JVC GY-DV500b Service Manual

SECTION 8

DESCRIPTION OF NEW CIRCUITRY

8.1 OUTLINE OF DV

8.1.1 Features of DV

Table 8-1-1 outlines comparisons between DV and other digital VCR formats.

D9 (Digital S) DVC Pro DV CAM DV

Material MP ME

Tape

Sampling frequency (MHz)

Sampling rate

Width (inch) 1/2 1/4

Track pitch (µm) 20

Y 13.5

R-Y/B-Y 6.75 3.375

NTSC

PAL

4 : 2 : 2

18 15 10

4 : 1 : 1

4 : 2 : 0 (Line Sequential)

Video

Audio

Error correction code

Modulation method

As shown in the above table, the major specifications of the DV are almost identical to the D9 (Digital S). Differences lie only in the tape used, sampling rate and number of executed pixels.

Samples per line

TV lines per frame

Quantization bits

Compression method

Compression

Sampling frequency (kHz)

Quantization bits

Channels

Compression

Y 720

R-Y/B-Y 360 180 : NTSC,360 : PAL

NTSC

PAL

Table 8-1-1 Comparison of Digital VCR Formats

482

578

1/3.3

480

576 : Y , 288 : R-Y/B-Y

In-frame DCT

1/5

48(32)

16(12)

2(4)

Non compressed

Reed-Solomon integration code

SI-NRZI , 24/25

(A) Video signal

High video quality thanks to the 4:1:1 (*4:2:0) component digital recording.

(Note) Descriptions marked * in the text and figures are PAL

data.

• Luminance signal

The Y signal is sampled at 13.5 MHz so the Y signal bandwidth is about half the sampling frequency, or 6.75 MHz. This provides better resolution than S-VHS and current TV broadcasting (Y bandwidth of about 4.5 MHz).

8-1

• Chrominance signal

By recording the chrominance signal in a form divided into color difference signals R-Y and B-Y (component recording method), recording with excellent color reproduction and little color blur is achieved.

MHz

Y Signal

First effective line in each field

Colour difference

Signal (CR, CB)

First effective line in each field

1/13.5

MHz

1/3.375

First pixel in the effective period

With VHS, the chrominance signal has been converted into a lower-frequency signal so that its bandwidth has been limited to 0.5 MHz. The DV has expanded the bandwidth by about 3 times to 1.68 MHz (3.375MHz/2) and records the chrominance components quite separately from the Y component. The sampling frequency of the R-Y and B-Y signals are 3.375 MHz (*6.75 MHz), which is 1/4 (*1/2) of the Y signal sampling frequency of

13.5 MHz. Based on this ratio, this recording method is referred to as 4:1:1 (*4:2:0 line sequential) digital component recording.

Line 285

Line 23

Line 286

Line 24

Line 287

Line 25

Line 285

Line 23

Line 286

Line 24

Line 287

Line 25

720 Samples

180 Samples

4:1:1 Sampling

: Effective sample

: Interlaced sample

480 Samples

Y Signal

First effective line in each field

Colour difference

Signal (CR, CB)

First effective line in each field

Fig. 8-1-1 4:1:1 Component Digital Recording

MHz

1/13.5

Line 335

Line 23

Line 336

Line 24

Line 337

Line 25

MHz

1/6.75

Line 335

Line 23

Line 336

Line 24

Line 337

Line 25

Line 338

Line 26

Line 339

Line 27

First pixel in the effective period

720 Samples

360 Samples

4:2:0 Sampling

: CR

: CB

576 Samples

288 Samples

8-2

Fig. 8-1-2 *4:2:0 Component Digital Recording

(B) Audio signal

The DV format also records the audio digitally. Two recording formats (48 kHz, 32 kHz) are provided to allow selection according to the purpose. Compatibility with 44.1 kHz sampling is also provided for use in the reproduction of software tapes.

• 48 kHz, 16-bit high quality stereo mode

This recording mode provides high audio quality equivalent to the DAT. Although compression technology is used in video recording, the audio recording does not use compression.

• 32 kHz, 12-bit mode with a stereo dubbing capability

This mode divides the audio area into two parts and records 12bit stereo audio channels to both of them, or a total of 4 channels. This makes it possible to record 2 audio channels at the same time as performing video recording and to dub 2 additional audio channels while leaving the original audio channels.

• Video signal output

The digital signal read out from a tape during playback is fed to the image memory to arrange the time axis before being output as the video signal. This makes it possible to reduce wow & flutter, which may be caused by head rotation irregularities or tape transport variations.

8.1.2 Tape Format of DV

Note) Descriptions marked * in the text and figures are PAL

data.

The DV records the video and audio signals independently, in the video area and in the audio area respectively. With the video signal, the data of each frame is divided into 10 tracks (*12 tracks) before being recorded. Namely, the data of a frame is input to a memory, divided into 10 (*12), and recorded into 10 tracks (*12 tracks) using a 2-channel head by rotating the drum by 5 turns (*6 turns) in the period of one frame (33.3 ms/*40 ms) of a normal NTSC (*PAL) signal. The sub-code area records data including the timecode, recording date, index and absolute track number, and the ITI area records the reference signal of the absolute height of track (SSA) ad tracking signals. The DV does not use the tracking control track. The tracking system of the DV is called the 3-frequency (F0, F1, F2) digital pilot system, which records digital pilot signals F0 (0Hz), F1 (465 kHz) or F2 (697.5 kHz) on every track during recording. When the F0 track is reproduced, the DV controls the tracking servo so that the crosstalk levels of the pilot signals (F1, F2) in the tracks on the left and right of the F0 track become identical. The pilot signals are recorded over all the tracks.

Sub

• Timecode enabling editing

The DV tape has a sub-code area for recording signals for use in editing. Timecodes are automatically recorded frame by frame in this area so that the video can later be edited on a per-frame basis. The sub-code area also includes the recording of index ID signals, which can be used in search and other operations later.

