Home
AGS
LVDC
Gemini
Download
Document library
Change log
Bug and issues
Developer info
Virtual AGC — AGS — LVDC — Gemini

Peripheral Components
for use with yaAGC

FAQ
yaAGC
yaYUL
yaDSKY
yaOtherStuff
Luminary
Colossus
Language Manual
Physical Implementations

Table of Contents



VirtualAGC

The Basics

You may recall seeing an extremely abbreviated description of the VirtualAGC program on the Virtual AGC project home page.

VirtualAGC
is a GUI front-end which allows you to choose almost any reasonable set of options related to the emulation, and then to run all of the emulation programs needed, in a way that allows those programs to intercommunicate properly.  VirtualAGC does little for you that you couldn't have done from a command line using the various programs and their command-line options described on this page and on the yaAGS, yaAGC, yaYUL, and yaDSKY pages of this website, but it's safe to say that it will almost always be easier to accomplish any given task using VirtualAGC than using equivalent command-line operations.

The basic VirtualAGC screen looks like this:

Screenshot of VirtualAGC program.

The screenshot above depicts the one and only window of the VirtualAGC program.  There are no menus, toolbars, hot-buttons, or other controls.  While a large number of options are presented, you don't necessarily need to change any of the selected options.  The defaults are as shown, and if you simply click the "Run!" button at the bottom of the window, the simulation will run. 

Running the Simulation

When you actually do run the simulation by hitting the "Run!" button, the large VirtualAGC screen shown above politely disappears in order to free up the screen space it's using, and instead the a tiny window that looks like this replaces it:

Simulation-running window.

This window is gimicked to stay atop all of the other windows open on the screen, but you can minimize it to make it disappear if you don't like seeing it.  Note, however, that while the simulation is running you won't be able use VirtualAGC's controls to open a new browser window for viewing the source code, so if you want to do that you'll need to do it before hitting "Run!". 

To end the simulation, simply exit from any of the visible elements of the simulation, such as the simulated DSKY.  Within a few seconds all of the other elements of the simulation will automatically terminate and the large VirtualAGC window will return.  On some platforms, there may be curious exceptions to this behavior that result in some windows needing to be explicitly manually closed, but closing the DSKY is a good recommendation for all platforms.  The small simulation screen depicted in the screenshot above cannot itself be closed manually, so do not expect that the simulation can be ended by closing this screen.

We'll discuss the "More", "Less", and "Uplink" buttons later.  You don't need to know about them for basic operation.

All About Settings

If you change any of the settings on the VirtualAGC screen, the program will remember those changes and the settings you've selected will be the ones that appear the next time you run VirtualAGC.   On the other hand, you could click the "Defaults" button at the bottom of the window to restore all of the settings to their defaults. 

There is, however, a subtle distinction between closing the VirtualAGC main window using the "Exit" button at the bottom of the window, and using the operating-system-supplied controls on the border of the window.  The settings are saved only when the "Exit" button is used.  They'll not be saved if the window's border controls are used instead.

All of the settings are intelligent, in the sense that not all settings are reasonable in combination with other settings, and so illegal settings are grayed out and disabled.  As you change some settings, it may cause other settings to become enabled or disabled.  So it's hard to choose a combination of settings that don't work reasonably well together.

The Simulation Type settings

The left-hand pane of the screen is used to select the specific software to be run on the simulated AGC.  Included are all of the versions of the Colossus and Luminary AGC programs that have been made available to the Virtual AGC project.  The grayed-out mission/spacecraft combinations in this area are the ones for which the original software isn't yet available, but in many cases is known to exist in museums or private collections.  But we maintain hope that they'll eventually be made available to the admiring public, so we put them on the list anyway! 

The "Validation Suite" choice was not, of course, used in any actual Apollo mission.  However, at one time there was a software validation suite that checked out the AGC CPU registers and instructions for proper behavior.  (I surmise that this code was originally intended to be included in the flight software as part of the built-in self-test, but was mostly removed due to memory constraints.)  While that software no longer exists, as far as I know, the program documentation for it does exist, and the Validation Suite program was created by closely following the documentation of the original validation suite.  And a good thing, too, since in doing so a number of implementation errors were uncovered that might have lingered to cause mysterious problems later! 

You'll also notice a box for selecting "custom" software.  This would be software you had written yourself, or perhaps acquired from enthusiasts.  Normally all of the selections in the Simulation Type pane are expected to be executable binary code which has been pre-assembled from source code, so if you have your own AGC source code you'll want to read below about how to assemble it.

The Interfaces settings

The middle pane of the screen is mostly devoted to selecting the particular set of peripheral devices—in most cases, interfaces to the simulated CPUs—which you wish to simulate. 

Here's a brief description of all of the devices appearing in the interface pane:
You'll also notice a couple of buttons in the Interfaces pane.  The "Novice" button is a short-cut that simply deselects everything except the minimal set of devices, namely the AGC and DSKY simulations.  Conversely, the "Expert" button is a short-cut that selects every device that's reasonable with the spacecraft and mission type you've selected in the "Simulation Type" pane.

The LM Abort Computer (AEA) settings

Here you can select the specific software set which will be run on the Abort Guidance System (AGS or AEA) simulated CPU.  It would have been more logical to include this in the "Simulation Type" area, but alas! there's only so much space available.  In the screen-shot above, this area is completely grayed-out because the "LM Abort computer (AEA)" box is not checked in the "Devices" area, so the AGS/AEA simulation would not actually be run.  If it were enabled, however, any AEA flight program to which the Virtual AGC project has been given access could be selected.  Other versions of the AEA flight software which are believed to have existed (and hoped to still exist) appear also on the selection list, but are disabled and grayed out until the happy time when the Virtual AGC project acquires access to them.

As with the AGC software selection, it's possible to select custom software in place of actual mission software if you wished to write your own AEA software.  As with the custom AGC software, it is expected that the software you select is a pre-assembled executable binary.  But also see below about assembling AEA source code.

The Options settings

The right-hand area of the screen allows you to set various minor options associated with the exact manner in which the various emulated devices are run.  I won't bother to explain most of these in great detail since they're pretty self-explanatory, but here are a few words of explanation anyway:
It's important to note that halting the simulation and then resuming it later is not like the pause feature in a video game, because Virtual AGC is not a computer game.  The ability to resume program execution is not something that has been layered atop the simulation ... rather, it is simply a characteristic of the computer memories used at that time: the AGC retained the contents of memory across power cycles, and therefore Virtual AGC does as well.

