MITEL PDSP16256A, PDSP16256AC, PDSP16256AC1R, PDSP16256B0, PDSP16256C0 Datasheet

PDSP16256/A

Programmable FIR Filter

DS3709 Issue 7.1 June 1999

Features

● Sixteen MACs in a Single Device

● Basic Mode is 16-Tap Filter at up to 25MHz

Sample Rates

● Programmable to give up to 128 Taps with

Sampling Rates Proportionally Reducing to
3·125MHz

● 16-bit Data and 32-bit Accumulators

● Can be configured as One Long Filter or Two

Half-Length Filters

● Decimate-by-two Option will Double the Filter

Length

● Coefficients supplied from Host System or local

EPROM

Applications

● High Performance Digital Filters

Description

The PDSP16256 contains sixteen multiplier ­accumulators, which can be multi cycled to provide
from 16 to 128 stages of digital filtering. Input data
and coefficients are both represented by 16-bit
two’s complement numbers with coefficients
converted internally to 12 bits and the results being
accumulated up to 32 bits.

In 16-tap mode the device samples data at the
system clock rate of up to 25MHz. If a lower sample
rate is acceptable then the number of stages can be
increased in powers of two up to a maximum of 128.
Each time the number of stages is doubled, the
sample clock rate must be halved with respect to the
system clock. With 128 stages the sample clock is
therefore one eighth of the system clock.

In all speed modes devices can be cascaded to
provide filters of any length, only limited by the
possibility of accumulator overflow. The 32-bit
results are passed between cascaded devices
without any intermediate scaling and subsequent
loss of precision.

Ordering Information

Commercial (0°C to 170°C)

PDSP16256A/C0/AC 25MHz, PGA package

Industrial (240°C to 185°C)

PDSP16256 B0/AC 20MHz, PGA package
PDSP16256 B0/GC 20MHz, QFP package

Military (255°C to 1125°C)

PDSP16256 MC/AC1R 20MHz, MIL-STD-883*

(latest revision), PGA package

PDSP16256 MC/GC1R 20MHz, MIL-STD-883*

(latest revision), QFP package

*See notes following Electrical Characteristics for further
information on MIL-STD-883 screening

Associated Products

PDSP16350 I/Q Splitter/NCO
PDSP16510A FFT Processor

The device can be configured as either one long
filter or two separate filters with half the number of
taps in each. Both networks can have independent
inputs and outputs.

Both single and cascaded devices can be operated
in decimate-by-two mode. The output rate is then
half the input rate, but twice the number of stages
are possible at a given sample rate. A single device
with a 20MHz clock would then, for example,
provide a 128-stage low pass filter, with a 5MHz
input rate and 2·5MHz output rate.

Coefficients are stored internally and can be down
loaded from a host system or an EPROM. The latter
requires no additional support, and is used in stand
alone applications. A full set of coefficients is then
automatically loaded at power on, or at the request
of the system. A single EPROM can be used to
provide coefficients for up to 16 devices.

PDSP16256

CHANGE

COEFF

POWER-ON

RESET

RES

PDSP
16256

EPROM

GNDSCLK

OUTPUT

DATA

INPUT

DATA

EPROM

ADDR DATA

Figure. 1 A dual filter application

ANALOG

INPUT

EPROM

ADDR DATA

COEFFICIENTS

CHANGE

COEFF

POWER-ON

RESET

RES

PDSP
16256

ADC

CLKOP

EPROM

GNDSCLK

OUTPUT

Figure. 2 Typical system application

DATA

2

PDSP16256

Signal

Description

DA15:0 16-bit data input bus to Network A.
DB15:0 Delayed data output bus in the single filter mode. Connected to the data input bus of the next device in a

cascaded chain. Input to Network B in the dual filter modes.

X31:0 Expansion input bus in the single filter mode. Connected to the previous filter output in a cascaded chain.

The inputs are not used on a single device system or on the Termination device in a cascaded chain. The

X bus provides the output from Network B in both dual modes.
F31:0 In single filter mode this bus holds the main device output. In dual mode it holds the output from Network A.
FEN Filter enable. The first high present on an SCLK rising edge defines the first data sample. The control register

and coefficient memory must be configured befor FEN is enabled.The signal must stay active whilst valid

data is being received and must be low if FRUN is high.
DFEN Delayed filter enable. This output is connected to the Filter Enable input of the next device in a cascaded

chain when moving towards the termination device and with multiple stand-alone EPROM-loaded

configurations. It is used to coordinate the control logic within each device.
SWAP Selects either the upper or lower set of coefficients for Bank Swap. A low selects the lower bank, a high the

upper bank.
FRUN In EPROM load mode, when high this signal allows continuous filter operations to occur without the need for

the initial FEN edge. If the device is not a single, interface or master device then this pin must be tied low.

