Video Compression for DBS (Direct Broadcast Satellites).......................................
hei.ca
revised November 12,1997
HEI has prepared this brief technical discussion about video compression as it relates to Hughes Communications and United States Satellite Broadcasting Company (USSB) DBS-1, -2 and -3 satellites used for the DirecTV and USSB pay-TV services.
Operating in the microwave Ku band (12.2 GHz to 12.7 GHz), the three high-power satellites form a constellation which can be simultaneously viewed by small, fixed, DBS antennas like those from Thomson/RCA or Sony. The RCA system's 18" x 20" (45.7cm x 50.8cm) antenna has an approximate gain of 34 dB at 12.5 GHz and a half-power beamwidth of about 3.5 degrees. Each satellite has 16 transponders, analogous to channels on a television receiver. Each transponder in turn operates at 40 megabits per second (40 Mbps).
<Picture>Forward Error Correction (FEC)
Because of the short wavelength of the Ku band microwave signals, about 2.5 cm or 1 inch, during periods of heavy rain, the raindrops themselves act like small antennas and can absorb more or less of the signals from the satellites.
The approximate rainfall attenuation at 12.5 GHz, assuming one mile or 1.6 Km of the path from the satellite to the receiver passes through rain is given in the table below. Rainfall is measured as depth per hour:
RainfallAttenuationDescription
0.02" (0.5mm)/hour
0.01 dB
drizzle
0.1" (2.5mm)/hour
0.1 dB
light rain
0.5" (12.7mm)/hour
0.9 dB
heavy rain
1.0" (25.4mm)/hour
2.3 dB
-
4.0" (100mm)/hour
12 dB
cloudburst
10.0" (250mm)/hour
43 dB
-
As the signals from the satellite are digital signals, the increase in signal attenuation (or greater path loss) results in an increase in the signal's bit error rate (BER). Video compression is used for all the DBS satellite's video signals. This means that each data bit received was originally a number of data bits in the original television signal before compression. (This type of compression is sometimes also known as entropy coding.) If special steps weren't taken, the signal received at many locations during rainfall would be unwatchable -- a patchwork of colored blocks or a frozen image. In this case, the use of forward error correction (FEC) saves the day.
Of the 40 Mbps (Megabits per second) available per satellite transponder, only 23 Mbps is used for the actual audio and video signals while the remaining 17 Mbps is used for forward error correction information. Choice of a suitable FEC ensures that DBS can work reliably with tiny receiving antennas. Without FEC, these small antennas would otherwise only provide marginal receive signals.
With conventional data transmission like that between your home or office and the Internet or BBS, your modem and the one at the other location carry out a two-way exchange of information called "handshaking" to ensure error-free data signal transmission. If something happens during transmission, one modem will request the other modem to re-send the block of data in error. That's a viable error control technique when two-way communications is possible and just two sites are involved.
However, in receive-only situations like a DBS home receiver, each receive site must be able to correct received errors on its own.
For example, when the signal path from the satellite to the home receiver passes through heavy rain, the signal is considerably attenuated or weakened to perhaps one-one thousandth of its fair-weather value. This, in turn, introduces errors into the received signal. The availability of the FEC information, transmitted together with the compressed video, provides the means to restore the signal to its error-free form.
The production of the FEC information-signal takes place during the actual transmission of the television signals to the satellites. A mathematical process generates information about the television data signal (known as a syndrome). When the signals are received at a person's home, the incoming digital television signals are processed, together with the syndrome. When there are digital signal bits in error, the syndrome provides information for the receiver to faithfully re-create the original signals, at least most of the time.
The syndrome also includes sufficient information to enable it to correct and fully recover its own original information. This is because under error conditions, both the FEC syndrome and the actual television signal are both usually impaired.
When the receive signal quality becomes very poor (an extremely high bit-error rate) such as during the heaviest rains, the FEC's automatic error correction capability may start to fail. This might appear to a viewer as incorrectly placed small picture blocks, blocks of colored snow, rainbows or other visual anomalies. Static-like sounds may be heard from the audio. When the signal finally fades to the point where the FEC completely fails, the receiver mutes both the sound and video automatically. Muted video usually appears as a frozen image on the screen.
As the rainfall reduces, and the signal level gradually increases, the receiver reaches a threshold where it can lock on to the incoming signals from the satellites (synchronization) and once again decode signals, restoring normal video and audio.