If your only interest is in the guidance computer itself, then this distinction is unimportant.  However, if you are interested in the system as a whole, then the distinction is important because it means that only the state of the AGC is preserved, and not the state of the entire simulation.  In particular, the states of the IMU, FDAI, propulsion system, discrete inputs, etc., from the LM-Simulator program are not preserved.  For example, if you stopped the simulation in the middle of a descent to the lunar surface, when you resumed it later the AGC would think it was still in the middle of the landing, but the orientation of the spacecraft would have changed, the fuel tanks suddenly would be full, etc.
Resizable
                telemetry screen
"Resizable" Telemetry Style
"Retro" telemetry
                screen
"Retro" Telemetry Style

AGC debugger
Screenshot of AEA debugger

Browsing AGC or AEA Source Code

Having chosen the AGC software suite which you wish to emulate, you might like to actually view the software listing as well.  That's what the "AGC" button under the "Browse Source Code" heading is for.  Pressing the button displays an assembly listing for the selected software in the default browser (such as Mozilla Firefox, Internet Explorer, or Safari) of your computer.  At present, this won't display the Apollo 15-17 CM software, because even though we have a verifiable executable for it, we do not yet have the full source code.  Nor will it display your custom software, since we assume you know how to display that yourself!  However, it will show the source code for any of the other valid software selections.

Similar comments apply to browsing the AEA source code.

Assembling AGC or AEA Source Code

A fun thing to do (in an über-geekish way) is to write your own AGC software, just as I have written the Validation Suite.  So naturally, one of the options mentoned above is to run your own custom software.   If you simply type a filename into the box which is supposed to contain the filename of your custom software, it's expected to be a binary file which has already been run through the assembler.  However, if instead of typing in a filename you use the "..." button button located next to the filename-entry box to select a file by means of a file-selection dialog, you can select either a binary file or an AGC source-code file.  If you choose an assembly-language source file, then VirtualAGC will thoughtfully run the yaYUL program to assemble it for you automatically and will place the name of the binary executable file it creates into the filename selection box.

Similar comments apply to selection of custom AEA programs:  If you select custom source code rather than custom binary code via the "..." button button in the AEA-program selection area, then VirtualAGC will automatically run the yaLEMAP program to assemble it for you and place the name of the newly-created executable binary in the filename selection box.

How to Create Equivalent Command-Line Shell Scripts or Batch Files

In the small simulation window that pops up while the simulation is running, there are two buttons labeled "More" and "Less". 

The "More" button can be used to expand the simulation window so that it displays the contents of a shell script (or a Windows batch file) that could be used to run the simulation from a command-line (or by other means) without needing to invoke VirtualAGC

Screen shot of simulation
        window expanded by using the "More" button.

This information would also be useful to you if you wanted to create a custom setup too complex for VirtualAGC to handle, such as running a Command Module simulation that had two DSKYs, or if you wanted to run the simulation on several PCs ganged together via a network. The displayed commands can be cut-and-pasted into a text editor for subsequent modification.  Clicking the "Less" button causes this extra display to disappear again from view.

Not only that, but running the simulation from the command line using scripts can be very helpful if troubleshooting needs to be performed.  In normal operation, VirtualAGC tries to hide as much complexity from you as possible, and one thing which it hides from you is the messages printed out by the various programs in the Virtual AGC suite.  These messages include such helpful information as socket connects and disconnects (as the various peripheral devices try to communicate with the simulated CPUs), and the joystick positions being reported by the ACA simulation to the AGC.  The latter are helpful to know because there are some joystick-related configuration differences from platform to platform and joystick to joystick.  When running the simulation directly from the command line via a script, all of these messages become visible.

You could, of course, reconstruct all this information yourself by examining the documentation of the various component programs, but it's much simpler to adapt an existing script rather than to create one from scratch.

Getting the Simulation to a Known State

One thing that quickly becomes apparent in running the AGC simulation is that setting up the AGC to a given state can be very time-consuming, so it is helpful to have a method of automating that process.   VirtualAGC actually provides two separate methods of setting up the simulated AGC to a known state:  Digital Uplinks and resumption from core-dumps.  Each of these methods has advantages and disadvantages relative to the other, so it is not possible to provide a blanket recommendation for just one of them.  There may even be cases where a combination of the methods is the best approach.  In general, the disadvantages of the Digital Uplink are that it is slower to use and more cumbersome to set up than core dumps.  The disadvantages of the core dump are that although the CPU state is restored, the states of peripheral devices are less likely to be in sync and that it is impossible to combine core dumps (whereas it is possible to combine different digital uplinks) containing initialization of different subsystems. 

Digital Uplinks

Digital Uplink was a method that could be used by mission control to transmit data to the AGC, and is the complement of the "digital downlink" for transmitting data from the AGC to mission control discussed in the yaTelemetry section below.

What the digital uplink implemented was essentially a stream of DSKY keycodes.  Thus, any condition in the AGC which could be set up via the DSKY could also be set up via digital uplink.  I'm not sure exactly how this was done in the original missions—a DSKY in the control room? a paper tape?—but in VirtualAGC it is accomplished by creating a file containing a script of DSKY keycodes.  The digital uplink is the single piece of functionality provided by VirtualAGC itself rather than by other yaPrograms, so if you need a digital uplink there's presently no way to provide it without interactively running VirtualAGC.

To send a digital uplink, you click the "Uplink" button in the simulation window, and then select the DSKY-keycode script you'd like to use from the file-selection dialog which pops up.  Once you've done so, the "Uplink" button changes to a "Cancel" button, and the simulation window acquires a new pane which shows the status of the uplink in progress:

Screenshot of digital-uplink
        status screen.

Eventually, this method may be used for uplinks to the AEA as well as to the LM or CM AGC, but at present I don't know enough about the uplink system for the AEA to implement it. 

Here are the rulea for creating DSKY-keycode scripts for digital uplinks to LM or CM AGC simulations:
If you try to relate this information to the screenshot above, showing an actual uplink in progress, it's pretty straightforward to do so.  The only point worth mentioning is that the dots appearing in the screenshot are where additional 0.5 second delays appeared in the script.

There are, however, a few subtleties which the rules above impose on the scripts which may not be obvious, so let's illustrate this with a real script:

L                    # For LM
V05N09E              # Show program alarms
"          "         # Delay 5 seconds.
V91E                 # Commence showing memory-bank checksums.
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank.
"          "         # Delay 5 seconds.
V16N36E              # Commence showing time.
"          "         # Delay 5 seconds.

The first subtlety is that since the space character is used to insert extra transmission delays, while the tab character is not, and yet tabs and spaces are visually indistinguishable, it's not immediately obvious in looking at the script where the extra delays have been inserted.  I've handled it in this script by enclosing the spaces in quotes.  The quote character is simply discarded when the script is parsed, and so its presence causes no harm.  There are many other methods which could be used, of course.