DCLR

A low on this signal on the SCLK rising edge will clear all the internal accumulators.

low for a single cycle, signal BUSY will indicate when the internal clearing is complete. After a clear the

need only remain

DCLR

device must be re-synchronised to the data stream using FEN. It is recommended that FEN is taken low

at the same time as clear. FEN may then be taken high to synchronise the data stream once BUSY has

returned low.
C15:0 16-bit coefficient input bus. In the Byte mode of operation, C15:8 have alternative uses as explained in the

text.
A7:0 Coefficient address bus. In the EPROM mode A7:0 are address outputs for an EPROM. In the remote host

mode they are inputs from the host. A7 is not used when coefficients are loaded as 16-bit words.
CCS This pin is similar in operation to A7:0 and provides a higher order address bit. When low the coefficients

WEN

CS
BYTE

EPROM

are loaded, when high the control register is loaded.

In the remote mode this pin is an input which when low enables the load operation. In the EPROM mode

it is an output which provides the write enable for other slave devices.

This pin is always an input and must also be low for the internal write operation to occur.

When this pin is tied low, coefficients are loaded as two 8-bit bytes. When the pin is high they are loaded

as 16-bit words. In the EPROM mode this pin is ignored.

When this pin is tied low coefficients are loaded as bytes from an external EPROM. The device outputs an

address on A7:0. When the pin is high coefficients must be loaded from a remote master. They can then

be transferred individually rather than as a complete set.
SCLK The main system clock; all operations are synchronous with this clock. The clock rate must be either 1, 2,

4, or 8 times the required data sampling rate. The factor used depends on the required filter length.
CLKOP This output, when used to enable SCLK, can provide a data sampling clock. It has the effect of dividing the

OEN

SCLK rate by 1, 2, 4 or 8 depending on the filter mode selected.

Tri-state enable for the F bus. When high the outputs will be high impedance.

device and does not therefore take effect until the first SCLK rising edge

OEN

is registered onto the

BUSY A high on this signal indicates that the device is completing internal operations and is not yet able to accept

RES

new data. The signal is used during automatic EPROM loading, reset and accumulator clearing.

When this pin is low the control logic and accumulators are reset. In the EPROM mode it will initiate a load

sequence when it goes high.

NOTES

1. Unused buses (e.g. X31:0 when the device is configured in single or termination mode) can be set to any value. They should however be
maintained at a valid logic level to avoid an increase in power consumption.

2. To ensure correct input voltage thresholds are maintained all the VDD and GND pins must be connected to adequate power and ground planes.

Table 1 Pin descriptions

3

PDSP16256

EXTRA PIN D4,

CONNECTED TO D3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

R
P
N
M
L
K
J
H
G
F
E
D
C
B
A

AC144

Fig. 3a Pin connections for 144 pin PGA package (bottom view)

PIN 1 INDEX

PIN 1

PIN 172

GC172

Fig. 3b Pin connections for 172 pin QFP (top view)

Figure. 3 Pin connection diagrams (not to scale). See T able 1 for signal descriptions and Table 2 for

pinouts.

4

PDSP16256

GG

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
31
32
33
34
35
36
37
38
39
40
41
42
43