• AUX data recording information on recording and shooting

The video signal recording area on the tape records the signal called the AUX data together with the video signal. The AUX data contains information on the recording date and shooting conditions such as wide-screen shooting, and is recorded automatically during recording. Parts of the AUX data information can be displayed on the screen as required. The AUX data on the record mode, etc. is also recorded in the audio area in a similar way to the video area.

• Digital interface for digital inputs/outputs (optional)

All track data can be input or output directly without altering the digital format (IEEE1394 compliant). DV terminals are provided to enable dubbing and editing using digital signals.

Video

Head writing

Audio

ITI

F0 F0 F0 F0 F0F1 F2 F1 F1F2

10 tracks/frame

Fig. 8-1-3 NTSC Format DV Tape

Sub

Video

Head writing

Audio

ITI

F0 F0 F0 F0 F0F1 F2 F1 F1F2 F0 F2

12 tracks/frame

Tape travel

Fig. 8-1-4 PAL Format DV Tape

8-3

(A) Mini-DV cassette tape

The DV cassette uses metallic magnetic particles so that stable, high-power signals can be obtained from even very thin tracks. Metal tapes include the metal particle coated (MP) tapes and metal evaporated (ME) tapes. The Metal tape used with the DV is the ME tape which features a thin magnetic layer and low demagnetization.

• ID board and memory-in cassette (MIC)

The DV cassette has 4-contact terminals on the back label side edge of the shell. These terminals are called the “ID board terminals”, and the values of resistance across the terminals express basic cassette ID (BCID) information such as the tape thickness, tape type and applications. The DV cassette can optionally incorporate a memory for storing the BCID as well as the content information. In this case, these 4 terminals are used as the serial communication terminals for the storage and readout of cassette memory contents.

As described above, the ID board terminals are used in two ways and the cassettes are available as those with or without internal memory according to the usage. The GY-DV500 and BR-DV600 are not avairable with this optional function, so information such as the content information cannot be written or read even when a cassette with internal memory is used with them. All these models do through these terminals is read the BCID of every cassette.

Upper shell

Top lid

ID board terminal

Terminal

thickness

Tape type

Tape grade

Front lid

Accidental erasure protection hole

Accidental erasure protection tab

Open

REC possible : Close REC impossible: Open

Lid lock

Fig. 8-1-5 External View of Mini-DV Cassette

Contents

Tape

7µm

Reserved

For cleaning

For consumer

For professional

Reserved

For PC

–

LED hole

Lower shell

Reel lock

MICID Board

Resistor value Function

Open

1.80kΩ ± 0.09k

VDD

Open

6.80kΩ ± 0.34k

1.80kΩ ± 0.09k

SDA

S/C

Open

6.80kΩ ± 0.34k

1.80kΩ ± 0.09k

SCK

S/C

–

GNDGND

Bottom lid

Tape

8-4

Table 8-1-2 ID Board Terminal Standard

(B) Main Standard of DV Tape Format (SD)

θr

Tape travel (Ts)

A ch 1 (ch 1, 2)

Sub

Code

A ch 1 (ch 1, 2)

MRG

α1

α2

174°

A ch 2 (ch 3, 4)

Video

0 F0

Opt. track 2

Video

ITI

9 F2

(PF1)

A ch 2 (ch 3, 4)

Audio

8 F0

Sub

Code

MRG

Sub

Code

Video

Head motion

Lr (θe)

(We)

(PF1)

Effective data area (NTSC: 134975 bit)

Audio

ITI

1 Frm (Pilot Frm 0) 525/60

ITI

5 F1

Opt. track 1

Video

4 F0

Audio

ITI

MRG

Sub

Code

Audio

3 F2

Opt. track 2

MRG

Sub

Code

MRG

Sub

Code

MRG

Sub

Code

MRG

Sub

Code

1 F1

(PF0)

Effective data area (PAL: 134850 bit)

Audio

ITI

Video

ITI

Audio

ITI

Video

1 F1

Video

0 F0

Audio

ITI

11 F2

Video

Opt. track 1

1 Frm (Pilot Frm 0) 625/50

(PF0)

Fig. 8-1-6 Main Standard of DV Tape Format (SD) (Track Pattern)

Head motion

8-5

8.2

MAJOR SIGNAL PROCESSING OPERATIONS OF DV

Fig. 8-2-1 shows the basic configuration of the major signal processing circuitry of the DV in the form of a block diagram.

SHUFFLE

SHUFFLINGAUDIO

VIDEO

AUDIO

SHUFFLE

MACRO BLOCK SHUFFLING

DESHUF.

DESHUFFLING

DESHUF.

DESHUFFLING

CODING

DCT VLC

DECODING

I-DCT

VLD

ECC

INNER

OUTER

SHUFFLING

Fig. 8-2-1 Basic Configuration of Major Signal Processing Circuitry of DV

8.2.1 Flow of Video and Audio Signals in Recording Circuitry

The video and audio signal inputs/outputs shown in Fig. 8-2-1 are sampled digital signals. The input digital video signal is shuffled block by block, so that deviations in the compression rate of the picture can be prevented. This is achieved by gathering data from various positions in the picture and compressing it. When the video signal is supplied to the ECC block for error correction later, de-shuffling is applied to return the data to the original positions in the video signal. The shuffled video signal is coded. The coding consists of replacing the image signal in the conversion block composed of multiple pixels with signals without correlation by means of discrete cosine transform (DCT). The DCT operation consists of dividing the picture into blocks (each composed of 8 x 8 pixels) and obtaining the transform coefficient, which indicates the amount of components of the previously determined basic image pattern (64 pixels) in each block. The VLC (Variable Length Coding) quantizes the DCT transformed signal and applies entropy coding to the quantized signal. The coding technique used here performs Run Length coding first, then applies Huffman coding. The audio input signal is also shuffled block by block in the same way as with the video signal, but the audio signal is not compressed. The video and audio signals are input to the ECC block, where error-correcting codes are appended to them. The error-correcting codes use product codes obtained by double coding of the Reed-Solomon integration codes. The DCI-R (Digital Channel Interface for Recording) block performs channel coding of the recorded signals. It performs what

IEEE1394

I/F DV TERMINAL

SERVO

DCI-R

SI-NRZI

24/25

Phase &

Amplitude

Compensator

ECC

OUTER

INNER

DESHUFFLING

DCI-P EQ

3 TO 2

1+D

VITERBI

TRACKING

Note : PAL model is not available the signal output.

