Don  Lancaster's Guru's Lair What's New? 2005 Archive
auctions bargains books contact email home library map ?new? rss

January 2, 2006

Refurb Log:

Items are typically very easy or impossibly difficult to
refurb, and there often is a very fine line between the
two. Sometimes a little patience and rethinking that is
combined with a very unusual tool or two can salvage
any marginals. Plus bunches of blind luck.

Ferinstance, I got this otoscope medical device that had
its threaded locking ring epoxy seal break away from
its battery case. And some epsilon minus apparently
cross threaded the ring while attempting a fix.

Stripped threads are often bad news, especially if
brass and large diameter fine threads. But by reversing
the ring and using the good "exit" threads and some light oil,
the ring was first gotten to work properly backwards, and then
even frontwards. Much of this was a combination of patience
and blind luck. And using just the right amount of force.

To reassemble the device, a special spanner wrench is
usually used that lets you twist two holes in a ring that
are spaced an inch apart at the bottom of a six inch
deep case. Such a tool normally costs a lot more than the
otoscope itself. But an ordinary stainless steel dinner fork
can be destructively bent into a tool that is just barely enough
to get the job done.

More on related topics in our Refurb Log.

January 1, 2006

Got one of the usual late night phone calls from an
individual who just invented a new way of marketing and
wanted to run out and get a patent on it to "protect" their
invention. The individual, of course, had never really
marketed anything themselves.

For openers, a patent in no manner prevents anyone from
stealing your ideas. All a patent does is give you a right to
sue someone should your idea in fact get stolen. As the
typical enforcement cost of patent litigation is in the
$300,000 range (yearly litigation insurance alone normally
costs more than $90,000), the cost of a getting a patent is
pocket change that simply does not enter into the economics.

There is, of course, not one patent in one thousand that cannot
be busted outright by a diligent enough search for prior art in
obscure enouth places. Nor does more than one patent in
five hundred ever return a net positive cash flow.

I may have mentioned this a time or two before, but any
involvement whatsoever by an individual or small scale
startup with the patent system is VIRTUALLY CERTAIN
to result is a HUGE net loss of time, energy, money, and
sanity.

Much more on these topics in our Patent Help library.
especially our Collected Patent Tutorial Reprints, and
specific files on When to Patent, The Idea Mortality
Curve, How to Bust a $650 Patent, and, of, course, our
classic Case Against Patents that started it all.

December 31, 2005

Here's some of the major Data Sheet Sites, courtesy of
the Sci.Electronics.Design newsgroup...

http://www.partminer.com (12 Million)
http://www.datasheets.org.uk (5.20 Million)
http://datasheetarchive.com   (2-3 Million)
http://www.alldatasheets.com (1.07 Million)
http://www.chipdocs.com (701,010)
http://datasheets4u.com (523,031)
http://www.datasheetcatalog.com (280,758)
http://chipcatalog.com (253,733)

And, of course...

http://www.tinaja.com/glib/numschip.pdf    (1)

December 30, 2005

Some eBay sellers seem obsessed with having as high a
sellthru rate as they can, attempting to sell everything
every time on its first listing.

In reality, there probably is no correllation whatsoever
with fast sellthru and optimal eBay profits. In fact, if
things are selling too fast, you are probably charging far
too little.

My own feeling is that if it takes a few weeks or even a
few months to sell something, the chances are you can
get a much higher price for it. And that your total return
even after the extra fees should be significantly higher.

Especially since your first relisting is free.

At least to me, a twenty one day cashout and a fifteen
month hang time appears about right for industrial items
acquired in quantity. Especially if a 30:1 sell/buy ratio
goal is in fact acheived.

More on our Auction Help page.

December 29, 2005

I've been doing more study into image pixel interpolation.
This is an utterly fascinating subject with some amazing
potential. Several very interesting links appear here, here,
here, and here.

It still turns out that your "best" image interpolation scheme
for minor size changes often remains the High Resolution
Cubic Spline Basis Function. Such as we are using in our
Universal Bitmap Manipulator and as we looked at back
here.

But if you are reducing an image, another problem rears its
ugly head. You might get Moire sampling artifacts or dropouts
of fine detail. The workaround is to combine a low pass filtering
(such as a Gaussian correllation) with your downsizing to
prevent missed lines or added artifacts.

When highly magnifying, there are newer and fancier techniques
that go beyond what you can do with high resolution cubic
splines. Many of these are based on Lanczos techniques, while
the present top dog just may involve a process called LAD
Deconvolulation.

Beyond these techniques are whole other worlds of content specific
and nonlinear transformations. A content specific image processor
might deal with a picket fence area differently from a blue sky area.
And nonlinear techniques can solve particular problems. For instance
an algorithm of "Replace each pixel with an average of its nine nearest
neighbors" dramatically reduces artifacts in a noisy image. But also
introduces very bad problems of its own.

Another consideration is that the fancier techniques often involve
great heaping bunches of computer horsepower and may not be suitable
for real time or even near real time uses.

December 28, 2005

Refurb Log:

Sometimes you can guess at a schematic simply by
noting which integrated circuits are apparently used
how. I was refurbing a Heath 4804 byte analyzer that
seemed to be behaving erratically. Docs appeared to
be unavailable anywhere. On closer checking, though
operation ended up pretty much as they intended.

We'll first note that you can read the date code on any
IC to get an age estimate. Chances are the unit was
sold about a year later. A typical date code might be
1389 which would be the thirteenth week of 1989. The
IC package style, its SMT vs thruhole mounting, and
its complexity can also give you major clues as to what
you are dealing with.

In the case of the 4804, the inputs obviously went to
a pair of 4-bit transparent latches whose true outputs
indicated their current or held logic state. This was
followed by a plain old 8-input NAND gate that allowed
"1" "0",, or "don't care" adress trapping for fancier
uses. The choice of chips and their position made it
obvious how the unit was supposed to work.

