Mitsubishi M65727FP Datasheet

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

Specifications (Ver. 2.0)

MPEG2 MOTION ESTIMATION LSI (M65727FP)

December 1, 1995

Mitsubishi Electric Corporation

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 1

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

CONTENTS

1. Outline

1.1 Outline

1.2 Characteristics

2. Pin Listing and Description

2.1 Pin Listing

2.2 Pin Description

2.2.1 Data I/O Ports

2.2.2 System Control Pins

2.2.3 Input Pins for Sync. Signals

2.2.4 Pins Specifying Operational Modes

2.2.5 Others

3. Outline of Functions

3.1 Outline

3.2 Block Configuration

3.2.1 Input Unit

3.2.2 Integer Pel Unit

3.2.3 Motion Detection Unit

3.2.4 Half-Pel/Dual-Prime Unit

3.2.5 Output Unit

3.3 Operation modes

3.3.1 Field/Frame/Field Dual-Prime/Frame Dual-Prime

3.3.2 Horizontal Search Range

3.3.3 Vertical Search Range

3.3.4 Search range expansion for Vertical Direction

3.3.5 Half-Pel Precision/Integer-Pel Precision

3.3.6 External Frame Memory Data Format

3.3.7 Operation Modes and Output Data

3.3.8 Operation Modes and Dynamic Control Signals

4. Operations

4.1 Reset Operation

4.2 Wait Operation

4.3 Motion Estimation Operation

4.3.1 Search Window Image Input for Field/Frame mode

4.3.2 Search Window Image Input for Dual-Prime mode

4.3.3 Template MB Data Input

4.3.4 Output Request

4.3.5 Activation of Execution Cycle

4.3.6 Dynamic Control Input

4.3.7 Number of Cycles Needed (from Data Input to Data Output)

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 2

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

5. Others

5.1 Macro Block Processing for Missing Search Range

5.2 Priority Order for Distortion Equivalent Vector

5.3 Detail of Vector Search Range

5.4 Formats for External Frame Memory (Search Window Image Memory)

5.5 Expansion of Search Range and Controlling the Validity of Search Range

5.6 Operational Timing

5.7 Treatment of Final MB

5.8 Mode Change during Processing

6. Electrical Characteristics and Others

6.1 Electrical Characteristics

6.1.1 Maximum Ratings

6.1.2 Operating Range

6.1.3 Electrical Characteristics

6.1.4 Switching Characteristics

6.1.5 Set-up and Hold Times

6.2 Package

6.3 Thermal Management

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 3

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

1. Outline

1.1 Outline

The M65727 is a highly efficient motion estimation LSI used to estimate motion vectors for real-time
encoding of dynamic images. M65727 can be used together with the frame memory, the M65721
(Controller LSI) and the M65722 (Pixel Processor LSI). It operates under the control of the M65721.
M65727 accepts the template macro block (MB) data inputs from the M65721 and accepts the search
window image data from the frame memory. It estimate the motion vectors by searching the minimum
value of the mean absolute error between template block data and search window image data. It outputs
the result to the M65721. The M65721 and the M65722 are able to generate prediction image using the
above mentioned motion vector. The M65727 is designed so that it is applicable to MPEG2, the
international video compression standard.

[Main Specifications]
Field Mode (per field)

*Block size 16x16 and 16x8 (Upper), 16x8 (Lower) simultaneously
*Search range Vertical direction: ±7.5(*A) / ±15.5 pixel (Switchable)

However, ±8.0(*A) / ±16.0 are used during expansion respectively
Horizontal direction: ±7.5 / ±15.5 / ±31.5 / ±63.5 / ±127.5 pixel

(Selectable)

*Search methods Integer-pel precision exhaustive search

Half-pel precision 9 points around the best integer-pel precision vector

(Including the integer precision location)
*Evaluation function Full sampled mean absolute error
*Execution cycle Dependent on the search range

