Optimize Air Chain (Page 1 of 3)
Fundamentals of Audio Processing
Peak Limiting & Clipping
Multi-Band Compression & Frequency Selective Limiting
Preserving the Waveform Fidelity of Processed Audio
Location of System Components
Requirements for Studio-Transmitter Links ("STL")
Peak Modulation Control
Overshoots in Composite STL Systems & FM Exciters
The FM Stereo System
Previous Non-Linear Solution
Broadcast audio processing is both an engineering and artistic endeavor. The engineering goal is to make most efficient use of the signal-to-noise ratio and audio bandwidth available from the transmission channel while preventing its overmodulation. The artistic goal is set by the audio processing user. It may be to avoid audibly modifying the original program material at all. Or it may be to create a distinct sonic signature for the broadcast by radically changing the sound of the original. Most broadcasters operates somewhere in between these two extremes, with the main goal of audio processing to increase the perceived loudness within the peak modulation constraints of a transmission channel.
Provided that the transmitted signal meets regulatory requirements for modulation control and RF bandwidth, there is no well-defined right or wrong way to process audio. Like most areas requiring subjective, artistic judgment, processing is highly controversial and likely to provoke thoroughly opinionated arguments amongst its practitioners. Ultimately, the success of a broadcast's audio processing must be judged by its results - if the broadcast gets the desired audience, then the processing must be deemed satisfactory regardless of the opinions of audiophiles, purists, or others who consider processing an unnecessary evil.
Loudness is increased by reducing the peak-to-average ratio of the audio. If peaks are reduced, the average level can be increased within the permitted modulation limits. The effectiveness with which this can be accomplished without introducing objectionable side effects (like clipping distortion) is the single best measure of audio processing effectiveness.
Compression reduces dynamic range of program material by reducing the gain of material whose average or rms level exceeds the threshold of compression. AGC amplifiers are compressors. Compression reduces the difference in level between the quiet and loud sounds to make more efficient use of permitted peak level limits, resulting in a subjective increase in the loudness of quiet sounds. It cannot make loud sounds seem louder. Compression reduces dynamic range relatively slowly in a manner similar to "riding the gain."
Density is the extent to which the amplitudes of audio signal peaks are made uniform (at the expense of dynamic range). Programs with large amounts of short-term dynamic range have low density; highly compressed programs have high density.
Peak limiting is an extreme form of compression characterized by a very high compression ratio, fast attack time, and fast release time. In modern audio processing, a peak limiter, by itself, usually limits the peaks of the envelope of the waveform, as opposed to individual instantaneous peaks in the waveform. These are usually controlled by clipping. Limiting and clipping reduce the short-term peak-to-average ratio of the audio.
The main purpose of limiting is to protect a subsequent channel from overload, as opposed to compression, whose main purpose is to reduce dynamic range of the program.
Peak clipping is a process that instantaneously clips off any part of the waveform that exceeds the threshold of clipping. While a peak clipper can be very effective to increase loudness, it causes audible distortion when over-used. It also increases the bandwidth of the signal by introducing both harmonic and intermodulation distortion into its output signal. Therefore various forms of overshoot compensation are used, which is essentially peak clipping that does not introduce significant out-of-band spectral energy into its output.
Limiting increases audio density. Increasing density can make loud sounds seem louder, but can also result in an unattractive, busier, and flatter denser sound. It is important to be aware of the many negative subjective side effects of excessive density when setting controls that affect the density of the processed sound.
Clipping sharp peaks does not produce any audible side effects when done moderately. Excessive clipping will be perceived as audible distortion.
These techniques divide the audio spectrum into several frequency bands and compress or limit each band separately (although some inter-band coupling may be used to prevent excessive disparity between the gains of adjacent bands). This is the most powerful and popular contemporary audio processing technique, because, when done correctly, it eliminates spectral gain intermodulation. This occurs in a wideband compressor or limiter when a voice or instrument in one frequency range dominates the spectral energy, thus determining the amount of gain reduction. If other, weaker, elements are also present, their loudness may be audibly and disturbingly modulated by the dominant element. Particularly unpleasant effects may occur if the dominant energy is in the bass region, because the ear is relatively insensitive to bass energy, so the loudness of the midrange is pushed down by the dominant bass energy seemingly inexplicably. The best results are obtained with steep crossover slopes allowing more consistency from various program sources. It can also give the "illusion" of an unprocessed "big" sound.
Another type of frequency-selective limiting uses a program-controlled filter. The filter's cutoff frequency, its depth of shelving, or a combination of these parameters, is varied to dynamically change the frequency response of the transmission channel. Such program-controlled filters are most often used as high-frequency limiters to control potential overload due to pre-emphasis in pre-emphasized systems.
Equalization is changing the spectral balance of an audio signal, and is achieved by use of an equalizer. In broadest terms, an equalizer is any frequency-selective network (filter) placed in the signal path. In audio processing, an equalizer is usually a device that can apply a shelving or peaking curve to the audio.
Equalizers are sometimes used on-line in transmission to create a certain sonic signature for a broadcast. Any of the types above may be used. Commercial audio processors may include equalizers for program coloration, or for correcting the frequency response of subsequent transmission links.
A typical audio processing system consists of a slow AGC followed by a multi-band compressor with moderate attack and release time. Correctly-designed multi-band processors have these time constants optimized for each frequency band; the low-frequency bands have slower time constants than the high-frequency bands. This multi-band compressor usually does most of the work in increasing program density.
