What Exactly Is 8-VSB Anyway?

By David Sparano

The U.S. invented it, the ATTC tested it, the FCC accepted it, everyone is talking about it, and soon we'll all get it in our homes--but what is 8-VSB anyway? Simply put, 8-VSB is the RF modulation format utilized by the recently approved ATSC digital television standard to transmit digital bits over the airways to the home consumer. Since any terrestrial TV system must overcome numerous channel impairments such as ghosts, noise bursts, signal fades, and interference in order to reach the home viewer, the selection of the right RF modulation format is critical. Being one of the most crucial aspects of the DTV system, the 8-VSB format has received a great deal of attention and scrutiny recently.

In the alphabet soup world of digital communications, there are two big names to remember when thinking about the complete DTV system: 8-VSB and MPEG-2. 8-VSB is the RF modulation format and MPEG-2 is the video compression/packetization format. To convert high definition studio video into a form suitable for over-the-air broadcast, according to DTV standards, two stages of processing are needed: MPEG-2 encoding and 8-VSB modulation. Accordingly, two major pieces of equipment are required: an MPEG-2 encoder and an 8-VSB exciter.

The MPEG-2 encoder takes baseband digital video and performs bit rate compression using the techniques of discrete cosine transform, run length coding and bi-directional motion prediction--all of which are discussed elsewhere in this book. The MPEG-2 encoder then multiplexes this compressed video information together with pre-coded Dolby Digital (AC-3) audio and any ancillary data that will be transmitted. The result is a stream of highly compressed MPEG-2 data packets with a data rate of only 19.39 Mbps. This is by no means a trivial task since the high resolution digital video (or multiple standard resolution video) input to the MPEG-2 encoder could easily have a data rate of one Gbps or more. This 19.39 Mbps data stream is known as the DTV Transport Layer. It is output from the MPEG-2 encoder and input to the 8-VSB exciter.

Although MPEG-2 compression techniques can achieve stunning bit-rate reduction results, still more tricks must be employed to squeeze the 19.39 Mbps DTV Transport Layer signal into a slender six MHz RF channel and transmit it (hopefully without errors) to the eager consumer waiting at home in front of the TV set. This is the job of the 8-VSB exciter.

Figure 1 is a block diagram of a typical 8-VSB exciter. In this section, we will walk through the major processes that occur in the 8-VSB exciter--identifying the major components of the 8-VSB signal and explaining how this signal is generated.

Data Synchronization

The first thing that the 8-VSB exciter does upon receiving the MPEG-2 data packets is to synchronize its own internal circuits to the incoming signal. Before any signal processing can occur, the 8-VSB exciter must correctly identify the start and end points of each MPEG-2 data packet. This is accomplished using the MPEG-2 sync byte. MPEG-2 packets are 188 bytes in length with the first byte in each packet always being the sync byte. The MPEG-2 sync byte is then discarded; it will ultimately be replaced by the ATSC segment sync in a later stage of processing.

Data Randomizer

With the exception of the segment and field syncs (to be discussed later), the 8-VSB bit stream must have a completely random, noise-like nature. This is because our transmitted signal frequency response must have a flat noise-like spectrum in order to use the allotted channel space with maximum efficiency. If our data contained repetitious patterns, the recurring rhythm of these patterns would cause the RF energy content of our signal to "lump" together at certain discrete points of our frequency spectrum--thereby leaving holes in other parts. This implies that certain parts of our six MHz channel would be over-used while other parts would be under-used. Plus, the large concentrations of RF energy at certain modulating frequencies would be more likely to create discernible beat patterns in an NTSC television set when DTV-to-NTSC interference is experienced.

In the data randomizer, each byte value is changed according to known pattern of pseudo-random number generation. This process is reversed in the receiver in order to recover the proper data values.

Reed-Solomon Encoding

Reed-Solomon encoding is a Forward Error Correction (FEC) scheme applied to the incoming data stream. Forward Error Correction is a general term used to describe a variety of techniques that can be used to correct bit errors that occur during transmission. Atmospheric noise, multipath propagation, signal fades, and transmitter non-linearities may all create received bit errors. Forward Error Correction can detect and correct these errors, up to a reasonable limit.

