More and more stations are getting program content to listeners via data-reduced means, such as netcasting and HD Radio. Fortunately, more audio processors are being built with feature sets meant to accommodate this trend.
Is there really any need to process audio for these two applications? Of course there is. Let's take a look at the similarities and differences between HD Radio, netcasting, and analog AM/FM radio.
For HD Radio (specifically the HD1 signal, or the digital version of the analog transmission) there is no technical reason for processing beyond keeping the program audio encoded in the AES data stream from hitting 0dBFS. However, in practice, as the HD Radio receiver fades audio from the analog version to the digital, and vice-versa, it is pleasing to the ear if the two audio paths sound the same. This requires some amount of processing for the audio encoded into the AES data stream for HD1.
The masking effects in the frequency domain: A masker inhibits perception of co-existing signals below the masking threshold.
Like its HD1 brother, HD2 doesn't technically need any audio processing either; however, many of the practical reasons for audio processing that are relevant to AM and FM are also of importance to HD2 and streaming audio (netcasting) as well.
One of the most important reasons for audio processing is simply to combat ambient noise levels in a typical listening environment. For in-car listening, we have to combat the effects of wind and road noise. No driver wants to have to continually move his volume control up and down on a source-by-source basis as he goes down the road. In a home listening environment (the kitchen, for example) we have to contend with multiple sources of background noise. In an office environment there are many sources of noise, and because the space is generally shared you have to consider your office mates as well. No one wants to hear the neighbor's radio blaring.
Another important reason for audio processing is to add that sonic signature that so many programmers (rightly) insist that a radio station have This can also be done for netcasting. Programmers and engineers alike are going to seek a signature sound for their HD2 stations as well.
The masking effects in the frequency domain. A masker inhibits perception of co-existing signals below the masking threshold.
So, many of the basic reasons for audio processing are as important in the streaming audio application, and in the HD Radio application, as they are in the more familiar AM and FM media. Yet there are major differences in the way processing is accomplished between the related media.
Don't cut the losses
The Wheatstone Vorsis is designed for production or on-air use.
With respect to netcasting and HD Radio, the lossy codecs that are used in the building of the data stream that we eventually hear coming out of speakers on the other end all work within a few basic precepts. First, in their emulation of the way human hearing actually works, the audio spectrum is broken into discrete critical bands, each having a differing bandwidth. At 100Hz, for example, the bandwidth is about 160Hz. At the 10kHz center frequency, the bandwidth is about 2.5kHz. Based on the level of the strongest signal in a particular band, and the center frequency of the particular band, the codec calculates the masking threshold, which is the level below which the human ear, while encountering the loudest signal in the particular band, will not be able to hear.
The codec then throws out signals below the masking threshold, and sets the number of bits used to encode the loudest signal in that particular band. The number of bits is reduced to the point that quantization noise in that particular band will not be perceptible to the human ear listening to the audio coming out of the decoder.
There is masking not only in the frequency domain (as described above) but in the time domain as well. A loud sound can mask quieter sounds for up to 200ms after it; this is called backward masking. A loud sound can also mask sounds that come before it — up to about 15ms. This is called forward masking. Sounds that would be masked in this way are effectively thrown out by the codec.
The Omnia 6EX is the premium version of the company''s series of DSP-based processors.
Lossy codecs take advantage of the way human hearing works, in the time and frequency domain, by not encoding sounds that are ultimately imperceptible, and by efficiently using the minimum number of bits to encode the sounds that are perceptible. The ultimate goal is to fit as many useful bits as possible through what is typically a data-bandwidth limited path between the program source and the far end — our listeners. In the case of HD Radio, the amount of payload data is limited (now) to 96kb/s. In the case of netcasting, the cost of providing the service is usually proportional to the data rate of the stream times the number of listeners. We all want to provide the highest data rate to an unlimited number of listeners, but in practice we can't afford to do that. The compromise is to limit the data rate of the stream. The effectiveness of the current generation of lossy codecs makes the listening experience acceptable (if not downright enjoyable) to the majority of users.
One of the most effective tools for audio processing in the analog domain is old-fashioned clipping. The amount of clipping used in the analog media of AM or FM is, of course, subjective and usually dependent on an agreement between programming and engineering. However, clipping is not effective when it comes to processing audio that will go through a lossy codec system. The reason behind this is quite simple. If you were to clip a pure sine-wave, what you would have on the output side of the clipper would be the fundamental signal (the sine wave you put in) plus harmonics that each have a level and time relationship to the fundamental. If you now pass that clipped sine wave through a transmission system that does not have the adequate frequency and phase response so that the harmonics arrive on the other end in the same level and time relationship, you will find that you no longer have a clipped sine wave; what was a nice flat-top at first is now slanted to one side. The clipping is negated. This is what would happen in the lossy codec. The lossy codec could simply not encode the harmonics at all, or, it might waste bits by encoding the distortion products (i.e. the harmonics) at a low bit rate, which could be noted on the far end by quantization noise in the critical bands that contain the harmonics. Not good.
Because 75µs emphasis is used in the transmission of analog FM, analog FM audio processors include a high-frequency limiter working along that emphasis curve. A typical trick to build loudness on analog FM is to “get in” to the HF limiter more. HD Radio and streaming audio applications are “flat” though, and therefore there is no reason for HF limiters in those applications.
The Orban Optimod-FM 8500 is the latest in a long line of the company's on-air audio processors.
