:onference Record
`
`IEEE Catalog Number
`76 CH 1149-4 CSCB
`
`Volume I
`
`1976
`National
`Telecommunications
`Conference
`
`Communications and Knowledge,
`Partners in Progress
`Copyright c 1976 by the Institute of Electrical and Electronics Engineers, Inc.,
`345 East 47th Street, New York , N.Y. 10017
`
`NTC 76 • November 29, 30 and 1 December 1976
`Dallas, Texas
`
`•
`
`APPLE EX. 1026
`Page 1
`
`

`
`Conference Record
`1976 National Telecommunications Conference
`(NTC 76)
`
`I K
`5101
`f!\\
`f\(~'74-
`1'11"
`'{, \
`
`Copyright ~ 1976 by the Institute of Electrical and Electronics Eng ineers, Inc.,
`345 East 47th Street, New York, N.Y. 10017
`
`Library of Congress Catalog Card Number: 57-20724
`IEEE Catalog Number 76CH1149-4 CSCB
`
`II
`
`APPLE EX. 1026
`Page 2
`
`

`
`COMPOSITE INTERFRAME CODING or NTSC COLOR TELEVISION SIGNALS
`
`Tatsuo Ishiguro, Kazumoto Iinuma, Yuldhiko Iijima,
`Toshio Koga, Shintaro Azami and Takayoshi Mune
`
`Nippon Electric Company Ltd.
`
`ABSTRACT
`
`This paper describes an interframe codec ~hich
`directly encodes composite NTSC color television
`signals at a transmission bit rate of 16 - 32 Mb/s
`for use as high quality television transmission. The
`coding principle is based on transmitting a higher
`order differential PCM signal of frame difference
`vith a variable length coding . As methods of
`suppressing un~anted significant picture elements,
`a compensation for sub- carrier phase inversion
`between frames and a nonlinear function in the coding
`loop are adopted. For signals with large amount of
`motions, a sub- Nyquist sampling is used to reduce
`coding bit rate .
`
`The codec is transparent for the composite
`video signal, providing high color fidelity no
`resolution loss and wide band (5 MHz) capability. A
`codec, which is named NETEC-22H, is constructed .
`
`INTRODUCTION
`
`Interframe coding is expected as a powerful
`method for realizing a television transmission vith
`very low bit rate.l-4 NETEC 6/16 system2,3 demon(cid:173)
`strated that, although pictures were restricted to
`those with a little motions as conference television
`signals, NTSC color television signals could be
`transmitted at 6 Mbit/sec and that, at 16 Mbit/sec,
`most of broadcast television pictures could be
`transmitted with good picture quality . However, the
`coding performances were not sufficient in the sense
`
`that a jerkiness was caused for violent motion and
`that the separate component coding produced small
`deteriorations in resolution and color fiderity .
`order to avoid these deteriorations, a composite
`coding must be applied to interframe coding. The
`composite coding has an advantage of preserving
`strictly the waveform of the signals; video, sync,
`color burst, VITS, standard test signals, etc.
`
`In
`
`This paper presents a composite interframe
`coding system aiming at the very high quality with a
`relatively high bit rate of 16 to 32 ~fuit/sec . An
`algorithm and coding performances are discussed . A
`codec, named NETEC-22H, is described briefly.
`
`ENCODING ALGORITHM
`
`System block diagram
`
`NETEC- 22H system block diagram is shown in
`Figure 1. An input composite NTSC color television
`signal is converted into an 8 bit PCM signal by an
`analog- to- digital converter. The PCM signal is
`encoded into a reduced rate coded words with 2 - 3
`bits/sample on an average through a digital signal
`processing followed .
`
`A pre-processing circuit makes compensation for
`sub- carrier phase inversion between frames, an inter(cid:173)
`frame predictive coder removes redunduncy from the
`signal, and a variable length coder removes re(cid:173)
`dunduncy from the codes representing the prediction
`error. The compressed data are once stored in a
`
`8BIT I SAMPLE
`
`i PRE-
`
`PROCESS.
`
`INTERFRAME
`CODER
`
`VARIABLE
`LENGTH
`CODER
`
`BUFFER
`MEMORY
`
`DATA
`OUT
`
`IN
`SOUND
`
`OUT
`
`VIDEO
`OUT
`
`POST(cid:173)
`PROCESS.
`
`INTERFRAME
`DECODER
`
`VARIABLE
`LENGTH
`DECODER
`
`BUFFER
`MEMORY
`
`Fig. 1 NETEC-22H system bleak diagram.
