Thanks, another interesting find... just to record some points from the document:
- author Fujiko Oguri of Ohio University
- August 1983
Quote:
The MOS Technology 6502 Super-Jolt does all of the microprocessor work. The software is written in 6502 assembler language
...
The Am9511A by Advanced Micro Devices is a peripheral mathematics processor which does the necessary floating-point calculations. It is designed to be used in microprocessor systems which have an eight-bit data bus. It can handle 16-bit and 32-bit fixed point arithmetic, 32-bit floating-point arithmetic and trigonometric functions using a stack-oriented operand storage (sixteen 8-bit words).
A 6820 PIA is used to interface with the 9511. Four 2k EPROMs hold the software.
Quote:
Total RNAV calculations require about 0.20 second to complete with the A149511A and the 6502 running at a clock speed of 1MHz.
They did 19 test flights, with real-time position calculations on the microprocessor system, later compared with results from a Fortran simulation running on an IBM 4341.
The nominal 1MHz clock of the 6502 system needs to be stable - they replaced the original and got much better results.
Interesting survey of navigation systems available at the time, including
- inertial
- Omega (VLF, global coverage with 8 stations)
- GPS:
Quote:
GPS is still being tested and is expensive; however, the implementation will begin in 1986 and will be completed approximately two years later. It is regarded by some as a future navigation system which has a capability to be a single universal system.
- Loran-C (LF, continental US coverage with 40 stations, 0.3 nautical mile accuracy)
The idea is to make a cheaper Loran-C receiver than the $40k state of play.
The nature of radio wave propagation close to the surface is such that a receiver needs a detailed map of geographical impedance. (GPS is much simpler in this regard!) Or, possibly, calculation can come to the rescue.
Some tens of multiplication and trig operations are needed per point, and the overall position calculation takes 1.5 seconds. (Possibly this is the time for a naive approach and the paper intends to do better?)
It's necessary to filter over 4 or 16 samples to get a reliable result.
Quote:
Computing the GS between the present point and the oldest 16 cycles just minimizes the random noise errors. More cycles might reduce random noise errors; however, they consume more memory space and cause large errors on turns because the distance between a straight line and a curved line becomes larger.
Their approach of filtering over 4 samples is better for rapid travel.
Quote:
Although the response is still slow for accelerated flight, it is adequate for most flight conditions.
Source code in Fortran and 6502 assembly language is included - they used a cross-assembler running on the IBM.
Login
Register
FAQ
Search