There is still a need to provide absolute peak protection in a digital system, though, you will not like the sound of your audio encoded in to an AES data stream as it hits 0dBFS.
This can be accomplished by the use of a look-ahead limiter. The HD Radio or netcast audio processor includes a delay line in the audio path and uses that delay to determine where in the time domain it needs to limit a peak, and then does so at the appropriate time.
So, you can see that some processing functions commonly used in analog processing, like clipping and pre-emphasis limiting, are of no use for HD Radio or netcast processing. On the other hand, wide-band automatic gain control (for level consistency, source to source), and multi-band gain control (for greater intelligibility, along with assertion of a sound signature) and absolute peak control are useful for HD Radio and netcast processing, as they are in analog processing.
Remote control of processors through RS-232 or other means has become common. The Broadcast Warehouse DSPX is one such unit.
The current crop
Wheatstone has introduced the Vorsis AP-3, a 1RU unit audio processor with built-in three-band AGC, parametric EQ, de-esser, downward expander and peak limiter. Control can be achieved via the front panel, via local computer control or via Ethernet.
Broadcast Warehouse offers its DSPX, a 1RU audio processor featuring the typical elements of a broadcast audio processor: wideband AGC, multi-band AGC and limiting, HF limiting and a stereo generator for FM applications; plus a separate output that makes use of look-ahead limiting (as opposed to HF limiting) for streaming audio applications. Control is accomplished via the front panel, RS-232 or Ethernet.
FM audio processors on a budget
The price of DSP-based audio processing has come down with more competition and maturing of the technology. There are several manufacturers building less expensive versions of their flagship models.
Omnia has the Omnia 3 Turbo, a DSP-based audio processor that included a three-band AGC and a multi-band limiter. A digital stereo generator comes standard. This unit succeeds the Omnia 3.
Orban makes the 2300, a digital audio processor with a two-band architecture and a built-in stereo generator. AES I/O are included, along with remote control via GPI, RS-232 or Ethernet. This unit is the successor to the Orban 2200.
Broadcast Warehouse makes the DSPX-mini, a 1RU simplified version of the DSPX. This unit includes AES I/O, four-band architecture, a DSP-based stereo generator with clipping, and RS-232 or Ethernet control.
You aren't limited to DSP architecture if you happen to be looking for new audio processing.
Inovonics produces the David-III, a multi-band audio processor with a digital-synthesis stereo generator. Also included are limiting and composite clipping.
CRL manufactures the Amigo, a FM processing system with dual-band AGC, stereo enhancement, a multi-band limiter and digitally synthesized stereo generator.
Broadcast Warehouse has also recently introduced the DSPXtra, which features the Ariane (by Translantech Audio) RMS leveler as its front end.
Omnia offers a DSP-based FM audio processor known as the Omnia-3; the Omnia-3net unit is specifically built for streaming audio applications. If you are about to delve into streaming audio, you may want to consider the Omnia A/X, a software-based processor that runs on a computer and works in conjunction with Windows Media, Real and MP3 streaming encoders.
Omnia's flagship processor is the 6EX (or 6EXI) and it retains the modern features that you would expect. Its processing engine is totally DSP based. Some of the most basic features are a six-band limiter with adjustable cross-over points; a built-in (DSP-based) stereo generator with a composite clipper; and the now-ubiquitous Ethernet connectivity. The 6EX features a simultaneous processing path for HD Radio; and the 6EXI option includes a diversity delay that allows the user to avoid the delay line built-in to HD Radio exciters, thus completely isolating the analog transmission from the HD Radio equipment.
Omnia offers AM processors as well; the 5EX and the 5EX+HD. These share many of the same features with the FM brothers — AES inputs and outputs, DSP-based processing and Ethernet connectivity.
Orban's flagship product is the 8500. Its processing functions are completed by DSP; AGC, multi-band, stereo encoding and composite clipping included. It also features a parallel processing path for digital radio included as standard, along with a built-in diversity delay. Communications with the 8500 is accomplished with something as simple as contact closures from your remote control, or via RS-232 (like an old-fashioned dial-up modem) or via its TCP/IP interface, which comes standard.
Orban also offers the Optimod 1100PC, which is a professional sound card designed specifically for streaming media applications. On-board DSP performs the typical audio processor functionality — AGC, multi-band, EQ and look-ahead limiting. It also can function as a mixer because it has one stereo analog input, two AES or S/PDIF inputs with sample rate converters, and one WAV input.
Orban offers Opticodec-PC, which is encoding software that makes use of MPEG-4 AAC/aacPlus encoding technologies. It's available in multiple versions, and able to run on all the standard platforms.
Orban has a new product to be introduced this spring: the 9400. This processor has two parallel processing chains (the wide-band AGC and stereo enhancer are all that are in common) allowing it to serve as an on-air processor and HD Radio (or streaming audio) processor simultaneously. Unlike its brother, the 9200, this unit serves as a stereo audio processor in both domains allowing the user to process for C-QUAM should they need to do so.
How important will audio processing be for the future of radio? The universe of choices that a tech-savvy listener now has is practically mind-numbing: aside from HD Radio, and the recently introduced HD2 channels and accompanying streams, there are countless other streams available from around the world. Wi-fi radios will soon be here en masse. No longer will you be competing against your buddy across town, or even a half dozen stations. The radio listener will no longer be limited to the carriers in town. With that in mind, the features of streaming audio, processing amongst them, will become more important than ever.
Manufacturers of audio processors and processing accessories
Irwin is director of engineering at Clear Channel, Seattle.