See note:

REC.AMP. HEAD

PRE-AMP.

HEAD

TAPE

is usually called modulation by transforming a series of digital data composed of “1” and “0” to match the properties of magnetic recording systems.

8.2.2 Flow of Video and Audio Signals in Playback Circuitry

When a digital VCR plays a recorded signal, the frequency response of the low-frequency and high-frequency components is degraded. The degradation in the frequency response of the low-frequency components is because the played signal becomes a differential waveform and a rotary transformer is used. That in the frequency response of the high-frequency components is caused by the performance of the recording tape itself and by the gap between the tape and head during recording/ playback. When a tape in which 1-bit pulses are recorded is played while the frequency response is degraded, the pulse duration is expanded and intersymbol interference results. The equalizer is used to suppress the intersymbol interference and reduce code errors. The DCI-P (Digital Channel Interface for Playback) block demodulates the signal, which has been coded for recording and turns it into a signal that can be subjected to error correction. The ECC (Error Correcting Codes) block corrects errors while performing shuffling. The audio signal is then de-shuffled and output. The video signal is sent to the decoding block where the signal compressed by coding is expanded by means of decoding, and returns to a digital signal. The digital video signal is then de-shuffled before being output.

8-6

8.3 VIDEO/AUDIO SIGNAL PROCESSING IN RECORDING CIRCUITRY

8.3.1 Division into Blocks

As shown in Fig. 8-3-1, data of each frame is divided into macroblocks (MBs) (8 x 8 pixels) which are the basis of the DCT circuitry. Since the luminance signal and two color difference signals are sampled with different frequencies, 4-luminance signal blocks and each of the color difference signal blocks occupy the same position and area in the picture. When data of any one of the 4-luminance signal blocks is lost, the data in other blocks becomes meaningless. Therefore, the signals of every 6 blocks are processed as a single processing unit and recorded onto tape.

(a) With a 525/60 system: 90 (22.5) blocks

(b) With a 625/50 system: 90 (45) blocks

8.3.2 Shuffling (Video)

The 5 macroblocks composing a single video segment is collected from a picture by means of shuffling according to a specified rule. The objective of shuffling lies in making uniform the quantity of information that is contained in the 5 macroblocks in a video segment. The image information of a frame is not usually distributed evenly in the picture, but there are segments with a large amount of information and those with a small amount of information as shown in Fig. 8-3-2.

5MB

60 (60) blocks

Figures inside ( ) are the number of color difference (CR, CB) blocks.

Macroblock

8 8 8

Six blocks occupying the same position and area in the picture form a macroblock.

72 (36) blocks

Macroblock

8 8

Fig. 8-3-1 Division into Blocks, Macroblock

The macroblock is a group composed of the 6 blocks. The data divided into blocks is processed by the DCT coding as described later, and the video-recording rate is compressed to 25 Mbps. In fact, however, the data of each frame is compressed to the specified number of bits, which are recorded onto the specified number of tracks on tape. Namely, when 25 Mbps is converted into the number of bits per frame, the following number of bits are recorded onto 10 (NTSC) or 12 (PAL) tracks:

(25 x 10 (25 x 10

)/30 = 833333 bits ............ (NTSC)

)/25 = 1000000 bits .......... (PAL)

As seen above, the data in each frame is compressed so as not to exceed the specified number of bits calculated based on the video-recording rate. This is referred to as length fixation. Although the length of the data of a frame is fixed so as not to exceed the calculated number of bits, the actual length fixation is applied per 5 macroblocks. In other words, the length of the data of 5 macroblocks is fixed so as not to exceed the following number of bits:

833333 x 5/1350 = 3086 bits ......... (NTSC)

1000000 x 5/1620 = 3086 bits ....... (PAL)

The 5 macroblocks used as the unit of length fixation is referred to as a video segment.

Video segment A (Flat, sky section)

Video segment A (Few information)

Video segment B (Much information)

MB1 MB2 MB3 MB4 MB5

Length fixation information

Video segment B (Fine branch section)

Coarse quantization for reducing the amount of information

Fig. 8-3-2 Distribution of Information in a Picture

As a result, if the fixed-length data obtained from 5 sequential macroblocks uses the same number of bits for each macroblock, distortion due to compression or expansion of data would be noticeable. This would depend on the screen segments (distortion is less noticeable in the segments with little information but noticeable in segments with much information). To prevent this by making the information uniform across the video segments and making distortion less noticeable, shuffling is performed in the picture. The shuffling is performed based on the rule shown in Fig. 8-3-

3. Each picture is divided vertically into the same number as the number of macroblocks in a video segment (5), and divided horizontally into the same number as the number of tracks used to record the data of 1 frame onto tape (NTSC: 10, PAL: 12). The blocks divided in this way are referred to as superblocks.

8-7

Sequence of MBs in a superblock

No.

8 pixels

2nd Field 1st Field

8 lines

Sequence of superblocks

Order of shuffling

12 34 5

MB1 MB2 MB3 MB4 MB5

Set at a fixed length of 5 MBs.

Fig. 8-3-3 Shuffling Technique (Example with 525/60)

With shuffling, the first macroblock (No. 0) in each of the 5 superblocks are collected to form a video segment as shown in Fig. 8-3-3, and the next video segment is formed by collecting the first macroblocks (No. 1) of the same 5 superblocks. When all of the macroblocks in the 5 superblocks have been collected, data collection of the next 5 superblocks starts.