The execution cycle mentioned here refers to the throw-in period of MB,

not the 1MB processing time.

ex. 1 (Horizontal ±7.5, vertical ±7.5) (550) cycles / ΜΒ
ex. 2 (Horizontal ±15.5, vertical ±7.5) (550) ∗2 cycles / ΜΒ
ex. 3 (Horizontal ±7.5, vertical ±15.5) (806) cycles / ΜΒ
ex. 4 (Horizontal ±15.5, vertical ±15.5) (806) ∗2 cycles / ΜΒ

*Search window image inputs

When vertical ±7.5 is searched: 512 pixels / (550) cycles

When vertical ±15.5 is searched: 768 pixels / (806) cycles
*Processing capability

27MHz operation: Processing the search range over horizontal ±7.5 and

vertical ±7.5 for the ITU-R 601 image size is possible.

40MHz operation: Processing the search range over horizontal ±7.5

and vertical ±15.5 for the ITU-R 601 image size is possible.

*A: The mode of vertical search range is ±7.5 can be used only under 27MHz
operation.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 4

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

Frame Mode (per frame)

*Block size 16x16 and 16x8 (Top*), 16x8 (Bottom**) simultaneously
*Search range Vertical direction: ±7.5(*A) / ±15.5 pixel (Switchable)

It corresponds to 2 sets of fields each ±3.5 pixel / ±7.5 pixel location for

16x8 blocks.

However, ±8.0(*A) / ±16.0 are used during expansion respectively

Horizontal direction: ±7.5 / ±15.5 / ±31.5 / ±63.5 / ±127.5 pixel

(Selectable)
*Search methods Integer-pel precision exhaustive search

Half-pel precision 9 points around the best integer-pel precision vector

(Including the integer precision location)
*Evaluation function Full sampled mean absolute error
*Execution cycle Depends on the search range

The execution cycle mentioned here refers to the throw-in period of MB,

not the 1MB processing time.

ex. 1 (Horizontal ±7.5, vertical ±7.5) (550) cycles/ ΜΒ
ex. 2 (Horizontal ±15.5, vertical ±7.5) (550) ∗2 cycles / ΜΒ
ex. 3 (Horizontal ±7.5, vertical ±15.5) (806) cycles / ΜΒ
ex. 4 (Horizontal ±15.5, vertical ±15.5) (806) ∗2 cycles / ΜΒ

∗Search window image inputs

When vertical ±7.5 is searched: 512 pixels / (550) cycles

When vertical ±15.5 is searched: 768 pixels / (806) cycles
*Processing capability

27MHz operation: Processing the search range over horizontal ±7.5 and

vertical ±7.5 for the ITU-R 601 image size is possible.

40MHz operation: Processing the search range over horizontal ±7.5 and

vertical ±15.5 for the ITU-R 601 image size is possible.

*, **: "Top" and "Bottom" mentioned here indicate the field parities. In the
following explanation, a frame with the top field side as the first line is assumed.

*A: The mode of vertical search range is ±7.5 can be used only under 27MHz
operation.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 5

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

Field Dual-Prime Mode

*Block size 16x16
*Search range Second field: horizontal ±0.5 pixel, vertical: ±0.5 pixel
*Search method Take the average with the first field data (half-pel is generated as needed)

and estimate 9 points over the second field (including the specified point)
*Evaluation function Full sampled mean absolute error
*Execution cycle The execution cycle mentioned here refers to the throw-in period of MB,

not the 1MB processing time. (550) cycles / MB
*Search window image inputs

First field: 432 (18x24) pixel / (550) cycles

Second field: 432 (18x24) pixel / (550) cycles
*Processing capability

27MHz operation: Processing the search range over horizontal ±7.5 and

vertical ±7.5 for the ITU-R 601 image size is possible.

