`
`By a. s. BUSBY, JR.
`
`Theprinciplesofslgnaldlgltlaafionsa they relatetodigitalteler'nionarepraeutedia
`such a tray as to lay groundwork for htturetllgltnl television papers. To clarify thestnpa
`tht are involved, a hypothetical “manual” conversion of an analog signal to Its digital
`equivalent hexamined. Forsamplhtgcohrtelevhionmthebulsaupflumeis
`cos-idem! to be exactly three dmesthe colors-humor frequency. Various ltethnlaof
`digitization are “Mind those suitable for widehnd alga-b undefined. Quanti-
`uflmmrandtheprohlemsotmlaeandnolsemmmtmdiscmethegmt
`advantage of handling manor. signals digitally isthatdlgitll quantifies are capableof
`being sent. received, switched. stored. recordul and delayed virtually without distortion.
`Whatever ball effectsaccruefromconversiontoandironthedigltalmodeaeedaccrue
`onlyonce.Whateterillacanhefefladlgitnlsignahshofloftotalloathcycanheude
`invisible and they do not accumulate.
`Introduction
`
`carrier.
`
`Actual Real-Time Signal Digitisation
`Now let us speed things up 60 million
`times and get back to “real” time. A real
`television signal must be sampled at me-
`gahertz rates. The voltmeter must settle
`in tens of nanoseconds. The records are
`stored in a high-speed electrical memory.
`The same sequence of sample. wait and
`record still applies.
`The rate that we repeat the sampling.
`measuring and recording is important.
`Harry Nyquist showed in 1928 that, as-
`suming a random signal. the original can
`be reconstructed only if the sampling
`rate is at least twice the bandwidth of the
`sampled signal.
`in the spectrum of a
`sampled signal. with the sampling rate
`normalized at two. there is a strong com-
`ponent at the sampling frequency. and
`like an AM carrier,
`it has sidebands
`above and below it (Fig. 3). For an input
`frequency equal
`to half the sampling
`rate, the input and the lower sidebend
`caused by it coincide and cannot be sepa-
`rated. lf the ratio of sample rate to sam-
`pled input is made greater than two, a
`low-pass filter can be made to reject the
`sample frequency and its sidebarlds. The
`higher the ratio. the easier the filter de-
`sign becomes.
`In a color television signal. there is a
`strong component around the color sub-
`carrier frequency. An argument can be
`made for sampling at an exact multiple.
`greater than two. of that frequency. Fig-
`ure 4 shows the sampling rate at three
`times submrrier. in the case shown, any-
`thing that might go wrong is most apt to
`do so once each subcarrier cycle. or twice
`or three times. If twice or three times,
`sit-m
`Into
`
`Digital signals are being used at this
`time to transmit many thousands of still
`pictures to ground stations from ERTS
`(the Earth Resources Technology Satel-
`lite). In Europe. experiments are being
`conducted by EBU members on digital
`television transmission from satellites.
`Digital time base correction is already a
`reality, and digital television signal pro-
`cessing equipment and digital videotape
`recorders cannot be long in coming.
`With so much worldwide interest in sig-
`nal digitization and reconstruction.
`the
`engineer who wants to stay up to date in
`the television field must understand the
`principles of digitization and have some
`sense of the probable impact of digitiza-
`tion on the television industry. The pur-
`pose of this paper is to discuss digiti-
`tion principles in such a way that elec-
`tronics and television engineers will gain
`a clear if not rigorous understanding that
`will enable them to more thoroughly
`profit from future digital television pa-
`pets.
`
`Doing Signal Digitization “Manually”
`To dramatize the proces of signal di-
`gitisation and reconstruction. we might
`consider how the job could be done
`“manually” if every step were slowed
`down 60 million times. On this scale, a
`microsecond becomes a minute and a
`wristwatch ticks every five months.
`Under such conditions. we could ana-
`lyze a signal picked up by an antenna by
`applying the signal to a digital voltmeter
`(let us say a simple three-digit one with
`high impedance) through a momentary
`pushbutton switch. with a capacimr
`across the input (Fig. I). When the but-
`ton is pushed. the capacitor charges to a
`voltage equal
`to the signal; between
`pushes. the voltage on the capacitor is
`stable and the meter has time to settle
`down to an accurate reading. It is tedi-
`ous and inefficient, but we could take a
`reading every 5.6 seconds: push the but-
`ton. read the meter. write down the volt-
`age — again and again.
