For the first generation of the INTRO Zumo robots, I have Pololu optical wheel encoders (see “Adding Quadrature Encoder to the Zumo Chassis“) which were available last year. It seems that Pololu heard my feedback, and are offering new encoders. Time to move things to the next level, using an optical encoder attached to the motor shaft:

First Optical Encoders

For the first robots, I used this Pololu optical encoder:

While this works, there are several disadvantages:

1. The encoders are rather expensive (one encoder costs US$14.95).
2. They need a hardware change to run at 3.3V.
3. It is difficult to mechanically mount them on the chassis.
4. Because they are between the chassis and the robot PCB, this increases the height of the robot.
5. The sensors tend to get polluted with dust from the ground/wheels, and then they do not run properly.

Shaft Mounted Encoders

So I was very excited when Pololu announced new motors with extended shafts (75:1 Micro Metal Gearmotor HP with Extended Motor Shaft), plus matching optical encoders (Optical Encoder Pair Kit for Micro Metal Gearmotors, 3.3V).

There are several advantages:

1. Higher resolution: now I get 12 or 20 ticks per motor revolution (before the gear!) instead of 48 per wheel rotation (after the gear). At full speed of the motor, I will get a signal change about every 200 µs.
2. The encoders do not conflict with the robot PCB.
3. There are 5V and 3.3V encoders available.
4. They are less expensive: US$8.95 per pair.

Of course there are disadvantages too, more about this later on.

Mounting the Encoder

The encoder gets soldered to the power leads of the motor, with the shaft going through the middle hole:

Wire Connections

Each encoder has 6 connections: Motor Power (2) and Quadrature Signals (4). I use wires soldered to the PCB. The encoders come with a 3-tooth or 5-tooth encoder wheels to be attached to the shaft:

The Challenges

As always, things have disadvantages too:

1. The speed of the signal is now very high, so I need enough processing power/time in the microcontroller.
2. The encoder is very sensitive to ambient or sun light. This will not be a big problem for me as it will be inside the robot.
3. Putting the encoder wheel at the right distance is important, otherwise the signal is not good.
4. But most important: the signal is not a digital one, it is a sinus one.

❗ While it would be technically possible to use the signal with a digital input pin, that only would work for small speed and if the signal is well-formed. The amplitude of the signal will vary making the need to post-process the signal.

Signal Processing

To generate a proper quadrature signal, a combination of D/A converter (Microchip MCP4728) and OP-Amp (Microchip MCP6004) is used:

Instead of a D/A converter, a ‘traditional’ approach with trimming potentiometer has been considered. But this was not ideal for many reasons:

1. Potentiometer require manual tuning
2. They need to be tiny, and can break
3. Temperature has an impact, hard to automatically recalibrate the values
4. They cost footprint and money too.

In general, I tink potentiometer are more a of ‘hack’ and not a solution. So instead we went for a MPC4728 hardware solution with software capabilities over I²C 🙂

Prototype Board

The MPC4728 from Microchip seems to be designed for excatly our needs. Even the data sheet features a similar application :-). The MPC4728 has an I²C interface: it has a non-volatile memory so the microcontroller can calibrate the sensor automatically or semi-automatically:

1. Measure the quadrature signal pulse width coming out of the system.
2. Adjust the signal with increasing/decreasing the analog voltage produced by the D/A converter.

To test the concept, a prototype board has been built:

Ambient Light Impact

One of the first results was that the ambient light (e.g. from room light or even more the sun light) has a big impact. This can be easily shown with protecting the encoder from the ambient light:

Then the analog sinus-like signals are pretty much ok:

But if the encoder is exposed to ambient light:

Then the signals are pretty much not usable and too weak:

With sunlight the impact is even bigger with signals not usable at all:

Signal Processing in Action

The job of the signal processing PCB is to convert the sinus-like analog signal into a digital quadrature signal. In the image below you see how the yellow analog signal gets converted into the purple quadrature signal:

The op-amps in the circuit get a voltage signal from the MPC4728 (blue signal below), and the yellow analog signal gets (inverted) and converted into a rectangular signal with a small hysteresis. That picture below shows nicely how the blue D/A output determines the ‘tipping’ point to build the purple quadrature signal:

This screenshot shows more details of the signal and threshold:

The next screenshot shows each analog signal converted to the quadrature signal:

And here both the quadrature and analog signals overlayed. Ideally, the quadrature signals have 50% duty cycle and are shifted by one quadrant/90 degrees:

The challenge will be to provide a good analog output signal with the D/A converter. More about this later.

Board Layout

After the concept worked, a new board layout has been designed in Eagle:

Connection to Base PCB

The quadrature signal processing board is hooked between the encoders and the main PCB:

💡 In a next iteration of the base robot PCB, I want to have that circuit integrated.

This is how the prototype looks like:

Zumo Blade

The board gets taped to the backside of the Zumo blade:

Then the motors get placed into the chassis:

Next, the motors get covered with the plastic plate:

And finally on top the robot PCB:

Calibration

The I²C signals of the MPC4728 are available on the outer expansion connector. The MPC4728 only needs accessible during calibration. The calibration values are then stored permanently in the MPC4728.