Frame Dual Prime Mode

*Block size 16x8
*Search range Second field: horizontal ±0.5 pixel, vertical: ±0.5 pixel
*Search method Take the average with the first field data (half-pel is generated as needed)

and estimate 9 points over the second field (including the specified point)

Not only the minimum evaluation value, but also all the 9 points are stored

and output.
*Execution cycle The execution cycle mentioned here refers to the throw-in period of MB,

not the 1MB processing time. (550) cycles / MB
*Search window image inputs

First field: 288 (18x16) pixel / (550) cycles

Second field: 288 (18x16) pixel / (550) cycles
*Processing capability

27MHz operation: Processing the search range over horizontal ±7.5 and

vertical ±7.5 for the ITU-R 601 image size is possible.

***: The Frame Dual-Prime Mode supports a part of Dual-Prime prediction specified

by the MPEG2.

Items common to modes

*Maximum operating frequency 40.5MHz (24.6ns)
*Two input ports:

Port 1 Search window image data, 32 bit
Port 2 Template MB data, 8 bit

*Output port 8 bit output port

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 6

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

1.2 Characteristics

The M65727 has the following characteristics.

*Highly efficient parallel architecture for high speed processing and data transfer, which
eliminates I/O bottleneck.

*Supports prediction modes for MPEG2, Field prediction, Frame prediction, Field Dual-Prime
prediction and Frame Dual-Prime prediction.

*For Field Mode and Frame Mode, it is possible to do simultaneous vector search over 16x16 and two
16x8 blocks.

*Implementing low cost image compression hardware is possible using DRAM frame memory in a 32
bit DRAM interface.

*Estimates half-pel precision vectors in a chip.

*The exhaustive search method is used over the integer-pel precision vectors in a search range.
Evaluation function is full sampled mean absolute error.

*The M65727's scalable architecture allows wider search range with multiple-chip configuration.
When expanding the horizontal search range, it is possible to use a common search window image
data into all chips.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 7

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

Number

Type

DSWI [31:0]

DMBI [7:0]

DOUT [7:0]

MODE [1:0]

HSIZE [2:0]

EXTND [1:0]

DCNT [3:0]

2. Explanation of Pins

2.1 List of Pins

Table 1 shows the list of pins of M65727.

Table 1 List of M65727 Pins

Pin names

Remarks

of Pins

32 I SEARCH WINDOW IMAGE DATA INPUT

8 I TEMPLATE MB DATA INPUT
8 O RESULTS OUTPUT

CLKI 1 I CLOCK INPUT

RESETC 1 I RESET

CEC 1 I CLOCK ENABLE

DENSWC 1 I DSWI INPUT DATA ENABLE

DENMBC 1 I DMBI INPUT DATA ENABLE

OEC 1 I OUTPUT ENABLE

SSYNC 1 I DSWI INPUT SYNC SIGNAL

MSYNC 1 I DMBI INPUT SYNC SIGNAL

ESYNC 1 I PROCESS EXECUTION SYNC SIGNAL
DSYNC 1 I DYNAMIC CONTROL SYNC SIGNAL
OREQC 1 I OUTPUT REQUEST SIGNAL

2 I SPECIFIES OPERATION MODE

FIELD, FRAME, FIELD DUAL-PRIME,
FRAME DUAL-PRIME

3 I HORIZONTAL SEARCH RANGE

±7.5, ±15.5, ±31.5, ±63.5, ±127.5

VSIZE 1 I VERTICAL SEARCH RANGE ±7.5, ±15.5

2 I VERTICAL SEARCH RANGE EXTENSION

HLFPL 1 I VECTOR PRECISION

FMFMT 1 I FRAME MEMORY FORMAT

4 I DYNAMIC CONTROL INPUT

TEST [6:0] 7 I TEST PIN

VDD 19 I POWER
GND 19 I GND

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 8

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

DENSWC

DENMBC

This is the output enable pin. It controls the tri-state of DOUT port. DOUT port.