Their hassle (and a really dumb mistake) was that any
ungrounded inputs could easily pick up the power line
or a radio station via body coupling. If you did not know
that All unused inputs MUST be grounded, you would
easily get erratic and confusing results.

If I were keeping the unit, I'd add eight 100K resistors
from each input to ground to discourage noise source
inputs. But I kept the unit as is on eBay, just in case
anyone wanted an authentic collectible.

Naturally, if any item has a use quirk, you should very
conspicuously point it out ahead of time.

More on related topics in our Refurb Log.

December 27, 2005

Someone had asked when the eBay spring slowdown starts.
This is nothing to worry about, because it is simply that little
dip between the winter slack period and the summer slump.

December 26, 2005

The fancier features of a Universal Bitmap Manipulator
demand a high quality true X-Y pixel interpolator. For
such utilities as a precision rotator, a true keystone
corrector, a general distortion remapper, or a scanner
"reperspector".

The "best" interpolation often involves High Resolution
Cubic Spline Basis Functions. As they involve a 4x4=16
correllation at every pixel in a megapixel image space,
these can easily get excessively time intensive. Most
especially when using raw PostScript to create corrected
.BMP images. The big question is what corners, if any
can be cut for faster performance.

The full 4x4 interpolation looks like this...

p03(b0y)(b3x) + p13(b1y)(b3x) + p23(b2y)(b3x) + p33(b3y)(b3x) +
p02(b0y)(b2x) + p12(b1y)(b2x) + p22(b2y)(b2x) + p32(b3y)(b2x) +
               evaluated pixel -----> X <----- is positioned here
p01(b0y)(b1x) + p11(b1y)(b1x) + p21(b2y)(b1x) + p31(b3y)(b1x) +
p00(b0y)(b0x) + p10(b1y)(b0x) + p20(b2y)(b0x) + p30(b3y)(b0x)

You can get this result by starting with four X interpolations that
are followed by one Y interpolation, or vice versa. Either way ends
up with the same result shown above. Twenty lookup-multiply-adds
would seem to be needed at first glance.

But terms like (b0y)(b3x) can be combined into a single table
lookup. Ferinstance 10X the x residue plus 100X the y residue
could give you a 100 entry lookup for each of sixteen tables.
Thus reducing you to Sixteen lookup-multiply-adds per pixel.

Further, as this plot shows, those four "corner" tables do not
do all that much. They only add two percent worst case and add
very little noticable in most other positions. (red on the plot is
a one percent or higher weight; green is two percent or higher.)

So it would seem tempting to bundle the corner table weights
with those of their diagonal neighbors. And getting us down to
Twelve lookup-multiply-adds per pixel.

A second corner cutting opportunity involves dealing with the
rare underflows. As this anaylsis shows, underflows can occur
just under two percent of the time at an average of a five percent
darkening of the blackest blacks.

Naturally, the underflows have to be trapped out somehow. The
obvious dup 0 lt {pop 0} if might be approximated by a simple abs.
This would incorrectly lighten a few of the lowest energy blackest
of blacks. While eliminating a time intensitive test on a centermost
processing loop.

I'll try to work up these concepts further.
Consulting services available.

December 25, 2005

Optimum bid strategy very much depends upon the
auction venue. As we've seein in EBAYBUY.PDF,
your best eBay bid strategy is to always proxy bid
your max once very late in the auction, doing so in
oddball penny amounts just above a currency
resistance threshold.

This may also seem best at a live auction on a big
ticket item that you really want, but I have found
compelling advantages to consistently offering
the auctioneer loball floor opening bids, especially
if you are subtle about it. The auctioneer is more
likely to notice you and accept your bids and may
even give you a fast hammer every now and then.
More in AUCTSCNE.PDF.

The situation with Government Liquidation auctions
is much more complex. I've cut back on my bidding
here because end-user bidders have gotten way too
competitive, because of an unaceptably high $50
minimum bid opening, their maddeningly infuriating
auto bid extensions, and problems with site access
and site manager customer treatment.

Trap #1 is any GL auction is that your optimal
bidding time is precisely 42 minutes into the last
hour of the bid offer. Anything less telegraphs
your intent, and anything later trips the ludicrous
bid extensions and really invites competition.

Trap #2 is that oddball penny amounts will immediately
tell the competition that you have just proxy bid your
max should you become high bidder. Thus one or two
even bid increments above a currency threshold may
in fact be your best bet.

December 24, 2005

O.K. Heres some more details on our "Use both
a camera and a scanner" ploy from two days back.

The basic idea is covered in EBAYFOTO.PDF. I'll
first take an ordinary picture using a Nikon Coolpix
5000 at full native resolution. The next step would
usually be going through our Swings and Tilts utility
to force Architect's Perspective. But this particular
item is squat enough that the persepective can
be forced by background knockout alone.

While I've got a new Fast Background Knockout
utility available, this time, I used our plain old
KNOCKOUT.BMP with a simple cut and paste.

As we've recently seen, fitting background to a
diagonal line can be greatly sped up by using the
lasso tool with a matching diagonal closure. The
rest of the image is post processed using my
usual tricks.

At this point, the space for the label gets carefully
measured. Label width is 561 pixels, its left height
is 120 pixels, its right height is 118 pixels, and its
right offset climb is 233 pixels.

The side label is than scanned. Any retouch or
lettering improvements are best done before
perspective distortion using our newly revised
Bitmap Typewriter tools.

Now for the tricky part. The image is resized to
120 x 561 pixels and rotated vertically. It is then
copied into the center of the top half of a new Paint
page of exactly double height or 1122 pixels. The
reason being that we want to keep the original left
side of the label the size it was, while progressively
applying climb and shrink correction to the rest of
the label.

Some playing around with NUTILT01.PSL is needed
to find the magic values that give you exactly the right
amount of climb and shrink. In these examples, setting
the tiltaxis to -1.15 and the howmuchtilt to 0.1 gave a
quire good fit.

