DasDvorak, part 3

In previous posts, we covered taking apart a DasKeyboard, mapping the keys to their corresponding signals on the internal connector, and getting that information neatly cataloged into a MySQL database.  In this part of the series, we’ll be looking at how we’re going to map each key to a unique value, or rather a pair of unique values so we can construct a proper lookup table.

If you recall, we have a table in our database that looks like the following:


The key that is a ‘W’ in the qwerty layout represents a comma character in the dvorak layout.  The signal combinations for each are laid out in the qwRow, qwCol, dvRow, and dvCol fields.  When this key is pressed, it will connect pins 7 and 19.  Our programmable logic is going to instead make it look like lines 1 and 21 are connected, thus fooling the host microcontroller into sending a different key code to the host computer.

To do this, we need a series of one bit memories with one memory location for each key.  The problem we now need to solve is to figure out how we’re going to create the mapping between the keys and the memory locations.

After giving the problem some thought, I decided that I would assign a unique number to each key.  To generate those numbers automatically, I created the following table:


For the moment, ignore the id column (it is a habit of mine to put one in even if redundant).  Basically, this table counts upwards as each row or column signal is encountered.  Since we know that 11, 12, 16, 17. 18, 19, 21, and 22 are the rows, they count in sequence 1, 2, 3, … ending with 8.  The columns do the same, counting from 0 through 17.

This table allows me to use the row and column data to compute a unique memory location for each key.  Even better, once you wrap your head around the next bit of sql, you’ll start to see that the database is going to generate our lookup table for us and it will do it in a way that eliminates manual mistakes that might occur if we tried to do it by hand.  Here is our query:

   (qwC.val * 8) + qwR.val lutAddr,
   (c.val * 8) + r.val tgtLatch
 from keymap k, pinMap c, pinMap r, pinMap qwR, pinMap qwC  
 where k.dvCol = c.pin and k.dvRow = r.pin and k.qwCol = qwC.pin and k.qwRow = qwR.pin
order by lutAddr;

The main idea here is that each of our columns contain 8 rows or keys.  Because of that, we are going to multiply the column value by 8 and then add the row value to it.  This is similar to how you might code a two dimensional array if your language of choice didn’t support two dimensional arrays.  The first element would be 1 and the first element of the second column would be 9, and the first element of the third column would be 17 … and so on.  Here is a snippet of the results:


For the first two rows, those keys have no translation going on.  They have the same memory address in both layouts.  When you get to the key assigned to column1, row 3 (pins 1 and 16) that key then gets translated to be something else.  In fact, if you look in the same table, you’ll see that the reverse mapping occurs two rows down.  In this particular example, the ‘=’ and ‘]’ keys have swapped locations.   The original idea was to use this lookup table to determine where to store the state of each key as it was scanned and then to provide a logical expression that would activate the alternate row/column combinations at the right time in the scan sequence.

If one was very well versed in verilog, I’m sure this implementation would be very easy.  Unfortunately, I’m just learning it and am no expert by any means.  Instead, I went through the column/row combinations and figured out a process for storing the information and reading it back to the host microcontroller.  Once I had a process figured out, writing the code was fairly quick, especially since the database had generated the mappings for me using the above query.

As for the current status of the project, I’ve successfully tested the code on edaplayground.com and it pretty much works as I expected.  I then brought the design into the Xilinx ISE, I had a couple of small problems to fix, and then tried to fit the design to a xc2c256 CPLD.  It turns out that the CPLD is a nice fit for the project with the code using just a bit over 60% of the devices resources.

I’m getting close to hardware at this point and will hopefully be providing a schematic soon to help demystify some of what I’ve described so far.  The code needs some tweaks here and there and that should also be coming soon.  The next post will show some rather newbie-ish verilog with some explanation of how we take the table above and turn it into code.

DasDvorak, part 2


Above is a graphic from wikipedia.org showing the keyboard layout we’re trying to create.  Note that only the central portion of the keyboard is different.  About half of the keys will remain right where they are.

In this post, we’re going to take a look at how I went about detecting errors in my keymap data.  Any problems we can eliminate early on will be problems we don’t have to deal with later and the database makes this rather easy to do.

We’ll start with identifying our rows and columns again.  We know that pins 11, 12, 16, 17, 18, 19, 21, and 22 are the rows for our key matrix.  Get very familiar with this set of numbers – we will use them a lot as the project progresses.  The first task is to query our table and make sure that both the qwRow and dvRow columns only contain these numbers.  If either column contains anything else, it’s a transcription error that crept into the data somewhere along the way and it needs to be fixed.  Here are a couple of queries that help do that:

select distinct qwRow from keymap;
select distinct dvRow from keymap;

Once those errors are corrected, then the same can be done for the column rows:

select distinct qwCol from keymap;
select distinct dvCol from keymap;

Again, the only thing that should show in the results pane are the columns: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 13, 14, 15, 20, 23, 24, 25, and 26.