2.2 Explanation of Pins

Functions and uses of M65727 pins are explained below. Refer to "2.1 List of Pins" for the bit
configuration of terminals and I/O attributes.
The term Execution cycle used in this explanation refers to 550 / 806 . It means that the above cycle is
capable of vector detection within a search range of -7 (-8) ~ +7 horizontally using integer precision.
When the horizontal search area is greater than or equal ±15, the integer precision operation requires
multiple execution cycles.

2.2.1 Data I/O Ports

DSWI This is the 32 bit wide search window image data input port. The search

window image input is processed in parallel with the arithmetic operation.
Therefore, the data inputted will be used in the next execution cycle.

DMBI This is a 8 bit wide template MB input port. The template MB input is

processed in parallel with the arithmetic operation. Therefore, the data inputted
will be used in the next execution cycle.

DOUT This is an 8 bit wide output port, during the field or frame mode, receives output

request, OREQC, and outputs the following information in the following order.
horizontal motion vector, vertical motion vector, minimum distortion,
distortion of vector (0,0), half-pel indication code
During the field dual-prime mode, the M65727 outputs minimum distortion and
dmv indication code. During the frame dual-prime mode, it outputs minimum
distortion, dmv indication code and distortions correspond to all estimation
points.

2.2.2 System Control Pins

CLKI Clock input.
RESETC RESET pin. Hardware reset. Asserted low. Not all registers are reset by

RESET. Before the normal operation, the M657272 requires RESET.

CEC Asserted low. This pin enables the input clock. This signal is sampled at up-

edge of CLKI. The next clock cycle is valid when this signal is asserted. The
invalid clock cycle is called "wait cycle". The chip is designed as static CMOS
circuits and the internal data will not be destroyed during wait cycles.

This pin enables DSWI port. This signal is asserted low. Data is not accepted
during not-active cycles.

This pin enables DMBI port. This signal is asserted low. Data is not accepted
during not-active cycles.

OEC

This signal is asserted low.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 9

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

This pin specifies the horizontal search range. The following five types of range

000: ±7.5, 001: ±15.5, 010: ±31.5, 011: ±63.5, 100: ±127.5, 101~ 111: reserved

2.2.3 Sync Signal Input Pins

SSYNC This is a sync signal for the DSWI port. It is asserted low. This signal must be

asserted when the leading data for DSWI is inputted

MSYNC This is a sync signal for the DMBI port. It is asserted low. This signal must be

asserted when the leading data for DMBI is inputted

ESYNC This is a sync signal for the block level pipeline. When this signal is asserted,

one execution cycle (550 / 806 cycles) is activated. It is asserted low.

DSYNC This is a sync signal for the DCNT port. It must be asserted when dynamic

control signal is inputted. See DCNT for the content of the dynamic control
signal. This signal is asserted low.

OREQC This is used to request output. M65727 starts output from DOUT port after this

signal is asserted. This signal is asserted low.

2.2.4 Pins Specifying Operational Modes

MODE This pin sets the mode of the M65727. The following four modes can be

specified.
00: Field mode, 01: Frame mode,
10: Field Dual-Prime mode, 11: Frame Dual-Prime mode

HSIZE

can be specified.

VSIZE This pin specifies the vertical search range. The following two types of range

can be specified. 0: ±7.5, 1: ±15.5

EXTND This pin specifies the vertical search range expansion. When expansion modes

are selected, the vertical search range is set to ±8.0 / ±16.0. It is possible to
expand a vertical search range using multiple chips. Depending on modes, the
order of priority regarding the vectors with same distortions is different.
00: non-expansion, 01: reserved,
10: upper-range of expansion, 11: lower-range of expansion

HLFPL This switches between half-pel precision search mode and integer-pel precision

search mode. 0: Integer-pel precision search, 1: half-pel precision search

FMFMT Switches between the external memory (SW image) formats.