The amount of gain reduction determines how much the loudness of soft passages will be increased (and, therefore, how consistent overall loudness will be). The broadband AGC is designed to control average levels, and to compensate for a reasonable amount of operator error. It is not designed to substantially increase the short-term program density (the multi-band compressor and peak limiters do that).
Modern audio processing systems usually add other elements to the basic system described above. For example, it is not unusual to incorporate an equalizer to color the audio for artistic effect. The equalizer is usually found between the slow AGC and the multi-band compressor. The multi-band compressor itself can also be used as an equalizer by adjusting the gains of its various bands.
Peak clippers decrease the peak to average ratio, increasing loudness within the peak modulation constraints of the channel. To decrease clipping-induced distortion, some processors use sophisticated distortion cancelling schemes that remove distortion in frequency bands most likely to be audible to the listener.
Various low-pass filters are often included in the system to limit the bandwidth of the output signal to 15kHz for FM, or to other bandwidths as required by the local regulatory authority. The final low-pass filter in the system is almost always overshoot-compensated to prevent introducing spurious modulation peaks into the output waveform.
Highly-processed audio contains many waveforms with flat tops that resemble square waves. The waveshape of a square wave is very sensitive to the magnitude and phase response of the transmission channel through which it passes. Deviations from flat magnitude and group delay over the frequency range containing significant program energy will cause the flat tops in the processed program to tilt, increasing peak modulation levels without increasing average levels. This increases the peak-to-average ratio of the wave, reducing the average level (and therefore the loudness) that the channel can accommodate.
Although the audio to the input to an audio processor may be high-pass filtered, the fast peak limiting or clipping processes occurring in the processor are non-linear, producing difference-frequency intermodulation components below the high-pass cutoff frequency of the unprocessed audio. Even if the audio has been high-pass filtered at 30Hz, these intermodulation products may extend down to 5Hz or less. To preserve the shape of the processed wave, these IM products must be passed through the system without being subject to significant magnitude or phase distortion.
Ordinarily, the audio waveform will overshoot less than 1% if the low-frequency cutoff of the transmission system is 0.16Hz or less. This ensures less than 1% tilt of a 50Hz square wave. Although the waveforms of the infrasonic IM products are affected more by this cutoff than power in the audio band, the audio-band power dominates, so the overall waveshape is still adequately preserved when system LF cutoff is 0.16Hz or less. One obvious consequence of this principle is that a system that passes sinewaves flat to 30Hz may severely distort the shape of processed audio unless its LF cutoff is, in fact, far lower.
The best location for the processing system is as close as possible to the transmitter, so that the processing system's output can be connected to the transmitter through a circuit path that introduces the least possible change in the shape of the carefully peak-limited waveform at the processing system's output. Sometimes, it is impractical to locate the processing system at the transmitter, and it must instead be located on the studio side of the link connecting the audio plant to the transmitter. (The studio/transmitter link ["STL"] might be telephone or post lines, analog microwave radio, or various types of digital paths.) This situation is not ideal because artifacts that cannot be controlled by the audio processor can be introduced in the link to the transmitter or by additional peak limiters placed at the transmitter. (Such additional peak limiters are common in countries where the transmitter is operated by a different authority than that providing the broadcast program.).
In this case, the audio output of the processing system should be fed directly to the transmitter through a link which is as flat and phase-linear as possible. Deviation from flatness and phase-linearity will cause spurious modulation peaks because the shape of the peak-limited waveform is changed. Such peaks add nothing to average modulation. Thus the average modulation must be lowered to accommodate those peaks within the carrier deviation limits dictated by government authorities.
This implies that if the transmitter has built-in high-pass or low-pass filters (as some do), these filters must be bypassed to achieve accurate waveform fidelity. A competent modern processing system contains filters that are fully able to protect the transmitter, but which are located in the processing system where they do not degrade control of peak modulation.
The audio received at the transmitter site should be as good quality as possible. Because the audio processor controls peaks, it is not important that the audio link feeding the processing system's input terminals be phase-linear. However, the link should have low noise, flattest possible frequency response from 30-15,000Hz, and low non-linear distortion.
If the audio link between the studio and the transmitter is noisy, the audibility of this noise can be minimized by performing the compression function at the studio site. Compression applied before the audio link improves the signal-to-noise ratio because the average level on the link will be greater. If the STL has limited dynamic range, it may be desirable to compress the signal at the studio end of the STL. To apply such compression, split the processing system, placing the AGC and multi-band compressor sections at the studio, and the peak limiter at the transmitter.
In some countries, the organization originating the program does not have access to the transmitter, which is operated by a separate entity. In this case, all audio processing must be done at the studio, and any damage that occurs later must be tolerated.
If it is possible to obtain a broadband phase-linear link to the transmitter, use the processing system at the studio location to feed the STL. The output of the STL receiver is then fed directly into the transmitter with no intervening processing. A composite STL (ordinarily used for FM stereo baseband) has the requisite characteristics, and can be used to carry the output of the processing system to the transmitter. Because use of a composite STL has so many ramifications, we recommend this only as a means of last resort - installation of the processing system at the transmitter is vastly less complicated.
Where only an audio link is available, feed the audio output of the processing system directly into the link. If possible, request that any transmitter protection limiters be adjusted for minimum possible action - the processing system does most of that work. Transmitter protection limiters should respond only to signals caused by faults or by spurious peaks introduced by imperfections in the link.