`
`6 . 4-1
`
`APPLE EX. 1026
`Page 3
`
`

`
`buffer memory and then transmitted through an error
`correcting coder to a line . At the decoder, the
`compressed data is decoded into the PC~1 signal
`through the reverse process, and the PCN signal is
`converted into the analog composite NTSC color
`television signal by a digital-to- analog converter .
`Since the digital processing is designed principally
`to preserve the waveform, the codec is transparent for
`the composite video signal ,
`
`Sampling fre g ~ency
`
`In the composite coding, the san:pling frequency
`must be higher than double the ~TSC television signal
`bandwidth of 4 , 2 NHz, and is cho en to be 10 . 76 MHz
`(684 fh , fh : horizontal sync frequency) in normal
`n;ode ,
`
`On the other hand, in interframe coding a sub(cid:173)
`sampling, or halving the sampling frequency, is
`usually used to avoid the buffer OYerflow when
`pictures have a large amount of motions . However,
`Lhe sub- samp l ing causes an apparent picture quality
`degradation, which vill not. be permitted in the high
`quality encoding system,
`
`In NETEC- 22H system, a sub-Nyquist encoding
`vit.h a sampling rate of 7 , 16 Hllz(·156 fb), about double
`the sub- carrier frequenc , is adopted in stead of the
`sub- sampling. Aliasing components can be removed by
`means of a comb filtering , and the picture quality
`degradations are hardly seen .
`
`Pre-processing , OTF
`
`As long as frame-to-frame prediction is limited
`to the use of only one frame delay, the sub- carrier
`chrominance component remains in the frame difference
`signal because of the sub-carrier phase difference
`of 180° between succeeding frames, Compensation
`for the phase inversion is made by first separating
`the composite signal into luminance and chrominance
`components and then inverting !.he polarily of the
`chrominance component frame by frame ,
`
`In the pre- processing circuit, the separation
`is made by an orthogonal transformation, OTF convert(cid:173)
`ing two horizontal scanning lines signal into a line
`sum end a line difference component . The line
`difference contains mainly the chrominance component,
`while the line sum contains the luminance . Figure 2
`It shoul d be noted
`shows a schematic diagram of OTF,
`that the waveform of the composite NTSC television
`signal is strictly preserved since OTF is a reversi(cid:173)
`ble processing.
`
`Predictive coding
`
`A block diagram of the interframe predictive
`coder and decoder is shown in Figure 3. The coder
`is composed of a frame - to-frame prediction loop and
`an in~rrame coder which encodes the frame difference
`signal. The in- frame coder is a higher order DPCH
`which can effectively encode simultaneous l y the
`luminance and the sub- carrier chrominance component .
`
`The prediction function is changed according
`to the operating modes as follows:
`
`(1) T mode interframe coding (normal mode):
`HO-DPCM of the frame difference (P(z)=z - 3).
`
`(a) Transmitting side
`
`±I
`
`(b) Receiving side
`
`Fig. 2 Orthogonal transformation , OTF.
`
`r-------- ------,
`I
`I
`I
`I
`
`(a) Coder
`
`FRAME
`DELAY
`
`(b) Decoder
`
`(2) T mode in- frame coding (refreshing or transient
`S/T) HO-DPCM (P(z)=z- 3)
`
`Fig. 3
`
`Interframe predictive coder and decoder
`block diagram.
`
`6. 4- 2
`
`APPLE EX. 1026
`Page 4
`
`

`
`(J) S mode in- frame coding ,
`Sub- Nyquist HO- DPCM (P(Z)=0.5z- 1+z- 2- 0 . 5z- J) .
`
`A nonlinear function NL is pl aced into the frame
`difference path, taking a key ro l e in improving the
`coding performances . The nonlinear function has an
`input-to-output relation as shown in Figure 4 . The
`transfer gain is less than unity for the small input
`amplitude and unity for the large amplitude .
`
`When the gain of the frame difference path is
`less than unity, the transfer function from the
`interframe predictive coder input to the quantizer
`has a recursive type lov pass characteristic a l ong
`the temporal axis. The lov pass filtering suppresses
`feedhacked quantization noise vhich causes unvanted
`significant picture elements even for still pictures .
`Random noise in the input signal is also suppressed
`through the temporal lov pass filtering . On the
`contrary, the temporal lov pass filtering gives rise
`to a bluring of the moving objects in the picture .
`
`The nonlinear characteristic as shown in Figure
`4 is useful because small amplitude noises on still
`part of picture are suppressed through the temporal
`lov pass filtering and large amplitude frame differ(cid:173)
`ences caused by motions are not almost affected .
`
`The nonlinear function causes a bluring of
`moving picture edges vith small brightness change .
`This effect is hardly perceived in NLl, but is
`occasionally seen in NL3.
`
`1-
`:::::>
`a...
