Measuring Audio -- Part One
On a number of occasions on this
site, I’ve alluded to my early days as an audio writer with a Canadian magazine
called Electron. It had been a general electronics publication throughout the
1960s, and was being transformed into an audio-only magazine when I joined the staff. When
the change was complete, it was re-named AudioScene Canada, and was generally
regarded as one of the best hi-fi books around.
One reason for that was that the editor of the day, Ernie
Welling, did his homework. Leading up to the shift in subject matter, he scrutinized the
successful sound publications in the US, Britain, and elsewhere to see what made them
tick. One thing the good ones had in common was a program of equipment test reports based
not on subjective evaluations alone but on hard measurements. Julian Hirsch was doing them
for Stereo Review, CBS Labs for High Fidelity, and a group of experts
including the likes of Len Feldman and Richard Heyser for Audio.
In his search for a Canadian source for such data, Welling
was introduced to Dr. Floyd Toole at the National Research Council in Ottawa, and began a
relationship that was to continue for two decades.
Toole is mostly associated with his landmark work on
testing speakers, but he undertook to measure all sorts of transducers for the magazine,
such as phono cartridges and microphones. In addition, thanks to his ability to tap into
the research of his colleagues at the NRC (notably Dr. Edgar Shaw), Toole was able to
provide what I’m sure were the only actual measurements on headphones published by
any audio magazine anywhere.
But the NRC was not equipped to handle all types of audio
gear. For other components -- amplifiers and tape recorders, turntables and FM tuners --
Welling turned to another similar organization closer to home: the Ontario Research
Foundation in Mississauga, Ontario, just outside Toronto. The audio testing program there
was run by Errol J. Byers, a lesser-known name than Toole, but nonetheless a vital part of
our early testing program.
A couple of years into that program I sat down with Toole
and Byers to discuss the problems and challenges of audio testing. What follows is the
report of the roundtable published in AudioScene in the fall of 1973. It will
continue next month and conclude in October.
Although I wrote the article, my role was mainly to
transcribe what these two experts said. Looking back, I still think it’s one of the
most insightful discussions of audio evaluation ever published by a consumer audio
magazine.
Bear in mind, however, that this conversation took place
more than 30 years ago, when the CD was well in the future, the LP was the high-quality
source of choice, and the cassette was only just becoming accepted as a true high-fidelity
medium. Still, in its context, the discussion below holds up amazingly well.
Ian Masters: To start off, when testing audio
gear -- and we're talking about just audio gear -- is there one parameter that is the
hardest to nail down?
Floyd Toole: Without any question it's distortion.
And I think this is so because it is very difficult to relate the distortion we can
measure to what you can actually hear.
Distortion measurements are a means of evaluating the
degree of non-linearity of a device, whether that device be a tape recorder, amplifier,
speaker, or what-have-you. But there are several methods of measuring that non-linearity.
The commonest one is harmonic distortion, where you apply a test tone to the device, and
measure the level of the harmonics introduced by the non-linearity. The distortion
products as a percentage of the fundamental plus the distortion products is the total
harmonic distortion of the device.
Another way is to measure intermodulation distortion. Some
people maintain that this test is the most useful, and certainly the most valid, method of
measuring the non-linearity of a piece of audio gear because the IM products are the most
objectionable to a listener. Others believe that the harmonic-distortion technique is just
as valid a measure of non-linearity, even though the harmonic-distortion components
themselves are not as objectionable as the IM products.
My attitude is that either technique will work, and
probably equally well. But they both suffer about equally from the difficulty in relating
the numerical results of the measurements to the listener's subjective reaction to that
distortion.
In other words, if you measure non-linearity by both
methods, and you get two numbers -- one a fraction of a percent, and the other, one or two
percent -- which one is the more objectionable? I don't know of anybody who can say with
certainty.
You may well be measuring exactly the same non-linearity in
different ways. The defect is the same, but the numbers that come out of the measurements
are different.
IM: So harmonic and intermodulation distortion
are just two ways of expressing the same thing?
FT: Generally, yes; although there are certain
situations where they can, in fact, measure different things.
Errol Byers: It's useful to note that neither of
these methods gives a complete picture of the non-linearity of, say, an amplifier. When
you measure harmonic distortion and obtain a single number, you have no information about
the frequency content of the harmonics -- what percentage of which harmonic.
This can affect considerably how a unit sounds. If it's 1%
of something ten times the frequency of the fundamental, it's going to be a lot more
noticeable than the second harmonic would be.
FT: Then what you're talking about is total
harmonic distortion, as opposed to discrete harmonic measurements.
EB: Yes, and this is what most people are measuring.
FT: That's right.
EB: That's not to say that a measure of total
harmonic distortion tells you nothing about a piece of equipment's performance. In the
case of an amplifier, absolute linearity is the ideal, and any measure of it is very
useful -- but there is still no information about what it sounds like.
This is where I think the greatest lack of information lies
at the moment -- how much of what kind of distortion can one tolerate in a system?