The final step is a bit more interesting.  Both the qwerty and the dvorak row/column pairs should correlate with each other.  It’s a bit hard to explain how that is, but in the end, we’re checking for consistency between the two views of the table.  These views help in ensuring that the keymap data is consistent.  Here’s the query:

select a.qwKey, b.dvKey
from keymap a, keymap b
where a.qwCol = b.qwCol and
a.qwRow = b.qwRow and
a.dvCol = b.dvCol and
a.dvRow = b.dvRow
order by qwKey;

The trick I’ve used above is an old one that has come in handy many times.  I purposely joined the same table to itself and forced the arrangement of the data so that each keyname from both layouts would be displayed together.  Once this is done, it’s fairly simple to visually scan through the results for problems.    Here is a snippet of the results:


If you take a look at a qwerty keyboard and then compare it to the graphic at the top of this post, you can see that where there is a G in a qwerty layout, there would be an I in the dvorak layout.  The Home and Insert keys would remain in their same position in both layouts.

Finally, lets try another query:

select a.qwKey, b.dvKey
from keymap a, keymap b
where a.qwCol = b.dvCol and
a.qwRow = b.dvRow
order by qwKey;

This one will yield a somewhat surprising result.  It will show you two columns of keynames and they should all be the same across each row:


This query puts the key translation into the where clause of the query, which makes the result set show if there are any inconsistencies.  At the end of the day, each key can only be represented once for each layout.  The forwards and backwards examination of that translation should be consistent.  If any problems do show up, you have an opportunity to debug it now before we’re about 500 lines deep in verilog code.

Next time, we’ll take a look at how we’re going to implement the key translation.  That post will be mostly about the theory behind the design.  Do note that if you’re not comfortable with sql queries, it might seem to get a little bit more complicated from here, but we’ll really only be using the same multiple-joins trick in a slightly different way.

DasDvorak, part 1

I’ve been typing on a dvorak layout keyboard for well over 5 years now and while OS based mechanisms for changing the keymap do work, the problem I keep running into is that things like the BIOS or UEFI don’t have mechanisms to change the layout from a qwerty based layout.  Furthermore, some applications tend to grab keyboard scan codes directly in the background, bypassing the preferred layout (Dell, please fix your iDrac software so it works correctly).

Normally this wouldn’t be bothersome, but I’m not the average user.  I do systems administration type things which usually take me into those BIOS screens and whatnot on a regular basis.  My home keyboard is an IBM Model M and I got it *because* I could rearrange the keycaps so they would reflect the actual dvorak layout.  At work, I use a DasKeyboard because it gives me the same tactile keyclicks, but there’s no labels on any of the keys.  I find it very frustrating to suddenly be forced into a keyboard layout I don’t use and neither keyboard helps me in that situation.  In those cases, I have to go find a regular keyboard to use temporarily until I’m done with the current task.

So, I wanted to learn verilog and I had a FPGA coming from kickstarter.  I went out and purchased another DasKeyboard and then promptly took it apart.  I think I did plug it in to make sure it worked, but that’s all I did.  Inside the new keyboard there isn’t a lot of space to add other components, so there wouldn’t be any way to make this particular hack invisible, but when I’m done I would have a hardware keyboard that spits out the scan codes in dvorak sequence.  Problem solved.

The way it would work is the host computer would think it has a qwerty keyboard attached, but all of the translation would be done in hardware.  The FPGA would examine the lines that were scanning the keyboard matrix and then either store the state of each key temporarily or it would pass the signal right through back to the keyboard’s own microcontroller.  What happens in the code really depends on what key gets pressed, but we’ll get to a discussion of that a bit later.

For those of you who don’t spend much time looking at USB specifications, the way a keyboard works is it sends sequences of scan codes to the host computer.  The scan code isn’t the same as the key you’re pressing, it’s a number that corresponds to a keymap that’s been published as part of the USB  HID specifications.  The host computer sees the scan code and then uses that same map to figure out what key was pressed.  Some applications, though, have this keymap hard coded to the qwerty keymap and that’s the problem.

So for this part of the write up, the main issue is that once the keyboard is apart, we then need to figure out what pairs of wires become connected when each key is pressed.  In my case, there are 26 lines running from the keyboard microcontroller to the key matrix.  I grabbed a multimeter and then carefully put together a listing of what column and row are connected as each key is pressed.  The engineers that put DasKeyboard together were nice enough to silkscreen the name of each key on the circuit board, so it wasn’t as difficult as I thought it would be.  From this, I learned that pins 11, 12, 16, 17, 18, 19, 21, and 22 are the rows.  The remaining pins are the columns.