The newly "perspected" label is then rotated back to
its normal position and cropped to size. The "white"
background may no longer be a pure white, so the
lasso tool is used to crop out all but the active label.

Merge time. The scanned and corrected label is then
pasted on top of the camera image of the rest of the
module. Some minor edge cleanup and treatment may
be needed to make the patch completely invisible.


One minor gotcha: The lettering across the label will
be uniformly spaced rather than perspectively spaced
where the distant letters are compressed. You are very
unlikely to notice this discepency at all. The problem
can be fixed at the cost of computation time.

December 23, 2005

We have pretty much decided to strongly discourage
premium shipping services. The day we use one is the
day the UPS relief driver forgets to stop or the day our
shipping help uses their flextime for dog dentistry.

More often than not, someone will piss around negotiating
for a month or two and then demand instant shipping. It
makes little sense to double or even triple their costs for
little apparent benefit.

Plain old UPS brown is surprisingly fast for most people
most of the time. And most of our eBay feedback praises
our fast shipping using this service. And the disruptions do
not in any way seem to justify the benefits.

December 22, 2005

The obvious answer to whether you should use
a digital camera or a scanner for your eBay
photography is to use both at once...

Note the infinite depth of field along the lettering
plane! Plus the total absence of any flash flare on
the label. Combined with our usual pixel locked 2-1/2 D
"Architects Perspective" and our normal shadowless
photography with JPG edge artifact reduction.

What you do is scan the label, resize and retouch as
needed, and then use my Swings and Tilts backwards
to distort it into the correct trapezodial shape. Then
cut and paste and retouch the borders a tad.

You can click expand the above image to get it up to
the normal eBay JPG size. The full size, full res
bimap image can be made available on request.
The latest Swings and Tilts version is found here.

I hope to expand our Universal Bitmap Manipulator
routines to simplify converting flat text into a
perspective paste. More on PostScript here.

We have the finest photos on eBay, bar none. Yes,
Consulting and Seminar services are available.

December 21, 2005

Refurb Log:

I can't emphasise strongly enough how important
careful inspection of auction items and continuous
inventory control is. Here's some recent examples
that have caused me grief...

   ~ A pricey community college instrument needed
      an outrageously expensive replacement part.

   ~ What I thought was a complete scanner turned
       out to be only the slidemaking attachment.

   ~ Most of a stack of large electronic breadboards
      was missing crucial inserts.

    ~ A pallet full of slot machine mechanisms turned
       out to be pulls needing extensive refurb.

   ~ Apparently new-in-original-packaging drafting
       table covers turned out to be the old covers
       stashed in new boxes.

   ~ Repairs to an oscilloscope plug-in were so
      awkward they could not justify resale price.

   ~ What appeared to be a real bagain was really
      a mix of useless halves of two wildly different
      Items.

    ~ An attempt at humor in an instrument description
      caused widly wrong assumptions by naive buyers.

   ~ A lab oven turned out to be a much lower value
      lower temperature incubator. Also flood damaged.

Much more in our ongoing Refurb Log and on our
Auction Help library pages.

December 20, 2005

If you divide the number of lots in a live auction by the number
of bidders, you can approximate an average number of lots that
each bidder will win. One way to do this is to keep track of the
first and last bidder registrations issued. But some auction houses
may purposely shuffle their numbers or preassign out-of-range
numbers to regular customers.

In reality, the usual 80-20 rule will apply in which twenty percent
of the bidders will likely end up with eighty percent of the lots.
And since your particular items will be a small subset of the total
offered, what really matters is how many specific interest bidders
are competing against you.

Ideally, of course, you want zero competition. Especially regulars
who are known to bid for spite just to prevent others from scoring
a lot. Such pissing contests, of course, should be studiously avoided.

Much more on our Auction Help library page.

December 19, 2005

I'll try and get our Sunraise thermography machine checked
out and up on eBay shortly. Thermography is a fascinating
process applied to traditional printing that gives a raised
lettering effect. Often for business cards and wedding
announcements Typical machines are $1800 new.

Normally, a special powder is sprinkled onto still wet ink.
When heated, the powder intumesces and swells, creating
the raised effect. There are several grinds of powder
for varying lettering sizes. Thermography works best if
all of the lettering is nearly the same size.

Fortunately, when using standard rubber base inks, the
thermography process can still be used up to three hours
after printing. This greatly simplifies things.

Sadly, thermography does not work with toner or laser
printing. One approximation is a spray product called
Laser Buddy. Another older route is called Bakerizing.
In which you place a sheet of mylar in contact with your
toner and run it through a heat and pressure machine.
Thus creating a shiny black calendaring effect.

December 18, 2005

There seems to be an unwritten rule of "No Christmas Auctions"
There are typically 65 to 130 auctions scheduled in Arizona at any
given time. This seems to consistently drop to less than one-sixth
at year end.

More on the Arizona Auction Scene in This Tutorial.
And more on live auctions in general are found here.

Your own custom regional auction finder can be created for
you per these guidelines.

December 17, 2005

A reminder that I have a pair of very rare 1908
commercial silent movie projectors available.
They are presently disassembled, so they would
easily be UPS shippable. They seem to be nearly
complete and should be eminently restorable.

Please email me if you have any interest in this unique
opportunity. Inspection welcome. We will shortly
be dramatically be expanding our refurb activities, so
I very much need the shop space back.

December 16 , 2005

I've sharplly reduced prices on some of our older eBay
items to clear them out for some exciting new offers.
This is sort of a one time sale, so when these are gone,
they are gone.

Several prices on our unique collectibles have been
sharply increased, because I got tired of giving away
items that I am the the world's sole source supplier for.

December 15 , 2005

Apparently Government Liquidation has just raised their
minimum bid to $50. I consider this really, really dumb
because many of their items clearly were not worth their
previous $35 minimum opening price.

I've personally cut way back on my GL involvement,
first and foremost because their recent prices are way
too high, caused by far too many competitive bidders.
Problems with site managers, site access, and all their
maddeningly infuriating auction extensions also remain.