AC

A15
B15
D13
C14
G15
C15
D14

J15
E13
D15
E14
E15
F13
F14
F15

­G14
G13
H14

­H15
H13

J14

K15

-

J13

K14

­L15
K13
L14

M15

L13

M14
N15

-

N14
M13

P15

­P14

N13
R15

Signal

F0
F1
F2
F3

V

DD

F4
F5

GND

F6
F7
F8

F9
F10
F11
F12

GND

F13
F14
F15
V

DD

F16
F17
F18
F19
V

DD

F20
F21

GND

F22
F23
F24
F25
F26
F27
F28

GND

F29
F30
F31
V

DD

FEN

DFEN

DCLR

GG

44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86

AC

R14

­N12
P13

­R13
P12
N11
R12
P11
R11

R9
N10
P10
R10

P9
R7
N9

P8
R8
N8

P7
R6

-

N7

P6
R5
N6

P5
R4

-

N5

P4
R3

P3
N4

-

R2

P2
N3

-

-

R1

Signal

SWAP

GND

OEN

CLKOP

V

DD

DA0
DA1
DA2
DA3
DA4
DA5

GND

DA6
DA7
DA8
DA9

V

DD

DA10
DA11
DA12
DA13
DA14
DA15

GND

C0
C1
C2
C3
C4
C5

V

DD

C6
C7
C8
C9

C10

GND

C11
C12
C13

V

DD

GND

C14

GG

87
88
89
90
91
92
93
94
95
96
97
98

99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129

AC

P1

N2
N1

M2

L3
M1
M3

L2

L1

K3

K2

K1

J2
J3

G1

H2

H1

J1

H3
G2

F1
G3

F2

E1

F3

E2

D1

E3

D2

C1

C2

D3

B1

B2

C3

Signal

C15

-

-

GND
GND

WEN

CCS

CS

-

V

DD

RES

SLCK

GND

-

V

DD

BYTE

EPROM

A0
A1
A2
A3
A4

V

DD

A5
A6

GND

A7
DB0
DB1
DB2

-

GND

DB3
DB4
DB5
DB6
DB7

-

V

DD

DB8
DB9

DB10
DB11
DB12
DB13
DB14

-

GND

DB15

-

V

DD

GG

130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172

AC

­A1
A2

­C4
B3
A3
B4
C5
A4

­B5
A5
A7
C6
B6
A6
B7
C7
B8
A9
A8
C8
B9

A10

C9

B10
A11
C10

-

B11
A12
C11

-

B12
A13
B13
C12
A14

-

B14

-

C13

Signal

GND

BUSY

X0

V

DD

X1
X2
X3
X4
X5
X6

GND

X7
X8

V

DD

X9
X10
X11
X12
X13
X14

GND

X15
X16
X17
X18
X19
X20
X21
X22

GND

X23
X24
X25
V

DD

X26
X27
X28
X29
X30

GND

X31
V

DD

FRUN

NOTE. All GND and VDD pins must be used

T able 2 Pin connections for AC144 and GC172 packages

5

PDSP16256

DA15:0 F31:0 OEN

SCLK FRUN

SWAP

A7:0

C15:0

CCS

WEN

CS

BYTE

EPROM

FEN
DFEN
DCLR

RES

COEFFICIENT

STORAGE

AND

CONTROL

CLKOP BUSY

Figure. 4 Block Diagram

Operational Overview

The PDSP16256 is an application specific FIR filter for
use in high performance digital signal processing
systems. Sampling rates can be up to 25MHz. The
device provides the filter function without any software
development, and the options are simply selected by
loading a control register. The device can be user
configured as either a single filter, or as two separate
filters. The latter can provide two independent filters for
the in-phase and quadrature channels after IQ splitting,
or can provide two filters in cascade for greater stop
band rejection.

The device operates from a system clock, with rates up
to 25MHz. This clock must be 1, 2, 4, or 8 times the
required sampling frequency, with the higher
multiplication rates producing longer filter networks at
the expense of lower sampling rates. Devices can be
connected in cascade to produce longer filter lengths.
This can be accomplished without the need for any
additional external data delays, and all the single device
options remain available.
Continuous inputs are accepted, and continuous results
produced after the internal pipeline delay. Connection
can be made directly to an A-D converter. The filter
operation can be synchronised to a Filter Enable signal
(FEN) whose positive going edge marks the first data
sample. The internal multiplier accumulator array can be
cleared with a dedicated input. This is necessary if
erroneous results obtained during the normal data ‘flush

6

NETWORK

A

DUAL

MUX

NETWORK

B

SINGLE

MODE

DB15:0 X31:0

MODE

through’ are not permissible in the system.
Coefficients can be loaded from a host system using a
conventional peripheral interface and separate data
bus. Alternatively, they can be loaded as a complete set
from a byte wide EPROM. The device produces
addresses for the EPROM and a BUSY output indicates
that the transfer is occurring. Up to sixteen devices can
have their coefficients supplied from a single EPROM.
These devices need not necessarily be part of the same
filter network.

