Understanding Clipper Induced Intermod Distortion
Quite often I’m asked if there will ever be the ultimate audio processor, or if the limitsin this unique combination of art and science will ever be reached? My answer has been, and continuesto be: “No, it will not!” Personally, I’m always playing around with some processing idea. Not a week passes by, when one of the Team-Omnia crew, or myself, aren’t burning the midnight oil oversome concept, either to raise the bar ofsignal processing, or investigating some new frontier.
Aslong asthere are new fresh innovative thoughts about this stuff, and we continue to find platformsto bring ideasto life, processing innovation will continue. It’s no different than the personal computer. No sooner do you purchase the latest whiz-bang PC, then that same company releases the next bigger-faster machine. Frustrating, I know! Processing is no different! Just as your latest laptop PC is an investment to improve your workload efficiency, your next processor will be an investment, yet again, to improve on the dial.
This paper is written as a “state of the audio processing union” message. Much has transpired over the last few years in processor design, along with the mediums that need these black boxes. As of this writing, broadcasting requires specialized processing for conventional FM/AM, HD Radio/DAB, Satellite, Streaming, and DTV. As new technologies emerge, the TO DO list continuesto grow.
Can You Hear Me Now?
To quote an obnoxious cellphone ad, “Can you hear me now?…Good!” Present generation on-air audio processors are capable of being competitively loud. It’s a safe bet that in most markets, where modulation levels are close to the same, the loudness difference among the dial-dominatorsis within 0.5dB of one another. Most audio professionals will debate that you cannot hear the level difference of a 0.5dB. Well we need to remember that moderate to dense broadcast processing yields significant RMS signal levels. Since the human is a very good RMS detector, it will easily recognize a 0.5dB difference, when moderate to dense audio levels are present. With moderate processing, a level difference of 0.5dB is perceptible between competing radio stations.
Loud audio processing exists today. This has been achieved using algorithms that reduce, or mask, harmonic distortion so the ear will accept more distortion than it’s actually hearing. That’s the premise behind most processors. As competition rises among broadcasters,so does the amount of processing employed. As the amount of processing increases, other forms of distortion creep in. Aside from harmonic distortion (THD), there are two other types of distortion that must be dealt with: aliasing and Intermodulation Distortion (IMD). Omnia wasthe first to tackle the digital aliasing problem, yet it still seems to be a challenge for others, but the nemes is that now plagues competitive processing isIMD.
Intermodulation distortion (IMD) has a distinctive sound. Whereas harmonic distortion (THD)sounds like a broken component ,IMD is that packed up, bizzy, rattling, warbled, texture that appears with dense/heavy processing.It can be created a few different ways. The easiest is when time constants in a dynamic section, like a compressor or limiter, are set too fast. The resulting pumping, bizzy, and smashed quality, that’s heard, in the intermod. This problem isfixable via adjustment to the timing networks, usually the release time, or a design that can mask IM distortion. Modern processors have these functions.
Clipper Induced IMD Explained
Another form of IMD that is very annoying is the effect created when moderately heavy clipping is used. In this case, lower frequencies that reach the clipper threshold, will force all other higher frequency signals, that are present, into the clipper as well. Even if these other signals are below the threshold of clipping, they have been modulated in and out of the clipper process. A good example is sustained audio content that contains a vocalist and bass instrument. As the bass signal reaches the clipper threshold, it will push the vocal range in and out of clipping, even if the vocal is below the threshold of clipping. This results in the vocal sounding warbled, asif the singer is underwater. The obvious fix to this would be a reduction in the clipper, but then competitive loudness is lost. Reducing bass, via equalization, might eliminate the problem, but then bass detail is lost.
Spectral analysisis an easy method for viewing harmonic content. Following are spectrographs that will illustrate a twin-tone situation that is used for observing IMD. The first graphic is of the fundamental frequencies of 50Hz and 1kHz. Notice that they are harmonically clean.