IM: There may be some confusion about distortion
being an expression of non-linearity. Many audiophiles think of linearity in terms of
frequency response, with distortion as a separate factor to be added to frequency response
to give an overall view of a unit's worth. Could an amplifier, for example, have a
relatively flat frequency-response curve, and still have high distortion levels?
EB: Yes. When you put a signal into a system, you
expect the same signal to come out -- amplified, but with the same waveform. Anything else
in that waveform could be considered distortion. It is distorted in that it is not the
same waveform that went in.
The resolution you have on a frequency-response measurement
is such that a 1% distortion level is 40dB down from the test signal's fundamental. If you
subtract this -40dB level from the frequency-response curve, you simply won't see the
difference on normal charts.
IM: So, essentially, distortion measurements
simply "zero in" on a very fine thing.
EB: Yes. Fine but noticeable.
FT: To give a specific example, you would require
something close to 50% total harmonic distortion before it begins to show significantly on
a frequency-response curve.
IM: So does that mean that very small amounts of
distortion are much more appreciable to the listeners than smallish non-linearities in the
frequency response?
FT: Well, that's not entirely clear, because they're
two different things. To a listener, a deviation from a linear frequency response produces
a shift in balance between the bass and treble -- and even a rather subtle shift in
balance may well be audible, particularly in a comparison test between two otherwise
identical systems.
IM: But that's not likely to be as annoying.
FT: No. I think one can adjust more readily to a
frequency deviation than one can to high distortion or large problems of other kinds.
There are numerous examples of this in our everyday
experiences. In home stereo systems, for example, one can adjust the bass and the treble
balance by means of tone controls, and small adjustments produce audible differences. But
if the adjustment is maintained for a period of time, one tends to acclimatize to it, as
one has over the years to gross spectral imperfections in speakers and phono cartridges,
which tend to have the more serious problems of this type.
IM: I would suppose transducer distortion would
be the hardest to track down by measurements.
FT: What puts the speaker, for instance, in a
special measurement category is that it transduces an electrical signal into an acoustical
one that comes out in three-dimensional form as a wave propagating into the room.
As we measure the speaker from different positions, we get
vastly differing frequency-response and distortion readings. So, in addition to the higher
levels of distortion that tend to be produced in the transducer, there is the accompanying
problem of actually measuring that distortion. You can make several measurements and get
different values.
IM: In a practical sense, would you think that
for a speaker it is more valid to go through this exercise of measuring from a number of
different places in a live environment, or to measure it anechoically and presumably get
just what is being produced by the transducer itself?
FT: Well, both have their problems. The anechoic
measurement has the distinct advantage in that it is repeatable. A live room measurement
presents severe problems because, in addition to changes in the directivity factors in the
speakers, there are standing-wave patterns in the room. The sound reflects from wall to
wall and from ceiling to floor and from pieces of furniture, backwards and forwards.
If one does a frequency-response measurement, in a live
room, for example, it produces a "mountain range" type of curve, in which the
peak-to-trough differences may be of the order of 15 or 20dB. This magnitude of
fluctuation wreaks havoc with the distortion measurement. It would be possible, by
judicious placement of a microphone, to create an unrealistically favorable or unfavorable
measurement -- particularly at low frequencies.
IM: In view of that, are speaker distortion
measurements valid at all? Given a particular location in a particular room, and the same
program material, the room will presumably do the same thing no matter what the speaker.
FT: As long as the relative position of speaker and
listener are maintained that is true. The problem is most serious at the low frequencies;
and this occurs because, at low frequencies in normal-sized listening rooms, the normal
modes -- the resonant frequencies of the room -- are rather widely spaced.
It is entirely possible -- in fact, demonstrable -- that a
problem in a particular speaker may not be audible. A speaker may produce measurable and
audible distortion in an anechoic chamber or in a given room; but, in a different room,
not exhibit this audible distortion at all, because of coincidence of the room mode with
the fundamental frequency.
If the fundamental happens to be enhanced by the room mode,
and the harmonics happen by chance to be diminished by the absence of a resonant
frequency, then that distortion may well not be audible. The converse also can be true. A
speaker in a certain room may produce audible distortion, but in another position in the
same room -- or in a different room -- it may produce distortion that is very low indeed.
IM: But all of this seems highly unpredictable
from room to room.
FT: It is, yes.
IM: To what extent, in that case, does even the
repeatability of the anechoic measurements relate to a real environment? If you have a
speaker with say, 1% distortion, measured anechoically, and another with 5%, there's
nothing to prevent the second from having less distortion in a particular live
environment.
FT: Yes, that's possible. But we're talking about
specific frequencies, of course -- frequencies that happen to coincide with particular
room modes.
IM: But how do you rationalize a meaningful
measurement of speakers?
FT: The rationale behind the clinical anechoic
measurement is that it is at least a measure of the potential of a transducer. A speaker
will be used in many different environments -- as many environments as there are rooms and
positions in them, really.
You have to look at it statistically. A speaker that, in
the anechoic chamber, produces low distortion at all usable frequencies and power levels,
will most likely sound better than one that has an inferior performance.
...Ian G. Masters
ian@mastersonaudio.com
|