0: Field format, 1: Frame format

DCNT This is a dynamic control input. (Dynamic control means the control which

differs in each execution cycle.) The following is required.
Control of valid or invalid for search range (SKILL)
Leading pixel location of search window image in the vertical direction when the
dual-prime mode (DVSPO)
The central position of the search window image used for the dual-prime mode
(DCNTR)

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 10

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

prearranged sequence by assuming DSYNC repeatedly. The control information

The above control signals are inputted within one execution cycle in the

are used at the next execution cycle.

2.2.5 Others

TEST Used for testing the M65727 and is not released to users.
VDD Power supply pin
GND Grounding pin

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 11

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

3. Outline of Functions

3.1 Outline

M65727 is configured as shown in Fig. 3.11. Its main components are the Input Unit, the Integer-pel
Unit, the Motion detection Unit, the Half-pel / Dual-Prime Unit, and the Output Unit. When the
Field/Frame mode is selected, the search window image data and template MB data are inputted to the
Input Unit from their respective input ports. The data is used as the source data of the Integer-pel Unit
after its order is changed. Then, the mean absolute error is calculated for each cycle at the Integer-pel
Unit and the result is given to the Motion detection Unit and the best integer-pel precision motion
vectors are estimated. Then, at the Half-pel Unit, the best half-pel precision motion vectors are
estimated. The results are output from the Output Unit. When the Field/Frame Dual-Prime mode is
selected, the template MB data and the search window image data of the first field and the second field
are sent to the Input Unit from their respective input ports and become the source data of the Dual-Prime
Unit. Then, the motion vector estimation is conducted at the Dual-Prime Unit and the results are
output from the Output Unit. The functions of each unit are outlined below.

3.2 Block Configuration

3.2.1 Input Unit

The function of Input Unit is to output the search window image data and the template MB data to the
Integer-pel Unit or Dual-Prime Unit with the necessary sequence and timing.
Having this block enables the user to input comparatively freely the needed search window image data,
using the sync signal (SSYNC) and the data enable signal (DENSWC), without regard to the motion
estimation sequence. Similarly, necessary template MB data can be inputted fairly freely using the sync
signal (MSYNC) and the data enable signal (DENMBC).
The search window image data is inputted from the highest line towards the lowest line scanning left to
right. The output sequence, on the other hand, starts from the leftmost column to the rightmost column
and scans top to bottom. The input and output sequence of the search window image data for Dual­Prime are from the highest line to the lowest line and scanning from left to right. The search window
image data is inputted as 4 vertical continuous pixels using the 32 bit input port. The Input Unit outputs
a pixel per cycle by parallel-serial conversion.

3.2.2 Integer-pel Unit

The function of this block is to calculate the mean absolute error using the template MB data and the
search window image data coming from the Input Unit. The Integer-pel Unit is composed of
processing elements arranged in parallel, allowing high speed processing of the data to be evaluated.
Three sets of calculated mean absolute error (corresponding to 16x16 block and two sets of 16x8 block)
are given to the Motion detection Unit. Three sets of search window image data and the template MB
data that correspond to the three sets of vectors are transferred from the integer-pel Unit to the Half-pel
Unit.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 12

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

3.2.3 Motion detection Unit

The function of this block is to select the vest motion vector of the integer-pel precision by comparing
the 16-bit mean absolute error coming from the Integer-pel Unit. In addition, this block stores
distortion of vector (0,0). In case when the distortions for multiple motion vectors are same, the most
suitable motion vector is determined according to the order of priority.
The output data depends on the modes. When the integer-pel precision search mode is specified during
the Field/Frame mode, the following items are output to the Output Unit. They are three sets of best
integer-pel precision motion vectors, the distortion for each, and the distortion for each vector (0,0). If
the half-pel precision search mode is specified when the Field/Frame mode, the three sets of best
integer-pel precision motion vectors and the distortion for each vector (0,0) are output to the Output Unit.
And, the distortions for each integer-pel precision vector are output to Half-pel Unit.