Raising the level of the signalsto engage a clipper function by 6dB, the following is a spectral analysis of intermod products that are created.
The intermod products are the added spursthat are paired up around each of the harmonics of the 1kHz fundamental. This same condition occurs even if both frequencies are at or below the threshold of clipping. The summed peak value of the two signals, if they are in phase, will crossthe clipper threshold and intermod products are generated. These are the signals that cause the annoyances from dense clipping.
Modestly speaking, we are our own worst critics when it comesto audio. Since the question we’re asked most often, is “How can Omnia achieve more?” We sat down, and tried to do just that. Taking the time to challenge ourselves, we observed that after really cranking on the processing, the signal was definitely loud, moderately low in perceived harmonic distortion, but the level of intermod was definitely apparent. In fairness, we observed this with all other processors we tested.(We try to be an equal opportunity offender!)
Reducing clipper induced intermod distortion is easier said than done. Research has shown a few attempts at it, but most of these are dynamics processor sections that are wrapped around a clipper function. They will work some of the time and on some material, but they fail to consistently reduce intermod on a broad range of content. Modifying the clipper to behave as a look-ahead limiter doesn’t do the trick either. All of the harmonic distortion is replaced with added intermod content, thereby creating even more bizzinessto the audio.
Transmission System Contributions
Before embarking on a new doo-dad, it’s worth examining a few other transmission system elements that are obstacles for an audio processor. The biggest challenge is the management of emphasis networks. Both FM and AM employ them to improve SNR and/or overcome receiver limitations. Preemphasis used in broadcasting establishes a +3dBbreakpoint at 2.2kHz and boost will escalate at either 50µs or 75µs, depending upon location about the globe. Even though a multiband limiter is used to manage the emphasis boost, there will remain spectral level uncertainties along the emphasis curve, although they will be radically reduced. As frequency and level rise, their clipper threshold is reduced by a like amount. This makes the upper vocal range much more sensitive to clipping, even if only by a few dB, as compared to the low and mid frequency range. Here is where trouble begins. Upper midrange and presence frequencies are pushed into clipping by large lower frequencies and IMD develops.
Another contributor isthe Gibbs Phenomena of the 15kHz low passfilter. In simple layman’sterms, this is what creates the overshoot characteristics of a low passfilter, when a square wave is passed through it. Under ideal conditions,such as a phase linear filter, overshoots are kept to a minimum of 1.09, or 9%, of the original peak level. Additionally, the upper 1/3 of the filter’s spectrum will be subject to this effect due to the removal of any third harmonic from any square wave that has a fundamental at 1/3 the low pass Each cluster of “spurs” represents IMD productsuality, Clarity, and Competitive On-Air Processing: Understanding Clipper Induced Intermod Distortion! filter’srange or above. Asthe third harmonic isremoved, the square wave changes to a sine wave. Following is an illustration of the Gibbs Phenomena.
Now consider the 15kHz audio spectrum for FM-Stereo broadcasting. The Gibbs Effect will consume just about all of the preemphasis curve asthe effect will kick in at 5kHz, which is 1/3 the cutoff frequency. Preemphasisincreases gain significantly within the upper third (5kHz – 15kHz) of the filter’s passband, in what can be called the Gibbs Region. This yields high levels of clipper activity producing significant density along with severe overshoot content. Note: This is reset to 10kHz for those AM broadcasters using NRSC response, and even narrower for those whom choose to use 5kHz.
Properly designed audio processors employ methodologiesto eliminate overshoots caused by Gibbs Phenomena. A subsequent clipper is not viable after the low passfilter asrecreates harmonic spectra that the low passfilter removed. Any processing system thatrelies on thistype of tech is poorly designed asit yields additional distortion due to the cascading clipper sections.
Of concern isthe signalrange between DC – 5kHz, especially the range between 1kHz and 5kHz where fundamental frequenciesfor vocals, and string instrumentsreside. Also, based upon the Fletcher-Munson curve, it isthe area of maximum hearing sensitivity. Note that the breakpoint of preemphasis occurssmack dab in the middle of this 1kHz – 5kHz range, thus making clipper induced IMD that much easier to perceive.