Each of the filter networks shown in Fig. 4 contains eight
systolic multiplier accumulator stages; an example with
four stages is shown in Fig. 5. Input data flows through
the delay lines and is presented for multiplication with the
required coefficient. This is added to either the last result
from this accumulator or the result from the previous
accumulator. The filter results progress along the adders
at the data sample rate. If the sample rate equals SCLK
divided by four, for example, then the accumulated result
is passed onto the next stage every fourth cycle. The
structure described is highly efficient when used to
calculate filtered results from continuous input data.
A comprehensive digital filter design program is
available for PC compatible machines. This will optimise
the filter coefficients for the filter type required and
number of taps available at the selected sample rate
within the PDSP16256 device. An EPROM file can be
automatically generated in Motorola S-record format.

PDSP16256

DATA

OUT

ACCUMULATE

EXPANSION

IN

DATA

DELAY LINE

COEFF

ADDER

2

1

Z

RAM

DATA

DELAY LINE

COEFF

ADDER

2

1

Z

RAM

Figure. 5 Filter network diagram

Single Filter Options

When operating as a single filter the device accepts data
on the 16-bit DA bus at the selected sample rate, see
Figs. 5 and 6. Results are presented on the 32-bit F bus,
which may be tristated using the
registered onto the device and does not therefore take
effect until the first SCLK rising edge. Devices may be
cascaded this allows filters with more taps than available
from a single device. To accomplish this two further
buses are utilised. The DB bus presents the input data to
the next device in cascade after the appropriate delay,
while, partial results are accepted on the X bus.

input. Signal

OEN

OEN

DATA

DELAY LINE

COEFF

ADDER

2

1

Z

RAM

DATA

DELAY LINE

COEFF

ADDER

2

1

Z

RAM

the higher frequency components present in the input.
The Nyquist criterion, specifying that the sampling rate
must be at least double the highest frequency compo­nent, can still then be satisfied even though the sampling
rate has been halved.

is

The system clock latency for a single device is shown in
Table 3. This is defined as the delay from a particular data
sample being available on the input pins to the first result
including that input appearing on the output pins. It does
not include the delay needed to gather N samples, for an
N tap filter, before a mathematically correct result is
obtained.

DATA

IN

RESULT

OUT

Single filter mode is selected by setting control register bit
15 to a one. The required filter length is then selected
using control register bits 14 and 13 as summarised in
Table 3. The options define the number of times each
multiplier accumulator is used per sample clock period.
This can be once, twice, four times, or eight times.

In addition a normal/decimate bit (CR12) allows the filter
length to be doubled at any sample rate. This is possible
when the filter coefficients are selected to produce a low
pass filter, since the filtered output would then not contain

CR Input Output Filter Setup

14 13 12 Rate Rate Length Latency

0 0 0 SCLK SCLK 16 Taps 16
0 0 1 SCLK SCLK/2 32 Taps 17
0 1 0 SCLK/2 SCLK/2 32 Taps 16
0 1 1 SCLK/2 SCLK/4 64 Taps 18
1 0 0 SCLK/4 SCLK/4 64 Taps 20
1 0 1 SCLK/4 SCLK/8 128 Taps 24
1 1 0 SCLK/8 SCLK/8 128 Taps 24

Table 3 Single Filter options Figure. 6 Single Filter bus utilisation

DA15:0 F31:0 OEN

NETWORK

A

DUAL

MUX

NETWORK

B

SINGLE

MODE

DB15:0 X31:0

MODE

7

PDSP16256

SPEED MODE 0 (Data input and output at f

SCLK

FEN

DA15:0

F31:0

CLKOP

123

ABC

) CR14:13 = 00, CR12 = 0. CLKOP held high.