3.2.4 Half-Pel / Dual-Prime Unit

This Unit calculates, during the Field/Frame mode, the mean absolute errors for half-pel precision
vectors and detects the minimum distortions using the partial search window image data around best
integer-pel precision vectors.
The search window image consists of 18x18 pixels and two sets of 18x10 pixels around the three sets of
integer-pel precision motion vectors detected at the Motion detection Unit. Eight kinds of interpolated
images are generated by the half-pel interpolation filter. This image is matched against the template
MB data given from the integer-pel Unit and the minimum distortions are detected from the above
results.
During the Field/Frame Dual-Prime mode, it detects the best dmv from the template MB data, the first
search window image data and the second search window image data.

The first and second search window image consist of 18x18 pixel (10) each. The interpolated
image from the first search window is generated according to the central position information
(Displacement based on 0.5 pixel from 16x16 (8) pixels contained in 18x18 (10) pixels). Similarly,
nine sets of interpolated images are generated through the interpolated filter from the second search
window. Then, the nine sets of averaging images of the first and the second images are obtained.
Next, the block matching between the template MB data is conducted and the best dmv is obtained. In
case of the Frame Dual-Prime mode, not only the minimum evaluation value, but all the evaluated
values for all the displacement are output.

3.2.5 Output Unit

This is the interface circuit related to the Output Port, DOUT. Necessary data comes from the
Motion detection Unit and Half-pel / Dual-Prime Unit and is output to through DOUT in sync with the
output request signal, OREQC.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 13

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

3.3 Operational Modes

M65727 has the following operational modes which can be switched externally. This chapter
outlines the operational modes of M65727.

3.3.1 Field / Frame / Field Dual-Prime / Frame Dual-Prime

M65727 is capable of 4 modes, namely Field mode / Frame mode / Field Dual-Prime mode / Frame
Dual-Prime mode, to work with the prediction mode of MPEG2. These modes are specified as shown
below using MODE pins. They must be specified before the chip operation and fixed during operation.

00: Field mode
01: Frame mode
10: Field Dual-Prime mode
11: Frame Dual-Prime mode

Field mode detects three sets of motion vectors simultaneously which work with 16x16 block, 16x8
(upper) block, and 16x8 (lower) block.
Frame mode detects three sets of motion vectors simultaneously which work with 16x16 block, 16x8
(top) block, and 16x8 (bottom) block.
Field Dual-Prime and Frame Dual-Prime mode detect the dmv.

3.3.2 The Search Range in Horizontal Direction

The search range in the horizontal direction can be selected from ±7.5 / ±15.5 / ±31.5 /± 63.5/ and
±127.5. The number of cycles needed for the motion vector estimation increases in proportion to the
size of the horizontal search range. When ±15.5 or larger is specified as the horizontal search range, it
is expected that multiple chips with interleaving manner are used. Refer to Chapter 5.5 for the number
of chips needed when the horizontal search range is ±15.5 or more. The horizontal search range is
specified as shown below using HSIZE pins. The horizontal search range must be specified before the
chip goes into operation and should be fixed during the operation.

000: ±7.5, 001: ±15.5, 010: ±31.5, 011: ±63.5, 100: ±127.5, 101 ~ 111: Reserved

3.3.3 The Search Range in Vertical Direction

The vertical search range can be selected from ±7.5 and ±15.5. This range detects a minimum
execution cycle (550 / 806) In the vertical expansion mode, the search range will be ±8.0 / ±16.0.
See Chapter 5.5 for the number of chips needed for the vertical expansion. The vertical search range is
specified as shown below using VSIZE pin. This must be done before the chip operation and it should
be fixed during the operation.