The Reed-Solomon encoder takes all 187 bytes of an incoming MPEG-2 data packet (the packet sync byte has been removed) and mathematically manipulates them as a block to create a sort of "digital thumbnail sketch" of the block contents. This "sketch" occupies 20 additional bytes which are then tacked onto the tail end of the original 187 byte packet. These 20 bytes are known as Reed-Solomon parity bytes.

The receiver will compare the received 187 byte block to the 20 parity bytes in order to determine the validity of the recovered data. If errors are detected, the receiver can use the parity "thumbnail sketch" to locate the exact location of the errors, modify the disrupted bytes, and reconstruct the original information. Up to 10 byte errors per packet can be corrected this way. If too many byte errors are present in a given packet, the parity "thumbnail sketch" no longer resembles the received data block, the validity of the data can no longer be confirmed, and the entire MPEG-2 packet must be discarded.

Data Interleaver

The data interleaver disturbs the sequential order of the data stream and disperses the data throughout time (over a range of about 4.5 msec through the use of memory buffers) in order to minimize the transmitted signal's sensitivity to burst-type interference.

This is the equivalent of spreading all of your eggs (bytes) over many different baskets (time). If a noise burst punches a hole in the signal during propagation and "one basket" (i.e., several milliseconds) is lost, many different segments lose one egg instead of one data segment losing all of its eggs. This is known as time diversity.

Data interleaving is done according to a known pattern; the process is reversed in the receiver in order to recover the proper data order.

Trellis Encoder

Trellis coding is yet another form of Forward Error Correction. Unlike Reed-Solomon coding, which treats the entire MPEG-2 packet simultaneously as a block, trellis coding is an evolving code that tracks the progressing stream of bits as it develops through time. Accordingly, Reed-Solomon coding is known as a form of block code, while trellis coding is a convolutional code.

For trellis coding, each 8-bit byte is split up into a stream of four, 2-bit words. In the trellis coder, each 2-bit word that arrives is compared to the past history of previous 2-bit words. A 3-bit binary code is mathematically generated to describe the transition from the previous 2-bit word to the current one. These 3-bit codes are substituted for the original 2-bit words and transmitted over-the-air as the eight level symbols of 8-VSB (3 bits = 23 = 8 combinations or levels). For every two bits that go into the trellis coder, three bits come out. For this reason, the trellis coder in the 8-VSB system is said to be a 2/3 rate coder.

The trellis decoder in the receiver uses the received 3-bit transition codes to reconstruct the evolution of the data stream from one 2-bit word to the next. In this way, the trellis coder follows a "trail" as the signal moves from one word to the next through time. The power of trellis coding lies in its ability to track a signal's history through time and discard potentially faulty information (errors) based on a signal's past and future behavior.

This is somewhat like following one person's footsteps through the snow on a busy sidewalk. When the trail becomes confused with other tracks (i.e., errors are received), the trellis decoder has the ability to follow several possible "trails" for a few footprints and make a decision as to which prints are the correct ones. (Note: change this analogy to "footprints in the sand on a crowded beach" if you are reading this in a warm climate.)

Sync and Pilot

The next step in the signal processing chain is the insertion of the various "helper" signals that aid the 8-VSB receiver in accurately locating and demodulating the transmitted RF signal. These are the ATSC pilot, segment sync, and frame sync. The pilot and sync signals are inserted after the data randomization, Reed-Solomon coding, data interleaving and trellis coding stages so as not to destroy the fixed time and amplitude relationships that these signals must possess in order to be effective.

Recovering a clock signal in order to decode a received waveform has always been a tricky proposition in digital RF communications. If we derive the receiver clock from the recovered data, we have a sort of "chicken and egg" dilemma. The data must be sampled by the receiver clock in order to be accurately recovered. The receiver clock itself must be generated from accurately recovered data. The resulting clocking system quickly "crashes" when the noise or interference level rises to a point that significant data errors are received.

