I recently had a crazy idea about how engineers should be educated.
They should be coached until they have reimplemented from scratch the
software inside the Hewlett Packard HP48 calculator. It’s that simple.
I discovered the HP48 just after graduating from engineering school
and I was sickened. I couldn’t believe this device which very clearly
was designed to help kids claw their way to an engineering diploma
existed. Every topic in my engineering studies was in that machine
ready to help. I had struggled through the entire ordeal unarmed,
oblivious to the existence of this heavy weapon.
You might think that a rewrite of all of that device’s superb software
would be especially heavy on the programming. Yes. That is precisely
why it is the perfect exercise for a modern engineer. And at the end
of it all, the graduate would have a magnificent collection of
self-made technology with which any engineering problem could be
tamed.
Ya… Funny story. So it turns out that’s kind of what I’m doing.
I loved that HP48’s architecture so much that it was one of the main
inspirations for my programming language, Geogad,
which I have used every day for about 15 years as my personal
calculator and heavy hitting math tool. Recently I’ve been
getting into electronics more than
ever and I’ve also been hitting a lot of broad topics in other kinds
of engineering. This all means that I’ve been doing a lot of
conversions from, say, picoFarads to microFarads and knots to meters
per second, etc. All of this has made me nostalgic for the superb units
handling features of the HP48. Of course I could not resist the overly
clever impulse to reimplement all of that functionality.
Yes, I do know about thousands of simplistic online conversion
utilities (e.g. "hey Google"). I also know about
GNU Units — it is an
incredible tool and if you have unit conversion problems, you should
know about it too. But I needed something that was going to ultimately
be a better fit with my programming language and so I got started
writing a quick study to see how feasible it was.
And wow, that was quite a rabbit hole. Eventually I got something I
feel good about and I thought that maybe I should also make it a
standalone program so that people can check it out. Maybe it will
prove to be helpful for someone. I packaged it up and called it
Unicon.
You never know quite how many problems you can run into until you get
a computer to help you uncover them; the number of bugs I had to deal
with on this project was impressively high. Just typing in the entire
HP48 units catalog by hand is going to produce more bugs than an
anthill. There were subtle weird problems all over the place. Finally
as I was starting to feel more confident that it was going to start
working pretty reliably, I stumbled across this question: what is an
electronVolt converted into kilocalories?
Fortunately I found this example on some random website that was
demonstrating how to use Google to do units conversions. I put that
value in my test harness and Unicon came back… wrong. Nothing
shocking there, but it was slightly wrong.
Unlike in the real world, in computer work when you’re wrong but
close, that is often quite weird. Usually when you’ve made a
computer mistake, you’re way off. But here I was getting a similar
looking answer that was not at all the right one. I checked some other
online resources. Google reaffirmed. Wolfram Alpha concurred.
For me, the last word on these questions is always my beloved HP48.
What did the old master say?
Amazingly, the HP48 and the new Unicon agreed! It’s fine if I am
wrong but the HP48? No. The HP48 is immaculate! It is never wrong! So
that was a puzzle.
Pause here if you’d like to contemplate the answer to this conundrum
before I tell you what the source of the error was.
I don’t know how I ever figured this out but I suspect a good deal of
internet research was going on when I happened upon
this
wikipedia article: 2019 redefinition of the SI base units. [Yes, sic,
just like that with weird capitalization.]
And sure enough, it turns
out that an electronVolt had changed value in May of last year.
Apparently all kinds of major changes were going on in the standards
of fundamental physics. I may have to stop making fun of
the
philosophical problems with Avogadro’s number now.
How’s that for very weird? And the lesson there I guess is that even
if your software is so badass and perfect that it never needs any
updates ever, the universe in general is not so stable. So when people
wonder why I don’t just stick with my old HP48 (or a very clever
emulator thereof), this
is as good of an answer as any.
There are no doubt more such mysteries to be solved and if you’re an
engineer or physics person, I’d love some help. Go to the
Unicon project page and give it a whirl.
