Hacker News Reader

Comment by blackhaz | Hacker News Reader

OK. Now please do the same with a Z-80 class CPU.

In my experience, if there was a well-defined problem in the 70s or 80s, some clever person came up with a workable solution using a tiny microprocessor with limited RAM.

The way I think about it is... when I was a kid I hated waiting at lights and imagined you could build a computer vision/ML system (this was in the '80s) that coudl recognize traffic and change the light to green when it was safe, rather than timing out. I mentioned this to some traffic engineer and he pointed out the simpler solution was to put magnetic detectors under the road and then just look for blips which are cars driving over.

And then someone rolls up with a bicycle, and because it doesn't have as much metal in it as the detector expects, they end up sitting at a red light forever. Been there, done that.

> And then someone rolls up with a bicycle

My local town has a roundabout with a cut-through lane for buses. The cut-through has a traffic light to help the bus emerge. Every so often, a (presumably non-local) car driver accidentally enters the bus lane, and then has to wait for the next bus to trigger the traffic light.

Compared to the rest of the engineering, I don’t think the computation is all that hard. It doesn’t need to be that fast since you only need to do a correction maybe every 10-20 minutes or so. The math was innovative in that they invented the Kalman filter for these types of computations. You pick your star such that it’s by far the brightest one in its neighborhood. So the image processing is just finding the brightest pixel.

From what I’ve read (mostly the excellent book “Inventing Accuracy” the main challenge is very precisely aligning the telescope with the “stable member” in the gyroscope.

Did the units in question even have pixels?

Not exactly. The SR-71 system had an expected position of the star, and then it would mechanically scan the telescope in a rectangular spiral scan pattern with a single pixel photodetector until it found a star with the expected brightness. This could take up to 22 minutes from a semi-cold start (roughly, you know your heading within 3 degrees, and your position within 1 degree lat/lon). [1] Once the system had shot a couple of stars and the IMU was dialed in, it could do an update in about 30 sec.

The Trident I used a Vidicon tube (what they used to use for TV cameras before CCDs were invented) [2] The Trident approach was to shoot a single star once in the flight.

The Trident II uses a CCD. The first version of the guidance system for that missile (The Draper Mark 6) used a 90x90 CCD [3]. That has been replaced with the Mark 6 Mod 1, which they say has increased resolution but they don't give a figure [4]

[1] NAS-14V2 ANS System Manual, Page 10A-49 https://airandspace.si.edu/webimages/collections/full/NAS-14...

[2] Inventing Accuracy, Page 290

[3] The Trident II (Mk-6) Guidance System, Page 1460 https://sci-hub.se/https://doi.org/10.2514/6.1991-2761

[4] Modifying the MARK 6 Guidance System Part 1 https://www.electronicdesign.com/home/article/21201730/modif...

I don't know the technical specs of the Z-80, but even on a minipc's ancient and super cheap CPU (I think it's a 2 point something ghz dual core) it can do a solve in seconds.

If you know roughly where you're looking, the lookup is really quite quick.

GHz? Try 2 MHz.

The hardest part would be doing the image recognition, but that wouldn't be that bad since the position (and therefore expected picture) is already known, a priori, to a good accuracy.

OK, now please do it without a computer or calculator

Reply on news.ycombinator.com