The second subtlety is that if you like to format the script nicely, by doing things like aligning the comments, you'd better do so using tab characters only or else you'll have lots of unwanted delays added to the transmission.

The extra delay at the very end of the script has no functional significance, but allows the status display to remain on the screen 5 seconds after transmission of the "V16N36E" command, rather than disappearing instantly on transmission of that command.  I did this merely to make the status message readable.

Note:  The AGC is not able to properly accept digital uplinks at arbitrary times, and must be prepared in advance to do so.  I'm not sure at this writing exactly what preparation is required, but (at least after a clean power-up) a DSKY-test a "goto pooh" command (V37E00E) entered directly from the DSKY keypad seems to do the trick.  After the DSKY test completes, a digital uplink is accepted.

Core Dumps

An alternative to Digital Uplinks in getting the simulated AGC CPU to a known state is to use the method of "core dumps" instead.

As mentioned above, when the simulated CM or LM AGC is running, it saves a snapshot of its simulated ferrite-core memory periodically.  Nominally, these "core dump" files are saved at 10-second intervals.  The core-dump files are named "CM.core" or "LM.core", depending on whether it is a CM or an LM simulation being run.  One of the items under the Options Settings, "Resume from ending point of prior run", allows you to start the next simulation run with the memory contents and CPU state of the last relevant core-dump file.  (In other words, a CM simulation can be started from the last CM.core, and an LM simulation from the last LM.core.)  Thus, to get the AGC into a desired state you could use the DSKY to set it up, and then resume the simulation later with those settings.

The difficulty, of course, is that after you had resumed the simulation, execution of the simulation might result in an alteration of your setups, so if you resumed the simulation still later you would no longer have the same setups.

This is handled by an additional item ("Custom") in the AGC Startup area in VirtualAGC's Option Settings pane.  What this item does is to let you save named copies of your LM.core or CM.core files (using the "Save" button), and then to later resume execution of the AGC simulation from your named core-dump files rather than from the generic LM.core or CM.core files.  The core-dumps made after resumption of the simulation are still called LM.core or CM.core, so your named core-dump file is not affected by the continuing simulation and could still be used later to exactly duplicate your setups.

You can make as many saved core-dump files as you like, and the area under Options Settings allows you to either type in the name of the file directly, or else to browse for it using the "..." button.

It is perhaps worth noting also that while LM.core and CM.core are created automatically at 10-second intervals, if you are running AGC under its debug monitor then you can use debug-monitor commands to manually save core-dump files (with names of your own choosing) that are accurate to the machine-cycle at which they are saved.  By default, VirtualAGC likes to save and look for core-dump files in the scenarios/ folder, so if you save core-dump files from the debugger it is best to prefix the names with "scenario/".

Scripted operation for tutorials, museum exhibits, and other demo purposes

(For related functionality that can be used interactively rather for unattended operation, see the yaDSKY page.)

The Digital Uplink capability described above has a special feature that could be used in a variety of instructional settings, of which a few are listed in the title of this section.  Specifically, it is possible to embed commands in the digital upload script which cause programs of your choice to be run when those positions in the script are reached.  Specifically, when the uplink script finds an '!' character, any text between the '!' and the end of line is treated as a command which would be suitably run from your operating system's command-line.

Interesting software which might be run in this fashion includes:
For example, you could imagine creating a tutorial on some particular sequence of operations on the DSKY, where at each critical point a window opened up to describe the operation that was being performed.  Or you could imagine a museum display with a simulated DSKY on an LCD screen, and an audio commentary synchronized to the operations being performed on the DSKY.

Here is the example digital uplink script used earlier, but with embedded commands for audio playback using the madplay mp3-player program found on many Linux computers:

L                    # For LM
!madplay DescriptionOfProgramAlarms.mp3

V05N09E              # Show program alarms
"          "         # Delay 10 seconds.
!madplay ShowingMemoryBankChecksums.mp3
V91E                 # Commence showing memory-bank checksums.
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank
"          "         # Delay 5 seconds.
V33E                 # Show next memory bank.
"          "         # Delay 5 seconds.
!madplay MonitoringTimeSincePowerup.mp3
V16N36E              # Commence showing time.
"          "         # Delay 5 seconds.


As of yet there is no feedback to the digital uplink that could be used to pause the uplink or branch to alternate possible uplinks, but if anyone has a serious use for this kind of capability, I'm sure it could be added.



Contributed Code

LM-Simulator, by Stephan Hotto

This is a kind of proof-of-concept program, which is in the process of becoming an IMU simulation.  It's pretty close to being complete.   But the program is useful even without a complete IMU simulation, as a nifty little monitor for the CPU's i/o channels.  You can use it to see the output on output channels, to generate input on input channels, or else to simply monitor or generate arbitrary messages on the socket interface.  At the moment, it's specifically targeted towards the Lunar Module simulation, but because there is a very great overlap between the way peripherals like the IMU were implemented in the LM and CM, you can profitably use LM-Simulator with the Command Module simulation as well.  Nevertheless, be aware that not everything you see in the CM version of the program is actually present in the CM!  Also, at present the program interacts only with the AGC simulation (yaAGC).

LM-Simulator is provided in the Virtual AGC developer snapshot and all of the binary packages, and is completely integrated into the VirtualAGC GUI.  So you don't have to do anything special to get it or to run it.  The program is written in the Tcl/Tk script language, however, so you do have to have Tcl/Tk installed.  Like most of the rest of Virtual AGC, LM-Simulator is licensed under the GPL. 

Here are some notes from Stephan on its use, as edited somewhat freely by me.  The program is a moving target, and I'm combining text from a number of different emails here, so you can blame me for any incoherence.

As a by-product [of the IMU development] I've created a small Tcl/Tk program to monitor the activity of the output channels and to manipulate dedicated bits of the input channels, which is probably  useful in developing the different interfaces.  Because it is written in an interpreter language you can use it on Linux/Windows/MAC without any change. The only pre-condition is an installed Tcl/Tk environment. For example under LINUX you can simply type "wish lm_simulator.tcl".   [The VirtualAGC GUI can automatically start LM-Simulator, so you really don't need to start it explicitly, but you can also explicitly start it as described.]

The monitor connects to the localhost on port 19801 (IMU reserved port) [or port 19701 for the CM] and is focused on the Luminary 131(1C) channel allocation.  If you, for example, want to simulate a "Temperature of Stable Member out of Design Limits" then you can set the input channel 30 to "100000000000000" and the DSKY Temperature warning will go on.  [After he wrote this, Stephan implemented a checkbox for this particular input, so it's actually much easier than he says, but what he says here still works.]