You just need some late-ish Python (probably 3.7+) and it’s easy to
play with. I’d love to hear about your serious units conversion
problems. If you find anything interesting, especially interestingly
stupid, definitely let me know!
Thermal imaging is
interesting and potentially quite useful. Though working examples of
it have been around since the 1920s, I think that it will become much
more prevalent in the next decade. For example, for between $250 and
$500, you can buy a neat
accessory for telephones. Or you can even buy
a
$350 telephone with it built in these days.
On my autonomous boat, I’ve collected a lot of video from a
thermographic camera designed for marine use. The fact that anything
that produces heat is usually glaringly obvious makes this kind of
sensor very compelling for boat use. Check out how well I stand out
in a typical water scene.
(Let’s save reflections for another discussion.)
Or how about this shot showing me tying up the boat at the dock.
If you’re in the business of rescuing people who are in the water,
there is a 99% chance that this camera will be incredibly valuable and
you should probably have one. But look again at the previous photo — that image shows a strange day where the sky, land, and water are all
practically the same temperature. If you want to use this kind of
imagery to choose where navigable water might be, there are plenty of
difficult situations.
Check out the amazing ability of the camera to pick out ducks far away
in the background.
But in this cold water on a cold day, the contours of the terrain are
completely lost. It’s not just cold days that can be problematic.
Check out these images where the shore and the water have some
ambiguous temperature readings.
As a general rule living things tend to glow pretty brightly but check
out this shot of me on a day where swimming is cold but standing
around in the wind and rain is colder.
It’s easy for humans to tell what’s going on and not impossible for
computers to make some sense out of it with some machine learning
effort. But naively assuming that the sensor itself will make the
underlying situation as black and white as the image itself is not a
safe assumption.
Sometimes I get a wild idea in my head - actually, I specialize in
that. This week’s wild idea was that I need to learn more about
oscilloscopes. I’ve been working on my electrical
engineering notes and my newly appreciated ignorance about
oscilloscopes was irking me.
Looking online for interesting resources that could educate me about
the topic, I wondered if electrical engineers ever acquire electronics
equipment the most honorable way possible: by making it themselves. I
always found that the best machinists had a lot of tooling that they
had machined themselves.
That sounded awesome. I had to give it a try and I ordered one. My
expectations were reasonable. I figured it had about even chances to
(A) be defective or missing a part, (B) fail because of some ineptitude
on my part, or (C) actually work. At the very least, I expected that I
would get some soldering practice. When it arrived, I set to
practicing!
According to the instructions, a completed unit should look like this.
However what actually showed up was this.
Gulp! Some extreme attention to detail was required to make sure all
the values were thoroughly checked and all components sorted. I
actually used all of the functions on my multimeter for this project,
several for the first time. Just knowing how to test transistors with
my meter was worth $22.
Finally, after a lot meticulous work, I was ready to power it up. I
did the preliminary assembly power checks and they came out good! Wow!
I attached the display and powered it up and it was… iffy. After
playing with it, I had seen it boot properly and sort of work and I
had also seen it act completely rubbish. Hmm.
Carefully studying every connection, I found some dodgy soldering. I
actually did all the resistors and capacitors with my beloved 25 year
old gas powered soldering iron. Unfortunately, its tip was so worn
that it wasn’t doing a great job. After fixing up some of that with a
newer iron I discovered a single lead that I had missed — no solder at
all. Oops. I fixed that and when I subsequently booted it, it came up
fine and worked flawlessly! It was a euphoric feeling to pull off a
correct execution and be so damn lucky.
Ok, cool. I built an oscilloscope. Uh, what now? What do I do with it?
After playing around with it I got the very good idea to hook it up to
a guitar. I just put the probes on one end of a 1/4" patch cable plug — other end on the guitar. Then I was able to see the sound waves.
Amazing! So cool.