December 14 , 2005

The "whipsocket effect" appears to be alive and well in
the digital camera arena. In which several decades of
automobiles continued to provide whip sockets.

We seem to have a cuious mix of outstanding improvements
in resolution, performance, noise, power, algorithms, and cost
with a dogged continuance of previous camera features that
now are clearly utterly worthless.

A digital camera can be any shape and form. Where the
"two cassettes, a film plane and a chunk of glass" clearly
and exactly defined what a 35 mm traditional camera had
to look like.

"Single Lens Reflex" cameras solved a parallax problem
that simply does not exist on a digital camera! It is
ludicrously absurd to offer a SLR digital. Especially if
the SLR no longer offers interchangable lenses.

The lenses themselves are far less critical than they used
to be. As you get up into the 10 and 15 megapixel arena,
digital zoom can replace most traditional lens needs. And
waiting in the wings are liquid lenses and a stunning new
breakthrough called "negative indicies of refraction" that
totally blows away traditional optic laws.

For studio photography, even the viewfinder makes no
sense whatsoever. Your composition should be done at
the same size as the final image or print! Preferably through
a wireless link to a 17 inch interactive monitor. One that
lets you pinpoint the focus distance and view depth of field.

What I'd really like to see for my eBay photography is a
crane-like camera in which you can remotely position your
image sensor over a wide three dimensional area. With total
interactive control of everything. All at final full size.

Chances are overwhelming that such a system would not
look at all like a whip socket. And might be a perfect
candidate for a hexapod solution.

December 13 , 2005

OK, here is the full solution to the "Bezier problem" in
which you want to relate the x0-x3 control points to the
A-D cubic equation coefficients. And their y equivalents.

Create some cubic basis functions by starting with an
obvious 1 = 1 and morphing it into a bizarre but highly
useful (1-t) + t = 1. Cube this expression to get...

         (1-t)^3 + 3t(1-t)^2 + 3t^2(1-t)  +  t^3  = 1
 
    Call each of these left four terms a basis function..

            B0(t) + B1(t) + B2(t) + B3(t) = 1.

x (and separately, y) are now related to our cubic t by...

       x(t) = x0(B0(t)) + x1(B1(t)) + x2(B2(t)) + x3(B3(t))

x(t) will also equal our cubic spline equation...

                  x(t) = At^3 + Bt^2 + Ct + D

Since t has to be allowed to vary from 0 to 1, the only
way these two equations can be equal is if the t coefficients
match for each power. Expand the terms...

    x0(1-t)^3             =    x0 - 3x0t  + 3x0t^2   -   x0t^3
    x1(3t (1 - t)^2)    =           3x1t  -  6x1t^2 +  3x1t^3
    x2( 3t^2(1-t))      =                        3x2t^2  -  3x2t^3
    x3t^3                  =                                           x3t^3

Now, think vertically upwards and regroup to get...

                      A =   x3 - 3x2 + 3x1 - x0
                      B = 3x2 - 6x1 + 3x0
                      C = 3x1 - 3x0
                      D =  x0  

You can easily show that the control points behave in
the expected manner. At t=0, only B0(t) is active and
thus x0,y0 defines the starting point of the curve. At
t=1, only B3(t) is active and thus x3,y3 defines the
ending point of the curve.

The initial x versus t slope is 3(x1 - x0) and the initial
y versus t slope is 3(y1 - y0). The initial xy slope is thus
(y1 - y0)/(x1-x0) and thus the x1y1 control point sets
the initial slope. Similarly, the final xy slope can be shown
to be (y3 - y2)/(x3 - x2) and thus the x2y2 control point
sets the final slope.

Finally, note that the x1,y1 point has its strongest
influence precisely and always at t=1/3. And similarly,
the x2,y2 point has its strongest influence precisely
and always at t=2/3. The "tension" or "enthuasiasn"
of the Bezier curve is thus determined by how much
change has to happen in x and y between t=1/3 and
t=2/3.

Q.E.D. Or something like that there.

December 12 , 2005

Magic Sinewaves are a newly discovered class of math
functions that dramatically can raise the efficiency and
improve the quality of induction motor controls, solar panels,
electric vehicles, telecom, inverters, and aerospace apps.

An intro tutorial appears here.

The key Magic Sinewave features are that any number of
low harmonics can be forced to extremely low values while
using the absolute minimum number of efficiency-robbing
switching transitions to do so. Yes, three phase variations
are also available.

As are demo chips and sourcecode.

Several readers have asked about the frequency limits to
magic sinewaves. These are primarily for 60 Hertz use but
might be extendable to 400 Hertz aerospace apps.

The reason for the low frequency limitiations is that as many
as 45,000 instruction cycles might be needed to properly
generate one magic sinewave cycle. This translates to a
10 MHz clock for a PIC generating ordinary 60 Hz. power
waveforms.

Consulting, Training, and Development programs available.

December 11 , 2005

We seem to have a coastal fog this morning. Perhaps
it is time to set out the lobster pots.

December 10 , 2005

Those Basis Functions are really neat when you start digging
deeper into Cubic Splines. What they really do is decide how
much contribution each control point gives you anywhere along
the t curve from 0 to 1.

Ferinstance, at t=0, x0 (and y0) completely dominate and give
you 100 percent control of position. At t = 1/3, x1 dominates
and also sets the initial slope of the x versus t (or y versus t)
curve. At t=1/3, the scaled contribution of the four control points
is 8/27ths, 12/27ths, 6/27ths, and 1/27th respectively.

At t = 1/2, each end point contributes 1/8th and each influence
point contributes 3/8. With symmetrical results for t=2/3 and
for t= 1. Other t values are similarly proportioned.

Additional plots are found here.

December 9 , 2005

Org. My first clue that today's live auction was not going to
go my way was when the first lot of pop rivets sold for $2200.00.