The program is split into the following modules which can be called by the menu of the main program:

lm_simulator.tcl Main Program for the LM version of the program
lm_simulator.ini
The default configuration file.
AGC_Crew_Inputs.tcl Nearly all Crew Switches connected to the AGC
AGC_Outputs.tcl All binary AGC outputs interpreted by showing their status
AGC_System_Inputs.tcl All binary LM->AGC System Inputs
AGC_IMU.tcl First steps into the IMU & FDAI (not completely working, need some more counters)
AGC_Attitude.tcl
For manually feeding acceleration or changes in physical attitude into the IMU.  (Eventually, of course, this will be accomplished by modeling the thrust and rotation of the spacecraft.)
AGC_DSKY.tcl
A DSKY simulation that can be used in place of yaDSKY.

Stephan has also provided a helpful tutorial.  The tutorial will be found in the Contributed/LM_Simulator/doc directory of the developer snapshot, or can be activated from the LM-Simulator main menu when actually running the program.

Here are some (possibly out-of-date) screenshots from the program.

Main screen
Main screen
System inputs
Inputs from system
Crew inputs
Inputs from crew
Outputs
Outputs
IMU stuff
IMU
Propulsion-system status screenshot
Propulsion-system status

DSKY LITE and
                    yaDSKY side-by-side
DSKY Lite with yaDSKY


The program has a configuration file called "lm_simulator.ini" which can be used to configure some aspects of LM-Simulator's operation.  The program has built-in configuration options, but these can be overridden with the configuration file, which in turn can be overridden by command-line options.  I won't bother to describe this configuration file in detail, since you'll see how it works if you look at it.  Basically, it tells you which port to use to connect to yaAGC, and determines which of LM-Simulator's windows to automatically open at startup.

As of 2005-06-19, the usage of LM-Simulator is as follows:

cd InstallationDirectory
lm_simulator [OPTIONS]

The currently-recognized command-line options are:

--port=PortNum
Changes the port number (by default, 19801) used to connect to yaAGC.  This can also be changed in the configuration file.

--cfg=IniFilename
Used to change the name of the configuration-file used.  By default, the file is lm_simulator.ini, in the installation directory.  With this option, you can change the name or directory of the file.

If you discover problems with LM-Simulator, or want to cooperate on it, you'll probably want to contact Stephan directly.  (His contact info is in the source code.) 


Game Pack, by John Pultorak

John Pultorak, of Block 1 AGC hardware-simulator fame, has provided a Game Pack containing a 1- or 2-player tic-tac-toe game and a Simon game.  This is in the source tree under Contributed/GamePack/.  Unfortunately, it does not yet run on VirtualAGC for a couple of reasons.  If anybody cares to fix it up, I'd be ever so grateful.
  1. It is written in the syntax of John's assembler rather than yaYUL, so it cannot be assembled; and
  2. It depends on the presence of Colossus rather than being a standalone program, but there is no room left in Colossus in which to fit it.
The reason it depends on Colossus is for such stuff as accessing the DSKY, which it could do by directly accessing the DSKY through i/o channels.


yaUniverse

yaUniverse is a program that physically models the motion of the spacecraft and of the heavenly bodies visible to the spacecraft or affecting it.  From knowing the initial time, position, and velocity of the spacecraft, and application of physical laws, yaUniverse is able to calculate the position and orientation of the spacecraft and heavenly bodies at all times relevant to the mission.  At present, the program exists and accounts for gravitational influences, but is not integrated with other Virtual AGC software such as the (currently non-existent) yaIMU program that would be the principal consumer of the data.

Eventually, all of the forces acting on the spacecraft might be accounted for, such as:
Originally I intended not to calculate the positions of heavenly bodies in real time, but rather to use pre-calculated or pre-tabulated ephemeris data.  However, the amount of ephemeris data is pretty large, so I've instead decided to use the laws of physics to track the heavenly bodies as well as the spacecraft.  Note that the initial positions and velocities of the heavenly bodies still need to be obtained somehow, but they can simply be gotten for any given mission epoch by downloading them from the Jet Propulsion Laboratory's HORIZONS system at
telnet ssd.jpl.nasa.gov 6775
I'll do all of the downloading, of course, though at present (2004-09-23) I include only Apollo 8 data in the development snapshot.

An additional complication is that even though I've spoken above of "the spacecraft", there is not just a single spacecraft.  Rather, there several spacecraft, which at any given time in the mission may be docked or separated:  the CM, the SM, the LM's descent stage, the LM's ascent stage, the Saturn stages, and various combined versions of these.  The motion of each must be tracked.  For example, it may be necessary in the course of the mission for the the astronaut to use the CSM's AOT (see below) to mark the position of the LM.   yaUniverse provides the data for this to yaAOT, and thus must be able to simultaneously track the CSM and the LM.

Like yaAGC, yaUniverse would be a server from which yaIMU and yaAOT obtain data.  A TCP socket interface is used for this, though the data protocol is presently TBD.

yaUniverse requires no user interface, other than a way to define the starting time, positions, velocities, and physical characteristics of the spacecraft.  However, it would be convenient to have at least some kind of running printout of positions, velocities, orientations, and masses.

The barest beginning of yaUniverse exists.  It is presently capable only of numerically integrating the positions of the heavenly bodies and spacecraft, but not of communicating this information to yaAGC.

The syntax is:
yaUniverse [OPTIONS]

The recognized options are:

--help
Displays textual info similar to that shown here.

--mission=Name
Selects the name of the mission, which determines the initial positions of the Earth, Moon, Sun, Venus, Mars, Jupiter, and Saturn for the mission, and thus the gravitational influences on the spacecraft.  The mission names, by convention, are "Apollo8" (without quotes), "Apollo9", etc.  The default is "Apollo8".  The actual ephemeris files used have names like "Ephemeris-Earth-Apollo8.txt", "Ephemeris-Moon-Apollo8.txt", and "Ephemeris-Sun-Apollo8.txt", but this is transparent to the user.  The Apollo8 mission is special, in that it contains complete ephemeris data (rather than mere initial conditions) and hence can be used for testing yaUniverse's ability to perform numerical integrations of planetary positions.

--ephem-read
Causes yaUniverse to display ephemeris data and then quit.  It's purpose is really just to test that it can correctly read ephemeris files.  It forces --mission=Apollo8.

