Copyright © 1997, 1998 Roger Jennings, all rights reserved. Written May 20, 1997; updated September 1, 1998. This document may be copied and/or distributed for non-commercial purposes only. Please send comments and corrections to the author at 70233.2161@compuserve.com.

Sony Corp.'s Betacam SP format is the standard of comparison of video tape recording formats. According to Sony Europe, more than 350,000 Betacam SP devices had been sold world-wide as of early 1997. The majority of broadcast electronic news gathering (ENG) operations currently use Betacam (oxide) or Betacam SP camcorders and VTRs. Virtually all broadcast stations require (or at least strongly prefer) Betacam SP source footage. Most clients of professional video production firms specify Betacam SP for industrial shoots and are likely to require videographers to use Sony or Ikegami camcorders. Although the M-II format from Panasonic Broadcast and Digital (formerly Television) Systems Company (PB&DSC) offers about the same performance as Betacam SP, Sony and Betacam SP are untouchable when it comes to brand recognition and status. As a result, all other video recording formats are ranked as "not up to Beta SP," "equal to Betacam SP," or "better than Beta SP." These comparisons, based on the beholders' perception of image quality, are reminiscent of the meaningless "broadcast quality" and "studio quality" bullet points on advertisements for consumer and low-end prosumer video gear.

The advent of the Digital Video (DV) format has ignited a controversy among current and prospective users of DV gear. Initially, arguments appeared regarding the "legality" of broadcasting NTSC DV's 480 instead of the "standard" 483 active lines of video. Reports circulated that broadcast station engineering personnel had counted active video lines and rejected 480-line content. Obviously, if broadcasting less than 483 active lines was illegal, all U.S. TV stations transmitting letterboxed movies would have by now lost their licenses. The subsequent controversy, DV's 4:1:1 vs. ITU-R BT.601-4 (formerly CCIR-601) 4:2:2 sampling, has generated thousands of messages in on-line forums, newsgroups, and listservers. This paper represents an attempt to dispel the rumor and innuendo surrounding the 4:1:1 versus 4:2:2 issue, especially as it relates to the "Is DV better (or worse) than Betacam SP?" controversy, and DV compression artifacts.

Horizontal Resolution and Bandwidth of Analog Recording Formats

Traditionally, resolution of analog video cameras, recording formats, and monitors has been expressed as "TV lines" or "horizontal lines." Originally, TV lines were determined by the ability to clearly distinguish sets of vertical lines in a standard TV test pattern. Subsequently, horizontal resolution became related to the bandwidth of the recording format. The generally-accepted "rule of thumb" is 80 lines per MHz of luminance bandwidth for recording formats. The following table lists the lines of horizontal resolution of various recording formats, based on Sony and other manufacturers' data sheets, with the NTSC broadcast signal added for comparison. Bandwidth is implied from the "rule of thumb."

Commencing with introduction of the Betacam format, Sony and its competitors ceased publishing horizontal resolution values for professional and broadcast recording formats. The reason for doing so is clear: Using 80 lines/MHz would have positioned oxide Betacam, with it's 4.0 MHz luminance bandwidth, at 320 lines (inferior to U-matic SP.) Even worse, metal-particle Betacam SP would fall below consumer-grade S-VHS and Hi8 TV-line ratings. Both oxide Betacam and Betacam SP produce image quality superior to S-VHS and Hi8 formats, primarily because of Betacam's separate recording tracks for luminance (Y) and chroma (Cr and Cb), rather than the color-under modulation method of S-VHS and Hi8. Thus the two licensees of Thomson's original Betacam technology, Sony and Philips BTS, and their erstwhile competitors (Panasonic and JVC with M-II) began using bandwidth and signal-to-noise ratio (SNR) specifications for the luma and chroma components, instead of specifying TV-lines. The following table lists the published values for luma and chroma bandwidth, plus SNR (in dB) for several Betacam and related VTRs.