SCLK

16 17 18

First data point (A)
is read on edge 1

SPEED MODE 1 (Data input and output at half f

SCLK

FEN

DA15:0

F31:0

CLKOP

123

AB

SCLK

16 17 18

First data point (A)
is read on edge 1

SPEED MODE 2 (Data input and output at a quarter f

SCLK

FEN

DA15:0

123

AB

45 23 24

A′ B′ C′

First valid result
including data point (A)
available after edge 16

) CR14:13 = 01, CR12 = 0

A′ B′

First valid result
including data point (A)
available after edge 16

) CR14:13 = 10, CR12 = 0

SCLK

20 21 22

31 32 33

A′′ B′′ C′′

34 35

D′′ E′′

Valid result contains
the first 16 data points
available after edge 31

78 79 80

A′′ B′′ C′′

81 82

Valid result contains
the first 32 data points
available after edge 78

272 273

274 275 276

F31:0

CLKOP

First data point (A)
is read on edge 1

SPEED MODE 3 (Data input and output at an eighth f

SCLK

FEN

DA15:0

F31:0

CLKOP

SPEED MODE 1 Decimating (Data input at half f

SCLK

FEN

DA15:0

F31:0

123

A

45

First data point (A)
is read on edge 1

123

AB

6789

B

SCLK

18 19 20

A′ B′

First valid result
including data point (A)
available after edge 20

) CR14:13 = 11, CR12 = 0

SCLK

24 25 26

A′ B′

27 28

29 30 31 32

First valid result
including data point (A)
available after edge 24

and output at a quarter f

21 22

B′

SCLK

A′′ B′′

Valid result contains
the first 64 data points
available after edge 272

1040

1041 1042 1043

A′′

Valid result contains
the first 128 data points
available after edge 1040

) CR14:13 = 01, CR12 = 1.

142 143 144

B′′

145

CLKOP

First data point (A)
is read on edge 1

First valid result
including data point (A)
available after edge 18

Valid result contains
the first 64 data points
available after edge 142

Figure. 7 Single Filter timing diagrams

8

PDSP16256

Dual Indipendant Filter Options

When operating as two independent filters the device
accepts 16 bit data on both the DA and DB buses at the
selected sample rate, see Fig. 8. Results are available
from both the F and X buses. The F bus may be tristated
using the

OEN

input. Signal
device and does not therefore take effect until the first
SCLK rising edge

Each filter must be configured in the same manner, and
multiple device expansion is not possible due to the pin
re-organization. The latter requirement can, of course,
still be satisfied by several devices configured as single
filters.
Dual independent filter mode is selected by setting
control register bits 15 and 4 to a zero. The required filter

CR Input Output Filter Setup

141312 Rate Rate Length Latency

0 0 0 SCLK SCLK 8 Taps 16 27
0 0 1 SCLK SCLK/2 16 Taps 17 ­0 1 0 SCLK/2 SCLK/2 16 Taps 16 28
0 1 1 SCLK/2 SCLK/4 32 Taps 18 ­1 0 0 SCLK/4 SCLK/4 32 Taps 20 36
1 0 1 SCLK/4 SCLK/8 64 Taps 24 ­1 1 0 SCLK/8 SCLK/8 64 Taps 24 40

Table 4. Dual Filter options

DA15:0 F31:0 OEN

OEN

is registered onto the

Ind Cas

length is selected using control register bits 14 and 13 as
summarised in Table 4, which also shows the resulting
latency. As in single filter mode normal or decimate-by­two operation can be selected using control register bit

12.

Dual Cascaded Filter Options

When operating as two cascaded filters the device ac­cepts 16 bit data on the DA bus at the selected sample
rate. Results are presented on the 32-bit X bus, see Fig.

9. Each filter must be configured in the same manner.
Multiple device expansion is not possible in this mode.

Dual cascaded filter mode is selected by setting control
register bit 15 to a zero and bit 4 to a one. The required
filter length is selected using control register bits 14 and
13 as summarised in Table 4, which also shows the
resulting latency. The decimate-by-two option is not
available in this mode.
The data for the second filter network is extracted as the
middle 16 bits from the first networks accumulated result.
For successful operation the first filter network must have
unity gain. See the section on filter accuracy for more
details.

The cascade option is used to increase the stop band
rejection in a practical filter application. Theoretically,
increasing the number of taps in an FIR filter will increase
the stop band rejection, but this assumes floating point
calculations with no accuracy limitations. In practice, with
fixed point arithmetic, better performance is achieved
with two smaller filters in series.

NETWORK

A

MUX

NETWORK

B

SINGLE

MODE

DB15:0 X31:0

DUAL

MODE

DA15:0 F31:0 OEN

NETWORK

A

DUAL

MUX

NETWORK

B

SINGLE

MODE

DB15:0 X31:0

MODE

Figure. 9 Dual cascaded filter bus utilisationFigure. 8 Dual independent filter bus utilisation

9