0: ±7.5, 1: ±15.5

The vertical search range of ±7.5 can be used only under 27MHz operation.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 14

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

3.3.4 Search range expansion for Vertical direction

The non-expansion mode of vertical search range is ±7.5 / ±15.5 as explained above. At this time,
the order of priority for vectors, which have same distortions, gives the highest priority to the vector
(0,0). When the search range expansion mode of the vertical direction is specified, the range becomes
±8.0 / 16.0. It becomes possible to expand the vertical search range using multiple chips. There are
two modes for vertical expansion mode, upper-range of expansion and lower-range of expansion. The
difference between the two is in the priority order for distortion equivalent vectors. If it is specified as
upper-range, the vector (0, +8 / +16) will have the highest priority. If it is specified as lower-range, the
vector (0, -8 / -16) will have the highest priority. See Ch. 5.2 for the order of priority for vectors.
See Ch. 5.5 for the number of chips needed for the vertical search range expansion.
This mode is specified by EXTND pin as shown below. This mode must be specified before the
chip goes into operation and it should be fixed during the operation.

00: non-expansion, 01: reserved, 10: upper-range of expansion (Expansion),
11: lower-range of expansion (Expansion)

3.3.5 Half-Pel Precision/Integer-Pel Precision

The motion vector searched by M65727 is integer-pel precision during the integer-pel precision mode.
In case of half-pel precision mode, half-pel precision vector is detected using interpolated search
window image. The order of output data is shown in Table 2 (See Ch. 3.3.7). The above two modes
have different outputs.
This mode is specified by HLFPL pin as shown below. This mode should be specified before the
chip operation and should stay fixed.

0: Integer-Pel Precision mode, 1: Half-Pel Precision mode

3.3.6 External Frame Memory Data Format

M65727 is capable of selecting the external frame memory (SW image) format only when the Frame
mode is ON. This format is specified by the use of FMFMT pin as shown below. In other modes,
only field format can be used. See Ch. 5.4 for details on formats. This mode must be specified prior
to the chip operation and should be fixed during the operation.

0: Field format 1: Frame format

3.3.7 Operation Modes and Output Data

Data output from M65727 differs according to the operational modes. During the Field / Frame
mode, three sets of data group, a group of 16x16 and two groups of 16x8, are output in sequence.
Outputs from half-pel precision mode and integer precision search mode are different.
When the Field mode is ON, a data group for 16x16 block is first output. A data group for 16x8
(upper) block is output next and a data group for 16x8 (lower) is output last. It takes 21 cycles.
When the Frame mode is ON, a data group for 16x16 block is output first. A data group for 16x8
(top) is output next. And a 16x8 (bottom) block is output last. It takes 21 cycles.
Table 2 shows a set of data outputs during the Field/Frame mode in the order they are output.
In case of the Field Dual-Prime mode, Dual-Prime vector specifying code and its distortion are output in
the order shown in Table 3. It takes 3 cycles to output data.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 15

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

Operational modes

When the Frame Dual-Prime mode is used, the Dual-Prime vector specifying code, its distortion, and
distortions for each displacement point are output in the order shown in Table 4. It takes 21 cycles to
output data.
The integer-pel precision motion vector outputs as motion vector corresponding to 16x16 block even
for 16x8 block. Therefore, during the Frame Estimation Mode, the vertical component of the motion
vector for the 16x8 block must be changed outside.
When multiple chips are used to expand the vertical search range, the vertical components of the
motion vectors must be changed for all blocks.
Fig. 3.3.7-1 shows correspondence between 16x8 block and vectors.

Table 2

Integer-pel precision search Half-pel precision search

Output sequence1Motion vector horizontal

component

2 Motion vector vertical

component

Integer precision motion vector
horizontal component
Integer precision motion vector

vertical component
3 Minimum evaluation value (upper 8bits) Minimum evaluation value (upper 8bits)
4 Minimum evaluation value (lower 8bits) Minimum evaluation value (lower 8bits)
5 (0, 0) evaluation value (upper 8bits) (0, 0) evaluation value (upper 8bits)
6 (0, 0) evaluation value (lower 8bits) (0, 0) evaluation value (lower 8bits)
7 all 0 (L) output Half-pel indication code

Note 1: The motion vector is a binary number in 2's complement. It is output after it is expanded to

8 bits.