`m 15 January [975 at the Society‘s Tech-
`nical Conference as“ Francisco. by E. Stanley
`Busby. Jr.. Ampet Corp. so: Broadway, Redwood
`City, CA gross. (This paper was first received on 15
`April 1915 and in final form on IS May 1975.)
`542
`
`struction of the original samples. but
`limited by the accuracy of measurement.
`Drawing a curve among the points is
`equivalent to putting the output through
`a low-pass filter. This curve is as close as
`we can get to the original signal.
`
`"Itt ' 'm- "or
`
`.I
`
`H
`
`II
`
`I
`
`a
`
`ml
`
`2.!
`
`I
`
`M
`
`u
`
`“3.1Spectrumotuampledslgnnl.
`
`Finally we could take some graph
`paper and plot each voltage reading as a
`point. Then we would draw a smooth
`curve among the points so that half the
`points are above the curve and half
`below it. The result might be a curve
`such as Fig. 2.
`let us give names to
`At this stage,
`what we have done. The switch and ca-
`pacitor performed the function of sam-
`pling. essentially melting the voltage
`seen during a quick look hold still long
`enough to be measured accurately. The
`duration of the button push determines
`whether we get a sharp or blurred "pic-
`ture" of that piece of the waveform: the.
`button-pushing rate determines whether
`we miss capturing any posits or valleys of
`the waveform.
`The voltmeter performed the task of
`analog-to-dfgiml (A/D) conversion. Al-
`though the input is analog (in that it can
`have any value over
`the operating
`range), the output is a number that, for a
`three-digit voltmeter. can have only a
`thousand values (0-999). [f the output
`reads 0.785 V. the input might have been
`anywhere between 0.7845 and 0.7855 V.
`It is in this step that the system accuracy
`or resolution is forever determined.
`The written record is a memory or
`stare. The numbers could as well have
`been stored in flip-flops. magnetic cores.
`or holes in punched cards or paper tape.
`Plotting points on the graph paper corre-
`sponds to making an digital -to-ano!og
`{Bl/A) comrsr‘an. The result is a recon-
`SAMPLING
`WITDH
`
`“s
`
`Hull“!
`Bflfll'flll
`
`I
`
`'r
`
`‘I
`
`Fig. I. Essential elementsofsaA/Doomert—
`8!.
`
`Fig. 2. Reconstructed analog waveform.
`
`July 1975
`
`Journal of the SMFI‘E Volume 84
`
`E
`
`I!
`
`II
`
`I.l
`
`ll.
`
`III
`
`fig.4.5supun¢uttmnmoemiut-
`
`PMC Exhibit 2030
`Apple v. PMC
`IPR2016-00755
`Page 1
`
`
`
`Fig.5.Princlpleofstmpleth-ee-
`deeadeD/Aeonverm.
`
`I'll IE “(HUHED
`
`I
`
`corona!
`
`Iestatesa
`
`""
`
`In}
`
`catartn man m
`_|_
`
`I
`
`’ I fill-Ill! lll'l
`“call! A
`
`( 3 IINIRV
`g DIGI‘IS
`
`543
`
`exception. A marginal television signal
`or poor SNR might be adequately pro-
`cessed in a six-digit system because fur-
`ther resolution would only serve to better
`define the noise and would be a waste. A
`good quiet signal is adequately handled
`by an eight-bit system.
`Statistically, when the original signal
`is sampled some samples will land direct-
`ly on the boundary between one incre-
`ment of measurement and the next. Such
`a borderline case can and will go either
`way . .
`. a half-increment too high or a
`half-increment
`too low. This random
`choice is measurable as if it were noise.
`It is sometimes called quantizing noise
`but is more properly called Quantizing
`error because it need not appear at all
`unless there is a signal present to cause
`it. Its magnitude is proportional to the
`size of the minimum increment. In a bi-
`nary system. adding one more digit, or
`hit. halves the minimum increment and
`reduces the noise measured by six deci-
`bels.
