Moulton Laboratories
the art and science of sound
How High Do We Really Need To Hear
Dave Moulton
April 2000

What Really Happens To Them Ultrasonic Signals, Anyway?

How High Do We Really Need To Hear

Our Story To Date . . .

Alert readers will recall that last month I ranted about audio resolution, and promised that this month I’d rant about audio bandwidth. The underlying issue that I have been ranting about for the past couple of months is the “audibility” of our new high resolution formats, specifically 24-bit word length and 96 kHz. sampling rates.

The issue of bandwidth is a fascinating one. On the one hand, the high-frequency limits of our hearing seem pretty well established at somewhere between 14 and 20 kHz., depending on our age, gender, exposure to race cars, etc. On the other hand, we KNOW that there is a lot of significant acoustical information extending well above 20 kHz. and speculations abound about the possible effect that such information may have upon our perceptions. Food for thought . . .

Meanwhile, the rules of digital audio bandwidth are pretty unequivocal. The Nyquist frequency (you know, 1/2 the sampling frequency – the upper limit for bandwidth in a digital system) has proved inviolable, and as we’ve gotten more experienced at digital stuff, we’ve come to notice that we’ve got to be really careful to get everything filtered out at some point BELOW the Nyquist frequency to head off sonic nasties. This means, of course, that if there IS something useful for us above 20 kHz., well, the digital audio system is for sure taking it out. Being the paranoid lot we are, we suspect the worst, to wit: that we are losing some ineffable ultrasonic signal illumination that reveals artistic truth and purity. Moan, groan, whimper, whine! Hence the big push for the 96 kHz. sampling rate. We ain’t sure we ain’t losing something, and we sure wanna be sure we ain’t!

Tales From the Audio Crypt

This is all fine and dandy. However, let me share with you two experiences I have had.

Tale #1. About two years ago, while attending an AES Convention, I was in a meeting of the Psychoacoustics Subcommittee of the Technical Council and we were discussing a proposal for a new wide bandwidth digital format (100 kHz.!) when one of the committee members asked a really simple question. “Hey,” he said, “does anyone here know of any studies that convincingly show that humans hear above 20 kHz.?” And we all just sat there. No one raised their hand. Not one of us knew of anything like that!

So much for what the people who are really interested in what and how we hear can tell us. Above 20 kHz? No evidence! Zero, zippo, zilch.

Tale #2. Last Fall, I attended the Sony/Phillips SACD demo at the AES convention in NYC. Now SACD is a 1-bit format with a sampling frequency of slightly less than 3 megahertz, which means the bandwidth is going to be really pretty wide. Before one of the recordings played, the presenter (an engineer from Phillips) murmured “And this is a copy of a special analog recording that was done on a specially tweaked analog deck flat to 50 kHz. . . .” That murmur got me to thinking. After the presentation (which was really pretty nice, by the way), I asked the presenter if anyone had bothered to measure the acoustic bandwidth throughput of the SACD system, which is this geek’s way of asking, “How high a frequency can really pass through this system, from microphone input to loudspeaker output?”

No, they hadn’t measured such a thing, he said, and he had no idea what the figure might be. However, he went on, the recording had been done with B&K 4006s, which were good to about 30 kHz., he figured. He went on to note that to get the bandwidth much higher, you’d have to seriously reduce diaphragm size, which in turn would increase the already troublesome equivalent input noise of the mic.

Now this IS seriously interesting. I’d never thought of it this way, but there is an inverse relationship between noise and bandwidth that, in microphones, is actually meaningful around the limits of our hearing. If we extend the bandwidth, we reduce the resolution! Yet another engineering tradeoff! Damn!!

Meanwhile, I’ve been hard at work on wide-dispersion lenses for tweeters, and am passionately interested in their power response. The default tweeter we use for prototypes is roughly flat to 30 kHz. However, we’ve found that at that frequency the tweeter’s beamwidth is down to a few degrees and its power (relative to, say, 15 kHz.) is practically nothing.

What About Total System Response?

What does this all mean from a system bandwidth standpoint? Well, we’ve got to define “the system.” It includes the microphone, all the audio devices, and finally, the loudspeakers.

Music recording microphones are, at their best, limited to perhaps 30 kHz., and the really quiet microphones that we’ve come to know and love are going to cave in at about half that. Analog audio gear, except for recorders, can go casually to >>100 kHz., no problemo. Digital gear is band-limited to 20 kHz., unless it’s 96 kHz. stuff, in which case it’s good to about 45 kHz. Good loudspeakers make it to about 15 kHz. with any kind of reasonable power output. That pretty much describes the high-frequency limits of the system – all the stages seem to do OK to 15 kHz., but it gets ratty and inconsistent above that. And humans seem to hear up to about 17 kHz. on average, but few of us hear past 20 kHz. at best.

The Straight Skinny!

The bandwidth limits of a system are, in essence, defined by the response of the most limited stage in that system, a fact we often fail to keep in mind. Given that most of the stages in the system begin to seriously fall off around 15 kHz., it is comparatively useless to tweak one stage out to 50 kHz.

The real engineering approach is to work on all of the stages and get control of what I call the throughput. When we have a SYSTEM with really smooth response (say, +/- .5 dB) from 30 Hz. to 17 kHz., from microphone diaphragm to eardrum, well, then we really have something to brag about. And UNTIL we can do this, it doesn’t do a lot of good to invest a lot of bucks in dramatically extending the response of a single stage.

This is particularly true given the irony of the microphone tradeoff the engineer from Phillips described. As he so eloquently pointed out, we can have wide bandwidth OR wide dynamic range, but we can’t have both, if we are going to include microphones in the system! And that, it turns out, is a fundamental system limitation. It’s worth keeping in mind.

Thanks for reading this.
Note: The following group of columns that I wrote for TV Technology are an attempt on my part to describe some of the issues surrounding our attempts to measure and evaluate the audibility of high-resolution formats. Together, I think they make an excellent short survey of these issues. I hope you find them useful.


UK     Jan 03, 2013 05:51 AM
Wireless World did a lot of reporting in the early 50,s on hearing ability. Some reports showed that the human ears can detect a difference in arrival time of as little as 1micro second, which is how people can locate a sound source accurately. You might be interested in locating some of these articles and incorporating the findings into your excellent notes. Telephone companies used to test telephone microphones and earpieces in minute detail, using groups of listeners for intelligibility. This was done for at least 20 years to my knowledge. The results should be published in one of the telecommunications journals.

I would like to see an analysis of TV/DVD sound audio bandwidths(not sampling rates), as I am convinced by my ears that the received bandwidth can be extremely narrow and the sound highly distorted.


Geoff Duddridge
Geoff Duddridge 
france     Jul 13, 2013 06:01 AM
I don't speak very well english, hope you will not judge it.
I think a important point to note is both in analog and digital worlds, problemes occurs often in the highest limit of bandwith. It's true for distortion in an amp circuit or a loudspeaker. And It's true for a plug-in dealing with dynamics or wathever, feed it with a 44.1 signal flow and it will be less accurate in the 15-20kHz band. Perhaps the only interest in increasing the bitrate is pushing the distortion and algorithms inaccuraties out of the earing range ? The differences I ear betwen 44.1 and 88.2 or 96 signals is for me just because most problemes have moved to an inaudible spectrum range.
Best regards, you learn me a lot and you are always a reference when I search a way to do it better.

Post a Comment


rss atom