--ephem-int
Causes yaUniverse to print a report testing its numerical integration algorithms and then quit.  It forces --mission=Apollo8.  Specifically, what it does is this:  From the initial positions and velocities of the supported heavenly bodies (Earth, Moon, Sun, Jupiter, etc.), it computes locations of all heavenly bodies for the complete Apollo 8 mission epoch.  It then compares these with the pre-tabulated ephemeris.  Only the error in the Earth/Moon positions is really interesting, since that's the region of space in which the Apollo spacecraft operated. At present, with the default settints, the cumulative error in Earth/Moon positions at the end of the 9-day epoch is about 0.35 km (which I consider acceptable, but which I'd like to improve in the future).

--runge-kutta=N   
The order of the Runge-Kutta numerical integration.  N=2 or 4 (default is 4).

--planets=N       
The number of planetary bodies used in the numerical integration.  N=3-15, and is 7 by default:
N=3   Earth, Moon, and Sun.
N=4   Same as N=3, plus Jupiter.
N=5   Same as N=4, plus Saturn.
N=6   Same as N=5, plus Venus.
N=7   Same as N=6, plus Mars.
N=11  Same as N=7, plus Ganymede, Io, Europa, & Callisto.
N=15  Same as N=11, plus Titan, Tethys, Rhea, & Dione.
The addition of Titan et al. makes a big difference in the error of Saturn's position, but has no obvious effect on the inner solar system.  Similar comments apply to Galileans and their effects on Jupiter and the inner solar system.  Mercury and Uranus also have no obvious effect at all.  Note:  If the Galileans are added, you will need to adjust the timestep (see below) downward, say to 7200, to account for the very short orbital periods of some satellites.

--timestep=T      
The time, in seconds, used as the timestep for the numerical integration when only gravitational effects present, and the spacecraft are not close to the planetary bodies.  The default is 6 hours (21600 seconds).  The value must be either an exact divisor or exact multiple of 3600.  Intermediate values (at times between the timesteps) are obtained by interpolation.


yaIMU

Since Stephan Hotto's contributed LM-Simulator program provides an IMU, a separate yaIMU program is no longer planned.  However, if you're interested in providing one, feel free to proceed.  (Independent implementations are always useful for verification.)


yaAOT

yaAOT would be a simulation of the Alignment Optical Telescope (AOT).   yaAOT would be a client of both the yaAGC and yaUniverse servers.   From the orientation of the spacecraft (obtained from yaUniverse), and from the orientation of the telescope with respect to the spacecraft (which is initially driven by yaAGC), yaAOT would be able to compute the direction which the telescope is pointing.  From starcharts, it would be able to display a star-field on the PC.  The astronaut could then manually adjusts the orientation of the telescope (i.e, the star-field) to point at the objects he is attempting to mark; these objects could be stars, reference points such as the Earth horizon, or the other spacecraft—i.e., the LM or CM.  The new orientation of the telescope could then read back by yaAGC.

No work on this program has yet been done, other than researching available star data.  Star-data is available online and, as nearly as I can tell, the Centre de Données astronomiques de Strasbourg is the generally-recognized place from which to download them.


yaACA, yaACA2, and yaACA3

The Attitude Controller Assembly (ACA)—also known as the rotational hand-controller (RHC)—is used by astronaut to affect the pitch, roll, and yaw of the LM.  José Portillo has described the interaction between the ACA and AGC in great detail in a paper (klabs.org/mapld04/papers/g/g202_portillo_p.pdf) which is the basis of all ACA-related work in Virtual AGC.  Refer to the developer page and to the assembly-language manual if you're interested in knowing more about the integration between the ACA simulation and yaAGC.  Note that yaAGC retransmits any input-channel data received from the ACA simulation, so other onboard devices (such as yaAGS) wishing to receive RHC data automatically receive this data even though not connected to the ACA simulation.

In Virtual AGC, the ACA is simulated by one of three essentially equivalent programs called (surprise!) yaACA, yaACA2, and yaACA3.  Since you are unlikely have an actual ACA unless you're very, very lucky indeed, the simulated ACA instead uses a 3D joystick like that used for many computer games.  The ACA simulation interacts with the joystick and sends the simulated AGC CPU information related to the displacement of the joystick from its neutral position.

Why three separate programs?  Well, it has turned out to be much more difficult (for me, anyway!) to write a joystick program that I can be confident works on all supported platforms than it has been to make sure that the remainder of Virtual AGC is portable across platforms.  Therefore, as time went on, I found the need to experiment and to create several such programs.  Technically, the three programs differ in using different underlying libraries (not written by me) to access the joystick, but what's important to you as a user is that if one of the yaACAx programs doesn't work for your operating system or joystick, another might.  So I pass all of them along to you and let you choose.

At the present time, yaACA3 is regarded as the default and best of the programs, since it appears to work on all supported platforms.  yaACA2 presently does not work on Mac OS X and yaACA presently does not work properly on Windows.  Neither yaACA nor yaACA2 works on FreeBSD, or at least I haven't jumped through whatever hoops might be needed to make them work.

As far as joystick selection is concerned, I use a Logitech Extreme 3D Pro, and all configuration defaults have been tailored for that model.  In theory, any twist-handle joystick should be usable, though possibly with some reconfiguration.  Only the most basic 3D joystick is needed, since only 3 degrees of freedom (roll, pitch, and yaw) are used, and no buttons are used.

Selection between yaACA, yaACA2, and yaACA3 and reconfiguration related to different joystick models can be done from a command-line, but to me that seems to be making a difficult chore even more difficult than it needs to be.  So I've provided a GUI front-end (jWiz) for this chore, as described in the next section.

jWiz, the GUI front-end for joystick configuration

As mentioned above, three separate joystick-handler programs are provided, and we hope that by default the best one of them is automatically used by VirtualAGC and that the default joystick settings of that joystick-handler program are correct.  But what if they're not?  Then you have to do a little work to correct the joystick settings.  On the VirtualAGC main screen, you may have noticed an unexplained button labeled "Handler":



In the screenshot above, this button is disabled because the "Attitude Controller Assembly" checkbox is empty.  If it was checked, then the button would be enabled, and if you clicked the button a screen like this would appear:



This is the main screen for a program jWiz, which helps guide you through the joystick reconfiguration process.  At "Step #1", "Step #2", and "Step #3" you can configure one of the three joystick handlers.  You only need to configure one of them, since only one of them can be used by the simulation at any given time.  Step #1 is the preferred handler, and if it doesn't work for you then you can proceed to the next-best handler at Step #2, and if that doesn't work for you then you can proceed to the fallback handler at Step #3.  Not all of these steps will necessarily be available to you, since some of the handlers don't work on some operating systems, and therefore may be grayed out and disabled.