8.3.3 DCT

The video segments formed by shuffling are DCT transformed per (8 x 8) blocks. There are two DCT transform modes including the stationary mode and the dynamic mode.

(1) Stationary mode (8 x 8 DCT mode)

This is the basic mode, executing (8 x 8) DCT transform to every (8 x 8) pixels in blocks.

(2) Dynamic mode (2 x 4 x 8 DCT mode)

The (8 x 8) blocks are divided into two (4 x 4) blocks called fields 1 and 2, and a (4 x 8) DCT transform is applied to every (4 x 8) pixels. The following paragraph explains this mode by taking the case in which an object moves toward the right as an example.

When the object moves horizontally:

The vertical high-frequency component increases.

4 lines

(2 x 4 x 8)

4 lines

8 pixels

(1st Field)

8 pixels

(2nd Field)

Fig. 8-3-4 (2 x 4 x 8) DCT Transform in Motion Mode

When the object does not move, the data in the horizontal direction contains the high-frequency component but the data in the vertical direction contains only the DC current. On the other hand, when the object moves in the horizontal direction, the position of the object at the time of 1st field scanning is different from that at the time of 2nd field scanning, so the picture of the object in a frame appears to have ridged edges. The highfrequency component increases in the vertical direction data as well as in the horizontal direction data. In such a case, the (8 x 8) blocks are divided into two pairs of (4 x 8) blocks for fields 1 and 2, and (4 x 8) DCT transform is applied to every (4 x 8) pixels. This method makes it possible to reduce any increases in the high-frequency component in the vertical direction data and prevents a drop in the compression rate. Whether a DCT transform is performed in the stationary mode or in the dynamic mode is decided by detecting the (8 x 8) blocks in each video segment by the motion detector circuit located before the DCT transform circuit. The data of a block in stationary mode is composed of one DC component and 63 AC components. But, in dynamic mode each one of the two (4 x 8) blocks is composed of a DC component and 31 AC components. To allow the dynamic mode data to be processed in the same way as the stationary mode data, the dynamic mode processing calculates the sum and differences of the factors of the same order and form (8x 8) blocks as shown in Fig. 8-3-5.

(2 x 4 x 8) DCT factors

(1st field)

(Sum of factors)

(Difference of factors)

8-8

(2nd field)

Fig. 8-3-5 Processing of DCT Transform Factors in Dynamic Mode

The above processing allows both the dynamic mode data and

0 0 1 2

0 0 1

0 1 1

1 1 1

1 1 2

1 2 2

2 2 2

2 2 3

1234567

(8 x 8) DCT

Vertical direction

0 0 1 2 3

0 1 1 2

1 1 2 2

1 2 2 2

1 2 2 3

2 2 3 3

2 3 3 3

3 3 3 3

1234567

(2 x 4 x 8) DCT

Horizontal direction

(Sum)

Vertical direction

(Difference)

Class No. Area No.

0 15 14 13 12 11 10

15 14 13 12 11 10

9 8 7 6 5 4 3 2 1 0

15 14 13 12 11 10

9 8 7 6 5 4 3 2 1 0

15 14 13 12 11 10

9 8 7 6 5 4 3 2 1 0

0 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 4 4 4 4 8 8

1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 4 4 4 4 8 8 8 8

2 1 1 1 1 1 1 1 1 2 2 2 2 4 4 4 4 8 8 8

8 16 16

3 1 1 1 1 1 1 1 2 2 2 4 4 4 4 8 8 8

8 16 16 16 16

Quantizer No.

Horizontal direction

stationary mode data to be handled as data composed of one DC component and 63 AC components in the subsequent quantization processing.

8.3.4 Quantization

(A) Data quantity estimation

After all of the blocks in a video segment have been DCT transformed, the video segment is stored in the buffer. At this time, the data quantity estimation block selects the quantizer for use in quantizing the video segment. With DCT transform, normalization is performed so that the DCT factors have a 10-bit dynamic range for the 8-bit pixel block. The DCT factors are divided by an integer called the quantization step and allocated to a smaller number of bits before being subjected to variable-length coding (VLC). This operation (division) is referred to as re-quantization, or simply quantization. The quantization reduces the values of the factors, turning many of 0 in the high frequencies. As the VLC in the subsequent stage performs code allocation by forming value 0 run lengths and the non-0 factor after it into a group, so the quantization improves the coding efficiency. As the VLC consists of simply allocating codes to factors, the code quantity (number of bits) is controlled by the VLC block.

(B) Quantization

Each quantizer is composed of a set of 4 quantization steps as shown in Fig. 8-3-6, and the factors in a block are divided into 4 areas. The data is quantized in the 4 areas using different quantization steps. The factors are shifted from low-frequency factors to high-frequency factors as the area number increases, so the quantization of high-frequency factors are coarser than for low-frequency factors. This utilizes the fact that the distortion of high-frequency factors is less noticeable to human vision even when they are coarsely quantized.

Fig. 8-3-6

Each block in a video segment is classified into one of 4 classes before being quantized. The quantization steps of the quantizers vary depending on the class numbers (see Fig. 8-3-6). Table 8-31 shows the definitions of the 4 classes.

Class No.

Block with which quantization distortion after compression is highly noticeable. The absolute value of the AC factor should not exceed 225.

Block with which quantization distortion after compression is less noticeable than Class 0. The absolute value of the AC factor should not exceed 225.

Block with which quantization distortion after compression is less noticeable than Class 1. The absolute value of the AC factor should not exceed 225.

Block with which quantization distortion after compression is little noticeable or the absolute value of the AC factor exceeds 225.

Definition

Table 8-3-1 Definitions for Classification

8-9

The Table 8-3-1 means that a block with a larger class number has a more approximate degree of definition (activity). The data in such a block is quantized relatively roughly to compress the data quantity. Rather approximate quantization of picture areas with low activity does not cause odd sensations for the human vision, but approximate quantization of those with high activity does cause odd sensations. The blocks with low activity in each segment are quantized as roughly as possible with a reduced number of allocated bits, unless the odd sensations are not felt. However, the blocks with high activity are quantized using an increased number of allocated bits to improve the picture quality experienced by the human vision. A block containing factors with a larger absolute AC factor than 255 is classified as Class 3, and subjected to an operation called initial shifting before quantization. This operation divides an AC factor over 255 by 2 so that errors do not occur in the VLC operation.