When NTSC was invented, the need was recognized to have a powerful sync pulse that rose above the rest of the RF modulation envelope. In this way, the receiver synchronization circuits could still "home-in" on the sync pulses and maintain the correct picture framing--even if the contents of the picture were a bit snowy. (Everyone saw the need for this except the French; sync there is the weakest part of the signal--vive la diffŽrence). NTSC also benefited from a large residual visual carrier (caused by the DC component of the modulating video) that helped TV receiver tuners zero in on the transmitted carrier center frequency.

8-VSB employs a similar strategy of sync pulses and residual carriers that allows the receiver to "lock" onto the incoming signal and begin decoding, even in the presence of heavy ghosting and high noise levels.

The first "helper" signal is the ATSC pilot. Just before modulation, a small DC shift is applied to the 8-VSB baseband signal (which was previously centered about zero volts with no DC component). This causes a small residual carrier to appear at the zero frequency point of the resulting modulated spectrum. This is the ATSC pilot. This gives the RF PLL circuits in the 8-VSB receiver something to lock onto that is independent of the data being transmitted.

Although similar in nature, the ATSC pilot is much smaller than the NTSC visual carrier, consuming only 0.3 dB or 7 percent of the transmitted power.

The other "helper" signals are the ATSC segment and frame syncs. An ATSC data segment is comprised of the 187 bytes of the original MPEG-2 data packet plus the 20 parity bytes added by the Reed-Solomon encoder. After trellis coding, our 207 byte segment has been stretched out into a stream of 828 8-level symbols. The ATSC segment sync is a repetitive four symbol (one byte) pulse that is added to the front of the data segment and replaces the missing first byte (packet sync byte) of the original MPEG-2 data packet. Correlation circuits in the 8-VSB receiver home in on the repetitive nature of the segment sync, which is easily contrasted against the background of completely random data. The recovered sync signal is used to generate the receiver clock and recover the data. Because of their repetitive nature and extended duration, the segment syncs are easy for the receiver to spot. The result is that accurate clock recovery can be had at noise and interference levels well above those where accurate data recovery is impossible (up to 0 dB SNR--data recovery requires at least 15 dB SNR). This allows for quick data recovery during channel changes and other transient conditions. Figure 2 shows the make-up of the ATSC data segment and the position of the ATSC segment sync.

An ATSC data segment is roughly analogous to an NTSC line; ATSC segment sync is somewhat like NTSC horizontal sync. Their duration and frequencies of repetition are, of course, completely different. Each ATSC segment sync lasts 0.37 msec.; NTSC sync lasts 4.7 msec. An ATSC data segment lasts 77.3 msec.; an NTSC line 63.6 msec. A careful inspection of the numbers involved reveals that the ATSC segment sync is somewhat more "slender" when compared to its NTSC counterpart. This is done to maximize the active data payload and minimize the time committed to sync "overhead."

Three hundred and thirteen consecutive data segments are combined to make a data frame. Figure 3 shows the make-up of an ATSC data frame. The ATSC frame sync is an entire data segment that is repeated once per frame (24.2 msec.) and is roughly analogous to the NTSC vertical interval. (FYI: The NTSC vertical interval occurs once every 16.7 msec.). The ATSC frame sync has a known data symbol pattern and is used to "train" the adaptive ghost-canceling equalizer in the receiver. This is done by comparing the received signal with errors against the known reference of the frame sync. The resulting error vectors are used to adjust the taps of the receiver ghost-canceling equalizer. Like the segment sync, the repetitive nature of the frame sync, and correlation techniques used in the 8-VSB receiver, allow frame sync recovery at very high noise and interference levels (up to 0 dB SNR).

The robustness of the segment and frame sync signals permits accurate clock recovery and ghost-canceling operation in the 8-VSB receiver--even when the active data is completely corrupted by the presence of strong multipath distortion. This allows the adaptive ghost-canceling equalizer "to keep its head" and "hunt around in the mud" in order to recover a useable signal--even during the presence of strong signal echoes.