The big red checkmark on "Step #1" indicates that the primary joystick-handler is currently set and will be used by VirtualAGC.  If you change the setup, then the checkmark moves to a different handler.

Wherever there is a "Default" button, you can use it to restore the settings for that handler to their original defaults.  Wherever there is a "Test/Set" button, it allows you to both test the joystick configuration for that handler and (at the same time) set it for use by VirtualAGC.  Where there are separate "Test" and "Set" buttons, it means that the settings for that handler can be tested without selecting that handler for use by VirtualAGC, and that you have to click the "Set" button as an extra step after testing.

When you click one of the "Test" or "Test/Set" buttons, one of the following windows will open up:


Step #1:  Configuring yaACA3

Step #1:  Configuring yaACA2

Step #1:  Configuring yaACA

You'll notice, I'm sure, that yaACA2, which had no adjustable parameters on the main jWiz screen, has adustable parameters as well as "Default" and "Set" buttons, so all three handlers have effectively the same kinds of adjustments.  One difference is that yaACA3 and yaACA2 have only two adjustable parameters ("Axis number"/"Axis" and "Polarity"/"Sense") while yaACA has four adjustable parameters ("Stick", "Axis", "Scale", and "Offset").  Thus, yaACA is harder to configure than the others, which is part of the reason the others are preferred.

But the three handlers are all similar in that they display the current pitch, roll, and yaw readings.  The aim is to find an appropriate combination of settings so that:
You can experiment with the settings directly in yaACA2 (by hitting the "Set" button for every combination of settings you want to test), but in yaACA3 or yaACA you need to exit and return to jWiz every time you want to change the settings.   In Mac OS X, by the way, these screens are shown to you inside of a program called Terminator, and you need to exit Terminator from its main menu.

For most joysticks, you should find that the pitch and roll settings are correct, or very nearly so, but that there is less certainty about the yaw settings.  Therefore, I'll discuss only how to change the yaw settings; for the pitch and roll settings, the principles are identical.  Here is the procedure I'd recommend:
  1. If there is no movement of the joystick that can cause the yaw value to change, then it probably means the wrong axis is set.  In yaACA3 and yaACA2, the joystick has a collection of axes, identified as 0, 1, 2, ....  The numbering is a little different in yaACA, where the joystick has a collection of "sticks" 0, 1, 2, ..., each one of which has two "axes" 0 and 1.  For example, if you look at the yaACAx screenshots above, you'll see that my Logitech Extreme 3D Pro has either 6 axes or else 3 sticks of two axes each.  Nevertheless, the principle is the same.  Any given axis (or stick/axis) provides only a single reading—in other words, if it was used for roll or pitch, it wouldn't provide a yaw reading—so step through the unused axes (or stick/axis pairs) until you find the one whose readings change with yaw.  If there is no such setting, try a different joystick handler.
  2. If the sense of an axis is wrong—i.e., if the yaw increases where it should decrease and vice versa—then the "Polarity" or "Sense" of yaACA3 or yaACA2 is wrong, or the algebraic sign of the "Scale" is wrong for yaACA.  Change plus to minus or minus to plus.
  3. If the readings are not symmetric about zero—e.g., if the readings were 0 to 255 rather than -128 to +127—then the "Offset" in yaACA is wrong.  Change the Offset to equal the center of the range you are observing.  (This problem cannot occur in yaACA3 or yaACA2, so there is no adjustment needed for them.)
  4. If the size of the range is wrong—e.g., if the reading varies from -128 to 127 rather than from -57 to +57—then the "Scale" in yaACA is wrong.  The Scale is a multiplier in units of 0.01, so a Scale of 100 has no effect, a Scale less than 100 (in magnitude) decreases the size of the range, and a Scale greater than 100 increases the size of the range.  (Again, this problem cannot occur in yaACA3 or yaACA2, so there is no adjustment needed for them.)
Once you're all done configuring within jWiz, a partial double-check can be made by running the simulation itself after doing this configuration.  One of the items in the LM telemetry display is the current octal value of input channel 31, part of which indicates the direction (but not the magnitude) of displacement of the ACA from its detent.  The last two octal digits are:
So you should be able to move the joystick and see the downlink telemetry change accordingly, keeping in mind of course that the telemetry downlink occurs at several-second intervals and so the telemetry display can't respond instantly to joystick movements.

How the selection and configuration process works

If you wish to bypass jWiz for some reason, and to configure joysticks from the command line, it's certainly possible to do so.  The principle involved is that each of the three programs yaACA, yaACA2, and yaACA3 store their configuration parameters in files (respectively, yaACA-0.cfg, yaACA2-0.cfg, and yaACA3-0.cfg).  These configuration parameters are selected by the yaACA2 GUI or from the command-line for yaACA/yaACA3, but once safely stored in the configuration files the command-line parameters are no longer needed.  (The details of the command-line parameters are described in the following 3 sections.)  What VirtualAGC does is simply to choose which yaACAx program is needed on the basis of what configuration files it finds:
Creating the configuration files is a matter of running the appropriate yaACAx program with the desired command-line switches.  Getting rid of configuration files is a simple matter of deletion.  However, actually running the appropriate yaACAx program isn't always so simple because you have to be in the proper working directory, and have to supply the proper path for the program, as follows:
  1. From a command line, 'cd' to VirtualAGC's working directory.  This happens to be the directory in which the configuration files mentioned above are stored.  If you have installed in the default locations, the command will be:
  2. Attach your 3D joystick.
  3. Run yaACA, yaACA2, and yaACA3 with your chosen command-line options.  The only relevant options are --pitch, --roll, and --yaw, as described below for yaACA or for yaACA3yaACA2 needs no switches, since you select the options from its own built-in GUI.  The appropriate command lines will be:
  4. Experiment with the pitch, roll, and yaw switches.  (Note that to change the switches in yaACA or yaACA3, you need to exit the program and re-run it.  To exit, hit the Ctrl-C key on your keyboard.  On Mac OS X, you will also need to close the Terminator program from its menu.)  You need to try different combinations until you see the following readings printed out when moving the joystick:
Note that yaACA2 and yaACA3 display joystick readings as translated to the range acceptable to the AGC (namely, -57-+57).  yaACA differs in that it shows both the raw readings obtained from the joystick (usually 0-255 or else -128-+128) and the translated readings.  The latter will appear in parentheses following the former, and are the ones which you need to check. 