In this image you can see I’m holding down the 5th fret (note dot) on
the E string which is an A (E - open, F - 1st fret, F# - 2nd, G - 3rd,
G# - 4th, A - 5th). That A is actually the
A440 pitch
standard described in ISO
16:1975 used to standardize all of music to a certain pitch.
Obviously if my guitar is in tune, the frequency of the sound wave
should be 440. And you can see on the top left that it is reading
Freq: 441.707Hz. You can also get a sense for the amount of voltage
produced by a guitar pickup. Actually I have an unusual
Yamaha
SLG Series and I have no idea how its invisible magic pickups work — with nylon strings no less. But we can see that the voltage while this
note rings goes between 16mV and -20mV, for (presumably) the Vpp (peak
to peak?) of 36mV.
Although this would have been even more fun back when I had enviable
eyesight, I’m still absolutely delighted with this project. I have
learned a lot about a lot of different topics and have added a very
powerful tool to my collection. What a cool little device!
UPDATE 2020-09-20
Having enjoyed that, I looked around for more such fun and found
this signal generator kit.
For only $12, I got to build this thing.
As you can see it does a nice job of giving my oscilloscope something
to do. I think one of the main use cases for a signal generator
(conceptually) is comparing this oscilloscope with a fancier, more
accurate (?) one. But for now I’m just content that it generates
square waves, sine waves, and the triangle waveform shown.
Regular readers may notice a decline in posts about autonomous
vehicles. Rest assured, I’m bringing to light everything interesting
in the field. The problem has been a conspicuous lack of anything
interesting. There are only so many times I can say "They’re doing it
wrong and will fail." But today some PR hype crossed my path that is
actually on the correct trajectory so I wanted to highlight it.
What’s somewhat unnerving to me is that I wonder if any of the people
involved in this project understand why it is the correct approach.
That sounds like a strange and pretentious thing to say, but note how
all the press releases stress the connected aspect of this project;
apparently CAVenue, their marketing name, stands for "connected
autonomous vehicle". Here is their
own
press release saying that.
They seem utterly fixated on the wrong thing. They claim they are
trying to "develop the world’s most sophisticated roadway" but that
can not be true. What they hope to build will ideally be less
sophisticated than any random road. No, I don’t care about the V2V
distractions and great network coverage so people can adore their
phones better. I’m talking about the brains here. Do you think their
roads will match the sophistication of the sum total of human minds
currently required to function on normal roadways? No. They will most
certainly not do that.
But that’s just it. What we need isn’t a more sophisticated road. We
need a less sophisticated road. Look at any of the PR shots of this
thing. What is really going on that is profound and useful is what I
call "idiot driver insulation".
That is the necessary and sufficient requirement to end human driving
completely. We are not waiting on technology. We are waiting to do the
simple things necessary to make the real problem not a problem.
I suspect that this move is shrouded in "connectedness" to distract
the Silicon Valley people while real progress is made.
The
connectedness of vehicles is utterly irrelevant. Weather sensors, a
perfectly good idea, are not essential. Platooning (as shown in the
image) may one day be helpful for many reasons but so might
teleportation. No, the real breakthrough here is simply clearing the
road of idiot drivers.
If you’re following my logic here, then one question should come up:
why not just build a light rail (like the kind my neighbors rightly
complain
about for the wrong reasons)? It is a good question and I think that
the value proposition works today because the physics benefits of
commuter rail are not as heavily leveraged as freight rail where steel
on steel is important to energy budgeting.
I believe that autonomous cars running on a closed single corrider
would still have some advantages. First is better on demand
scheduling; with rail the granularity is an entire passenger car.
Starting and stopping disadvantages rail but is easier for smaller
shuttles and even very small transportation pods. It is also easier to
integrate small freight shipments into the closed corridor. For
example, how would Amazon send a couple of delivery vans down a light
rail corridor? They can’t.