AM Modulation

Our eight-level baseband signal, with syncs and DC pilot shift added, is then amplitude modulated on an intermediate frequency (IF) carrier. With traditional amplitude modulation, we generate a double sideband RF spectrum about our carrier frequency, with each RF sideband being the mirror image of the other. This represents redundant information and one sideband can be discarded without any net information loss. This strategy was employed to some degree in creating the vestigial lower sideband in traditional NTSC analog television. In 8-VSB, this concept is taken to greater extremes with the lower RF sideband being almost completely removed. (Note: 8-VSB = 8 level--Vestigial Side Band.)

(There are several different ways to implement the AM modulation, VSB filtering, and pilot insertion stages of the 8-VSB exciter, some of which are completely digital and involve direct digital synthesis of the required waveforms. All methods aim to achieve the same results at the exciter output. This particular arrangement was chosen in the interest of providing a clear, easily understandable, signal flow diagram.)

Nyquist Filter

As a result of the data overhead added to the signal in the form of forward error correction coding and sync insertion, our data rate has gone from 19.39 Mbps at the exciter input to 32.28 Mbps at the output of the trellis coder. Since 3 bits are transmitted in each symbol of the 8-level 8-VSB constellation, the resulting symbol rate is 32Mb / 3 = 10.76 Million symbols/sec. By virtue of the Nyquist Theorem, we know that 10.76 Million symbols/sec can be transmitted in a VSB signal with a minimum frequency bandwidth of 1/2 X 10.76 MHz = 5.38 MHz. Since we are allotted a channel bandwidth of 6 MHz, we can relax the steepness of our VSB filter skirts slightly and still fall within the 6 MHz channel. This permissible excess bandwidth (represented by a, the Greek letter alpha) is 11.5 percent for the ATSC 8-VSB system. That is, 5.38 MHz (minimum bandwidth per Nyquist) + 620 kHz (11.5 percent excess bandwidth) = 6.00 MHz (channel bandwidth used). The higher the alpha factor used, the easier the hardware implementation is, both in terms of filter requirements and clock precision for sampling.

The resulting frequency response after Nyquist VSB filter is shown in figure 4.

This virtual elimination of the lower sideband, along with the narrowband filtering of the upper sideband, creates very significant changes in the RF waveform that is ultimately transmitted. For the NTSC-hardened veteran, there is a great temptation to imagine the 8-VSB RF waveform as being a sort of "8-step luminance stairstep" signal transmitting the eight levels of 8-VSB. Unfortunately, there is a fundamental flaw with this notion. As figure 5 illustrates, such a crisp stairstep signal with "squared off" abrupt transitions would generate a frequency spectrum that is far too wide for our single 6 MHz channel. A "square symbol pulse" -type signal generates a rich spectrum of frequency sidelobes that would interfere with adjacent channels.

We know that this type of RF waveform is incorrect since our Nyquist VSB filter has already pared our RF spectrum down to a slender 6 MHz channel.

As any video or transmitter engineer knows, when a square pulse is frequency bandlimited, it will lose its square edges and "ring" (oscillate) in time before and after the initial pulse. For our digital 8-level signal, this would spell disaster as the pre- and post-ringing from one symbol pulse would interfere with the preceding and following pulses, thereby distorting their levels and disrupting their information content.

Fortunately, there is still a way to transmit our 8-VSB symbol pulses if we observe that the 8-level information is only recognized during the precise instant of sampling in the receiver. At all other times, the symbol pulse amplitude is unimportant and can be modified in any way we please--so long as the amplitude at the precise instant of sampling still assumes one of the required eight amplitude levels.

If the narrowband frequency filtering is done correctly according to the Nyquist Theorem, the resulting train of symbol pulses will be orthogonal. This means that at each precise instant of sampling, only one symbol pulse will contribute to the final RF envelope waveform; all preceding and following symbol pulses will be experiencing a zero crossing in their amplitude. This is shown in figure 6a. In this way, when the RF waveform is sampled by the receiver clock, the recovered voltage will represent only the current symbol's amplitude (one of the eight possible levels).