FEC may sound like a pretty complex technique but it is commonly used for many forms of mass digital communications. For example, the audio CD, which is really stored one-way digital communications, uses a form of FEC called Reed-Solomon coding, allowing reliable playback even with sticky gobs of peanut butter on your disks (don't try it!)
The next time it rains, think of FEC hard at work . . .
For those of us in the frostier climes, the occasional winter blizzard won't cause a loss of your DBS signal. Only water in the liquid phase will attenuate the satellite microwave signals. However, snow when it is melting can attenuate signals significantly.
<Picture>Coding the Video Signals
The 23 Mbps available for signal transmission in each transponder is managed by a device known as a statistical multiplexer. This in turn is connected to several MPEG2 video codecs (short for COder-DECoder). Codecs take the uncompressed analog audio and video signals, convert them to a digital signal at about 140 Mbps and then reduce the actual bit rate required to between 1.5 Mbps and 15 Mbps for DBS applications.
For comparison. international standard H.320 videoconferencing codecs can compress the video plus audio into as little as 56 Kbps, about 1/2,500th of the original. Such low bit rate operation is really just adequate for military thread-of-life visual communications, providing a low-detail image at about one video frame per second. Business grade videoconferencing typically requires at least 256 Kbps transmission for good quality pictures at about 15 video frames per second. Excellent motion handling near 30 frames per second is possible at transmission rates of 768 Kbps and above.
The gamut of motion video compression techniques in common use include:
four versions of ISO-MPEG (Motion Picture Experts Group), ITU-T H.320, H.323 and H.324 for videoconferencing
For still image transmission, the following are in common use:
ISO-JPEG (Joint Photographic Experts Group), ITU-T JPEG (T.84)
All of these approaches make a greater or lesser use of two-dimensional discrete cosine transform (DCT) coding, developed by Dr. Wen-hsung Chen, Chief Scientist at Compression Labs, Inc. (CLI) in San Jose, CA, USA, a company that manufactures MPEG2 and other types of codecs. The DCT technique works by removing redundancies from the images. How profound the compression possible depends upon the available transmission bandwidth and the scene content.
As an aside, so-called motion JPEG is not an ITU or ISO standard method of image transmission but is really a proprietary approach in the guise of a standard.
There are other motion video compression techniques used for very specialized video compression applications -- Wavelet and Fractal compression are two of the most common. However, neither are used for DBS applications.
I mentioned earlier that the uncompressed video signal needs about 140 Mbps for transmission of a virtually unimpaired broadcast-quality signal -- 30 video frames per second (excellent motion-handling) about 5.2 MHz luminance bandwidth, about 416 lines video horizontal lines resolution (not related to the number of scanning lines) and about 55 dB signal-to-noise ratio. In more digitally-oriented terms, in the North American and Japanese NTSC universe where the horizontal scanning rate is 15.734 KHz, the 5.2 MHz bandwidth equates to about 544 picture elements (PELs or pixels) on the active portion of each horizontal scanning line (each about 52 microseconds in duration). These same parameters, set out in the CCIR (now ITU-R) 601 standard, also accommodate the PAL and SECAM video systems with very similar performance.
A short review. The term "lines of resolution" was developed to test the resolving power of a television system including lens, pickup device, transmission network and its equipment (which could be a short cable or a billion miles of space), and the display. To perform a test, a test chart is televised which includes a calibrated vertically-oriented wedge-shaped group of black and white lines of varying spacing and width. Calibration values are placed beside the wedge. Values shown typically begin at "200", increasing in increments of 100 to about "800". These values are "lines of resolution".
These values represent the resolving power of the system in terms of the number of black and white lines which can occupy a space equal to the picture height. In practice, when viewing the wedge on a monitor, a point is noted where you can't see the individual adjacent black and white lines on the wedge and it appears gray instead. The point where the transition occurs gives the number of lines of resolution.
For the NTSC system, one MHz of bandwith is required for the transmission of each 80 lines of resolution. This means that the absolute maximum resolution possible with over-the-air NTSC television with its 4.2 MHz bandwidth is 336 lines of resolution if the entire equipment chain is ideal. This value has no relation whatever with the number of scanning lines (525 for NTSC).