But the huge advantage and why I very much hope this project doesn’t
hype-fizzle away is that this arrangement provides the launch pad for
mixed mode cars. The reason you can’t sleep (alone) in your moving
Tesla (or Cadillac for that matter) is because it is not on one of
these roads. There are no other impediments. When people can choose
a road where they can stop driving completely, I believe they will.
Sure, 99.999% of all other roads require the dangerous and tedious
chore of driving, but if there’s one shining example of how it could
be, maybe people will start to think differently about how every road
might be organized. I certainly hope this will spark some discussion
and thinking about the correct way forward.
If you really want to understand this news you can go back and revisit
my 2016 post where I predict
this exact development. Indeed, although we have finally managed to
claw our way back to exactly where
we
were in 1997, in the world of autonomous vehicles, that is immense
progress. Let us hope we don’t retrogress again.
The bad news is obvious, right? It’s everywhere. Plague! House arrest
for everybody! Great Depression or Greatest Depression Ever?! Riots! A
president who once savagely beat a puppy to death with a golf club.
(Actually, I just made that last one up — but what is worse, that
false "fact" or a president who doesn’t mind citing such flagrantly
false facts? Feel free to make up your own "facts" about the president — he doesn’t care!)
Enough has been said about the bad news. I want to focus on the good
news. Before that can make sense, however, we need to consider the
ugly news — the news that makes no sense and is a muddled mess.
Unfortunately, that has been literally all news about the details of
this pandemic.
The current coronavirus disease, Covid-19, has been called a
once-in-a-century pandemic. But it may also be a once-in-a-century
evidence fiasco.
At a time when everyone needs better information, from disease
modelers and governments to people quarantined or just social
distancing, we lack reliable evidence on how many people have been
infected with SARS-CoV-2 or who continue to become infected. Better
information is needed to guide decisions and actions of monumental
significance and to monitor their impact.
That was back in March. Plenty of time for everybody to start to
figure out what sensible data looks like. And yet, it never came!
Every day there would be some fatuous headlines like these:
Another grand example of terrible data is the CDC. Check out their
CDC COVID Data Tracker.
The default that comes up is "Total Cases". What is so stupid about
this? Imagine a prolific creator of false facts using this "data" to
"manage" the pandemic. First of all one can have fewer "total cases"
by
simply
nerfing testing. But a really accomplished false fact artiste would
say something like, "COVID-19 cures most natural causes of death! Just
look at how the population is thriving under the therapeutic influence
of this wonderful virus. Which was my idea by the way." Go on, look at
that "data" in the CDC chart and tell me how that claim can be
refuted. It can’t.
That’s the ugly news. I feel like I have been watching an endless game
of Numberwang.
But I have seen a glimmer of hope! Let’s turn now to the good news.
Back in March, some rando on the internet — who also happens to be a
faculty member of the University of Cambridge’s Risk and Evidence
Communications unit — wrote a sensible article about
How
much normal risk does Covid represent?
The answer, summarized, was that your risk of dying pretty much
doubled.
The way I would explain that is to imagine playing Russian roulette
with a six-shooter that had two bullets instead of one. If your gun
had 800 chambers (equivalent odds for a normal person in their 30s),
going from 1 bullet to 2 wouldn’t be quite so problematic. Note that
everybody’s gun gets smaller as they age.
My real point there was that article way back in March was the last
example of good data considered sensibly that I have seen during this
entire mess — until today.
I feel too financially insecure to afford The Economist, but I do get
their email newsletter and sometimes their content is not paywalled.
Today they tipped me off to this unusually clear and cogent resource.
Stare at this and soak it in and you’ll understand what is wrong with
all other sources.
The first thing to notice is they’ve cut through the bullshit of
bullshit testing. The last time I saw such a muddled testing situation
was either standardized academic testing or maybe even IQ. Testing
sensibly for Covid pathologies and dangers is even worse than normal
CYA medical diagnostics. This data from The Economist understands that
and moves on to what can be sensibly inferred about the situation.