December 8 , 2005

Larger industrial auctions may be either the usual "walk
around" or the theater-style "sit down". It is super important
to know which is which ahead of time since previewing may
be difficult or even forbidden during a sit-down auction.

Sitdowns may be used because of the enormity of the site,
climate, or security. A walk-around auction is overwhelmingly
more in your favor since it is a lot rougher on the other bidders,
and because quick lot mergings or choices are likely to occur on
the fly at the auctioneer's whim.

If you must participate in a sitdown auction, be absolutely certain
that you carefully preview the "contents of cabinet" and "contents
of room" lots and have fairly approased their value to you. For
their hidden value may give you opportunities others miss.

Online bidders may have a slight edge on the sitdown auctions as
their bids may end up being entered faster and more reliably. But
there is rarely a time when "being there" is not overwhelmingly to
your advantage.

Much more on our Auction Help library page.
Custom Consulting also available.

December 7 , 2005

Some more background on the "smoke and mirrors"
behind Bezier cubic splines...

Any cubic spline relates x to t or y to t using a formula
of x(t) = At3 + Bt2 + Ct + D. t is a parameter that goes
from zero to one. The trick is to introduce methods to
externally and conveniently control the shape of the
curve.

A Bezeir Cubic Spline uses four x control points, two
at the ends of the curve and two that set the initial
and final slope and the enthuasiasm of slope travel
of the desired curve. Four similar y control points are
separately and independently involved.

How do you get from the control points to the A thru
D spline values? Bezier understanding can use the
following trick involving Basis Functions that center
on Bernstein Polynonials...

Start with an obvious 1 = 1. Morph this into a bizarre
but highly useful t + (1-t) = 1. Cube this expression to
get...

           t^3 + 3t^2(1-t) + 3t(1-t)^2 + (1-t)^3 = 1
 
    Call each of these left four terms a basis function..

            B0(t) + B1(t) + B2(t) + B3(t) = 1.

x (and separately, y) are now related to our cubic t by...

         x(t) = x0(B0(t) + x1(B1(t) + x2(B2(t) + x3(B3)t

from which the A thru D values as a function of x0 thru
x3 are eaily calculated by expanding and regrouping.

Two of the really neat features of basis functions are
that their peaks occur at the spline control points and
that any t value is the sum of the four basis functions
evaluated at (t) scaled by their control point values.

Further B1(t) and B2(t) always peak exactly at t=1/3
and t=2/3 respectively. Setting initial and final slopes
of the x versus t curve at 3x1 - 3x0 and 3x3 - 3x2
respectively.

Basis function plots appear here.

December 6 , 2005

Expanded and updated the Arizona Auction Resources on
 our Auction Help library page.

Your own custom regional auction finder can be created
for you per these guidelines.

December 5 , 2005

Dopeler Effect: The tendency of stupid ideas to seem
smarter when they come at you rapidly

December 4 , 2005

As previously noted, I picked up a bunch of linear position
sensors at the Rubbermaid auction. Some of these are huge,
being as much as five or six feet long.

There are two basic styles, the Waters Longfellow type,
which is a linear resistive potentiometer; and the Tempsonic
style which is a much more sophisticated and accurate device
that uses magnetostriction along a rod for super precise
contactless measurement.

I'll be placing these up on eBay in the next few days.
Condition varies from new to refurbed and guaranteed
usable, and costs of these superb robotics and automation
devices are a fraction of new.

email me if you have any special needs in these areas.
Note that a longer slider can be used for shorter strokes.

December 3 , 2005

Thought I'd once again start a list of live auction strategy
and tactics...

   ~ Agressively know about all regional auctions.

    ~ Seek out auctions where your items of interest are
       only a small portion of the total lots offered.

   ~ If more than five percent of your bids win, you
      are bidding way too high.

    ~ Bring extra copies of your tax certificate and your
       business cards. Be sure you know the acceptable
       payment terms and bid deposit requirements.

    ~ Make sure the auction is in fact being held before
       committing to a long trip. Always ask about preexisting
       bulk bids.

   ~ One simple way of dealing with "certified funds only"
       is to get a pile of certified checks made out to yourself
       in $100 - $200 - $400 - $800 denominations

    ~ Always seek out a sell/buy ratio of 30:1 or higher.
       Avoid ever paying more than a fraction of a cent
       on the dollar.

    ~ Focus on your areas of expertise where you see
       value that others do not.

   ~ Do not let the size of your vehicle determine your
      bids. But do not let a single trash item force you
      into a larger truck or trailer or other expense.

   ~ At larger auctions, tightly center on a few to a few
      dozen items only.

    ~ NEVER get into a pissing contest with another bidder!
       ALWAYS stick to your predetermined max price.

   ~ Best deals often happen very late in the auction,
       so be very patient.

    ~ Assume that items will look a lot worse when you
       get them home, compared to "heat of the moment"

    ~ Best deals often involve "contents of shelf",
       "by the multiple pallet" or "contents of room".

    ~ Continually split your bid increment in half by
        waving your hand palm down across your chest.

    ~ Never gloat or show ANY emotion over any
       spectacular win. Keep a poker face at all times.

    ~ Always carefully preview all items. Preferably
        at least three times and once in reverse sequence.

    ~ If you do not make at least a $25 mistake, you are
       not being aggressive enough.

    ~ Avoid speaking to other bidders, but listen carefully.

    ~ On-site auctions are overwhelmingly more productive
       than Auction House or Auction Barn events.

    ~ Be willing to return with nothing. Never bid just
       because you are there.

    ~ If heavily crowded, stay by your upcoming item and
       let the auctioneer and crowd come to you

    ~ Bring a dolly or a cart with you.

    ~ Become ultra aggressive if the auction clearly
       starts going overwhelmingly your way.

    ~ Be in the auctioneer's face when bidding and
      totally invisible otherwise.

   ~ Watch for "poisioned" or combined lots. These
       can be a superb deal if hidden value remains.

   ~ Keep lowball "floor" bids subtle and seen only
       by you and the auctioneer.