`Assuming a perfectly quiet input sig-
`nal of a random nature. there is a rather
`simple expression for the approximate
`apparent SNR due to digitization. It is:
`peak-to-pealt signalfRMS noise = -[11
`+ (6 x No. of bits)], where the p-p sig-
`nal includes 28.6% sync. In practice. the
`assumptions have to be adjusted because
`television signals are not quiet but noisy
`and they certainly are not random. Be-
`cause much of the energy of a color tele-
`vision signal is at a high frequency. most
`of the apparent noise added by digitizing
`lies outside the video passband. In care-
`fully designed systems.
`the apparent
`noise contribution is 8 to 10 dB less than
`the formula would predict.
`Importance of the A/D Conversion
`The most critical part of a digital
`video system is the A/D converter. in-
`
`Ia
`
`} I
`
`IIn} "I—
`
`Fig. 'l. Counter-type AID converter.
`
`Fig.9.0ue-looltAIDemerter.
`
`cluding the sampler used with it -just
`as in Fig. I. The source impedance of the
`signal must be low enough and the size
`of the storage capacitor small enough
`that when the switch is momentarily
`closed.
`the capacitor can charge to a
`voltage insignificantly different from the
`input. This will require a time several
`times longer than the product of the re-
`sistance and the capacitance (the RC
`time constant). At the same time. the ca-
`pacitor must be large enough that the
`current required to operate the voltmeter
`during the measurement period doesn‘t
`significantly discharge the capacitor.
`01' the many sorts of A/D converters.
`I will describe three basic ones and one
`which is a combination of two of these.
`The first sort employs a counter which
`provides numbers to a digital-to-analog
`converter to produce a voltage which is
`compared to the input to be measured
`(Fig. 1). At the beginning. the counter is
`set to zero. It is then allowed to count
`until the output of the D/A converter ex-
`ceeds the input. whereupon the count is
`stopped and the number is read from the
`counter. This method is simple. but is
`much too slow for our purposes. An
`eight-bit system might have to progress
`through 255 counts to take a measure-
`ment.
`The method shown in Fig. 8 is similar
`in that a DIA converter is used. but
`dissimilar in that there is individual con-
`trol over each of the binary digits. At the
`start, the most significant bit, responsi-
`ble for a half-scale voltage. is. set active.
`and the resulting voltage subtracted
`from the input. lithe input equals or ex-
`
`Busby: Prllelplea of Digital rv Simplified
`
`I ll LI. Ill
`
`1 HILLIMP—.
`
`H'1“
`".150
`m Sill
`Fig. EI‘brce-hlt binary D/A convenertrlthl
`partinSreaolutloa.
`
`the filter will remove the disturbance. if
`once. at least the disturbance is dot-in-
`terlaced and therefore of minimum visi-
`bility on the screen.
`The color subcarrier in the American
`NTSC system is quadrature modulated
`in both amplitude and phase. A short
`burst of the subcarrier at a reference
`phase and amplitude is transmitted dur-
`ing horizontal retrace for use in the re-
`ceiver to recreate a carrier for use in de-
`modulation. A sampling rate which
`bears a fixed relationship to the refer-
`ence burst can serve as a continuous ref-
`erence signal not otherwise available.
`This can be useful in time-base correc-
`tors used on tape playbaeks.
`In many cases. the signal processing is
`done digitally and the final numbers are
`converted back to a signal used to devel-
`op a display. This requires digital—to-
`analog (BIA) conversion. and if we were
`doing the job “manually” as we did ear-
`lier.
`three lO-position rotary switches
`connected in series could constitute a
`
`hand-operated D[A converter capable of
`handling three digits (Fig. 5). Numbers
`in, voltage out.
`If we were doing the conversion elec-
`tronically — say with integrated cir-
`cuits —-it would be more convenient to
`use binary digits (hits) instead of deci-
`mal digits. This can be done by using the
`integrated-circuit-equivalent of a single-
`pole double-throw switch to represent
`each binary digit. Since such a switch
`has only two possible states. one state
`can be used to represent a binary “one”
`and the other a binary “zero.” A two-
`switch. two-bit D/A converter. then. is
`capable of four levels of output. Adding
`another switch provides eight levels of
`output. doubling the resolution (Fig. 6).
`Four switches yield 16 levels, five switch-
`es 32. six switches 64. and so on. Thus, in
`a typical eight-bit system. the voltmeter
`reading can be represented by eight bi-
`nary digits. or bits. to an accuracy of one
`part in 256.
