Intersil Corporation HSP43891 Datasheet

HSP43891
Data Sheet May 1999
Digital Filter
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-20 0 to 70 100 Lead MQFP Q100.14x20 HSP43891VC-25 0 to 70 100 Lead MQFP Q100.14x20 HSP43891VC-30 0 to 70 100 Lead MQFP Q100.14x20 HSP43891JC-20 0 to 70 84 Lead PLCC N84.1.15 HSP43891JC-25 0 to 70 84 Lead PLCC N84.1.15 HSP43891JC-30 0 to 70 84 Lead PLCC N84.1.15 HSP43891GC-20 0 to 70 85 Pin CPGA G85.A HSP43891GC-25 0 to 70 85 Pin CPGA G85.A HSP43891GC-30 0 to 70 85 Pin CPGA G85.A
RANGE (oC) PACKAGE PKG. 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.
FILTER
9
CELL 3
SUM0 - 25
DF
26
MUX
26
OUTPUT
STAGE
DF
FILTER
9
CELL 4
26
26
http://www.intersil.com or 407-727-9207 | Copyright © Intersil Corporation 1999
DF
FILTER
9
CELL 5
26
DF
FILTER
9
CELL 6
26
DF
FILTER
9
CELL 7
9
26
COUT0 - 8
COENB
SS
99
26
2
Pinout
HSP43891
85 PIN GRID ARRAY (PGA)
2173 4 5 6 8 9 10 11
V
RESET
A
V
SS
B
VCCCOUT7 ERASE DIN1 DIN2
C
COUT5 COUT6
D
COUT3 COUT4
E
COUT1
V
F
SS
G
ADR2 DCM0 CLK
H
ADR1
V
J
CC
SENBH
K
L
CC
COENB
COUT8 DIN8
ALIGN
PIN
COUT2
V
SS
COUT0 SHADD
SUM25
V
SUM24
SS
DIN7
DIENB
HSP43891
V
SUM19
CC
TOP VIEW
PINS DOWN
SUM14SUM18SUM21SUM22SUM23DCM1
SUM23 SUM22
V SUM21 SUM20 SUM19 SUM18
V SUM17 SUM16
V SUM15 SUM14 SUM13 SUM12
V SUM11 SUM10
SUM9 SUM8 SUM7
1234567891011
V
DIN6 DIN3 DIN0 CIN8
CIENB
DIN5 DIN4 CIN5 CIN3
SUM16SUM17SUM20
SUM15 SUM12 SUM10 SUM8 SUM6
V
SS
SUM13
V
CC
V
CC
CIN7 CIN6 CIN4
CIN2
CIN0
CIN1
V
CCVSS
SUM0
SUM3 SUM2
SUM1
SUM5 SUM4ADR0
SUM7
V
SS
SS
V
CC
SENBL
V
SS
SUM9SUM11
L
DCM1SUM23SUM22 SUM21SUM18 SUM14 SUM13 SUM11 SUM9
K
SENBH SUM24 SUM19 SUM15 SUM12 SUM10 SUM8 SUM6
J
V
SUM25
CC
H
G
F
E
D
C
B
A
ADR0
ADR1
ADR2 DCM0 CLK
V
COUT0 SHADD
SS
V
COUT1
COUT3COUT4
COUT5COUT6
V
CC
V
COENB
SS
84 LEAD PLASTIC LEADED CHIP CARRIER (PLCC)
SS
V
DCM1
SUM24
SENBH
SUM25
111098765432184838281807978777675
12 13 14
CC
15 16 17 18 19
SS
20 21 22
CC
23 24 25 26 27
SS
28 29 30 31 32
33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53
SS
DCM0
ADDR1
VCCADDR0
V
HSP43891
TOP VIEW
ADDR2
CLK
COUT0
SHADD
COUT1
VSSCOUT3
SS
COUT2
VSSV
COUT2
ALIGN
PIN
V
CC
COUT4
V
CC
CC
V
SS
SUM20 SUM17 SUM16 SUM7
HSP43891
BOTTOM VIEW
PINS UP
DIENB
DIN5 DIN4
ERASECOUT7 DIN8COUT8
RESET
CC
COUT5
V
74
COUT6 COUT7
73
V
72
SS
COUT8
71
COENB
70
V
69
CC
ERASE
68
RESET
67
DIENB
66
DIN8
65
DIN7
64
DIN6
63
DIN5
62
DIN4
61
DIN3
60
DIN2
59
DIN1
58
DIN0
57
CIENB
56
CIN8
55
V
CC
54
DIN2DIN1
V
SUM1 SUM3 SUM2
SUM0
CIN1 CIN0
CIENB
CIN8
DIN0DIN3DIN6DIN7
SS
SUM5 SUM4
V
V
CCVSS
SENBL
V
CIN2
CIN5 CIN3
CIN4CIN6CIN7
V
CCVSS
SS
CC
V
SUM6
SS
SUM4
SUM5
CC
V
SUM3
SUM1
SUM2
SS
V
SUM0
CIN0
SENBL
CIN1
CIN3
CIN4
CIN5
SS
V
CIN7
CIN6
CC
V
CIN2
2
HSP43891
Pinout
(Continued)
DCM1
SUM24
V
SS
V
SS
SUM23 SUM22
V
CC
V
CC
SUM21 SUM20 SUM19 SUM18
V
SS
V
SS
SUM17 SUM16
V
CC
V
CC
SUM15 SUM14 SUM13 SUM12
V
SS
SUM11 SUM10
SUM9 SUM8 SUM7
NC
SUM6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