Class No.

Initial shifting

Before initial shifting

MSB=1

(ACmax > 255)

After initial shifting

Not applied1Not applied2Not applied

9 bits

MSB LSB

1-bit shift

The value is accommodated within 8 bits.

Applied

Fig. 8-3-7 Initial Shifting

While the AC factor before quantization consists of 10 bits, or 9 bits excluding the sign bit, the VLC in the subsequent stage can codify a non-0 factor value only into an 8-bit code. Initial shifting is performed to deal with this. As a result, a factor that has a value in the 9th bit (SMB) (i.e. a factor over 255) can be accommodated in 8 bits and processed by VLC. Fig. 8-3-6 shows that an increase in the class No. increases the quantization steps in the same quantizer No. or that quantization of a larger class No. is more approximate. Although the quantization steps of Class 2 look less than for those of Class 3, this is because the Class 3 data has been subjected to initial shifting before quantization. The actual quantization steps of Class 3 are more approximate than in those of Class 2.

8.3.5 VLC

After division into (8 x 8) blocks by DCT transform, the energy of pixels are concentrated in the DC component and the vertical and horizontal vertical factors become almost null. To code these factors, coding uses a technique for varying the code length according to the incidence of the factors. This coding technique varying the code length is referred to as variable length coding (VLC).

(A) Entropy coding

As the VLC achieves efficient coding by allocating short codes to data with a high DCT factor incidence and long codes to data with a low incidence, it is sometimes called the entropy coding. The entropy coding can be expressed with an entropy function.

H = 1 (max.) when Pn = 0.5 (when 0 and 1 both occur at the same probability)

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 0.1

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Fig. 8-3-8 Entropy Function

When VLC approaches the above function, it is regarded as coding with a high compression rate.

(B) Huffman coding

The DV uses Huffman codes to improve the compression efficiency. The Huffman codes determines the minimum average code length of information with a known incidence probability. Table 8-3-2 shows the method of construction of Huffman codes. The symbols in the table should be regarded as the levels of pixel values.

Symbol Incidence Code Word

0.49

0.14

0.07

0.04

0.02

0.01

Construction Procedure

100

101

1100

1101

1110

11110

111110

111111

8-10

Table 8-3-2 Construction of Huffman Codes

The Huffman coding uses the following procedure.

Mode Channels Sampling Frequency

48 k

44.1 k 32 k 32 k

48 kHz

44.1 kHz 32 kHz 32 kHz

16-bit linear

12-bit nonlinear

Quantization

(1) Information source symbols 1 to 9 are arranged in the order

of incidence probability.

(2) For the symbol with the lowest incidence and that with the

second-lowest incidence, “0” is allocated to one and “1” is allocated to the other. (Which of the two symbols is “0” and which is “1” can be decided arbitrarily.)

(3) Then, the above two symbols are considered to be joined

together as a single symbol. The incidence probability of the joined symbols should be equal to the sum of the incidence probabilities of the two symbols before joining.

(4) By considering the joined symbols as new individual sym-

bols, these symbols and other symbols (which should exclude the two symbols which were joined) are rearranged in the order of incidence probability.

(5) Steps (2) to (4) are repeated until the number of symbols

becomes 1.

(6) The values (0 or 1) allocated to the original symbols in step

(2) are read in the reverse order. The read values form the code word for the symbols.

255

Factor Value <Absolute Value>

1 2 3 4 5 6 61 62 6378

Order of Zigzag Scanning

Factor series = {0, 12, 5, 0, 0, 0, 4, 3,... 0, 0, 0}

Transform into (0 run length + non-0 factor values)

(1, 12), (0, 5), (3, 4), (0, 3)... (EOB)

A Huffman code constructed in this way can be decoded instantaneously because, by following the bits one by one from the first bit at the time of decoding, the end of a code word can be determined without referring to the head of the next code word. As the VLC used with the DV is based on Huffman coding combined with run length coding, it is called “modified 2-dimensional Huffman coding”.

Horizontal frequency

Vertical frequency

Fig. 8-3-9 Zigzag Scanning of DCT Factors

Among the factors transformed by 2D DCT, the 63 AC factors other than the DC component factor are rearranged by zigzag scanning as shown in Fig. 8-3-9. When the AC factors are rearranged in the order of scanning, the result is as shown in Fig. 8-3-10, where the scanned factor series is divided into a group of sequential 0 factors followed by a group of non-0 factors. A code called the EOB (End Of Block) is appended after the last non-0 factor scanned. After the factors are transformed into a group of 0-value run length and a group of non-0 factor values, a Huffman code table is compiled according to the incidence probabilities of these groups and VLC is applied based on this.

VLC

(111011s), (100s), (11011s),... (1110)

s: Sign bit

Fig. 8-3-10 2D Huffman Coding

8.3.6 Shuffling (Audio)

Unlike the video signal, the audio signal is not compressed. It is subjected only to shuffling as preparation for the ECC. Table 8-3-3 shows the basic modes of the DV.

Table 8-3-3 Basic Audio Modes

In the 2-channel modes, the quantization is 16-bit linear and the sampling frequency can be selected from 48 kHz, 44.1 kHz or 32 kHz. One of the 2 channels is recorded in the first 5 tracks of the 10 tracks of an NTSC frame (first 6 tracks of the 12 tracks of a PAL frame) after shuffling. Therefore, when an error occurs with a track, 1/5 (or 1/6) interpolation is applied.