What transpires asIMD is a summation of preemphasis, effectsfrom Gibbs Phenomena, and where human hearing is mostsensitive. Throw in some moderately aggressive processing in front of the clipper, and trouble arises.
Out Of The Box Thinking…LoIMD Clipper!
In giving this scenario deeper thought, the ultimate objective would be a clipper that had enough brainwave activity to understand they type ofsignal content and density that was applied. Allow the clipper to truncate peaksthat exist in a specified dimension, but also know when to adaptso that sustained signals do not intermod among otherspectra. An algorithm of this nature allowsthe clipper to retain power and impact, yet reduce or eliminate IMD from moderate to aggressive levels of clipping. The overall effect would be retained loudness and punch, but more importantly restoring clarity to the signal.
This is not the same as reducing the drive to the clipper like a dynamic section would do, or modifying a multiband limiter to manipulate the frequency domain. There is a difference. A dynamics processor will modify the overall RMS/Average level, and affects loudness consistency. Additionally, the timing of the dynamics section is no tfast enough to adjust levels to eliminate IMD imperceptibly. If adjustment is Quality, Clarity, and Competitive On-Air Processing: Understanding Clipper Induced Intermod Distortion! performed in the frequency domain using multiband limitng, the spectrum is modified to a point where holes in tonality are heard. Intermod is reduced, but at the same time, processing consistency is eliminated.
Investigation of the clipping mechanism used in Omnia process or sreveal that the conditions to create IMD are present in the same signal used to eliminate processing induced aliasing distortion. Through further analysis of this, IMD symptoms are monitored. Then, the gathered information is used to modify the clipper algorithm to reduce, and under many conditions, totally eliminate IMD.
This lead to the new LoIMD Clipper design that suppresses IMD when moderate to heavy clipping is used. End result: Preservation of clarity and detail in the audio, even with aggressive processing! Much easier said then done, but the diversity delay requirementfor HD Radio gave reason to rethink audio processing in a different dimension.It makessense to put thistime to use forreal work and not juststatic delay. Intellectual property protection does not allow further disclosure about LoIMD, but the following spectrographsreveal performance capabilities. For the purposes of discussion, 100Hz and 5kHz are used.
The first illustration is during initial impact of all frequencies on the clipper. High levels of IMD are easily observed with sideband pairs around each of the 5kHz harmonics. Thisis what occursin a clipper design, even when it contains distortion-canceling methods, asthat pertainsto harmonic distortion.
As density to the clipper increases, the algorithm beginsto modify clipper behavior and IMD productsstart to reduce.
After prolonged clipping activity, approximately 3ms, further behavior modification of the clipper has greatly reduced IMD.
These illustrations are based upon the sustained signal conditions as described earlier between bass and midrange/presence content. No longer will lower frequencies cause midrange signalsto warble. Instead, midrange content is held at, or just under, the main clipper threshold, and clarity is restored. When content contains wider dynamics, the algorithm allows the clipper to operate more historically. If observing the dynamic audio signal on a scope, less squaring of the waveform is observed, even during moderate levels of clipping. This further improves transient detail on short duration peaks.
Remember the earlier spectrograph depicting IMD among 50Hz and 1kHz? Well, here are the before and after pictures once this new algorithm is applied.
In this age of bigger, better, and who’s louder, broadcasters need to realize that their audience now has many more choices of what to listen to. If they choose to process for the sake of loudness at all costs, the results might find listeners migrating to better sounding alternatives. It’s vitally important to remain competitive sounding with respect to otherstations on the dial, but it’sjust as crucial, if not even more so, to offer quality along with the competitive loudness quotient. For this to happen, processor design must raise the bar, yet again, with fresh innovation that yields clarity, quality, and competitive loudness.