`I(cid:173)
`~
`0
`
`INPUT
`
`Fig. 4
`
`Input-to-output characteristic of
`nonliniear function , NL .
`
`Sub-Nyquist encoding , S mode
`
`In the sub- NYquist encoding, the sampling time
`is chosen so that the sampled points in the picture
`are under an interleaving condition as shown by the
`circles in Figure 5 . At the receiving side , aliasing
`components removal is made by using a comb filtering
`as shown in Figure 5 . The samples designated by x
`are interpolated from the adjacent samples .
`
`6 . 4- 3
`
`~ ---- HP: -------
`
`~X~X--o--­
`y
`y
`
`INTERPOLATION COMB FILTER
`
`Fig. 5 Sub-Nyquist encoding , S mode .
`
`Variable length coding
`
`In the interframe predictive coding, the pre(cid:173)
`diction error is quantized and coded into 6 bits/
`sample code . The codes are transformed into reduced
`bit rate data through a variabl e length coder.
`
`The code vord length is from 1 through 12 . The
`zero amplitude representing insignificant pi cture
`element is given by the unit bit code, and significant
`pels are given by J - 12 bits code .
`
`Continuations of the insignificant picture
`element code are deleted by a block addressing. The
`picture e l ements are divided into unit block s with
`9 samples in T mode and 6 sampl es in S mode .
`If a ll
`the samples in the unit block are insignificant, the
`unit block is deleted . The positions of the deleted
`unit blocks are represented by block address codes .
`
`According to computer sjmulati on , a transmission
`bit rate corresponding to 2 bits/sampl e is possible,
`when the number of significant picture el ements is
`less than about 25 % of the vhole sampl es ,
`
`Adaptive mode control
`
`The quantizati on is made by a dead z one circuit
`followed by a quantization circuit . The dead zone
`threshold level is changed from 1 thr ough 3, and the
`quantum step size of the quantizer for small input
`is set to be 1,2 , 4 and 6 , The overall quantization
`mode is represented as QmDn in the folloving dis(cid:173)
`cussion , vhere m and n represent the quantum step
`size and the dead zone threshold level, respectively .
`The nonlinear functions of NLl, NL2 and NL3 are used ,
`
`The choise of the parameters is made depending
`on the buffer memory occupancy. As tbe buffer
`occupancy increases, the quantization parameters is
`changed from the fine to the coarse ones .
`
`APPLE EX. 1026
`Page 5
`
`

`
`When the buffer occupancy exceeds a predetermined
`level, S mode in-frame coding operates. When back to
`T mode, T mo~e in-frame coding is made for one frame
`interval to refresh the frame memory with T mode
`sampling rate .
`
`Upon overflowing, the frame memory data is
`frozen untill the end of the frame .
`
`ENCODING PERFORMANCES
`
`Effect of OTF and NL
`
`The eff~cts of the subcarrier phase inversion by
`OTF and the nonlinear function NL are measured ex(cid:173)
`perimentally. Figure 6 shows the number of signifi(cid:173)
`cant picture elements with and without OTF and NL
`as a function of the dead zone threshold level.
`In
`this measurements, picture is a still and colorful
`one, and the quantization step size is kept minimum,
`It is seen that the number of significant picture
`elements is halved by OTF and further reduced with NL.
`The combination of OTF and NL makes
`it possible to operate under the low dead zone
`without excess generation of significant picture
`elements .
`
`-*
`
`....J w a..
`~ z
`<l
`u
`~
`
`z
`
`<.:)
`
`WITH OTF
`
`(/)
`
`00
`
`5
`4
`3
`2
`DEAD ZONE THRESHOLD
`
`6
`
`Fig. 6 Effect of OTF and NL on the number of
`significant picture elements .
`
`Input random noises produce significant picture
`elements, increasing coded data generation rate . A
`measurement is made of the coded data rate as a
`function of signal- to-added noise ratio with and
`without NL function.
`
`In this experiment, the source signal is a color
`telPvision signal taken by a TV camera with SNR of
`about 48 dB unweighted. When the quantization is
`fixed to Q2D2, the bit rate increases by about 8 and
`14 Mbit/sec for the low SNR of 40 and 35 dB, re(cid:173)
`spectively. These increases are reduced to only 2
`and 5 Mbit/sec, respectively, by applying the non(cid:173)
`linear function ~~. This indicates that possible
`operating range is extended to below 40 dB input SNR,
`covering almost all signals practically encountered.
`
`Picture quality
`
`Subjectively evaluated signal-to-noise ratio
`(SNR) are shown in Figure 7. The coding parameters
`are fixed to QlDl, Q2Dl, Q2D2, Q4D2 and Q6DJ. The
`fine solid line indicate a theoretical SNR of PCM
`with the same quantum step size as Qn .