A partial double-check can be made by running the simulation itself after doing this configuration.  One of the items in the LM telemetry display is the current octal value of input channel 31, part of which indicates the direction (but not the magnitude) of displacement of the ACA from its detent.  The last two octal digits are:
Finally, note that (in theory) multiple joystick controllers could be attached to the computer, and that each of the yaACAx programs allows access to these different joystick controllers.  The configuration files are separate for the different joystick controllers so that (for example) if joystick controller 1 was used, then the configuration files would have a name like yaACA3-1.cfg rather than yaACA3-0.cfg.  However, VirtualAGC make no such provision:  it always assumes that joystick controller 0 is used.  Therefore, to use a different controller you will have to bypass VirtualAGC and use your own startup procedure for the simulation.

Command-line usage of yaACA3

The syntax is:
yaACA3 [OPTIONS]

The recognized options are:

--help
Displays textual info similar to that shown here.

--roll=N
--pitch=N
--yaw=N
These options allow you to configure how the roll/pitch/yaw degrees of freedom map to the "axes" of the joystick as recognized by the computer's operating system.  In general, the operating system views a 3D joystick as possessing a certain number of axes, identified as axis 0, axis 1, etc.  I deduce from my readings that axis 0 will almost always correspond to roll and axis 1 will almost always correspond to pitch.  However, the axis used for yaw is likely to vary.  For example, I have seen cases where it is 2 and cases where it is 3.  I have chosen defaults based strictly on my own convenience (i.e., for my Logitech Extreme 3D pro joystick), and I have no theoretical basis for assuming that they'll work for you.  In addition to choosing which axis belongs to which degree of freedom, these command-line switches also allow you to choose the sense.  For example, you may indeed find that axis 1 corresponds to pitch, but that it pitches up where you expect it to pitch down, and vice-versa.  In this case, simply put a minus sign before the number, as in "--pitch=-1".  (And if you have to do this for axis 0, don't worry:  "--roll=-0" treated differently than "--roll=0"!)

--ip=Hostname
The yaACA3 program and the yaAGC Apollo Guidance Computer simulation exist in a "client/server" relationship, in which the yaACA3 program needs to be aware of the IP address or symbolic name of the host computer running the yaAGC program.  By default, this is "localhost", meaning that both yaACA3 and yaAGC are running on the same computer.

--port=Portnumber
By default, yaACA3 attempts to connect to the yaAGC program using port number 19803.  However, if more than one instance of yaACA3 is being run, or if yaAGC has been configured to listen on different ports, then different port settings for yaACA3 are needed.  Note that by default, yaAGC listens for new connections on ports 19697-19706, but that the recommended port range when using yaAGC in the LM is 19797-19806.

--delay=Milliseconds
Adds a delay at start-up, so that yaACA does not immediately begin attempting to communicate with yaAGC.  The current defaults are 0 ms. in Linux/Mac OS X and 500 ms. in Win32.  This "feature" has been added as a temporary work-around for problem report #23, and probably has no other sensible purpose.  Even on Win32 it isn't usually needed, but it's here for the 10% (or whatever) of the time it's needed.

--controller=N
In case more than one joystick controller is attached to the PC/Mac, this allows selection of just one of them.  The default is N=0.

Command-line usage of yaACA2

The syntax is:
yaACA2 [OPTIONS]

The recognized options are:

--help
Displays textual info similar to that shown here.

--ip=Hostname
The yaACA3 program and the yaAGC Apollo Guidance Computer simulation exist in a "client/server" relationship, in which the yaACA3 program needs to be aware of the IP address or symbolic name of the host computer running the yaAGC program.  By default, this is "localhost", meaning that both yaACA3 and yaAGC are running on the same computer.

--port=Portnumber
By default, yaACA3 attempts to connect to the yaAGC program using port number 19803.  However, if more than one instance of yaACA3 is being run, or if yaAGC has been configured to listen on different ports, then different port settings for yaACA3 are needed.  Note that by default, yaAGC listens for new connections on ports 19697-19706, but that the recommended port range when using yaAGC in the LM is 19797-19806.

--delay=Milliseconds
Adds a delay at start-up, so that yaACA does not immediately begin attempting to communicate with yaAGC.  The current defaults are 0 ms. in Linux/Mac OS X and 500 ms. in Win32.  This "feature" has been added as a temporary work-around for problem report #23, and probably has no other sensible purpose.  Even on Win32 it isn't usually needed, but it's here for the 10% (or whatever) of the time it's needed.

--controller=N
In case more than one joystick controller is attached to the PC/Mac, this allows selection of just one of them.  The default is N=0.

Command-line usage of yaACA

The syntax is:
yaACA [OPTIONS]

The recognized options are:

--help
Displays textual info similar to that shown here.

--roll=J,S,A,F,O
--pitch=J,S,A,F,O
--yaw=J,S,A,F,O
These options allow you to configure how the roll/pitch/yaw degrees of freedom map to the characteristics of the joystick as recognized by the computer's operating system.  J is the joystick device number (in case multiple joystick devices are installed), S is the stick number within the joystick, and A is the axis within the stick.  F is a factor which the joystick reading is multiplied by, and O is an offset added to the joystick reading (after multiplication is completed).  The factor is useful (for example) in swapping right-to-left, back-to-front, or clockwise-to-counter-clockwise.  The offset would be useful when the the joystick provides unsigned readings (0-255) rather than the desired signed readings (-127 to 127).  A reading of -127 represents maximum left roll, downward pitch, or counter-clockwise yaw; a reading of +127 represents maximum right roll, upward pitch, or clockwise yaw.  (Actually, maximum values of 127 seem to occur in Linux, whereas maximum values of 128 seem to occur in Win32.)  The defaults are:
Roll = 0, 0, 0, 1.0, 0
Pitch = 0, 0, 1, 1.0, 0
Yaw = 0, 1, 0, 1.0, 0 (Linux) or 0, 1, 0, 1.0, -128 (Win32) or 0,2,0,1.0,0 (Mac OS X)
These defaults are based strictly on my own convenience (i.e., for my Logitech Extreme 3D pro), and I have no theoretical basis for assuming that they're any good generally.  Incidentally, the explanation above would seem to imply that you could theoretically use different joystick controllers for different axes; you really can't.

--ip=Hostname
The yaACA program and the yaAGC Apollo Guidance Computer simulation exist in a "client/server" relationship, in which the yaACA program needs to be aware of the IP address or symbolic name of the host computer running the yaAGC program.  By default, this is "localhost", meaning that both yaACA and yaAGC are running on the same computer.

--port=Portnumber
By default, yaACA attempts to connect to the yaAGC program using port number 19803.  However, if more than one instance of yaACA is being run, or if yaAGC has been configured to listen on different ports, then different port settings for yaACA are needed.  Note that by default, yaAGC listens for new connections on ports 19697-19706, but that the recommended port range when using yaAGC in the LM is 19797-19806.