Note 2: Upper 8 bits of the evaluation value is first output and the lower 8 bits are output next in

natural binary

Note 3: Half-pel indication code is specified by the lower 4 bits as shown below. The upper 4 bits

are for L output.

0000: Most suitable for integer-pel precision motion vector (0.0, 0.0)
1010: Upper-left direction Half-pel of integer-pel precision vector (-0.5, -0.5)
1001: Upper-right direction Half-pel of integer-pel precision vector (+0.5, -0.5)
0110: Lower-left direction Half-pel of integer-pel precision vector (-0.5, +0.5)
0101: Lower-right direction Half-pel of integer-pel precision vector (+0.5, +0.5)
0010: Left direction Half-pel of integer-pel precision vector (-0.5, +0.0)
0001: Right direction Half-pel of integer-pel precision vector (+0.5, +0.0)
1000: Upper direction Half-pel of integer-pel precision vector (+0.0, -0.5)
0100: Lower direction Half-pel of integer-pel precision vector (+0.0, +0.5)

Note 4: The (0,0) evaluation value is an evaluation value corresponding to the no-motion. When

specifying the upper range of expansion, the evaluation point of (X, Y) = (0, +8 / +16) is
used as the position; when specifying the lower range of expansion, the evaluation point of
(X, Y) = (0, -8 / -16) is used as the position.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 16

DISTRIBUTION RESTRICTED. COPYRIGHT RESERVED 1995

Table 3 Relationship between Field Dual-Prime Estimation Mode and Its Output Data

Output sequence

Minimum evaluation value (Upper 8bits)
1
2 Minimum evaluation value (Lower 8bits)
3 dmv indication code

Table 4 Relationship between Frame Dual-Prime Estimation Mode and Its Output Data

Output

sequence

1
2 Minimum evaluation value (Lower) 13 Left evaluation value (Lower)
3 dmv indication code 14 Right evaluation value (Upper)
4 Center evaluation value (Upper) 15 Right evaluation value (Lower)
5 Center evaluation value (Lower) 16 Left lower evaluation value (Upper)
6 Left upper evaluation value (Upper) 17 Left lower evaluation value (Lower)
7 Left upper evaluation value (Lower) 18 Lower evaluation value (Upper)
8 Upper evaluation value (Upper) 19 Lower evaluation value (Lower)

9 Upper evaluation value (Lower) 20 Lower right evaluation value (Upper)
10 Right upper evaluation value (Upper) 21 Lower right evaluation value (Lower)
11 Right upper evaluation value (Lower)

Minimum evaluation value (Upper) Output sequence12Left evaluation value (Upper)

Note 1: The evaluated values are output using the natural binary number. First, the upper 8 bits are

output and the lower 8 bits are output next.

Note 2: The dmv indication code is specified using the lower 4 bits as shown below. The upper 4

bits are for L output.

0000: The center point vector is optimum (+0.0, +0.0)
1010: Upper left from the center point vector (-0.5, -0.5)
1001: Upper right from the center point vector (+0.5, -0.5)
0110: Lower left from the center point vector (-0.5, +0.5)
0101: Lower right from the center point vector (+0.5, +0.5)
0010: Left of the center point vector (-0.5, +0.0)
0001: Right of the center point vector (+0.5, +0.0)
1000: Upper direction from the center point vector (+0.0, -0.5)
0100: Lower direction from the center point vector (+0.0, +0.5)

3.3.8 Operational Modes and Dynamic Control Signals (for each processing cycle)

M65727 has controls which need to change every execution cycle. These controls differ according
to operational modes as shown below. They are input to the chip through DCNT pins when DSYNC is
asserted. One assertion is needed for each information write into the chip. Therefore, when a mode
needs multiple control information, DSYNC must be asserted multiple times. DSYNC is asserted
low.

CONTACT MITSUBISHI ELECTRONICS REGARDING DISTRIBUTION 17