Video recording bandwidth is determined by the modulation method, tape particle size, record/reproduce head design, internal electronic circuitry, and a variety of other factors. Maximum luma bandwidth is reported by the manufacturers at the point at which the signal falls to 63 percent of its original value (-4.0 dB). The high frequency content of the image, such as very small objects with sharply-defined outlines (leaves of trees in backgrounds, blades of grass, and the like) is attenuated in the recording process. The visible effect of attenuation is a combination of lower brightness levels and blurring of tiny objects. Maximum chroma bandwidth is reported at the 63 percent or 71 percent (-3.0 dB) loss point. The human eye is less sensitive to variations in color than in brightness, so the reduced chroma bandwidth gives acceptable perceived image quality. Betacam SP and Panasonic M-II represent the upper limit of quality for commercial analog video recording technology. These formats sustain several reproduce-record generations without obvious image degradation.

Signal-to-noise ratio is a small-signal effect; thus SNR is important for low-light scenes shot without camera gain boost. Small differences in SNR dB ratings are deceptive-in linear terms, 51 dB (355:1) is about 1.3 times better than 49 dB (281:1). The luma SNR of professional Hi8 VTRs is > 45 dB (Sony EVO-9720 and EVO-9850) and most pro S-VHS decks claim > 46 dB. Noise levels are cumulative in dubbing, so the better SNR of Betacam SP contributes to its multi-generation capability.

Digital Sampling, Analog Bandwidth, and SNR

In the uncompressed digital video realm, sampling rate is conceptually analogous to bandwidth. The standard ITU-R BT.601-4 (D-1, 4:2:2) luma sampling rate for NTSC and PAL video is 13.5 MHz and chroma sampling occurs at half the luma rate or 6.75 MHz. There are two chroma signals (Cr and Cb), so the resulting digital bitstream is composed of equal quantities of luma and chroma data. The Nyquist theorem states that the sampling frequency must be at least twice the highest frequency component of the sampled analog signal. Thus the theoretical maximum luma bandwidth of D-1 is 6.75 MHz and the chroma bandwidth of the two components is 3.375 MHz each. Low-pass filtering is required to prevent aliasing in the digitized signals, so the resulting bandwidths are less than the Nyquist limit. Digital filters using high-performance DSPs (digital signal processors) can provide an analog-equivalent maximum luma bandwidth of about 6.0 MHz, +0.5/- 3.0 dB and a maximum chroma bandwidth in the range of 3.0 MHz, +0.5/-3.0 dB.

DV samples luma at D-1's 13.5 MHz and chroma at one-half the D-1 rate or 3.375 MHz. DV-NTSC uses 4:1:1 sampling in which the chroma is sampled once for every four horizontal luminance samples. DV-PAL uses 4:2:0 sampling, which uses the average value of the chroma signal between successive horizontal samples and scan lines. According to the Japanese Digital Video Consortium, 4:2:0 sampling gives better perceived image quality with PAL's increased number of vertical scan lines. (Panasonic's DVCPRO components use 4:1:1 sampling for both NTSC and PAL.) The ATSC MPEG-2 formats for DTV, including HDTV, also use 4:2:0 sampling. Figure 1 compares 4:2:2, 4:1:1, and 4:2:0 chroma sampling.

The bandwidth values listed in the above table compare more than favorably with mid-range Betacam SP and M-II VTRs, and luma bandwidth exceeds that of the popular UVW-1800 Betacam SP editing recorder by a substantial margin. Reversing the bandwidth analogy, Betacam SP might be considered the equivalent of a 3:1:1 digital format. (4.5/5.8 MHz * 4 = 3.10).

Theoretically, the SNR of an ideal n-bit analog-to-digital or digital-to-analog conversion process is given by the formula: SNR = 6.02n + 1.76 dB (Watkinson, The Art of Digital Video, 2nd Edition, p. 138, Focal Press, 1994.) For DV, which records 8-bit samples, the theoretical maximum SNR is 49.92 dB, about the same as for Betacam SP's luma SNR. Simple analog-to-digital conversion, however, results in unacceptable contouring. Contouring causes distinct bands of brightness in images when recording a uniformly-reflective surface with slightly-varying lighting intensity or insufficient lens coverage. Adding Gaussian dither to eliminate contouring results in the formula: SNR = 6.02(n - 1) + 1.76 dB (ibid., p. 147), which gives an 8-bit SNR of 43.9 dB. The assumption is that oversampling (or other magic) on both the analog input (ADC) and output (DAC) is responsible for the 55 dB SNR quoted by Sony and Panasonic for their YCrCb -> DV -> to YCrCb tests. The more important SNR value, quoted only by Panasonic, is for DV -> YCrCb, because recording to an analog component format (usually Betacam SP) is necessary until DV, DVCPRO, or DVCAM gear becomes commonplace in broadcast stations and postproduction facilities.

