Last updated: 11 March 2004

Autostar - Optimising Training Values

Sent:	Friday, January 18, 2002 5:10:40
From: (Martyn Edwards)
I've watched the skies for quite a few years through binos, and bought
my first telescope - an ETX-90 - a few months ago. Before I bought it I
had a good look around your site, and therefore knew of the excellent
work that you, Dick Seymour and Clay Sherrod were doing.

What really aroused my interest was Dick's patch to modify training
values. I had always felt that the training method was a one-off single
shot attempt, and I wasn't too happy with my results. When Dick posted
his patch I started playing around, based on the following.

As I understand it, training just takes up (calibrates) the backlash.
Once the backlash is taken up during the training process then,
providing your ETX is not a complete dog's breakfast, it shouldn't
matter how far you slew to your target - and the same applies when you
return to the target from the opposite direction. I felt the kick which
is imparted by the Az and Alt Percent values was a bit of a red herring
(as well as being somewhat subjective), but might just do something
which was unwanted - therefore I needed a training procedure which
started off with zero percent values and allowed me to take a number of
readings to eliminate random errors.

Dick's patch enabled me to do what I wanted, so I developed the attached
spreadsheet. In order to use it you need to install Dick's patch; as a
matter of courtesy I emailed him before contacting you, and he is happy
for me to ask you to publish this post.

The spreadsheet is based on calculating averaged errors for a number of
Az and Alt training values, and then predicting the optimised Az and Alt
training values which would lead to zero error.

I have built it around four training values - the graphs at the bottom
of the spreadsheet should be straight lines, and thus any gross error in
user entered values will stand out a mile and allow the user to recheck
that particular area of data entry only. Three values wouldn't have
allowed you to identify the oddball, and five or more values gets
boring! The values I have used are 20%, 40%, 60% and 80% of the training
values that you start off with - for no particular reason ( you could
use any four values set between zero and 32767 - it's just that these
figures work for me. By the way, the spreadsheet has real figures from
my own ETX in it).

If you feel this email and the spreadsheet could help the ETX community
then would you please post it under the Patches (ASU 3.0 ROM file) area
of your Autostar Information page, in the hope that others might use it,
abuse it and hopefully improve it!

Best regards,

Martyn Edwards

Subject: Training your ETX scope
Date: 3/8/04, 11:45
From: Niall Saunders (
Hi Martyn,

Thanks for your excellent spreadsheet, and for giving me another way of
looking at what is happening inside the cute little Autostar. (And
thanks also to Dick Seymour for patching the code to allow us to see
what is actually happening when we train the drives - perhaps MEADE
might realise the power of this information, and make it 'official' in a
new release.)

Firstly, I would like to send you a slightly modified spreadsheet -
where the primary differences are in that the training exercise is
repeated SEVEN times. I know that you state that four iterations are
sufficient but, when a user is faced with a scope that has perhaps been
'playing up', then they will (and should) take the time to convince
themselves that at least one reason for misbehaviours has been
eliminated (or identified as a culprit !!).

I have also modified the spreadsheet calculations slightly to indicate a
'Standard Deviation' figure and a 'Correlation' figure. Whilst Standard
Deviation 'per se' might not be easily grasped by many users, the
Correlation figure, expressed as a 'percentage', could be used instead -
giving them at least a figure of merit for, or confidence in, the
results of their recently completed training exercise.

Also, to make the acquisition of the data easier, I have created a
simple 'blank form' that can be printed, and used 'in the field'.

In fact, the only part that I have NOT addressed is the creation of the
INITIAL training values for each axis. The way I have been doing my
excursuses (now implemented on both an ETX105 and an LXD55 8" SCT) is to
first determine the CURRENT training values, then to carry out a
'normal' single-iteration "Train Drives" as proscribed by Meade. The
only thing I do now make sure of, is that ANY 'Train Drives' routine
takes into consideration the advice of Dick Seymour, whereby you bring
IN when the Autostar implements the next stage if the process (thus
ensuring that the gearbox backlash has already been eliminated for the
'slew-away' phase of each of the four stages.) From the CURRENT, or
SINGLE-ITERATION, training values I think that it would be better to
provide a 25%, 50%, 70% , 100%, 133%, 200% 400% spread for each axis -
thus generating good statistical data on each side of the 'expected'

Finally, I found that, in order to be truly reliable, I had to make sure
that I 'slewed away' by an amount equivalent to 3/4 of one of the
'segments' in the FINDERSCOPE each time. This meant that the
calculations were based on AT LEAST 2 degrees of scope motion away, and
two degrees of return drive, on both axes, for each stage of the
routine. I also got into the habit of slewing away at no more than
'Speed 5', and came back at 'Speed 5' until I knew that the target would
have been re-acquired in the 26mm SP eyepiece. At that point I dropped
to 'Speed 3' and then 'Speed 1' for final location - remembering at all
times NOT TO OVERSHOOT !!!!!

