[net.audio] Crossover analysis

rdp@teddy.UUCP (10/11/85)

[]

I have, for a number of years, been using SPICE to perform analysis
of a variety of audio-oriented circuits, most notably crossovers. (I
have also analyzed a variety of things like phono preamps and one
MOSFET power amp as well).

SPICE is a complex electrical circuit analysis tool developed by
Lawrence Nagel, et al, at the U. of C. at Berkeley. It provides for
all the standard linera components (R's, C's, L's, sources) as well
as non-linear components, and handles most active devices such as
BJT's JFETS and MOSFETS.

I have had the opportunity to examine quite a few different networks
and would be glad to share some of my findings in some detail if there
is enough interest. Also, for the sole purpose of information gathering,
I might be coerced into analyzing networks that I haven't seen before.
(This presumes that I am seeing them now, of course :-)).

One general conclusion can be drawn from my experience. The vast majority
of loudspeaker "designers" haven't got a clue. Most of the networks I have
seen fall into two categories: cookbook and hacked. The former makes the
mistaken (albeit implied) assumption that the drivers are completely linear
in their output and provide a constant, non-reactive electrical load. The 
second seems to be as a result of breadboarding blindly until it sounds 
correct, a practice that, in my own experience, is doomed to failure.

In my analysis, I attempt to model fully the characteristics of the driver.
(usually, the best that I can get is a true representation of the electrical
load presented by the driver, but this is enough to infer quit a bit about
how the final system will interact).

The best thought-out networks seem to, in the most part, come from some of
the British manufacturers. Two come to mind immediately, KEF and B&W. Both
seem to base the design of their networks on the physical reality of the
drivers, both in acoustical characteristics and presented electrical load.
The very worst seem to be the almost exclusive domain of the American
"high-end" nutso manufacturers (Symdrek, are you still out there?).
On the other hand, by far the most bizzarre network I ever saw also was
British in origin. It appeared recently in Wireless World, and was designed
for the venerable KEF B-110 (a 5-inch bass-midrange driver) and the
Audax dome tweeter. It had 12 components in the tweeter portion of the
network and  16 in the woofer portion, totalling 28 components altogether!
There were resonant traps and this thingy and that. The network itself
suffered from a minimum constant loss accross the band of 3 db, and the
output was just soooo srtange. Couple all this with the fact that it just
did not do well when combined with the drivers is was alleged to associate
itself with, well...

Too few of these sorts of design tools are used (maybe because they are not
readily available, but also because the designer has no knowledge of them,
nor of normal electrical design methodologies), and much of the stuff on
the market is such junk because of it. Note that I am not trying to imply
that there use will ensure success. Here I must relate the incident I 
came across at JBL (whose name I won't mention). Back in the mid-70's, they
were using an HP dual channel digital audio spectrum analyzer for QC on
the woofer production line. On one channel they had stored the response
of a "standard" woofer, and one the other, was the unit under test. After
subtracting the two, if the resultant trace deviated outside of a critical
area, then the woofer was rejected. Granted, it was a trifle simpleminded
and very limited in the scope of things tested, but it was better than nothing.
Or so it seemed at the time. Dick Marks (I think that's his name) was the man
responsible for much of the medium-size woofer design at JBL. One day, he 
wanted to build a pair of speakers for himself, so he went don and got two out
of inventory. They sounded absolutely atrocious (but identical!)!. He
went to the QC department, and enquired as to why they were so bad. "We
can't understand", said QC, "they absolutely passed inspection, see, here's
the charts". Sure enough, the deviation from the "standard" was well within
tolerances. He ran the drivers through QC again, and they failed, miserably.
On a hunch, he went and got a driver from the previous day's production run.
They also failed! All of today's drivers passed with flying colors. A 
mystery! He asked the director of QC how the standard was selected. The
reply sent us all either laughing or barfing, "Well, we take the first
driver off the production line every morning, and use it as the standard
for the day." "How long", he was aked, "have this policy been in effect?"
"Oh, " came the reply, "a couple of years, at least". "What if, " came the
inevitable final question, "that driver is terrible?" This was greeted
by a long, puzzled silence.

In any case, if this sort of discussion is of interest to many of you,
then I shall collect my notes and fire up SPICE.

Otherwise, I won't!

Dick Pierce