`When any kind of signs] is to be pro-
`cessed,
`the problem of SNR must .be
`considered — and television camera sig-
`nals._being accompanied by noise:are no
`
`PMC Exhibit 2030
`Apple v. PMC
`IPR2016-00755
`Page 2
`
`
`
`I-‘Ig. III. Two-look, successive Ipp'mtimlliol AID converter.
`
`I5
`amt
`II'IIIS
`JLISI III
`Kiflfii
`
`-l WLT
`
`Fig. 12. Improved D/n comerter with ladder
`attenuator.
`
`tr:
`
`1M
`
`us
`
`rm In:
`
`Fig. 11. Simple D/A converter with switches
`and summing operational amplifier.
`
`Fig. 13. Nonideal DfA output.
`
`insignificant compared to a normal sig-
`
`carrier. Each eight-bit sub-group of the
`word is called a byte.
`Eventually. we will want to recall our
`numbers. assemble their hits into the
`original eight-hit byte and present them
`to a DfA cunverter to reconstruct an an-
`alog signal. In the first of two popular
`D/A methods. switches (electronic, not
`electromechanical), each one operated
`by its associated binary digit, connect
`appropriate resistors across a voltage
`supply. producing a current proportional
`to the size of the number (Fig. 11). The
`amplifier shown is one way of converting
`the current to a voltage. Unfortunately.
`the resistor values cover a wide range,
`making accuracy difficult
`to achieve.
`The other method (Fig. 12) uses identi-
`cal switched current sources. each con-
`nected to its own tap on a ladder attenu-
`ator. Each stage of the attenuator divides
`by two. The least significant digit suffers
`the most attenuation, the next least sig-
`nificant half that much. and so on. The
`range of resistor values in the ladder is
`small, varying only over a two to one
`range.
`At the D/A stage in a system. we
`begin to see some of the plagues of the
`analog world. D/‘A converters are beset
`with switching transients. ringing, over-
`shoots. crosstalk, etc. Figure 13 shows
`how a D/A output might vary from the
`ideal.
`While an output filter would remove
`most of the fast irregularities. some of
`the discrepancies can show in the filtered
`output. One method of avoiding these is
`called "re-sampling." As shown in Fig.
`l4.
`the D/A output is examined by a
`sampler after it has settled down and just
`before a new number is presented to it.
`The output of the sampler is then lil-
`tered. The filtered result of a sampled or
`
`ceeds one-half scale, the bit is allowed to
`remain active. If the input is less than
`one-half scale the bit is set inactive. The
`next most significant bit. respOnsible fer
`one-quarter scale. is then set high and
`the output of the subtractor determines
`whether this bit should be allowed to re-
`main active. After the least significant
`bit is tested. the number is read from the
`memories. This method is faster. requir-
`ing only eight “looks” for an eight-bit
`system.
`The third basic method (Fig. 9) cm-
`ploys a separate voltage comparator for
`each possible voltage increment. The
`number is derived by logic. It is very fast
`because the number is obtained with
`only one "look." For an eight-bit system,
`however. 255 comparators would be re-
`quired. This method has been used for
`digitizing video signals. but is cumber-
`SOME.
`
`The fourth method (Fig. ID) is a com-
`promise bettveen the second and third
`basic methods. A lo-level one-look type
`of A] D converter is employed. yielding a
`four hit number. These four hits are put
`into temporary store (memoryland will
`become the most significant four hits of
`the final number. In the next step. the
`four bits drive a D/A converter whose
`output
`is subtracted from the input,
`yielding a difference or error. This error
`is amplified 16 times and again exam-
`ined by a similar one-look A/D convert-
`er. yielding the least significant four hits.
`This method is used in a popular digital
`time-base corrector for videotape play-
`backs.
`
`Handling the Numeric Information
`The amount of numeric information
`generated by digitizing a television sig-
`nal is large. For example. an eight-bit
`system. sampling at three times subcar-
`rier. generates about 86 million hits (86
`Mb) per second. Moving these numbers
`from one place to another can be done
`bit-by-bit on one wire, much as a tele-
`type signal is sent. Eight wires may be
`used. one for each bit position. at 10.?