    ~ Suppress your ego at all times!

   ~ Never give any ambiguious signal or bid.

~ Avoid theft by loading your highest value items as
       quickly as possible.

   ~ Dress down to the point of being shabby but
       always wear a distinctive hat or whatever.

    ~ Stick to your bid strategy. I prefer continual lowball
       openings to single max bids. Have the auctioneer
       return to you when momentum lags.

    ~ Unwanted items can usually be given away after
        the auction, but carefully spell out who gets what.

     ~ NEVER piss off the auctioneer!

   ~ Pay particular attention to lots that are outside,
      in obscure rooms, or transition areas.

    ~ Always emphatically overstate your body language.

   ~ Have a pocket full of five and ten dollar bills for
      quick help and side deals.

   ~ Bring tools to split larger or heavier items into more
       managible pieces.

   ~ Personally thank the auctioneer after the event.

Much more on our Auction Help library page.
Custom Consulting also available.

December 2 , 2005

We are down to our last two Adept Linear Sliders. And
are unlikely to find any more any time soon. Remaining are
a pair of 950 mm linear acutators with encoders that are
repositionable to a tiny fraction of a mil.

email me if you need any further details. I expect these to
be gone in a week or two, judging by how fast the previous
fifty sold.

December 1 , 2005

The key secret to becoming a cubic spline expert is to
understand how you get from graph space to equation
space. Per this tutorial or these summary results.

You can think of a cubic spline as a "snake in a box"
that lies in x-y-t space. Parameter "t" (almost) always
goes from 0 at the beginning to 1 at the end. Although
x and y are cubic functions of t, they are independent
of each other. Letting you build fancy curves, cusps,
and even loops where you precisely control the entry
and exit positions and slopes.

In graph space, x0 sets your initial position and x1
the slope and "enthuasiasm" with which your spline
leaves your entry point. Similarly, x3 sets your final
position while x2 sets the exit slope. Ditto for y.

In equation space, the cubic equation is, of all things,
a cubic equation of x = At^3 + Bt^2 + Ct + D. Now, to
become a cubic spline epert, all you have to do is
relate x0 through x3 to A through D.

What gets tricky is understanding where all those
wierd formulas come from. The simplest way is to
take these equations on faith...

   A = x3 - 3x2 + 3x1 - xo    E = y3 - 3y2 + 3y1 - y0
   B = 3x2 - 6x1 + 3x0         F = 3y2 - 6y1 + 3y0
   C = 3x1 - 3x0                   G = 3y1 - 3y0
   D = x0                              H = y0

   x0 = D                              y0 = H
   x1 = D + C/3                   y1 = H + G/3
   x2 = D + 2C/3 + B/3        y2 = H + 2G/3 + F/3
   x3 = D + C + B + A         y3 + H + G + F + E 

You can simply use plain old algebra  to prove that
the equation space formulas are inverses of the graph
space formulas. And easily use PostScript to prove that
the equations are in fact valid and perform as stated.

But, if you must know where these came from, recognize
that a Bezier curve is a very specific type of cubic spline
that is definied by...

    x(t) = x0(1 - t)^3 + 3x1t(1-t)^2 + 3x2t^2(1-t) + x3t^3
    y(t) = y0(1 - t)^3 + 3y1t(1-t)^2 + 3y2t^2(1-t) + y3t^3. 

Expand all these terms and regroup them, and all those
funny "3" and "6" coefficients in the equation space will
pop right out.

Those strange (1-t) expressions involve Basis Functions,
more on which you will find here. Curiously, the influence
points always have their strongest action at t = 1/3 and
t = 2/3. Which is where the "threes" in the slope equations
come from.

November 30, 2005

Just picked up a Sunraise tabletop thermography machine
at a public utility auction. It seems to be in real clean used
condition and is perfect for making raised print business cards
or fancy premium grade letterheads. email me if you want to
get in ahead of the hoarders on this one.

November 29, 2005

Latest GuruGram #57 is on Curve Fitting Cubic Splines to
Circles and Ellipses. It turns out you can do better than
some previously published web results. And that the math
can end up very much simplified and more obvious.

Sourcecode can be found here. More in our Cubic Spline
and PostScript libraries. Consulting Services available.

November 28, 2005

I continue to be utterly amazed at the lack of plain old
common sense on many alternate energy projects. Along
with the failure to ask fundamental questions.

Ferinstance, a recent post to the Sci.Energy.Hydrogen
newsgroup had a building they were adding a 150 kW
wind plant to. Since there was 100 kW of excess
electricity "going to waste", they wanted to know which
brand of electrolysizer to use for hydrogen storage.

Were the following issues even addressed...?

   ~ A building in a very windy area is likely to be
       very little in demand by tenants.

   ~ Building in a windy area dramatically increases
      heating and cooling costs. Wouldn't earth berming
     and planting trees save a lot more energy a lot
     cheaper than can be recovered?

   ~ Living inside a windmill would probably be an utter
      disaster because of continual noise, vibration, and
      subliminal stress issues. Kawhoosh kawoosh kawoosh.

   ~ The nameplate rating of a wind generator has nothing
       whatsoever to do with the amount of electricity that
       can be produced. The latter is determined by long term
       wind stats. Wind energy is also highly nonlinear with
       lower velocities. I'd predict something like twenty kWH
       per day geninely and actually long term recoverable as
       "excess" electricity from the proposed system. Electricity
      worth two dollars per day.

~ The recovered energy must significantly exceed the recovery
      costs, or the project makes no sense whatsoever. Two dollars
      per day of electricity would get totally consumed by a ten
      year, ten percent $4500 investment. Which wouldn't be even
      enough for union bribes and permit payoffs.

    ~ Because both electrical kilowatt hours and the dimes that
       pay for them are both fungible and interchangible commodities,
       there is absolutely no known energy storage sytem that is
       even remotely comparable to grid buyback for simplicity,
       economy, safety, and reliability. Not by a country mile.

