins of the device.
COENB input signal
The C and D registers are enabled for loading by
Loading is synchronous with CLK when
that
CIENB is latched internally. It enables the register for
CIENB is low. Note
loading after the next CLK following the onset of
CIENB.
CIENB low.
Actual loading occurs on the second CLK following the onset
of
CIENB low. Therefore CIENB must be low during the clock
cycle immediately preceding presentation of the coefficient
on the CIN0-8 inputs. In most basic FIR operations,
CIENB
will be low throughout the process, so this latching and delay
sequence is only important during the initialization phase.
When
CIENB is high, the coefficients are frozen.
The C and D registers are cleared synchronously under control
of
RESET, which is latched and delayed exactly like CIENB.
The output of the C register (C0-8) is one input to 9 x 9
multiplier.
The other input to the 9 x 9 multiplier comes from the output
of the X register. This register is loaded with a data sample
from the device input signals DIN0-8 discussed above. The
X register is enabled for loading by
synchronous with CLK when
DIENB. Loading is
DIENB is low.Note that DIENB
is latched internally. It enables the register for loading after
the next CLK following the onset of
DIENB low. Actual
loading occurs on the second CLK following the onset of
DIENB low; therefore, DIENB must be low during the clock
5
HSP43891
cycle immediately preceding presentation of the data sample
on the DIN0-8 inputs. In most basic FIR operations,
DIENB
will be low throughout the process, so this latching and delay
sequence is only important during the initialization phase.
When
DIENB is high, the X register is loaded with all zeros.
The multiplier is pipelined and is modeled as a multiplier core
followed b y two pipeline registers, MREG0 and MREG1
(Figure 1). The multiplier output is sign extended and input as
one operand of the 26-bit adder. The other adder operand is
the output of the 26-bit accumulator. The adder output is
loaded synchronously into both the accumulator and the
TREG.
The TREG loading is disabled by the cell select signal,
CELLn, where n is the cell number.The cell select is decoded
from the ADR0-2 signals to generate the TREG load enable.
The cell select is inverted and applied as the load enable to
the TREG. Operation is such that the TREG is loaded
whenever the cell is not selected. Theref ore, TREG is loaded
every clock except the clock following cell selection. The
purpose of the TREG is to hold the result of a sum-ofproducts calculation during the clock when the accumulator is
cleared to prepare for the next sum-of-products calculation.
This allows continuous accumulation without wasting clocks .
The accumulator is loaded with the adder output every clock
unless it is cleared. It is cleared synchronously in two ways.
When
RESET and ERASE are both low, the accumulator is
cleared along with all other registers on the device. Since
ERASE and RESET are latched and delayed one clock
internally, clearing occurs on the second CLK following the
onset of both
ERASE and RESET low.
The second accumulator clearing mechanism clears a single
accumulator in a selected cell. The cell select signal, CELLn,
decoded from ADR0-2 and the
ERASE signal enable
clearing of the accumulator on the next CLK.
The
ERASE and RESET signals clear the DF internal
registers and states as follows:
ERASERESETCLEARING EFFECT
11No clearing occurs, internal state remains
same.
10RESET only active, all registers except ac-
cumulators are cleared, including the internal pipeline registers.
01ERASE only active, the accumulator
whose address is given by the ADR0-2 inputs is cleared.
00Both RESET and ERASE active, all accu-
mulators as well as all other registers are
cleared.
The DF Output Stage
The output stage consists of a 26-bit adder, 26-bit register,
feedback multiplexer from the register to the adder, an output
multiplexer and a 26-bit three-state driv er stage (Figure 2).
The 26-bit output adder can add any filter cell accumulator
result to the 18 most significant bits of the output buffer. This
result is stored back in the output buffer.This operation takes
place in one clock period. The eight LSBs of the output
buffer are lost. The filter cell accumulator is selected by the
ADR0-2 inputs.
The 18 MSBs of the output buffer actually pass through the
zero mux on their way to the output adder input. The zero
mux is controlled by the SHADD input signal and selects
either the output buffer 18 MSBs or all zeros for the adder
input. A low on the SHADD input selects zero. A high on the
SHADD input selects the output buffer MSBs, thus,
activating the shift-and-add operation. The SHADD signal is
latched and delayed by one clock internally.
6
Loading...
+ 12 hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.