`bes on each wire. Or. 24 wires could
`be used. in three groups or eight. at 3.58
`Mb/s on each wire. In computer par-
`lance. the 24-bit group would be called a
`word, each word representing all
`the
`measurements taken on one cycle of sub-
`544
`
`m outru‘l
`
`I
`
`ll
`IMEI’IK
`fl. is nuts
`Fig. 14. Princlpteof resounding.
`
`reconstructed signal exhibits some loss of
`response at high frequencies that is of
`the same form as the loss caused by the
`finite width of the optical soundtrack
`scanning slit of a sound projector or the
`finite gap length of a magnetic tape re-
`producer. The response for any given set
`of sampled and sampling frequencies is
`given by sin its/tn. where x in radians is
`the frequency being sampled divided by
`the sampling frequency. In realistic sys-
`tems. the loss at the highest frequency is
`only a few decibels and is easily compen-
`sated.
`
`Evaluating Digital Televisiol Systems
`Some of today's system measurement
`practices are ill adapted to digital televi-
`sion systems. The practice of measuring
`noise by shorting out the input and look-
`ing at the output does not reveal how the
`system will behave in the presence of a
`real signal with real noise on it. Some
`test signals are of small amplitude and
`hence are defined by only a small num-
`ber of digitization levels. The errors en-
`countered are large compared to the test
`signal and can be interpreted as large
`system errors. when in fact the errors are
`
`July 1915
`
`Journal of the SMP'I'E Volume 84
`
`PMC Exhibit 2030
`Apple v. PMC
`IPR2016-00755
`Page 3
`
`
`
`Television image enhancement has long been accomplished by using analog circuits to
`operate on the analog television signal and thereby emphasise both horizontal and verti-
`cal transitions. Now it is feasible to perform comparable image enhancement by digital
`techniques operating in real time on a pulse-code-modulated NTSC television signal.
`Several algorithms have been developed to generate vertical and horizontal video details
`from PCM NTSC teletlsion signals encoded at a loft-MHZ rate (three times the color
`subearrier frequency) and at a H.3—Ml-le rate [four time the subcarrier frequency).
`Practical design concepts are presented to implement digital image enhancers. in addi-
`tion to ensuring the stability. accuracy and reliability that are the usual advantages of
`digital systems, digital image enhancensmt will enable video processing techniques to be
`used that can actually be performed best by digital means. in particular. more nearly op-
`timum coring or “crispenlng” circuits will be possible and the detail signals will be easin
`measurable, which may permit automatic adaptive enhancement.
`1. INTRODUCTION
`2. ENHANCEMENT WITH
`DIGITAL COMB FILTERING
`Image enhancement of television sig-
`nals by digital methods usually involves
`comb filtering to generate vertical details
`and also to remove chrominanee compo-
`nents before generating horizontal de-
`tails. A typical comb filter for NTSC
`color television signals combines three
`consecutive (all odd or all even) televi-
`sion lines which we designate T. M and
`B—for top. middle and bottom—in the
`following
`proportions
`to obtain the
`chrominance (C) and luminance (Y) sig-
`nals:
`
`5‘5
`
`ill? MHZ}. the digital samples in adja-
`cent
`lines are vertically misaligned, as
`shown in Fig. 2. Here codeword BM in
`the middle line is not in vertical align-
`ment with codewords liq- of the top line
`and B; on the bottom line. Comb filter-
`ing under these conditions requires the
`interpolation of codewords 31-. CT. 33.
`and Ca before matrising them with
`codeword BM of the middle line. This in-
`terpolation modifies the typical comb fil-
`ter response in an undesirable manner.
`Another method of encoding the PCM
`NTSC signal at a 10.7-Ml-lz rate is with
`phase alternating line encoding (PALE).
`it involves reversing the encoding phase
`on alternate lines to vertically align the
`codewords as shown in Fig. 3. The PALE
`
`10!:
`AM"
`
`Mil‘DLE
`LINF
`
`ant TOM
`LINE
`
`A
`
`3
`
`H
`
`JMNSI‘"
`Fig. 1. Digital samples at lea-Mil: rate from
`three adjacent scanning lines.
`
`TOP
`LINE
`
`MIDDLE
`LlNE
`
`BOT TOM
`LINE
`
`n
`
`3
`
`3
`
`3
`
`46.5
`i
`;e— NS
`—e|95 NS L—
`Fig. 1. Digital samples at loft-MHz rate from
`three adjacent scanning lines.