SNR is not applicable in digital dubs between any of the three formats, assuming the digital data format doesn't change during the process. Using DV-over-IEEE-1394 (FireWire) to make dubs creates a clone of the original DV footage. (The audio data, however, may be bumped from 12-bit to 16-bit or from unlocked to locked in the process.) Similarly, transferring DV footage to a fixed disk using an IEEE-1394 adapter card and then recording back to DV results in a clone of the original content. No decompression or recompression of the DV data is involved in the tape -> disk -> tape process. Neither is SNR a factor in adding transitions, titles, or graphics. Conversion of 24-bit RGB graphics to the DV format, and vice versa, is a mathematical manipulation with a predefined and predictable outcome.

4:1:1 and Chroma Key Uncertainty

Assuming one accepts the preceding argument regarding the equivalent chroma bandwidth of 4:1:1 DV and Betacam SP, the issue of spatial uncertainty of chroma-keyed graphics remains. Casual comparison of the 4:1:1 and 4:2:2 diagrams of figure 1 would lead one to conclude that the horizontal uncertainty of chroma-keyed RGB graphics underlaid on 4:1:1 YCrCb video (+/- 2 pixels) would be twice that of D-1's 4:2:2 format (+/- 1 pixel). This is not necessarily the case because of the definition of the luma component by ITU-R BT.601 as follows: Y = 0.299 R + 0.587 G + 0.114B. The RGB components of the luma signal are weighted to correspond to the sensitivity of human vision to each color. If green is used as the key color the higher luma sampling rate reduces the uncertainty (roughly) as follows:

For 4:2:2: 0.299 * 2 + 0.587 * 4 + 0.114 * 2 = 0.598 + 2.348 + 0.228 = 3.174

For 4:1:1: 0.299 * 1 + 0.587 * 4 + 0.114 * 1 = 0.299 + 2.348 + 0.114 = 2.761

A more accurate comparison requires a multi-dimensional matrix transformation, which is beyond the scope of this paper. However, it's reasonably safe to say that the spatial uncertainty with 4:1:1 encoding when using green as the key color is only about 15 percent greater (3.174/2.761 = 1.15) than that of 4:2:2 D-1 and, based on comparative chroma bandwidth, is likely to be about the same as that for Betacam SP. Chroma green paint, paper, and cloth is as readily available from video accessory vendors as is chroma blue.

Compression and Other Artifacts

Up to this point, the comparison of DV and component analog formats has disregarded the issue of DV's 5:1 DCT-based video compression and the issue of compression-based artifacts. The first issue to lay to rest is the often-reported but seldom-demonstrated "stair-step" artifact shown in figure 2.

The dramatic artifacts shown in figure 2 occur only occasionally and usually happen with well-lighted, high-contrast image elements having an angle of 15 degrees or less to the horizontal. Relatively uniform motion, such as zooming in or out, increases the probability of occurrence. (The lower image was captured while zooming out from the upper image.) Repeated attempts to reproduce this atrocious artifact have been unsuccessful. Stairstep and other gross artifacts are defects in codec implementation or other camcorder circuitry problems, not in the design of the DV codec or in the DCT compression process. Luma undershoot (false black horizontal lines above the upper edge of the white paper) is an analog problem, not a digital artifact.

More subtle compression artifacts appear in high-frequency image content, such as the previously-mentioned leaves on background trees and bushes, as well as blades of grass, especially when placed in motion by wind. DCT quantization discards very high-frequency information, which the untrained eye sees only with difficulty, resulting in "blockiness" (visually evident 8 x 8-pixel DCT blocks.) The perceived blockiness is due to DV's higher luma bandwidth; higher frequency components suffer less attenuation in DV than by the high-frequency rolloff of the luma component of Betacam SP, which also softens high-frequency image components. DV artifacting is more evident to a trained observer and appears to be ameliorated (but not totally eliminated) when recorded Betacam SP with a Sony UVW-1800.