And, of course, you CANNOT do this without a cross-hair reticule
eyepiece. But, you can make one of these for NOTHING if you have some
card, glue, very thin wire (or human hair) - and a little bit of
ingenuity. Hint: Unscrew the chromed mounting barrel from your 26MM
Meade SP, then also unscrew the next (matt-black) ring as well. At this
stage, take care as you could end up disturbing the glass lens assembly
- just be careful, and set down the remainder of the eyepiece so that it
cannot be accidentally 'knocked over'. The matt-black ring has an
internal 'ridge' at one end, the end closest to the chrome ring. The
focal plane of the lens is just inside this ridge - at the lens side -
and so if you place a reticule in this plane, it will be in clear focus
whilst you are observing. Now all you need is a paper 'tube', with four
tiny 'razor-blade' slits at one end. Wedge, and secure with glue
(cyanoacrylate 'super' or 'crazy' glue seems best), one end of your two
'wires' (or 'hairs') into two adjacent slits. Stretch them, quite
tightly, across the tube, and secure them in the slit at the opposite
side. Fit the paper tube into the black ring, such that the reticule is
closest to the internal ridge, and reassemble all the rings.  Job done!!

[Note: Mike, if you want a series of pictures showing how I achieved
this, please let me know. Also, please ensure both Dick and Martyn are
FULLY credited for their initial efforts if you feel that this addendum
needs to be published.]

Hope this helps.


Subject: re: ETX Training values
Date: 3/9/04, 12:06
From: Anke 

I read your update article about training of the Autostar. I think the
data you show give a very good picture of the dependency between the
training values and the errors measured. The data range cover very well
positive and negatives errors from under and overshooting.

If I write you this e-mail is to suggest a further small improvement. If
you look carefully the graph I think you would agree that a linear
regression is not the best fit to the measured data. I believe a
exponential relation may give a much better fit.

As example I took the value for the AZ training. One of your data shows
a very small error (-0.00055 o less than 20 seconds!) for a train value
of 832. However your linear fit provide a optimum value of 936 (12.5%
higher). If you adjust a exponential equation to the data you will
obtain the following expression [AZ value] = 857 EXP(-4.42 x [Error])
(the R^2 = 0.997 if this make any sense to you). So if the error should
be zero the optimum value is 857 (3% higher)

May be you are familiar with the term R^2 to know how good is a
regression. As close to 1 better is the regression. So, for the linear
regression we obtain R^2 = 0.949 and for the exponential R^2 = 0.997.
That's all!

I am not a mechanical engineer but may be there is a practical reason
why the errors follows better a exponential relation than a linear
trend. I ignore it.

Sorry to disturb you. I just want to help a little bit.

Hi David,

I am glad that the information was of some use. I cannot say for sure
whether the data conforms to a linear or exponential regression. I
simply took the data as supplied by Martyn, and extended the number of
samples from 4 to 7. However, even at 7 samples there is probably not
enough statistical information to determine whether the function is, or
is not linear.

What I can say is that, in practice (and this is no doubt borne out by
Martyn) the Autostar behaves MUCH, MUCH better when trained using this
method. I have implemented the training exercise on two separate
telescopes - an ETX-105 and an LXD55 8" SCT. After using the suggested
training values, the scopes both performed SIGNIFICANTLY better. In
saying that though, moving between two Terrestrial Landmarks was still
not 'perfect'. It may be that the exercise needs further refinement,
perhaps with the OTA at a totally different angle (such that the
gearboxes are in a totally different orientation, and hence different

The sample data on the spreadsheet example is NOT a record of the actual
exercise I carried out on my ETX-105. (Sorry for any confusion!!). When
I implemented my version of Martyn's spreadsheet exercise, I used what
'I believed' were a suitable range of training values, picking numbers
that seemed suitable to give me a 'spread' around the Y-axis. However, I
simply 'jiggled' some of the readings on my posted example of the
spreadsheet, to fit with the fact that I have now entered formulae into
the spreadsheet to give 50%, 66.7%, 75%, 100%, 133.3%, 150% and 200% of
the 'initial starting values' seen by a user after the scope is first
trained (using the 'accurate method' as described by Dick Seymour). The
R^2 values are therefore NOT as good as I have seen from REAL exercises.