8-11

8.3.7 De-shuffling

The compressed video data is recorded on tracks on tape. The video data recording area of each track is divided into sub-units called the sync blocks (SBs). The number of SBs where video data is recorded is 135 per track, and the number of SBs per frame is:

135 x 10 = 1350 ............................. (NTSC)

130 x 12 = 1620 ............................. (PAL)

These values are equal to the number of macroblocks per frame. Therefore, the compressed video segment data (5 macroblocks) is packed into 5 SBs and recorded. Fig. 8-3-11 shows the packing of video segment data into 5 SBs.

• The macroblock data is packed into the specified area (fixed area) on a per-DCT block basis.

• Overflow data is stored in the macroblock memory.

• This step is performed for each macroblock.

• The data in MB memory is packed in the vacant area of the same SB.

• Overflow data is stored in the VS (Video Segment) memory.

• This step is performed for each macroblock.

• The data in theVS memory is packed in the vacant areas in the 5 macroblocks.

The section corresponding to the data area of each SB(unit) is divided into the number of DCT blocks per macroblock (fixed areas) and the data in the DCT blocks is packed in priority in the fixed areas. When data packed in 5 units is recorded, it is not recorded in 5 sequential SBs. After the data in each video segment has been packed in 5 units, the data is de-shuffled before being recorded. The data of a frame, composed by the de-shuffling, is recorded sequentially on the SBs on tape as shown in Fig. 8-3-12.

Data of video segment (5 MBs)

De-shuffling

Track 2 Track 3 Track 4 Track 5 Track 6 Track 7 Track 8 Track 9

De-shuffling memory

Data is recorded on the track in this order.

Track 10

Y block 1

Y block 2

Y block 3

Y block 4

CR block

(14 byte)

(10 byte)

CB block (10 byte)

5B1 8B

Overflow data is stored in the MB memory. (Step 1)

Data of

MB memory 1

block 1

Data of block 3

Data in the MB memory is packed in the vacant areas of the same SB. (Step 2)

Overflow data is stored

VS memory

Data of MB1

Data of MB3

in the VS memory.

Data in the VS memory is packed in the vacant areas in the 5 MBs. (Step 3)

MB1

MB2

MB3

MB4

MB5

Fig. 8-3-11 Packing of Compressed Data in SB

1 frame (10 tracks)

Fig. 8-3-12 De-shuffled Recording (NTSC)

The reason why the shuffled data is not recorded on tape but de-shuffled before being recorded is to make the picture quality in variable-speed playback easier to view.

Head

Areas updated per head scan

8-12

Areas forming a group on the picture are updated simultaneously.

Fig. 8-3-13 Picture Updating in Variable-Speed Playback

Fig. 8-3-13 shows the picture updating by data reproduced during variable-speed playback. The picture is easy to see because the de-shuffling before recording causes the reproduced data to form a sequential group.

8.3.8 Error Correction

Error correction consists of correcting errors occurring in the data. For this purpose, data is provided with regularity by adding redundant parts. The error correction codes are the mathematical systematization of how to add the redundant parts, and the theory of this is referred to as the coding theory.

(A) Principles of error detection and error correction

With the DV, it may happen that data recorded as “0” is reproduced as “1” or data recorded as “1” is reproduced as “0” due to thermal noise of the recording/playback amplifier or dust or scratches on the tape surface. The error correction detects and corrects such code error produced in the process of recording or playback. In the following, the principles of error detection and error correction will be explained by using codes composed by adding a 3-bit redundant part to every 2 bits of data as shown below.

Information part Information part Parity check part

(00) → (00 110) (01) → (10 101) (10) → (10 011) (11) → (11 111)

With the coding theory, the part for data is called the information part, the redundant part is called the parity check part, and the total of information part and parity check part is called the code word. The operation of transforming the information part into code word is referred to as coding. Let us assume that a code word (00110) corresponding to data (00) has been recorded and is reproduced with an error in 1 bit (01110). When the reproduced series is compared with previously defined 4 code words, it may be known that the former coincides with none of the latter. These are the principles of error detection. When the above situation is verified in more detail, it can be established that the reproduced series (01110) is different by 1 bit from the code word (00110) and by 2 or more bits from other code words. The fact that a code word is most similar to the reproduced series means that the probability that it is the recorded code word is highest. Therefore, coding errors can be corrected by considering that the most similar code word to the reproduced series is the recorded code word. These are the principles of error correction. The operation of correcting errors in the reproduced series is referred to as decoding.

(B) Hamming distance

The error detection and correction capabilities of error correction coding can be defined in terms of the hamming distance between the code words. The hamming distance is the number of dissimilar components when two code words are compared component by component. For example, assuming that code word (00110) is C tance dH (C (11111) is C

1 and code word (01101) is C2, hamming dis-

1, C2) is 3. If code word (10011) is C3 and code word

4, hamming distance dH (C3, C4) becomes 2.

Fig. 8-3-14 shows the scheme of the relationship between the error detection and correction capabilities and the hamming distance.

Min. hamming distance

Code word

Code word Cj

2t+1

Fig. 8-3-14 Expressions of Error Detection and Correction

Capabilities by Hamming Distance

To enable error correction coding, detect all of S or more errors, the minimum value of the hamming distance between code words (minimum hamming distance) should be S + 1. This is because, if the minimum hamming distance is less than S, code word Ci could be turned into another code word Ci if there are S error items. Similarly, to correct t or fewer errors, the minimum hamming distance between code words should be 2t + 1 or more. This is because, if the minimum hamming distance is less than 2t, code word Cj which has a shorter hamming distance than Ci would exist if t error items occur with Ci. The hamming distances between the code words of the above-mentioned code are:

H ((00110), (01101)) = 3

H ((00110), (10011)) = 3

H ((00110), (11111)) = 3

H ((01101), (10011)) = 4

H ((01101), (11111)) = 2

H ((10011), (11111)) = 2

Therefore, the minimum hamming distance between these code words is 2. Thus, as the number of errors that can be detected is:

S = 1 (because S + 1 = 2) This means that the error of a bit can be detected. It was described above that the error of a bit in code word (00110) can be corrected. This is because the minimum hamming distance between only code word (00110) and other code words is

3. Therefore:

t = 1 (because 2t + 1 = 3) However, correction of error of a bit is not possible with other code words, with which only error detection is possible.