SUM25
SENBH
99 98 97 96 95 94 93 91 89 87 85 84 83 818286889092100
CCVCC
V
100 LEAD MQFP
TOP VIEW
SSVSS
V
ADDR1
ADDR0
ADDR2
DCM0
CLK
CC
SHADD
VCCV
COUT1
COUT0
SSVSS
V
COUT3
COUT2
80 79 78 77 76 75 74 73 72 71 70 69 68 67 66 65 64 63 62 61 60 59 58 57 56 55 54 53 52 51
COUT4 COUT5 V
CC
V
CC
COUT6 COUT7 V
SS
V
SS
COUT8 COENB
V
CC
V
CC
ERASE RESET DIENB DIN8 DIN7 DIN6 DIN5 DIN4 DIN3 DIN2 DIN1 DIN0 CIENB CIN8 V
CC
CIN7 CIN6 V
SS
SS
32 33 34 35 36 37 38 40 42 44 4647 48 50494543413931
SUM5
SUM4
V
CC
SUM3
SUM2
SUM1
VSSV
VSSV
SUM0
SS
SENBL
CIN0
CIN1
CC
V
CIN2
CIN3
CIN4
CIN5
SS
V
3
HSP43891
Pin Description
PIN
SYMBOL
V
CC
V
SS
CLK G3 I The CLK input provides the DF system sample clock. The maximum clock frequency is 30MHz.
DIN0-8 A5-8,B5-7, C6,
DIENB C5 I A low on this input enables the data sample input bus (DIN0-8) to all the filter cells. A rising edge of the
CIN0-8 A9,B9-11, C10,
ALIGN PIN C3 Used for aligning chip on socket or printed circuit board. This pin must be left as a no connect in circuit.
CIENB B8 I A low on this input enables the C register of every filter cell and the D (decimation) registers of every
COUT0-8 B2, B3, C1, D1,
COENB A2 I Alow on the COENB input enables the COUT0-8 outputs. A high on this input places all these outputs
DCM0-1 L1, G2 I These two inputs determine the use of the internal decimation registers as follows:
NUMBER TYPE NAME 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.
I These 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.
I These 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.
O Thesenine 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.
DCM1 DCM0 DECIMATION FUNCTION
0 0 Decimation registers not used 0 1 One decimation register is used 1 0 Two decimation registers are used 1 1 Three 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
SUM0-25 F9, G9-G11,
H10, H11, J2,
J5-J7, J10, K2,
K5, K7-K11,
L2-L6, L8, L10,
SENBH K1 I A low on this input enables result bits SUM16-25. A high on this input places these bits in their high
SENBL E11 I A low on this input enables result bits SUM0-15. A high on this input places these bits in their high im-
ADR0-2 G1, H1, H2 I These three inputs select the one cell whose accumulator will be read through the output bus (SUM0-
SHADD F3 I The SHADD input controls the activation of the shift and add operation in the output stage. This signal
RESET A4 I A low on this input synchronously clears all the internal registers, except the cell accumulators It can
ERASE B4 I A low on this input synchronously clears the cell accumulator selected by the ADR0-2 signals. If RESET
(Continued)
PIN
NUMBER TYPE NAME AND FUNCTION
O These26 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:
ERASE RESET CLEARING EFFECT
1 1 No 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.
0 0 Both 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