    ~ A utility may elect three or more types of buyback. Net
       Metering, of course, is an outright ripoff as it severely taxes
       other utility users while obscenely undervaluing storage
       services. With Avoided Cost buyback, the utility pays you
       the same as they do everybody else. And any differential
       between what you buy and what you sell is more than
       justified by the additional storage services provided.

    ~ Finally, the utility may elect zero buyback because the
       project makes no economic sense to them. Which also
       suggests that if the utility cannot profit from the fiasco,
       then nobody else is likely to either.

   ~ Electrolysis to hydrogen, of course, is one of the worst
      possible energy storage schemes because of its instant
      destruction of exergy.  A kilowatt hour of electrical
      energy is ridiculously more valuable than a kilowatt hour
      of unstored hydrogen gas. Because its thermodynamically
      reversible recovery is insanely higher. And that is all
      before any amortization or storage or safety issues.

More in our Energy Fundamentals and Electrolysis tutorials.

November 27, 2005

The Onion reports major problems with the Panda
exhibit at the Chicago Aquarium. Apparently there
are issues with the masks and the regulators.

November 26, 2005

If you are making more than four or so newsgroup
posts per day, you probably should cut back and place
your comments in your own blog instead.

Newsgroup posts have about a seventeen hour attention
lifespan, while your blogs can run forever. You can edit
and rethink your blogs. Ad Hominem attacks and thread
hijaking cannot happen. Plus your posts are likely to be
more well thought out and focused.

Spend the time and energy where it will do the most good.

A reminder that we now have a RSS feed up. Included are
these sort-of-a-blog What's New pages and quick access
to our more popular tutorials.

November 25, 2005

The errors in a four spline circle approximation can be
dramatically reduced by going to a six spline one. I don't
have the exact error reduction figure yet, but I suspect it
is at least ten times lower as it seems completely invisible
at the highest Acrobat magnification.

My current prediction: eight splines is probably good enough
for most normal machine shop use, and m might be overkill.

I hope to have a GuruGram on Circle and Ellipse approximation
with Cubic Splines up on this shortly.

November 24, 2005

I find it more than annoying to be in a caving area or a
wilderness and have continuous and loud aircraft noise.
And was wondering why this is not a problem where I
live here in the Greater Bonita-Eden-Sanchez metropolitan
area.

Apparently this is one of the very few areas in the country
where commercial overflights are extremely rare. If you
go to Google Maps and magnify till you only get a few states
and plot the obvious city-to-city commercial routes, most of
them seem to miss the Gila Valley by at least thirty miles.

The quiet sure is welcome, though.

November 23, 2005

There's a lively discussion over on Slashdot on how the
new and highly-in-demand X-boxes routinely blow up. The
cause is apparently bad thermal design of the external
third party power supply.

Simply getting the supply off of the floor and providing some
circulating air seems to help bunches. Various "pieces of
string" suspension solutions have also been proposed.

Deep pile carpets (especially those where you cannot see the
dog) are very much a no-no.

I had a similar problem with a Cisco cable modem that was
partially my fault and partially theirs. Although they deny
there is a serious thermal design issue on the BEFCMU10.

Their vent holes are obviously too small. I compounded the
problem by putting a router on top that both blocked the
vents and actually forced heat into them. But when this DUH
was fixed, there still was an infuriating intermittent that
neither I nor the cable company could find despite repeated
testing.

The problem has long gone away. Simpy by placing the cable
modem on a 45 degree angle so the bottom vents also had
reasonable access to the ambient air. But larger holes makes
a lot more sense. You can easily feel the difference.

November 22, 2005

Always be very careful with humor on any eBay listing!

If someone does not get the joke, you will be out a refund
and its associated bad vibes. We had been selling these ultra
sophisticated earthquake simulation machines to qualified
researchers who needed their proof mass actuator, LVDT,
and super performance accellerometer pairs for shaker table
testing.

Two remain available.

Trying to add humor led to wildly wrong conclusions that these
were superaudio kickers or subwoofers, which they are not.

The rule is to be suspect if any inquiry doesn't get it. And to
flush the humor entirely on a second problem of any sort.

More on auction techniques on our Auction Help page.

November 21, 2005

Here's some more info on a recent solar cell paper that might
prove quite interesting:

Light enters via a glass substrate with a transparent tin oxide
conductive coating on it. An ultrathin layer of Cadmium Telluride
nanorods is spin cast in place. These are single crystals rougly 8
nanometers in diameter by thirty long. A second similar layer of
Cadmum Sulfide nanorods are overcast. A final aluminization
provides the back contact and an optical reflector.

Unlike a conventional doped p-n junction, a diffussion assisted
heterojunction is formed between the two layers. Carefully
controlled bandgaps between the two crystal matrials determine
the rectification and the photocurrent.

The materials are much more stable than oganic semiconductors
and actually improve with aging. The cells are exceptionally thin
and apparently lend themselves to simple manufacturer. They
apparently lend themselves to spincasting and possibly inkjet
technologies.

The efficiency of the demo cells is unacceptably low in the three
percent range. The paper does not address what the physical
theoretical limits of the technology are. But with heterojunctions,
choices in bandgraps should make multilayer and spread energy
devices possible. Some sort of alignment or orienteering should
be able to improve the fill factors as well.

This appears to be a genuine breakthrough development. It
seems to me that conventional silicon pv will NEVER be able to
acheive sustainability and renewability due to its intractability
and its single workfunction efficiency barrier. This new development
just may be the right horse to bet on

The original paper plus additional source materials that discuss
materials prep and such are available from Science.

November 20, 2005

A curious property of the PostScript arc and arcn operators:
These work as expected for smaller angles but are only a four
spline approximation for full circles.

Regardless of the circle size or the chosen Distiller quality
level!

The error is only one part in a thousand worst case. This is trivial
for most graphics, but this could easily cause problems in
CAD/CAM and other PostScript as Language applications.

Ferinstance, this error would be unacceptble in machining a
three inch cylinder for an engine.