For example (and you can simply enter these values into the small table
at the bottom of the spreadsheet - making sure that you are working with
a BACKUP copy, because you WILL be overwriting the simple 'lookup'
formulae otherwise in these cells), here are the results from the two
telescopes I have access to:-

For the ETX-105
Az Error        Az Training
0.104027778     600
0.047083333     800
-0.000555556    1000
-0.025694444    1200
-0.081527778    1400
-0.125138889    1600
-0.1725     1800

Alt Error       Alt Training
-0.040972222    100
-0.026111111    150
-0.017916667    200
-0.004305556    250
0.00625     300
0.010416667     350
0.02375     400

Giving the following final solution:-
Best Fit Az Intercept       1038        Best Fit Alt Intercept      283
Standard Deviation of Result    13.4                            5.2
Correlation of Result (%)   99                          98
(note that I have taken the R^2 value, and expressed it as a percentage,
to suggest a level of confidence for non-statisticians!!)

For the LXD55 8" SCT
Az Error    Az Training
0.015972222     300
-0.001944444    350
-0.009166667    400
-0.021805556    450
-0.0325     500
-0.040833333    550
-0.052638889    600

Alt Error   Alt Training
-0.070277778    500
-0.041944444    600
-0.008333333    700
0.016944444     800
0.040416667     900
0.068194444     1000
0.0925      1100

Giving the following final solution:-
Best Fit Az Intercept       357     Best Fit Alt Intercept      748
Standard Deviation of Result    5.7                         4.6
Correlation of Result (%)   99                          99
(note that I have taken the R^2 value, and expressed it as a percentage,
to suggest a level of confidence for non-statisticians!!)

>From a purely mechanical point of view, it would seem logical to assume
that the relationship between 'Error' and 'Training Value' would be
LINEAR, but without access to the algorithm used by Meade when
implementing the Training Values, it would be impossible to say.

If you consider what is happening during the 'Train Drives' routine
(some of the following comments being paraphrased from Dick Seymour, Dr.
Clay Sherrod and Martyn Edwards' previous postings), then things may
seem clearer:-

1.) Starting from 'scratch' you set out to "RESET", "CALIBRATE MOTORS"

2.) The "RESET" stage is no longer as important as it may well have been
in the earlier versions of the AutoStar code. Certainly, when I upgraded
from v2.6Ec (and v2.6Ed) to Meade's 'official' v3.1Ee, the 'setup
parameters' in the Flash Memory were apparently retained without any
corruption. The same applied when I then subsequently upgraded
'sideways' to Dick Seymour's v3.1e4 patch. Even the values held in the
AutoStar for the 'CALIBRATE MOTORS' stage were carried over without
corruption. But, providing you are happy to re-enter all your user data,
then, yes, I would recommend a full RESET - just to be on the safe side.

3.) The 'CALIBRATE MOTORS' stage is quite important - especially at any
point where you switch to a different power source for your scope.
Theoretically, it should NOT be that critical, as the action that
'calibration' has on each axis takes place on circuitry that has ALREADY
been supplied with a 'constant' voltage source from an on-board
regulator. More specifically, each axis has its own micro-controller (on
the circuit board associated with the axis, over and above the
microcontrollers and microprocessors that are in the AutoStar itself).
What happens when a 'calibrate' command is issued is that each axis, in
turn, has its microcontroller send a 'full power' voltage level to the
LED transmitter that shines through the slotted wheel (fixed directly to
the motor output shaft). Each axis also has TWO sensors to detect this
incoming beam of light - the intention being that the detectors will
sense the rotation of the disk. However, in order to save money, the
disk is made from translucent plastic, and so the intensity of the LED
output has to be 'cut back' until a happy medium is achieved. The
microcontroller for each axis therefore spins the motor (at full speed)
whilst reducing the power, in steps, until a 50% 'mark-space' ratio (or
'square-wave') is detected by the sensors. The microcontroller has 64
possible 'levels' it can use for this, and the source voltage, prior to
being 'stepped down' is from the on-board 5V regulator feeding the
microcontroller. (In actual fact, the microcontroller is regulating the
CURRENT into the LED. which is what actually determines the intensity of
output, but the theory is still the same). This happens twice, first for
the RA/Az axis, then for the DEC/Alt axis. The values, once determined,
are sent back to the main microprocessor, in the AutoStar, where they
are stored in non-volatile Flash EEPROM memory. Probably the MOST
IMPORTANT thing you should consider when 'calibrating' the motors is to
do this IN THE DARK (or in the ambient level of lighting that you will
most likely be using the scope in.) A higher level of ambient lighting
causes the LEDs to need to output MORE light in order to produce a 50%
square-wave pulse train. Calibrate your motors in bright daylight, and
then use them at your dark skies sight, and you could well be
'over-driving' the LEDs, causing false pulses to be counted by the
microcontrollers. By the way, if you are an LXD55 user, things are only
SLIGHTLY better - at least you have a solid, stainless steel, opaque,
slotted disc, instead of the cheap translucent plastic toothed wheel
used on the ETX scopes. But a close inspection of this will show that it
is almost certainly NOT running 'true' and will 'wobble' all over the
place - again the reason for making sure that your CALIBRATE MOTORS
exercise was carried out with SOME knowledge of what is happening inside
the system.