At this point in the process, we’ve taken apart the keyboard and gotten a map of where each key fits into the signals used to scan the key matrix.  The next step was then to sit down with a dvorak layout and create something that shows what the signals would look like if it were a dvorak layout.  For this, I started doing it with a spreadsheet, but that wasn’t flexible enough.  I then went to using a MySql database to enter the qwerty key name, the pins used to locate that key, and the dvorak equivalent.  For instance, the letter W in the qwerty layout is identified by signals on pins 7 and 19 of the connector but that key is a comma in the dvorak layout.  To finish this row of data, we’ll also need to know what signals represent a comma in the dvorak layout.  When done, you have something that looks like the following:


In essence, when I hit the W key, I want to activate the lines which correspond to the comma instead.  The keyboard’s microcontroller will think I’ve hit a comma and it will send that scan code to the host computer.  The FPGA will see lines 7 and 19 become active, but it will instead activate lines 1 and 21.

This is only part 1 of the project.  There are several other posts coming which describe how I used the database to search for errors in my keymap and how I basically used the computer to generate the logic necessary to do this.  I’m currently writing the verilog code on EDAplayground.com and should be able to code up a testbench soon.  If you have a DasKeyboard and want the keymap data for your own project, let me know.

Asus Nexus 7 … dead?

I have a Google/Asus Nexus 7 tablet and I think it’s the 2013 model (black rubber like back without a chrome bezel).  This is the 32GB model and for the most part, I’ve been very happy with it.  I’ve had very few problems and the size is just about perfect for dropping it into the inside pocket of my jacket.

During the past couple of days, I had turned on the wifi and forgot about it.  As you might imagine, the battery didn’t last long and the device went completely dead.  I found it this morning and plugged it into the original wall charger to charge the battery.  About an hour later, I went to start the device and it just sits there with a white Google logo on a black screen.  I didn’t have time to mess with it as I was trying to get out of the house to go to work.

After I got to work, I gave it a proper charge and yet the tablet still wouldn’t boot past the Google logo.  I contacted Google support and they had me try to boot the device into recovery mode.  That didn’t work either.  When you confirm you want to go into recovery mode, the screen goes black for 1/2 a second and then you get the white Google logo.  It’s supposed to show some kind of secondary screen where you can do a factory reset, but that never shows up and I’ve waited over half an hour for it to do its thing.

Ok, I can live with doing some extra work.  I installed the Android SDK on my machine and downloaded a factory image.  After figuring out a not-so-obvious problem with USB drivers, I can’t seem to get the device to unlock the boot loader.  In fact, any operation in which the system would try to format or clear any of the system partitions usually results in a perpetual wait that never ends.  Again, I’ve waited well over half an hour for it to format or clear the cache partition, which is only 500MB in size.

I’m still researching the issue, but I’m thinking a dead battery shouldn’t cause a product to brick itself.  Maybe the problem was there for a long time and I didn’t notice it because rebooting my tablet is fairly rare.  The unfortunate thing is that I’m not the only person who’s posted about this problem with the 32GB model of this tablet.  At least one other post I read from another user indicated that Asus hardware support had determined that the main board in the tablet was dead and in need of replacement.

I’m wondering if there are more failed tablets out there and users are getting stuck with replacement costs for a device that’s less than 2 years old.  I admit the sample size so far is rather small, but on the other hand, there are a lot of people who don’t opt to complain or post such issues online…

Update: I tried going over this with a hot air rework station to see if it was a cold solder joint and that didn’t fix the issue.  Either it still has a cold solder joint somewhere where I didn’t work on it or it’s truly dead.

I then went to ebay to find a replacement logic board.  I eventually bought a 16GB logic board for about $25.  Once it was installed, the tablet was back up and running.

While I was waiting for parts, I spend a lot of time looking at alternatives.  Far too many 7 inch tablets just don’t fit in an inside coat pocket, but the Nexus 7 does so easily.  I’m not looking forward to the day I have to replace it.

How does MS Windows search for a DLL?

Once upon a time, I was programming in Windows 3.1 using, I think, a Borland product called Turbo C++.  Back then, the operating system didn’t provide true multitasking and each application had to cooperatively give up control so that other programs could run at the same time.  Compared to today’s programming environments, it seemed like the stone age.