`
`mp
`LINE
`
`MIDDLE
`l F“!
`
`Will YUM
`UM
`
`—o———e-——o—9—n—+
`All
`Hit
`1 a
`
`"LT-MHZ rate
`Fig. 3. Digital samples at
`PALE from three adiacent scanning lines.
`
`an] amplitude and are invisible on the
`picture tube. Test signals having no im-
`portant component less than about 30%
`of the available signal swing usually pro-
`duce measurements free of misinterpre-
`tation. Unlike typical analog systems.
`digital systems are at their best when
`handling full-amplitude signals.
`At
`the cost of bandwidth and some
`complexity. is digital video system offers
`
`in trade some substantial values: whatev-
`er is lost in the system need be lost only
`once . .. in the initial digitization. After
`that it is all numbers. A binary bit
`is
`sturdy. You can't distort
`it. You may
`change it completely. or even lose it. but
`not distort it. You do not adjust a num-
`'-ber with a screwdriver.
`It is implicit in what we have said that
`the numbers representing a television
`
`signal can proceed through an unlimited
`number of manipulations without being
`degraded each time. The computer peo-
`ple have long known that a computer
`cannot do much with faulty input data;
`they express
`this with the acronym
`“GIGO” standing for “Garbage in. Gar-
`bage Out." The converse of this. of
`course. is “Clean Input. Clean Output"
`and this is the promise of “Doing it by
`the numbers."
`
`Digital Television Image Enhancement
`
`By JOHN P. ROSSI
`
`Television image enhancement has
`long been recognized as a means to
`subjectiver improve television pictures.
`Horizontal enhancement is accomplished
`by boosting the high frequencies with
`phase correction for symmetrical over-
`shoots. Vertical enhancement is obtained
`by the comparison of adjacent lines. low-
`pass filtering the difference signal and
`adding the resultant to the main signal.
`At present. this is done with analog cir-
`cuits operating on the analog television
`signal. Recently pictures have been en-
`hanced with digital techniques by com-
`puters. but this method is too slow for
`application to real-time television sig-
`nais.
`integrated
`in digital
`Advancements
`circuit technology new make it feasible
`to design a. digital image enhancer capa-
`ble of operating in real time on a PCM
`NTSC television signal. A digital image
`enhancer costs more than an equivalent
`analog image enhancer but offers better
`accuracy. stability and reliability. A dig-
`ital enhancer also can provide selectable
`frequency peaking for horizontal details
`and more nearly optimum coring or
`“crispening.” The present work analyzes
`digital image enhancement of NTSC sig-
`nals encoded by pulse code modulation
`at three times the color subcarrier (10.1
`MHz). at four times the color subcarrier
`(14.3 MHz] and at three times the color
`subcarrier with phase alternating line en-
`coding (PALEM
`Preceded. by title only. on it November 19‘“. at
`the Society's Technical Conference in Toronto: this
`paper first submitted on 30 September 1914. and in
`final form on 2| March 1975 by John P. Rossi. CBS
`Technology Center. High Ridge Rd. Stamford. CT
`06905.
`
`C-M — .l,(T +3)
`
`Y-M+§(T+ B)
`A closer scrutiny of this comb filter rc-
`veals that the technique is equivalent to
`sampling and averaging. with certain
`weighting coefficients. three picture ele-
`ments from three sequential
`lines. This
`sampling and averaging is repeated for
`all picture elements. A picture element is
`an infinitesimal
`image sample that
`is
`equivalent mathematically to the Dirac
`delta function 6. Thus. a pulse-code-
`modulatcd
`(PCM)
`television
`signal
`would seem ideal for comb filtering be-
`cause each digital codeword describes
`the instantaneous amplitude of the ana-
`log signal at
`the encode time. The re-
`quired IH separation between each of
`the three video samples is met when the
`television signal
`is encoded at a 14.3-
`MHz rate, or any even multiple of the
`color subcarrier. This is shown in Fig. 1.
`Whenever the signal is encnded at odd
`multiples of the color subcarrier (e.g.
`
`lineal: Digital TV Image Enhancement
`
`PMC Exhibit 2030
`Apple v. PMC
`IPR2016-00755
`Page 4
`
`