4.) Now you are ready to 'TRAIN DRIVES'. Ideally, you will follow Martyn
Edward's procedure, and will have used Dick Seymour's patch, and will
have upgraded your AutoStar to v3.1e4. But you still need to carry out
at least ONE 'Meade' training routine, to give you a good 'starting
point' for my 7-stage adaptation of Martyn's process. But, when you set
out to do this, you must be aware that the 'normal instructions' for the
process are NOT GOOD ENOUGH. As Dick Seymour states, you must visualise
what the AutoStar is going to ask your motor drive gearbox train to do
AT EACH STAGE OF THE PROCESS. Normally, users will 'Train' the RA/Az
axis first (it's the first menu option after all). An object is chosen,
and centred, and then the AutoStar slews off in the '<' direction, and
asks you to re-centre the object USING ONLY THE '>' key. For this
process to be meaningful, the gear-train MUST already be FULLY MESHED in
the '<' direction BEFORE the AutoStar 'slews away'. This means that you
MUST initially 'centre' the reference object using ONLY the '<' key. If
you overshoot during the initial phase, you will HAVE TO START OVER
AGAIN, moving away at least quarter of the finderscope AFOV before
slewing back to centre. The Training Values are also more reliable if
you do not have to adjust the 'other axis' at the same time. Note also
that, in later version of the AutoStar software, you CANNOT use any
other direction keys when returning the object to centre - but you are
NOT protected from this possible error when INITIALLY bringing the
object on-centre. So, start by Training the RA/Az drive, and bring the
object 'on centre' using ONLY the '<' key. AutoStar will the 'slew-away'
in the SAME direction, and as you to re-centre in the '>' direction. The
next stage is that AutoStar slews off in the '>' direction (you ought
not to have to do anything between these stages, as you will ALREADY
have centred the object by your previous actions, AND in the correct
direction). You re-centre in the '<' direction. Then you work on the
DEC/Alt drive, initially moving off in the '^' direction, before
re-centering in the 'v' direction. AutoStar takes over, and 'slews away'
in the same 'v' direction. You bring the object back to centre in the
'^' direction. Finally, AutoStar 'slews away' in the '^' direction, and
you bring it back in using the 'v' key. I only ever use speeds 5, 3 and
1 when carrying out ANY of my training - these speeds seem to give the
most control.

5.) And if you do NOT have an eyepiece with a reticle, you will only get
'average' (at best, more likely 'poor') results. Take the time to roll
up a piece of card and make a reticle for yourself. You will save
yourself I don't know HOW many beer tokens - but it is so simple to do
that it seems almost criminal to pay someone else to make a reticled
eyepiece for you (unless you need specifically high accuracy, or an
illuminated reticle for astrometric purposes).

6.) And, finally, the NEXT time you find yourself doing a Train Drives
exercise, at least you will have the PREVIOUS results to compare
against. If the values appear to be significantly different, or differ
each time, then it would be wise to suspect an anomaly with the gear
train of the appropriate axis. On the ETX-105, the motor control PCB is
attached to the gearbox subframe (plastic assembly) with self-tapping
screws, and these can, and do, work loose. Moreover, the gear-box
sub-frame is attached to the cast aluminium chassis with further
self-tapping screws - and on my ETX-105, these were almost falling out.
The ETX-105 has a better method of adjusting back-lash on the worm drive
than was apparently the case with the early ETX models (the new ETX-125
chassis uses the system developed for the ETX-105, or vice versa -
anyway, they are both FAR MORE RELIABLE) - but it is still worth
checking that the machine set-screws have not worked loose. If you have
an LXD55 GEM mount, then it is definitely worthwhile checking all the
bolts and screws in the gear mechanisms. There appears to be NO
'thread-locking' compound used on these, and a BRAND-NEW mount from
MEADE exhibited 'slop' all over the place. Twenty minutes with a set of
Allen keys and a Philips screwdriver will fix BOTH type of mount -
PAINLESSLY. And a little bit of 'light-duty' threadlock might keep the
problem at bay for years to come.

Hope this helps.

Niall Saunders
(Cloudy Skies, hence the time available to write this lengthy tome!)

Go back to the Autostar Information page.

Go back to the ETX Home Page.

Copyright ©1999-2004 Michael L. Weasner /
Submittals Copyright © 2002, 2004 by the Submitter