The workaround is to build your circle from at least six arc or
arcn spline calls. This should get you well within acceptable
tolerances for normal machine shop use. Virtually all of the
observable error disappears when you go from four spline
circles to six spline circles.

A GuruGram is in the works and should be available shortly.

November 19, 2005

There's some lively discussion going on in the newsgroups
over a new "trucker's hydrogen injection system" based on
electrolysis via an alternator. To me, the numbers simply
do not add up. Not by a country mile. Time will tell how much
of this is a scam, how much is the placebo effect, and how much
is real.

First, there definitely is credible peer reveiwed research that
shows that a modest (typically 5%) hydrogen injection can
improve combustion to significantly raise fuel economy. It
is also possible that lower hydrogen injection levels can reduce
carbon and other deposits. The latter has yet to be convincingly
shown.

Two key questions are whether any new mechanical load needed
to produce the hydrogen is greater or less than the benefits
derived; and whether the fully burdened cost amortization will
ever let the system pay for itself.

Few people realize how ludicrously inefficient a conventional
alternator driven electrolysizer would be. The odds are utterly
overwhelming that there is no way in hell that the benefits
could remotely approach the input loading penalty.

First, the fanbelt itself may be limited to the three the five
horsepower range. Its efficiency is probably around 97
percent, caused by flexing and air turbulence. Car alternators
are a lot less efficient than most people suspect, typically
being in the 70 percent range. While the largest losses are in
the rectifier diodes, the wider air gap, less-than-optimal regulator,
cheaper magnetic materials, and field losses all make car
alternator efficiency a secondary consideration. A typical 100 amp
12 volt car alternator is also only a 1-1/2 horsepower device.

For decent efficiency, you cannot simply connect an electrolysizer
to the output of an alternator. Because the alternator is a somewhat
constant voltage device, and an efficient electrolysizer demands a
carefully controlled current. The fancy switchmode electronics
required to do this would probably end up in the 80 percent efficiency
range. Analog control circuits or direct connection, of course, would be
much worse.

While an electrolysizer can theoretically be fairly efficient (neglecting
the staggering exergy hit, of course), most will operate well into their
exothermic range for decent gas volumes.

This introduces at least another 70 percent efficiency drop. Further,
most add-on manufacturers might tend to cheat and use stainless
steel or nickel rather than the platinized platinum required for decent
efficiency. this can add up to another 60 percent efficiency hit due to the
hydrogen overvoltages involved. Details in any electrochem book.

Finally, the engine itself is probably only 35 percent efficient at
converting fuel to shaft horsepower. Let's see. If we start with
1000 watts at the crankshaft, we get 970 watts at the alternator
input and 679 watts at the alternator output. And 543 watts at
the switchmode regulator. And 380 watts for the exothermic
drop. The stainless overvoltage puts us down to 228 watts. And
the engine efficiency finally delivers a mere 87 watts of mechanical
power!

The dilemma is this: The amount of hydrogen producible with one
fan belt and an unmodified alternator seems uselessly low and is in the
"homeopathic dose" range. While significant hydrogen production
seems to me to demand multiple fanbelts, custom heavier and much
more efficient alternators, and other complex modifications.

While other means of hydrogen injection (such as exhaust gas
reformation or a refillable on-board tank) might eventually be
shown to be useful, I strongly feel that the insane efficiency losses
in onboard alternator electrolysis absolutely guarantee that this
flat out ain't gonna happen. And that's BEFORE amortization.

In short, both the engineering economics and the thermodynamics
suck. More in our Electrolysis and Energy Fundamentals tutorials.

November 18, 2005

Approximating circles or ellipses with four cubic splines
is apparently not that big a deal. Go through all the hairy
math and a magic constant of 0.55228475 pops out. Also
known as four thirds of one less than the square root of two.

This magic number is how far you go on your influence points,
normalized to unity. And times unity for circles and directly
scaled for the minor axis on an ellipse.

The results give fractional pixel accuracy for larger than full
screen circles and ellipses, so should be more than useful
for most people most of the time. Smaller is, of course,
better.

Actually, the "best" magic number is something like 0.00044
lower. The original one gives a perfect fit every 45 degrees
but has an always positive error. The minumum rms error
is 24% lower with mixed positive and negative errors. And
gives three perfect fits per quadrant rather than just two.

I'll save exact details as an exercise for the serious student.
The difference is utterly negligible for most users.

I'll try to work up a GuruGram on this.

November 18, 2005

That big HSL font may not exist at all since it probably
has a "sign painter" history rather than a "hot lead"
one.

At any rate, a very useful font finder appears here.
It does lead to a Miehle Condensed font that gets
the "S" and the "H" right, but misses badly on the
"N" and on not being a fixed stroke width. Its 1905
design frame also sounds contemporary.

If one were serious about a historical restoration, it
would not be that big a deal to write the proper
font from scratch. The fixed with lets you just stroke
and then patch in seriffs. Which is probably the
"paintbrush" method originally used.

We can do this for you as a consulting service.

November 17, 2005

Sold two of our five remaining Adept Linear Sliders.
We are unlikely to find any more beyond these.

November 16, 2005

I took the HSL logo found in the 1914 Pittsburgh
Directory and cleaned it up a bit with Paint...



Yeah, that wrong spelling of Pittsburgh was theirs,
not mine. I also was not sure of the color, so I used
sort of a maroonish Tuscan for everything.

But this item clearly cries to be done right using
fancy mid-level PostScript procs. New stuff needed
might include a spline-to-ellipse converter and some
updated variations on a banner nonlinear transform
and an elliptical path nonlinear transform. All but two
characters seem fairly easy to do. These last two can
be curvetraced or built directly from splines.

Finding a perfect fit for the large font may be tricky
as the H centers are high and the S transition is very
much sloped. Seriffs are emphasized. The N slope is
also truncated.

More on this whenever. Please email me if you know
of a good candidate font. Can the project be done in
less than 999 total file characters?