Examples of simple error correction codes will be described in the following. (1) Simple parity check code

The parity check code is composed of a k-bit information part and 1-bit parity check part. The code with which parity is selected so that the number of “1” contained in a code word is an even number is referred to as the even parity code. That with which parity is selected so that the number of “1” contained in a code word is an odd number is referred to as the odd parity code. For example, a 3-bit code may have even parity codes as follows:

(00) → (000)

(01) → (011)

(10) → (101)

(11) → (110) All of the hamming distances between the 4 code words are 2, so a 1-bit error occurring in any code word can be detected.

8-13

(2) Repetition code

The repetition code consists of n times of repetitions of 1bit information parts, and the number of code words is 2. For instance, a 3-bit code has two repetition codes as follows:

(0) → (000)

(1) → (111) The minimum hamming distance between these codes is 2, so a 1-bit error occurring in a code word can be corrected.

(3) (7, 4) hamming code

The (7, 4) hamming code consists of a 4-bit information part (i

3, i2, i1, i0) and a 3-bit parity check part (P2, P1, P0) which is

determined according to the following rule.

P0 = i3 + i2 + i0 P1 = i2 + i1 + i0

.... (8.3.1)

P2 = i3 + i2 + i1

In expressions (8.3.1), operators “*” represent exclusive OR, so 0 + 0 = 0, 0 + 1 = 1, 1 + 0 = 1, and 1 + 1 = 0. The (7, 4) hamming code has 4 information bits, or 2

= 16

code words. Table 8-3-4 shows all of the code words.

Information Part

0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1

Parity Check Part

Table 8-3-4 List of Code Words of (7, 4) Hamming Code

This table shows that the minimum hamming distance of the code is 3 and that a 1-bit error can be corrected.

(4) Basic knowledge on algebra

The examples of error correction codes used in the description of the previous section are beginner-type codes, and their error detection and correction capabilities are insufficient for application in the DV. The error correction codes with practical error detection and correction capabilities for use with the DV are constructed based on abstract algebra. In the following, the minimum required knowledge on the algebra for understanding the error correction will be described. If you want to gain more knowledge, please also refer to books on mathematics.

(a) Galois field

Here, the term “field” refers to the set of elements with which the four arithmetical operations of addition, subtraction, multiplication and division can be defined. The term “field” is used because its functions are working organically, and have come from German word “Körper”. For example, a real number or complex number is a field. However, when the term “field” is used with error correction codes, it does not mean a field with an infinite number of elements such as a real number, but means a finite field composed of a finite number of elements. The “element” is what belongs to a set. For example, if there is “a” which belongs to set “M”, it is said, “a is an element of M”. The fact that “a” belongs to “M” is expressed using a symbol as “a ∈ M2”. “∈“ is the symbol of the initial letter of “element”. If “a” does not belong to “M”, it is expressed as “a ∉ M”.

A finite field is also called the Galois field, after the French mathematician Evariste Galois (Oct. 25, 1811 - May 31, 1832) who created one of the basic algebra theories. A Galois field composed of P elements is expressed as GF(P). The simplest Galois field is a GF(2) composed of elements (0, 1). The addition in GF(2) is defined as exclusive OR and the multiplication can be defined as multiplication of ordinary integers. Table 8-3-5 shows the operation tables of GF(2).

(a) Addition (b) Multiplication

Table 8-3-5 Operation Tables of GF(2)

8-14

(b) Extension field

The complex number field is created by adding root “i” of

“x

+ 1 = 0”, the polynomial which cannot be factorized any more in a real number field (irreducible polynomial) to a real number field. In the same way, by adding a root of an irreducible polynomial in Galois field GF(P) to GF(P), it is possible to create a field GF(P is equal to P

which is a P’s power. This operation is re-

) where the number of elements

ferred to as a field extension. GF(P) is called the ground field and GF(P

) is called the extension field.

For example, assume that there is an irreducible cubic polynomial as follows:

+ x + 1 = 0,

If there is an extension field GF (2

) created by a root of the above polynomial “α“ to GF (2), “0”, “1” and “α“ are the elements of extension field GF (2

). Then, since multiplication should be definable in a field, all of α‘s powers become the elements of GF (2 nomial x

). As for the α‘s powers, since “α“ is the root of poly-

+ x + 1 = 0,

= α + 1

Therefore, this relationship means the following:

= 1

= α

α3 = α + 1

= (α + 1) α = α2 + α

= (α2 + α) α = α2 + α1 + 1

= (α2 + α + 1) α = α2 + 1

= (α2 +1) α = 1 : (Repeated)

α‘s power after α

are repetitions of the above. Fig. 8-3-15 shows this relationship from a different viewpoint. In this figure, since the cycle is closed, α‘s power after α replaced by values of α

α = α

1 = α

or below.

are

element with a period of 2

- 1 to GF(2). The hamming code is a cyclic code which uses a primitive polynomial as the generation polynomial, and its parameters are as shown below. Code length : n = 2

- 1 Information amount : k = n - m Min. hamming distance : d

MIN = 3

A hamming code with code length of n and information amount of k is called a (n, k) hamming code. Since the minimum hamming distance is 2, any 1-bit error occurring in the code word can be corrected.

(7) BCH code

The BCH (Bose-Chauduri-Hocgenghem) code extends the error correction of a Hamming code to 2 bits or more. The BCH codes with t-bit correction become the cyclic codes of code table n with which generation polynomial G(x) is a minimum-order polynomial having α, α assuming that α is a primitive root of GF(2

- 1.

,... α2t as the roots,

) and that n =

(8) Reed-Solomon code

The BCH code described in above (7) is defined on GF(2). The Reed-Solomon code is an extension of the above to

GF(2

). Its error detection and correction are applied per symbol, which is an element of the extension field. In general, a Reed-Solomon code with code length of n and information amount of k is called the (n, k) RS code.