`
`60
`
`55
`
`OlD I
`
`Q2D\
`
`50
`
`45
`
`Q4D2.
`
`Q603
`
`Fig. 7 Subjective signal-to-noise ratio.
`measured with the coding parameters
`fixed.
`
`Statistical distribution of the operating modes
`is measured for variety of television programs. The
`encoded data generation rate is greatly dependent of
`not only the amount of motions but also picture
`character. This is becuase the use of OTF produces
`line prediction error into the frame difference
`signals. Picture containing many diagonal edges give
`rise to much informations . For such signals, pro(cid:173)
`bability of Q4 - Q6 mode occurence increases and SNR
`is decreased. Al hough it is very difficult to
`represent quantitatively the performance, at present,
`roughly speaking, average SNR lies in the range of 50
`io 54 dB weighted at around 22 Mbit/s .
`
`It is confirmed that there occurs no decrease in
`spacial resolution and color fidelity is very good.
`
`CODEC HARDWARE
`
`Figure 8 shows a photograph of the coder and
`decoder of NETEC-22H system. The major specifications
`are listed in Table 1. Each bay measures 2000HX500W
`x250D, A half of the bay is occupied by the frame
`memory and buffer memory. These memories are composed
`of 4 kbi ts NOS RAM, ppD411. When a 16 kbi ts RAM or
`64 kbits CCD memory element will be available in
`the near future, the size and cost of the memories
`will be greatly reduced,
`
`6.4-4
`
`APPLE EX. 1026
`Page 6
`
`

`
`CONCLUSION
`
`REFERENCES
`
`A composite interframe coding of NTSC color
`television signals has been developed amiog at a high
`quality transmission with a bit rate of 16-32 Mbit/sec .
`The codec is designed to preserve strictly the wave(cid:173)
`form of the composite video signal including VITS,
`providing high ~olor fidelity no decrease in reso(cid:173)
`lution and wideband (5 MHz) capability.
`The use of an orthogonal transformation for the
`sub-carrier phase inversion and a nonlinear function
`in the predictive coding loop makes it possible to
`realize the high SNR in the composite coding .
`
`ACKNOWLEDGMENT
`
`The authors would like to thank Hisashi Kaneko
`and Y. Kato for their helpful discussion and encour(cid:173)
`agement. The authors also wish to thank A. Hirano for
`computer simulation programming and Haruo Kaneko and
`M. Nishiwaki for the hardware development support.
`
`(1) B. G. Haskell, F . V. Mounts, and J. C, Candy,
`"Interframe coding of videotelephone pictures,"
`Proc. IEEE, vol . 60, pp. 792-800, July 1972.
`
`(2) K. Iinuma, Y. Iijima, T. Ishiguro, H. Kaneko, and
`S. Shigaki , "Interframe coding for 4 MHz color
`television signals," Interframe coding for 4 MHz
`color television signals," IEEE Trans , Communica(cid:173)
`tions, vol. COM- 23, pp . 1461-1466, December 1975.
`
`(3) T, Ishiguro, K. Iinuma, Y. Iijima, T. Koga, and
`H Kaneko, "NETEC system: Interframe encoder for
`NTSC color television signals," The Third Inter(cid:173)
`national Conference on Digital Satellite Communi-
`cations Record, pp.
`, November 1975.
`
`(4) B. G. Haskell, P. L. Gordon, R. L. Schmit, and
`J . V , Scattaglia, "Frame-to-frame coding of 525
`line monochrome television," 1975 IEEE National
`Telecommunications Converence Record, pp. 22-23
`22-25, December, 1975.
`
`. -----
`~ ----
`.!-::-.= ==
`I
`1
`
`1
`
`-
`
`I
`I
`
`a
`
`Fig. 8 NETEC-22H coder and decoder .
`
`Table 1
`
`NETEC-22H codec specifications
`
`Video signal input/output
`NTSC color television signal
`1 vpp/75 n unbalanced
`Sound signal input/output
`~lono/stereo sound program
`12 dBm maximum/600 n balanced
`Video A/D and D/A conversion
`Signal bandwidth: 5 MHz
`Sampling frequency: 10.76/14.35 MHz
`Number of bits: 8(A/D), 9(D/A)
`Sound A/D and D/A conversion
`Signal bandwidth: 15 kllz
`Sampling frequency: 31. 5 kllz
`11 bits nonlinear coding
`Error correcting codec
`239/255 double error correcting BCH code
`
`Transmission bit rate
`16 - 32 Mbit/s continuously variable
`Power supply
`-4~ V DC , about 600 watts each .
`
`6.4-5
`
`APPLE EX. 1026
`Page 7