One thing that was very important back then was to know how Windows goes about finding custom DLL libraries.  The problem was that there was no version control for DLL’s and it was common for developers of different applications to create DLL’s with the same name.  If the DLL being loaded turned out to be the wrong one, the application would likely crash before the first window even showed up on the desktop.  Knowing the order in which Windows searched for a DLL was important because it was a critical component of fixing the problem.

Today with abstract frameworks handling most everything for you, there isn’t much attention paid to the way DLL’s are loaded and there have been some improvements to the OS which help with the name collision problems of the past.  However, it’s still an important distinction to be aware of when building an application.

For example, at work I have a web server which has about a dozen different custom dotnet websites running on it.  Some of them use the Oracle Data Access Components(ODAC) to be able to read data from backend databases.  We’ve been having a problem with some of these applications crashing at start up because the original application developers deployed the website incorrectly.

Instead of identifying the specific DLLs from the ODAC that were linked with the website, the ODAC was installed on the webserver directly.  The reason why this is a problem is because not every developer uses the exact same version of the ODAC.   If I download the current ODAC, compile my website, and verify it works correctly, it will crash when it’s deployed to the webserver.  The reason for the crash is most likely because my ODAC version is newer than what is installed on the server.  If I then upgrade the version of the ODAC on the webserver, older applications will fail.  I need a solution that lets me deploy new code without having to recompile and redeploy every website on the server.

The solution to this problem is to extract the necessary DLL’s from the ODAC and place them in the bin directory of the deployed website.  The reason this fixes the problem is because Windows will first check the current working directory for any needed DLL’s.  If the correct DLL’s are found, then no further searching is done.  If the DLL’s are not found, then the system will use the %PATH% variable to search for the needed DLL until it is found.  If an older version of the ODAC comes first in the %PATH% variable, then the older applications that were linked with that DLL will work fine and newer applications will crash.  The only easy way to get around the potential version conflict is to put the needed DLL’s into the application’s working directory.

[Note: I’m not going to show the specific details for the ODAC here – there are many online posts that show which DLL’s need to be copied.  Failing that, an analysis of the program’s link structure should reveal what’s required.]

I’m writing this because as time has moved forward, these details seem to be getting lost to time.  Fewer and fewer developers have this knowledge and current programming training tends to skip over OS dependent behavior in favor of ‘run this program on everything and you’ll see dancing unicorns’.  Of course, I’m no expert on the matter, so if you have further information that would be helpful, please let me know.

Migrating to a not-misspelled-domain…

This blog used to be aperature.org and while I used that domain name for some time, it was a fairly bad misspelling and needed to be corrected.  If that wasn’t enough, the name didn’t really reflect the content very well.  When I got the original domain, I was very much into photography.  I migrated into other interests and didn’t change the name.  Besides all that, finding something in a .com, .org, or .net that was original and memorable was very difficult at the time.

So, this is now going to be stderr.info.  I like it because it’s short and it refers to the standard error file descriptor from the standard C library.  While some may think this refers to a medical condition, that’s not the case.  The great thing about TLA’s is that you can always come up with something different than what was intended.  🙂

Complete code for sun_switch project

Some time ago, I started documenting my project to automatically switch on/off some external lighting on the house. You can read the full rationale behind the project in this post. I’ve been delayed in getting this posted because I’ve had other personal projects that have been a higher priority lately.

Today, I’m posting the (nearly) final code for the entire project. All you need to play with this is a WWVB receiver, an Arduino, the TimeLord library, and an optional LED. The LED would be attached to the ssrPin output (with a current limiting resistor) to give an indication of whether the lights would be on or off. Initially, the output defaults to having the lights on. This is so that a power outage in the middle of the night will default to the on state until the system receives a valid time from the WWVB receiver. The LED that’s built into the Arduino is not used for this because I used it in another part of the code to indicate the stability of the received signal. That LED should turn on/off at a rate of once per second if the received signal is clean and free of errors.

There are probably a few bugs in the code at this point, but my testing has shown it to be fairly reliable – enough so that I’m going to be working on a project enclosure as well as some modifications to the electrical in the garage to put this project to good use. I think this code demonstrates a wide range of the capabilities of the arduino. There is some use of the default libraries as well as some custom programming that utilizes the atmega168 hardware directly.

I don’t think I’ll draw the ire of the Arduino haters on this one, but you never know. Personally, I think the flexibility of the platform is something that most of ‘haters’ don’t understand. You don’t have to use any of the libraries if you don’t want to and you can pretty much just skip forward to writing straight C for the mega168 if that’s what you want to do. Personally, I wouldn’t pass up an opportunity to use a development platform with those capabilities; especially when you can avoid spending extra time and money on creating custom boards.

If you want more details and explanations of the code, please read my prior posts.

Source: sun_switch