At all times in-between the instants of sampling, the total RF envelope waveform reflects the addition of the "ringing" of hundreds of previous and future symbols (since all symbols have non-zero amplitudes between sampling times). Note that, in the interest of simplicity, figure 6a shows our narrowband symbol pulses as ringing for only 10 sampling periods; in reality they ring for a much longer time. These non-zero values (between sampling times) from hundreds of symbols can add up to very large signal voltages. The result is a very "peaky" signal that most closely resembles white noise. This is shown in figure 6b. The peak to average ratio of this signal can be as high as 12 dB, although RF peak clipping in the transmitter can limit this value to 6 to 7 dB with minimal consequences.

8-VSB Signal Constellation

In 8-VSB, the digital information is transmitted exclusively in the amplitude of the RF envelope and not in the phase. This is unlike other digital modulation formats such as QAM, where each point in the signal constellation is a certain vector combination of carrier amplitude and phase. This is not possible in 8-VSB since the carrier phase is no longer an independent variable under our control, but is rather "consumed" in suppressing the vestigial lower sideband.

The resulting 8-VSB signal constellation, as compared to 64-QAM, is shown in figure 7. Our eight levels are recovered by sampling an in-phase (I channel) synchronous detector. Nothing would be gained by sampling a quadrature channel detector since no useful information is contained in this channel. Our signal constellation diagram is therefore a series eight vertical lines that correspond to our eight transmitted amplitude levels. By eliminating any dependence on the Q-channel, the 8-VSB receiver need only process the I channel, thereby cutting in half the number of DSP circuits required in certain stages. The result is greater simplicity, and ultimately cost savings, in the receiver design.

The Rest Of The 8-VSB Chain

After the Nyquist VSB filter, the 8-VSB IF signal is then double up-converted in the exciter, by traditional oscillator-mixer-filter circuits, to the assigned channel frequency in the UHF or VHF band. The on-channel RF output of the 8-VSB exciter is then supplied to the DTV transmitter. The transmitter is essentially a traditional RF power amplifier, be it solid state or tube-type. A high-power RF output system filters the transmitter output signal and suppresses any spurious out-of-band signals caused by transmitter non-linearities. The last link in the transmitting chain is the antenna that broadcasts the full-power, on-channel 8-VSB DTV signal.

In the home receiver, the over-the-air signal is demodulated by essentially applying in reverse the same principals that we have already discussed. The incoming RF signal is received, downconverted, filtered, then detected. Then the segment and frame syncs are recovered. Segment syncs aid in receiver clock recovery and frame syncs are used to train the adaptive ghost-canceling equalizer. Once the proper data stream has been recovered, it is trellis decoded, deinterleaved, Reed-Solomon decoded, and derandomized. The end result is the recovery of the original MPEG-2 data packets. MPEG-2 decoding circuits reconstruct the video image for display on the TV screen and Dolby Digital (AC-3) circuits decode the sound information and drive the receiver loudspeakers. The home viewer "receives his DTV" and the signal chain is complete.

Conclusion

The goal of this section has been to provide some insight into the inner workings of the 8-VSB transmission system. Like many things in life, 8-VSB can appear formidable at first, but is really quite simple "once you get to know it." Hopefully the knowledge conveyed in this section will dispel some of the fear factor that many NTSC engineers experience when faced with the unknown world of digital TV broadcasting.

So what then is 8-VSB? Simply put, 8-VSB is the future of American television. And the future doesn't have to be such a scary thing.

References

[1] Davis, Robert and Twitchell, Edwin, "The Harris VSB Exciter for Digital ATV" NAB 1996 Engineering Conference. April 15-18, 1996.

[2] Citta, Richard and Sgrignoli, Gary, "ATSC Transmission System: 8-VSB Tutorial" ITVS 1997 Montreux Symposium. June 12-17, 1997.

[3] Totty, Ron, Davis, Robert and Weirather, Robert, "The Fundamentals of Digital ATV Transmission" ATV Seminar in Print. Harris Corporation Broadcast Division, 1995.

Acknowledgements

Special thanks go to Joe Seccia, Bob Plonka and Ed Twitchell of Harris Corporation Broadcast Division for their contributions of time, material and assistance in writing this section.