--delay=Milliseconds
Adds a delay at start-up, so that yaACA does not immediately begin attempting to communicate with yaAGC.  The current defaults are 0 ms. in Linux and 500 ms. in Win32.  This "feature" has been added as a temporary work-around for problem report #23, and probably has no other sensible purpose.  Even on Win32 it isn't usually needed, but it's here for the 10% (or whatever) of the time it's needed.


yaTelemetry

LM
            telemetry monitor

In normal use the AGC periodically transmitted telemetry information which was displayed on monitors in Mission Control.  Naturally the virtual yaAGC periodically transmits telemetry data as well, using virtual radio waves, and with appropriate software to simulate a ground-based receiver it is possible to view this information.  The characteristics of the digital uplink and downlink can be explored by reading section 2 of the Guidance System Operations Plan (GSOP) for the LM or the CM.  The data consists of a conglomeration of many of the internal state variables used by the AGC, and is capable of conveniently giving a much more detailed picture of the state of the AGC and of the spacecraft than the AGC can give to the astronauts.

Prior to the creation of a dedicated yaTelemetry program, I cobbled telemetry-viewing capability onto the yaDSKY program, even though it is completely unrelated functionality.  If yaDSKY is started with the switches "--test-downlink" and/or "--test-uplink", then you can use it to view the digital downlinks or to create uplink data, respectively.  Using yaDSKY in this way has several drawbacks.  One drawback is that some people like to use an alternate DSKY simulation (such as the "DSKY Lite" built into Stephan Hotto's LM-Simulator described above), rather than to use yaDSKY, and in those cases yaDSKY isn't present.  Another drawback is that to see the downlink data, yaDSKY must be started in a console of an appropriate size and on some platforms (Windows, Mac OS X 10.4) this is either not possible, or else (Mac OS X 10.5) not possible without running the X-window system.

yaTelemetry, on the other hand, is a dedicated program that does nothing other than to provide a monitor for displaying telemetry information downlinked from the AGC. 

The accompanying photo is actually not a photo of a mission-control CRT; rather, it is a screen capture from the "Apollo 11" episode of the great HBO mini-series From the Earth to the Moon.  You can appreciate the care that the creators of the show must have taken with this, since many of the items on the screen clearly do relate to downlinked data.  If you enlarge the image by clicking on it, and squint at the upper left-hand corner of the display, you'll note that it refers to LM099; Luminary version 099 was the AGC software build used for Apollo 11.  Nevertheless, however convincing the film-makers' art was, this display does not seem to match any of the documented records we've found of the appearance of the actual mission control CRT displays.  At present, what you get with yaTelemetry (or yaDSKY) is simply a list of all variables and their values appearing in the downlinked data stream.  I hope to provide more accurate display formats in the future.

In addition to telemetry downlinks—i.e., reception by ground control of data from the AGC—digital uplinks are also possible.  Uplinks were (and continue in the virtual system to be) handled by the simple expedient of transmitting DSKY keycodes, encoded in a triply-redundant format to allow detection of errors.  The AGC flight software treats DSKY and uplink keycodes in a very similar fashion, so ground control could remotely perform any task which the astronaut could perform at the DSKY keypad, including data entry, entry of short program patches into memory, and activation of programs.  At the present time, yaTelemetry doesn't attempt to duplicate this uplink capability, but VirtualAGC does provide a handy mode for uplinking.  Also, the "--test-uplink" switch for yaDSKY causes keycodes to be transmitted to the AGC via the digital-uplink data-stream rather than via the virtual wiring it normally uses, and thus provides a proof-of-concept that the uplink is implemented correctly.  It would be interesting to know what kind of equipment mission control used for this purpose.  Perhaps they used an actual DSKY.

But enough about yaDSKY already!  As far as yaTelemetry is concerned, the usage-syntax is:

yaTelemetry [OPTIONS]

The recognized options are:

--help
Displays textual info similar to that shown here.

--delay=Milliseconds
Adds a delay at start-up, so that yaTelemetry does not immediately begin attempting to communicate with yaAGC.  The current defaults are 500 ms. in Win32 and 0 ms. otherwise (Linux/Mac OS X).

--ip=Hostname
The yaTelemetry program and the yaAGC Apollo Guidance Computer simulation exist in a "client/server" relationship, in which the yaTelemetry program needs to be aware of the IP address or symbolic name of the host computer running the yaAGC program.  By default, this is "localhost", meaning that both yaTelemetry and yaAGC are running on the same computer.

--port=Portnumber
By default, yaTelemetry attempts to connect to the yaAGC program using port number 19800.  However, if more than one instance of yaTelemetry is being run, or if yaAGC has been configured to listen on different ports, then different port settings for yaTelemetry are needed.  Note that by default, yaAGC listens for new connections on ports 19697-19706, which is the recommended range for virtual CM equipment, but that the recommended port range when using yaAGC for the LM is 19797-19806.  Thus, yaTelemetry would normally used either with port 19800 (LM) or 19700 (CM).

--spacecraft=CM
--spacecraft=LM
Self-explanatory, I think.

--font-size=Points
The yaTelemetry display window is not provided in a way that allows it to be resized.  This means that if downlink data does not fit within the display window, the only option is to change the size of the text, and not to change the size of the window.  The default font size is chosen with the expectation that adjustment isn't needed.  However, it's possible that the default won't be appropriate on some platforms, so adjustment can be achieved via the --font-size switch.  Because of platform variations, the default size actually varies from platform to platform (Linux, Windows, Mac) and according to whether or not the --simple switch (see below) is used.

--undecorated
yaTelemetry is gimicked-up to look something like a CRT display.  However, this takes a lot of space on the computer screen, and you may be unwilling to expend the extra space needed to achieve this retro appearance.  The --undecorated switch removes the decorations, and reduces the telemetry display to little more than a rectangle filled with text.  If the screen size is too small to display the decorations, this switch is activated automatically and so you do not need to do so manually.  Also, the switch is always in effect in Mac OS X, because the decorated display has not (yet) been made to work correctly.

--simple
Although by default yaTelemetry displays its downlink-data within a console intended to resemble a CRT, and although the --undecorated switch is available to remove that decorative trimming, the --simple switch is probably a better choice in practice.  The --simple switch overrides both the default and the --undecorated mode to provide a simple textual display with no flourishes except for two buttons which can enlarge the font size or decrease it.  The size of the display is always automatically adjusted so that the downlink data always appears within it without scrollbars being needed, regardless of the font size.  I would recommend always using the --simple switch in place of the defaults.


Last modified by Ronald Burkey on 2010-01-30.

Virtual AGC is hosted
              by ibiblio.org