In our last installment (February 2002), we discussed some of the realities and limitations inherent in image resolution. This time we will cover the bandwidth implications of those realities. Although many say that broadband Internet connections are the wave of the future, considerable evidence suggests that not everyone wants, needs, or — possibly more to the point — is willing to pay for that much bandwidth. In a larger sense, the bandwidth available to broadband subscribers is relatively small. Like it or not, bandwidth is expensive. Signals that use a lot of bandwidth are costly to capture, store, transmit, and display.
Compared to television and radio, the Internet is a low-bandwidth medium, especially without a broadband connection. Although it is unlikely that the future of professional audio and video depends on Internet broadcasts, they certainly offer an additional outlet for professional productions. Understanding why the signals we create require significant bandwidth can help with compromises that must be made to reduce bandwidth requirements and provide quality playback over a variety of delivery channels, including the Internet.
Audio is a relatively low-bandwidth signal (20KHz) but prior to the 486 series of microprocessors, computers had difficulty handling it. Today, the same can be said for uncompressed, studio-quality video (20MHz). Granted, today's computers can handle uncompressed video, but the average desktop system can still be overwhelmed by it. To some extent, computers are less than ideal for handling large, continuous data streams such as audio and video. Other devices, such as VCRs, can handle these large-bandwidth signals with ease. A relatively inexpensive VCR easily stores the 6MHz bandwidth of the television signal, complete with color and stereo audio. Regardless, computers are here to stay, and their use in the audio and video production process is a given. Let's keep that in mind as we look at some of the numbers produced by today's imaging technology.
Analog and digital bandwidth
In today's world, bandwidth is typically described in one of two ways. Analog bandwidth is referenced in terms of frequency — for example, 6MHz or 20KHz. One hertz is equivalent to one cycle per second. Digital bandwidth is normally more of a “speed” term, such as 1.6 megabits per second. This is different from 1.6 megabytes per second. Eight bits make a byte, so 1.6 megabits per second is equal to 0.2 megabytes per second.
It is not uncommon to see these numbers confused, especially when abbreviated. There is no official abbreviation for bits or bytes. But many times (including within these pages), “bits” is abbreviated “b” while an uppercase B represents “bytes.”
When comparing analog and digital bandwidth, one easy way is to say that 1Hz is equivalent to one bit per second. In reality, various techniques can be employed to increase the number of bits that can be transmitted over an analog circuit. One good example of this is the 56k modem. Using standard modulation techniques, 9600 baud modems (baud is a measure of symbol rate, not bit rate) were about all standard phone lines could support. However, advanced coding techniques allow more bits per symbol, and therefore higher bit rates. Signal compression is a technique that can increase the overall amount of information conveyed over a transmission path.
Another consideration is path bandwidth vs. payload bandwidth. Most digital transmission systems require some number of bits to be dedicated for overhead purposes. Payload bandwidth is what is left after the number of overhead bits is subtracted from the path bandwidth. For some signals, a missing bit here or there is no problem. For others, however, a single missing bit can mean disaster. In those instances, additional bits must be included in the data stream to allow for error-checking as well as error correction. These extra bits reduce the number of payload bits, sometimes by as much as 50%. Bi-directional communications also reduce the number of bits that can be sent, as do party-line technologies such as Ethernet that allow multiple devices to communicate over a common set of cables.
Image bandwidth
In the last installment we discussed sampling. This time we need to consider quantization. Quantization is the act (or art) of converting samples of an analog signal into numeric values. In nearly all cases that value is an integer. Normally, a specific number of bits is used. For example, if a single bit is used, the value is either 1 or 0. For imaging purposes this correlates nicely with black or white. Using two bits allows for four values (22), so in addition to black and white, a light and a dark gray would be available. Using 8bits/pixel allows for 256 different values (28) and offers reasonably good quality. Each time a bit is added, the number of quantization levels doubles. Much of today's video equipment allows for 12bit or 14bit quantization with processing performed using at least two additional bits.
Taking a black-and-white 640×480 computer image, we can calculate that it requires 8bits/pixel multiplied by 640 pixels per line multiplied by 480 lines. The result is 2,457,600 bits! If we wanted 30 of those images every second, it would require 2,457,600×30, or about 70Mbps. And that's just for black and white. Bumping the number of bits per sample to 10 and adding color could easily double that number. As a matter of fact, uncompressed digital video (SMPTE 259M) requires 143Mbps — that will fill a 20GB hard drive in slightly more than two minutes. The physical connection used for SMPTE 259M runs at 270Mbps, but it allows for 8bit or 10bit quantization and room is set aside for audio and auxiliary data. The specification for high-definition video calls for uncompressed data rates of 1.5Gbps.
Those numbers translate to large bandwidth requirements, especially when compared to the transmission system bandwidths available today. Digital television offers a fairly large bandwidth of 19.4Mbps, but a T1 connection from the local telco offers only 1.5Mbps. When you compare the data rates mentioned in the previous paragraph with the typical data rates available for transmission, it becomes obvious why compression systems are needed. The need for compression is not new. The fact that film runs at only 24 frames per second is a form of compression, as is interlace scanning.
Bandwidth and compression
Compression, or bit-rate reduction, provides a means to reduce the overall number of bits needed to convey signals. Compression systems can either be lossless (WinZip is a good example) or lossy (most MPEG or JPEG applications). Lossless compression, as the name implies, allows the original signal (or file) to be reconstructed precisely. Lossy compression literally throws away data. This irretrievable data is removed via several techniques, including sub-sampling, reducing the number of bits used for quantization, removing redundant data, and techniques that take advantage of the way the human visual and auditory systems perceive light and sound.
By taking advantage of how we perceive images and sound, the signal can be greatly distorted based on the original, and for the most part we perceive the distorted image as being very close to the original. This type of selective signal processing keeps the important data without wasting bandwidth on information that will not be noticed. High-quality compression allows the 1.5Gbps HD signal to be compressed to the 19.4Mbps datastream that is broadcast by today's digital transmitters. To the viewer, the signal looks much better than today's analog transmission, despite the compression.
Finding better ways to move images and sound of sufficient quality through smaller and smaller pipes — bandwidth — will allow new technologies to deliver video and audio to broader audiences. Despite the seemingly primitive displays on many of today's handheld devices, there is every reason to believe that quality audio and video will be compressed, downloaded, and displayed on future generations of cell phones and handheld devices such as PDAs.
