March 2005, Issue 176

Joystick Technology

GOT BRAINS?

Many digital potentiometers are available with serial-type interfaces because of the small number of I/Os required for communication. SPI and I2C are the most popular interfaces. These devices have expanded from the first 32-tap potentiometers to up to 256 taps, which allow the wiper to be positioned in increments of 0.4% of the total resistance, almost an order of magnitude better than their 32-tap predecessors.

Instead of the aforementioned simple interface, these parts now require some smarts so a simple switch input device like the JS1100AQ can control them. A microcontroller with at least eight I/Os can interpret the switch inputs and set the digital potentiometer. Five inputs monitor the JS1100AQ switches. You can use three output bits to communicate with the digital potentiometer (see Figure 2). I ended up with nil after a cursory search for small I/O microcontrollers with a serial interface, so the SPI interface had to be a software function. This wasn't a big deal because little else required lengthy time requirements.

The microcontroller's basic functions (besides communicating with the digital potentiometers) are to monitor the JS1100AQ for up, down, left, and right contact closures and to increment or decrement the x and y variables. The algorithm used for this might be as simple as incrementing or decrementing the variable once for each contact closure. You may wish to implement a more complicated scheme to automatically step the variable's value if the contact remains closed.

This will become more complicated if you dynamically change the step size based on time. Much of this would depend on how you'll use the output. If it's used as a simple volume control, then an adjustment response in the seconds range might be suitable. Note that 1-s responses might be required for pointing purposes. Sub-second moves are necessary for gaming because a perceivable time lag makes gaming difficult. Whether position feedback is a bar graph, a cursor, or a paddle, your eye, brain, and finger close the feedback loop by automatically compensating for errors.

Figure 3 shows a simple program loop that looks at the JS1100AQ input switches and determines which way the JS1100AQ is being tilted—position 1 (north) to position 8 (northwest). This allows you to assign a position based on any combination of switches, and you can compensate for a JS1100AQ rotation or the swap of an axis. Depending on the position, subroutines are called to adjust XVal or YVal. These values are used to set the taps on digital potentiometer 1 and potentiometer 2 every time through the main loop during the call to the SPI subroutine. The SPI subroutine requires 500 µs, which is the longest length of time to execute. It's executed every time to act as a timer. Therefore, timers aren't necessary.

The X_INC, X_DEC, Y_INC, and Y_DEC subroutines are all similar. Figure 4 shows the X_INC subroutine. A flag for each switch is cleared unless the switch is held in the closed position. The XSkip and YSkip variables begin with a value of 255. These variables count down each time an X or Y subroutine is executed (potentially each time through the main loop), allowing a subroutine to be skipped until the variable reaches zero. If the subroutine is skipped 256 times, then a change to XVal or YVal can happen only every 128 ms (256 × 500 µs), which is the automatic repeat rate.

In this simplest form, 256 increments would take 33 s (256 × 128 ms). So, even though the time for a single increment seems short (128 ms), waiting for a maximum change takes a long time (33 s). By substituting different algorithms in the Calculate New XInit in the X INC/DEC subroutines, you can affect the number loaded into the XSkip variable (reloaded from XInit). The simplest algorithm is to decrement XInit each time. This would reduce the automatic repeat rate from 128 ms by 500 µs for each change in Xval in automatic Repeat mode. For a maximum change, this cuts the total time in half to approximately 16 s.

You can see that a linear change might not be the best approach. If XInit is reduced by 12.5% of itself each time XInit is divided by eight (three rotations), the total time is reduced to less than 2 s. This has a side effect of limiting the minimum step time because 12.5% of anything less than eight equals zero, and XInit is no longer reduced. With this algorithm, the step time is reduced for the first 32 taps in 1 s, while the remaining 223 taps are stepped at 4 ms each over the next 1 s.