Because lines of resolution is really a qualitative measurement, it can be stated that the number of PELs needed per active video line (of 52 microseconds) for the NTSC system is approximately equal to 1.3 times the number of lines of resolution.
<Picture>Encoder Hardware
One of the important aspects of MPEG2 system design is that most of the compression effort is done by a single expensive encoder used at the signal's transmitting location. This in turn allows the set-top decoder to be a relatively simple and inexpensive device.
DirecTV's Castle Rock, CO and USSB's Oak Dale MN broadcast centers are equipped with more than 200 Compression Labs, Inc. (CLI) (San Jose, CA) Magnitude MPEG2 encoders. These encoders use an array of encoder chips from Milpitas, CA semiconductor manufacturer C-Cube Microsystems. An important part of the encoders is the inclusion of a dynamic statistical multiplexer, allowing the operating companies to best tailor the image quality to the demands of the program material, helping to optimize both bandwidth utilization and video quality.
In 1996, the broadcast division (including DBS coders) was sold to General Instrument who added these former CLI products to its product line. They did not. however, retain the CLI R&D staff.
In the spring of 1997, the balance of CLI was merged into videoconferencing company VTEL which has no broadcasting interests.
For this DBS service, the standard used is known as ISO-MPEG2, which provides high quality stereo audio and video transmission at a variety of bit rates from 1.5 Mbps to 15 Mbps. For DirecTV and USSB, two basic encoding rates initially chosen were centered on 3 Mbps and 7.5 Mbps. High-motion program material like sports was allocated a nominal 7.5 Mbps while more sedate material such as talk shows used a bit rate of about 3 Mbps. Several channels were then combined using a statistical multiplexer, a device which accommodates a number of digital signals simultaneously.
This approach makes efficient use of the bandwidth while optimizing signal quality by sending (buffer-fullness state feedback) control signals to the connected encoders.
Video codecs, in an otherwise unconstrained environment, produce an instantaneous bit rate more-or-less proportional to the differences between video in adjacent video frames (similar in concept to frames of film). This results in a bit rate which drops to very low values during static scenes. Conversely, the bit rate will momentarily change to high values when a total scene change takes place. As far as the average bit rate is concerned, the greater the amount of motion in an image, the greater the bit rate required. Added to that, the finer the picture detail of moving objects, the more data will be generated, but only up to a point -- the human eye cannot discern detail in rapidly moving objects so one method of compressing the needed bandwidth is to reduce the detail in rapidly moving parts of the scene -- the viewers will never see it!
However, in the constrained context of DBS, there is only a limited amount of bandwidth available among the group of codecs sharing any particular transponder. For example, if a codec only needs half of its available nominal assigned bandwidth for a period of time, the rest of its allotted bandwidth can be borrowed by another (or several) codec(s). This will provide better motion handling and/or picture definition for one or more other signals. Such dynamic bandwidth allocation is a very effective way of providing what seems like more bandwidth than the 23 Mbps available, if the types of services sharing each transponder are carefully chosen. On the other hand, if all the channels on a given transponder are assigned to sports, gains from bandwidth sharing could be non-existant.
Picture resolution is also programmable in each codec, allowing fine tuning of the available bandwidth for any given type of program. This may also help reduce some decoding artifacts. Most programs are encoded at the ITU-R 601's Nyquist sampling rate of 544 PELs by 480 lines (NTSC) (576 lines for PAL), a good-quality picture. However, resolution can be adjusted from a minimum of 352 PELs x 240 lines (NTSC) (288 lines for PAL) (about the same picture quality as H.320 videoconferencing) to a snappy full ITU-T 601 resolution of 720 PELs by 480 lines (NTSC), (576 lines for PAL). By the way, it's especially in the 544 PEL and 720 PEL resolutions that the use of s-video connections from the receiver to your TV set really pays off.
<Picture>Summer 1995, Transition to MPEG2
Although all set-top boxes, technically known as Integrated Receiver/Decoders or IRDs, include full MPEG2 hardware, the actual signals being transmitted until the fall of 1995 were in the older MPEG1+ format, but using the MPEG2 syntax. The IRDs are backward-compatible with MPEG1 which, for practical purposes, is a subset of MPEG2. IRDs automatically operated with full MPEG2 compatibility without any change required as the MPEG2 encoders were put on-line during the late summer of 1995.