To achieve 5:1 compression and a constant 25-Mbps data rate, DV uses adaptive quantization during the recording process. Adaptive quantization chooses the appropriate DCT quantization table for each frame of the sequence to maintain the constant data rate. Figure 3 demonstrates that "Ishmael" on the face-out book cover (about 6 meters from the camera lens) and "Abiding ..." on the poster (despite the reflection) are clearly readable. The almost-readable text (approximately 48-point script at a distance of roughly 5.5 meters) at the bottom of the poster is "Gretchen Lemke-Santangelo," the name of the author of Abiding Courage. The clarity of the image is remarkable, considering that the Sony DCR-VX1000 uses the not-so-remarkable lens and CCDs from the venerable Sony CCD-VX3, a 3-CCD Hi8 camcorder.

How Much Compression is Too Much?

Many detractors of the DV format arbitrarily categorize 5:1 compression as "excessive" for broadcast and corporate video production. Yet Sony is proposing a format with a higher compression rate for high-end video production. Betacam SX is a proprietary MPEG-2 derivative format (4:2:2 Profile@Main Level) proposed by Sony to ultimately replace Betacam SP. Betacam SX competes directly against Panasonic's DVCPRO product line, which has been adopted by a number of networks (most recently CBS) and local broadcast stations for ENG applications. Betacam SX uses adaptive quantization, and MPEG-2 I (inter) and B (bidirectional) frames (I-B-I-B-I..., GOP = 2) to achieve a constant 18 Mbps data rate with 4:2:2 chroma sampling. Sony appears to have picked 18 Mbps to allow uplinking of two simultaneous data streams (or a single stream at twice the data rate) to a single satellite transponder. A simple bit-rate ratio (5 * 25 / 18) results in a 7:1 compression figure. Use of 4:2:2 sampling, however, increases the precompression data rate by a factor of 33% (16 bits/sample versus 12 bits/sample), giving a 9.25:1 compression ratio. Betacam SX gains some compression credit for the use of B-frames and compressing the audio stream, but it's highly likely that the effective spatial compression of Betacam SX is greater than 5:1. Sony Europe claims on its Betacam SX Web site that DV is "designed for the professional corporate market where broadcast quality is not a prerequisite." If the DV format doesn't meet "broadcast quality" standards, it's highly unlikely that CBS would adopt DVCPRO for network news production.

JVC's Digital-S and Panasonic's forthcoming DVCPRO-50 use 3.3:1 DCT compression and 4:2:2 chroma sampling, requiring a data rate of 50 Mbps. These formats are intended to compete with Sony's Digital Betacam product line, which uses 2:1 DCT compression, and Sony's BVW and PVW Betacam SP offerings. Digital-S gear is priced in the Betacam SP range. Panasonic hasn't announced firm prices for DVCPRO-50 equipment, but it's a reasonable assumption that pricing for DVCPRO-50 and Digital-S products will be similar.

Overcoming Silk Purse Syndrome

Many comparisons of DV versus Betacam SP as an acquisition format are flawed as a result of comparing camera sections rather than recording formats. Comparing source footage shot by a $3,500 Sony DCR-VX1000 and a $21,000 DXC-637L/PVV-3 with a $15,000 Fujinon lens is a futile exercise. There is no known substitute for good glass and 2/3-inch CCDs. Fortunately, dockable DV-format recorders are available for Sony (DVCAM) and JVC (mini-DV) cameras. Side-by-side shooting with the same camera or, better yet, simultaneous recording to a DV-format dockable and a portable Betacam SP deck is the only technique that permits objective comparison of these recording formats.

Video acquisition and editing inevitably involves cost-performance compromises. Camcorders with high-quality lenses and 2/3-inch CCDs range in price from about $20,000 to more than $70,000. The most popular DV-format camcorders are an order of magnitude less expensive. D-1 VTRs offer pristine quality, but cost $150,000 or so; Digital Betacam decks, with 2:1 DCT compression, have list prices in the $50,000 range, as do the BVW series of Betacam SP recorders. The popular UVW-1800 Betacam SP editing recorder has a MSRP of $10,350. Sony's DHR-1000 consumer DV and DSR-30 DVCAM decks have list prices in the $4,000 range and a street prices close to $3,500. Evaporated-metal DV and metal-particle DVCPRO tape stock is considerably less expensive than Betacam SP's metal particle tape.