If we stop and ask ourselves what we care about, number one on the
list is not getting killed. So The Economist asks the simple question,
are we getting killed more than normal? Don’t worry about why. It
could be people dying of asbestosis because masks are hard to acquire
for construction work. Doesn’t matter. Step one is to answer the
question - is there a problem? And The Economist’s methodology is so
simple that it’s actually plausible. They simply look at how many
people used to die and how many people are dying now and see how/if
that has changed.
That’s so simple, but really it is a relief to me that someone is
looking at this in a way that is not deeply flawed to the point of
being meaningless. They even have a
nice
Github repo with all their data and scripts if you’d like to quibble
with their methodological details. (For example, I would make minor
adjustments based on
natural
changes of population, but really, not strictly necessary to
appreciate these numbers as better than anything we’ve seen thus far.)
It is "good" news in the sense that it is of good quality. It is also
good in the sense that it looks pretty good. I am not terrified by
this data.
At first I misunderstood the "Absolute" setting on the graphs to mean
they were like the fatuous CDC data. But really this data is much
more useful. It is not showing absolute deaths - which no one really
cares about. It is highlighting the validity of testing statistics.
The difference in Covid-19 attributed deaths and real deaths is
directly measuring the effectiveness of the testing programs. They are
frankly not as abysmal as I thought — though I live in NY where, like
NJ, Spain, Holland, Italy, and non-WEIRD nations, it is pretty
abysmal.
It’s interesting to note the negative rates after the spike. The
article cautions that some of this more recent data may just reflect
reporting lag. My intuition is that many victims were probably close
to death anyway and there were less people to die in June if they had
their exit schedule moved up to May.
Reports
of a "second wave" seem pretty overblown. Again, could be reporting
lag, but this data isn’t terrifying me. It’s looking like people got
sick, like they do when something is going around. All the vulnerable
and unlucky people got it, and it doesn’t really have much more room
to cause trouble. That could be a very wrong guess, but this data,
real actual data that is not meaningless, is in no way contradicting
that assessment.
I’m actually pretty impressed that a lot of states really are not
seeing anything like what New York and New Jersey saw. Look at
Wisconsin or Idaho or Missouri.
I would actually say that the danger for them is
greater. It’s greater because to my thinking they have not had the
full force of the pandemic come and hit their vulnerable populations
yet.
One could make the case that this shows that protective measures,
lockdowns, and hygiene theater are working. I’m not sure if there is
enough data for that by, say, comparing Sweden with Switzerland. But
this kind of data could tell us exactly what works. It’s not testing
as I had previously imagined. Testing is almost as elusive as a
vaccine; let’s just rule sensible tests out and focus on real and
obvious effects.
What really works is simply tracking deaths promptly and accurately
like any 21st century society should be doing anyway. With sensibly
tracked deaths, accurate studies to gauge the efficacy of various
measures are actually plausible, even using "natural" experiments with
different sides of the state line doing different things. It would be
really helpful to see these plots with the same plots superimposed
from other flu seasons and outbreaks.
As one of the early adopters of Covid-19 I can tell you that if you
get it, it’s no picnic. (I am feeling a bit better than last month,
thanks.) But this data is not only good in the sense of not being
stupid; it’s good because it doesn’t show an impending apocalypse.
It’s exactly data like this that we need.
UPDATE 2020-08-10
A perceptive reader sent me this NYT article from late May:
Putting
The Risk Of Covid-19 In Perspective. It tries to focus on the
comparative risk side of things. That would be valuable and correct if
there was sufficient data to support such aspirations. Note what this
article uses for data.
Using data from the [CDC], New York City experienced approximately
24,000 excess deaths from March 15 to May 9, when the pandemic was
peaking.
…
…using the C.D.C. data, Michigan had approximately 6,200 excess
deaths during this same time period.
Note he’s not comparing those directly but using excess deaths — the
only metric we really have right now — to go on to put the risks into
perspective. Excess deaths is finally some data that makes sense.