<Picture>April 1997, A loss of Quality
With much fanfare, both DirecTV and USSB added additional channels to their DSS service. DirecTV increased the number of channels carrying pay-per-view. More barker (preview) channels with motion video were also added to the system.
The firmware (software) of the codecs was changed.
Prior to this, the millions of satisfied customers received images of a quality which rivaled videodisks.
However, the same time as these channel additions took place, customers began experiencing significant reductions in picture quality. Blocking (blocks of pixels in error) are seen on all channels periodically because of errors introduced into the video compression chain, insufficient forward error correction allocation (insufficient transmit power) and numerous other problems became common. Bandwidth for many movie channels was reduced to 2 Mbps with no bandwidth sharing and reduced FEC which exacerbated the system's ability to faithfully reproduce the high quality of the original program material.
A significant difference in the impairments produced can be seen between DirecTV and USSB channels because of the engineering trade-offs made by the system operators. However, many viewers have noticed that the quality has dropped to that of cable TV or commercial VHS tapes. Its quality certainly no longer rivals video laserdisks. The loss of quality has also been noted by the professional press.
In September 1997, some of the entertainment channels had their bandwidth increased to 2.5 Mbps from about 2 Mbps. Regrettably, the improvements are not enough to bring the quality back to the levels advertised.
Remember that in the realm of video compression, bit rates in the region around 2 Mbps are considered good for talking heads, like newscasts. No wonder that action films like the blockbuster Independance Day cause the system to produce serious coding errors. But at the same time, the modes of operation chosen, together with the software employed, have rendered the otherwise robust and forgiving MPEG2 into a system fraught with video coding problems. It also seems that the statistical multiplexing is no longer functional in any meaningful way.
In 1998, HDTV (high definition television) is touted to appear on DBS as is Dolby AC-3 audio. This will create an interesting dichotomy as these improvements to quality will require more of the scarce bandwidth, which will further reduce the quality of the services.
Perhaps the DBS industry should implement P.800 et al subjective MOS quality testing as we do for a number of clients to determine just how many channels of what quality can really be squeezed into just so much bandwidth.
<Picture>Other DBS Systems
In addition to the groundbreaking DirecTV and USSB DBS systems, a number of other competing and incompatible systems are planned or implemented for the USA. The following is a brief description of each:
The AlphaStar system uses an MPEG2-compliant encoding scheme, incompatible with everything else because of a different encryption scheme. This Canadian-based partnership includes both Samsung and Tee-Com, known for C-band satellite receivers. Together with Philips/Magnivox, they will be the main receiver manufacturers for this system. Because this system will not initially use a high-power satellite, approximately 30 inch (76.2 cm) diameter receiving antennas will be necessary. The Canadian service went bankrupt in mid-year with only several thousand customers signed up. The EchoStar system will be similar to the AlphaStar system above but incompatible, and will be manufactured by Sagem, SCI and Philips. The PrimeStar system, the major DBS competition, is targetted at those users who don't want to purchase the hardware (although this is an option). Owned by five major cable TV companies, this system uses Channel Master dishes of from 32 inches (81cm) to 39 inches (~1 metre) in diameter, for control of reception quality depending on location, together with General Instruments receivers.
<Picture>About Video Compression . . .
I have been involved with video compression R&D and applied compressed video or videoconferencing, for more than 25 years. I remember the days when a codec occupied three eight-foot high racks of equipment and could operate at bit rates as low as 28 Mbps, a 5:1 compression ratio! Don't ask the price. Ten years ago, codecs had shrunk to the size of a mid-size refrigerator and cost about $US160,000. The DBS and the IRD are truly modern technological miracles -- and it's all thanks to some mathematical equations . . .
If the above has piqued your interest in video compression, you might like to read Basil Halhed's article on the subject for video teleconferencing -- the techniques are virtually the same, only the degree of processing is changed. The article is called Multipoint Videoconferencing, starting on page 339 in Volume 11 of The Encyclopedia of Microcomputers, published by Marcel Dekker, Inc., available in most good technical libraries. More articles by Basil Halhed will appear in 1998 in The Encyclopedia of Telecommunications and a supplement to The Encyclopedia of Microcomputers, both also published by Marcel Dekker, Inc.
c Copyright 1995-97 HEI |