(9) Reed-Solomon product code

Assuming that C

1 is (n1, k1) codes and C2 is (n2, k2) codes,

the product code of them can be obtained by coding k umns, each composed of k

2 rows, each composed of k1 symbols with C1. Fig. 8-3-16

2 symbols, with C2, then coding

shows the construction of product code.

1 col-

Fig. 8-3-15 Cycle of GF (2

)

(5) Cyclic code

The cyclic codes are important codes that can be coded or decoded easily. The cyclic code has the following properties:

• The sum of arbitrary code words is a code word.

• The series obtained by cyclic shifting of code words be-

comes a code word.

(6) Hamming code

When one of the roots of an irreducible m-order polynomial in GF(2) is assumed to be α, the maximum period until α‘s power returns to the original number is 2

- 1 (see Fig. 8-3-

15). An irreducible polynomial which has a root α with which the period is 2

- 1 is called the primitive polynomial, and α in

this polynomial is called a primitive element.

GF(2

) is an extension field created by adding a primitive

Information part

C2’s parity check part

C1’s parity check part

C2’s parity check part on C1’s parity check part

Fig. 8-3-16 Construction of Product Codes

The C2’s parity check part on C1’s parity check part is identical to the C1’s parity check part on C2’s parity check part. As a result, the same product code can be obtained by coding C1 before coding C2. The DV uses the Reed-Solomon product code obtained by a double coding of the Reed-Solomon code.

8-15

8.3.9 Data Structure

Random errors and burst errors due to dropouts on tape occur with the magnetic recording/playback circuitry of the DV. To deal with this, the DV records data by turning into sync blocks. Fig. 8-3-17 shows the DV data structure.

Sync

ID code

area

ID0,ID1,ID

135

Sync

ID code

ID0,ID1,ID

area

Sync

ID code

area

ID0,ID1,ID

Sync block length: 90 bytes

VAUX

Outer code parity

AAUX

Outer code parity

Sync block length: 12 bytes

Sub-code data

77 8

Video data

77 8

VAUX

Video data

72 8

Audio data

Inner code

parity

Inner code

parity

Inner code

parity

Inner code

parity

Inner code

parity

Inner code

parity

GF (2

)

GF (2

)

GF (2

)

Unit: Byte

Fig. 8-3-17 Data Structure

The video and audio uses the same sync block configuration to reduce the scale of hardware. The video or audio data forms a sync block together with the sync area, the ID code indicating the data attributes and the inner code parity.

The byte counts of the sync area and ID code in Fig. 8-3-17 are the values before the 24/25 conversion (by a converter circuit which produces digitally the pilot signal for use in tracking during playback). In practice, however, they are 24/25-converted and become scrambled interleaved NRZI-converted signals before being recorded onto tape. The following 17-bit patterns have been defined as sync patterns.

• Audio/Video sectors Sync-D : 00011111111110001 Sync-E : 11100000000001110

MSB LSB

• Sub-code sector Sync-F : 00000111111111101 Sync-G : 11111000000000010

MSB LSB

The ID code is composed of 3 bytes including two ID bytes and a parity byte. The ID bytes of a video or audio sector are: 4 bits of sequence number for indicating the continuity of frame, 4 bits of track pair number indicating the track number, and 8 bits of sync block number indicating the arrangement of sync blocks in each track. (1) Error correction in the ID code area

In the ID code area, (12, 8, 3) BCH codes are used for common error correction to the audio, video and sub-code sectors. The configuration is as shown below.

Primitive polynomial: x4 + x + 1 ID

0:C15 C14 C13 C12 C11 C10 C9 C8

ID1:C7 C6 C5 C4 C3 C2 C1 C0 IDP:P7 P6 P5 P4 P3 P2 P1 P0

MSB LSB

ID-CW

0: C14 C12 C10 C8 C6 C4 C2 C0 P6 P4 P2 P0

ID-CW1: C15 C13 C11 C9 C7 C5 C3 C1 P7 P5 P3 P1

The parity can be expressed as follows.

7 =C15 +C11 +C7 +C5

P5 =C15 +C13 +C9 +C5 +C3 P3 =C15 +C13 +C11 +C7 +C3 +C1 P1 =+C13 +C9 +C7 +C1 P6 =C14 +C10 +C6 +C4 P4 =C14 +C12 +C8 +C4 +C2 P2 =C14 +C12 +C10 +C6 +C2 +C0 P0 =+C12 +C8 +C6 +C0 As the contents of ID code have regularity (inertia) between successive sync blocks, it is also possible to apply only the error detection.

(2) Error correction in the video and audio areas

The video data and audio data are composed of Reed-Solomon product codes of GF (2

) as shown in Fig. 8-3-17. With internal codes, a common code length is used to reduce the burden on the hardware.

(a) Inner codes: same video and audio (85, 77)

α : Primitive element Primitive polynomial : x Generation polynomial : g = (x + 1)(x + α)(x + α

+ x4 + x3 + x2 + 1

)...(x + α7) As 8 bytes of parity are added, up to 4 errors can be corrected.

(b) Outer codes: Video (149, 138), audio (14, 9)

Primitive polynomial: x

+ x4 + x3 + x2 + 1 Generation polynomials: Video g = (x + 1)(x + α)(x + α Audio g = (x + 1)(x + α)(x + α

)...(x + α10)

)(x + α3) (x + α4) As 11 parity bytes are added to the video data, the maximum burst error correction length assuming that all random errors are corrected by inner code is 85 x 11 = 935 bytes, which corresponds to correction of error due to a defect of about 0.3 mm in the widthwise direction of tape. Similarly, 5 parity bytes are added to the audio data so the maximum burst error correction is 85 x 5 = 425 bytes, which corresponds to correction of error due to a defect to about

0.14 mm in the widthwise direction of tape.

8-16

+ 36 hidden pages