The DV format offers a virtually unbeatable performance/cost ratio. Mass-production economics of scale for consumer DV camcorder and DVTR record heads and drive mechanisms result in lower-cost professional products that use common components. An example is the almost-identical pricing of Sony's consumer DHR-1000 and DVCAM DSR-30 DVTRs. Today's administered pricing and artificial scarcity of professional DV equipment ultimately will succumb to competitive pressures as more Japanese and, especially, Korean manufacturers climb on the DV bandwagon.

IEEE-1394 (FireWire) digital I/O gives a dramatic boost to DV's performance/cost ratio for non-linear editing. FireWire I/O eliminates the analog-to-digital conversion (ADC) process of M-JPEG video capture cards and doesn't require ADC for audio inputs. The $999 DPS Spark ($699 without the full version of Premiere 4.2) is priced competitively with low-end M-JPEG capture cards. The Spark approach creates a hard-drive clone of the DV source data with headers added to accommodate Video for Windows' AVI file format. (Codecs aren't involved in DV tape-to-disk or disk-to-tape transfers.) Adaptec proved at NAB '97 that its software DV codec running on a high-end dual Pentium Pro PC outperforms by a substantial margin the Sony DVBK-1 hardware codec (used by FAST's DVMaster IEEE-1394 adapter) for adding transitions, special effects, and titles. When Adaptec optimizes their DV codec for ActiveMovie 2.0 and MMX, it's likely that the performance advantage of the Adaptec DVSoft codec over the DVBK-1 will increase. The constant 3.6 MBps data rate of DV data is well within the capabilities of the most common Ultra-Wide SCSI disk drives, such as Seagate Barracudas. Recent tests demonstrate that FireWire capture to low-cost 4G and 6G Ultra-DMA (a.k.a. Ultra-ATA) EIDE drives is practical; chances are that Ultra-DMA playback will become feasible shortly but may require a specific system board/drive combination.

Today, virtually all broadcast stations, advertising agencies, and tape duplicators houses accept Betacam SP tapes. Few broadcasters and even fewer agencies, postproductoin facilities, and duplicators currently have DV, DVCPRO, or DVCAM gear in house. Duplication firms are likely to be the first to make the investment to handle DV, DVCAM, and DVCPRO masters, eliminating dubs to Betacam SP for high-quality VHS dupes. As the DV format gains ground, local TV broadcasters will be the next to take DVCPRO and DVCAM source footage from news stringers and independent producers. Agencies, if history is a guide, will resist DV submissions to the bitter end, just as they resisted the move from 1-inch Type C to Betacam SP. Cash-starved cable MSOs and independents, striving to eke the remaining hours from their aging U-matic players, aren't likely to be able to handle DV footage until the turn of the century (or later), so don't throw away your 3/4-inch decks.

Sony saw the handwriting on the wall: Compressed digital video is a clear threat to the Betacam SP cash cow. Sony's response to the pending demise of new orders for Betacam SP gear is a proprietary, highly-compressed MPEG-2 format and SDDI (a SMPTE 259-M variant) I/O. But Sony's also hedging its Betacam SX bet with DV-compliant DVCAM and industry-standard IEEE-1394 I/O. The most likely scenario is that Betacam SX quickly will join the ranks of niche formats, such as M-II and D-5, and that the majority of new purchases will go to the three DV-compatible formats. Time will tell.

Despite the preoccupation of the television broadcast industry with DTV, HDTV, and other MPEG-2-related issues, the DV format (primarily in its DVCPRO incarnation) continues to make headway for conventional NTSC operations, especially for ENG. Press release wars between Panasonic and Sony, announcing major buys of DVCPRO or Betacam SX gear, erupt fortnightly in Broadcasting & Cable magazine's "Technology" section. The August 1998 issue of Television Broadcast magazine includes the result of its fourth annual facilities survey, "Facilities Embrace Digital Tools," carrying a "DVCPRO Top Requested Tape Format" subhead. The following table list the results of the survey of the relative popularity of video recording formats in the installed VTR base, including two formats (Digital Betacam and consumer DV) that didn't make the top 10.