I use the shell to inspect and change the D/A converter values:

Then the motors get kicked slowly with the motor driver. The Logic Analyzer gives me the signals. By default, there might be no signal at all, but setting the D/A registers to something around 0x200 should give already a pretty good signal:

Then I gradually increase/decrease the D/A converter register values until I have a 50% duty cycle of every signal:

At a minimal speed of 3% (without load on the motors), each pulse is about 10 ms wide. At 50% motor duty this gets down to about 450 µs or 1.25 kHz:

And with full speed (100% motor duty) it gets down to about 225 µs, or about 2.5 kHz per signal:

With Nyquist/Shannon, this means that I need to sample each encoder with 10 kHz.

Auto-Calibration Software

To tune the D/A voltage manually to get a quadrature signal is doable:

1. Let the motors turn slowly with a low duty cycle, e.g. 5%-10%.
2. Monitor the digital quadrature signal with a logic analyzer.
3. Increase/Decrease the voltage of the 4 D/A channels until each quadrature signal is having a 50% duty ratio.

That works, but requires hooking up a logic analyzer and manually entering the values. So I wrote a software which can do this in a few seconds, without the need to attach a hardware probe:

1. Turn on each motor with a low duty ratio (10%)
2. Start with a 0V output of the D/A converter
3. Measure the low/high time of the signal
4. Gradually increase the D/A converter output voltage until either the low or high duty of the quadrature signal reaches 50%
5. Do this for all four signals.
6. Stop the motors.
7. Store the determined values in the MPC4728 EEPROM
8. done!

The whole process is automated with a single command:

CMD> app tune
Tuning....
Channel: 0 DAC: 0x0000 No signal
Channel: 0 DAC: 0x0010 No signal
Channel: 0 DAC: 0x0020 No signal
Channel: 0 DAC: 0x0030 No signal
Channel: 0 DAC: 0x0040 No signal
Channel: 0 DAC: 0x0050 No signal
Channel: 0 DAC: 0x0060 No signal
Channel: 0 DAC: 0x0070 No signal
Channel: 0 DAC: 0x0080 No signal
Channel: 0 DAC: 0x0090 No signal
Channel: 0 DAC: 0x00A0 No signal
Channel: 0 DAC: 0x00B0 No signal
Channel: 0 DAC: 0x00C0 No signal
Channel: 0 DAC: 0x00D0 No signal
Channel: 0 DAC: 0x00E0 No signal
Channel: 0 DAC: 0x00F0 38% high, low 61%
Channel: 0 DAC: 0x0100 44% high, low 55%
Channel: 0 DAC: 0x0110 49% high, low 50%
Set!
Channel: 1 DAC: 0x0000 No signal
Channel: 1 DAC: 0x0010 No signal
Channel: 1 DAC: 0x0020 No signal
Channel: 1 DAC: 0x0030 No signal
Channel: 1 DAC: 0x0040 No signal
Channel: 1 DAC: 0x0050 No signal
Channel: 1 DAC: 0x0060 No signal
Channel: 1 DAC: 0x0070 No signal
Channel: 1 DAC: 0x0080 No signal
Channel: 1 DAC: 0x0090 No signal
Channel: 1 DAC: 0x00A0 No signal
Channel: 1 DAC: 0x00B0 No signal
Channel: 1 DAC: 0x00C0 No signal
Channel: 1 DAC: 0x00D0 No signal
Channel: 1 DAC: 0x00E0 No signal
Channel: 1 DAC: 0x00F0 39% high, low 60%
Channel: 1 DAC: 0x0100 46% high, low 53%
Channel: 1 DAC: 0x0110 48% high, low 51%
Channel: 1 DAC: 0x0120 49% high, low 50%
Set!
Channel: 2 DAC: 0x0000 No signal
Channel: 2 DAC: 0x0010 No signal
Channel: 2 DAC: 0x0020 No signal
Channel: 2 DAC: 0x0030 No signal
Channel: 2 DAC: 0x0040 No signal
Channel: 2 DAC: 0x0050 No signal
Channel: 2 DAC: 0x0060 No signal
Channel: 2 DAC: 0x0070 No signal
Channel: 2 DAC: 0x0080 No signal
Channel: 2 DAC: 0x0090 No signal
Channel: 2 DAC: 0x00A0 No signal
Channel: 2 DAC: 0x00B0 No signal
Channel: 2 DAC: 0x00C0 No signal
Channel: 2 DAC: 0x00D0 No signal
Channel: 2 DAC: 0x00E0 No signal
Channel: 2 DAC: 0x00F0 No signal
Channel: 2 DAC: 0x0100 40% high, low 59%
Channel: 2 DAC: 0x0110 41% high, low 58%
Channel: 2 DAC: 0x0120 46% high, low 53%
Channel: 2 DAC: 0x0130 48% high, low 51%
Channel: 2 DAC: 0x0140 46% high, low 53%
Channel: 2 DAC: 0x0150 48% high, low 51%
Channel: 2 DAC: 0x0160 49% high, low 50%
Set!
Channel: 3 DAC: 0x0000 No signal
Channel: 3 DAC: 0x0010 No signal
Channel: 3 DAC: 0x0020 No signal
Channel: 3 DAC: 0x0030 No signal
Channel: 3 DAC: 0x0040 No signal
Channel: 3 DAC: 0x0050 No signal
Channel: 3 DAC: 0x0060 No signal
Channel: 3 DAC: 0x0070 No signal
Channel: 3 DAC: 0x0080 No signal
Channel: 3 DAC: 0x0090 No signal
Channel: 3 DAC: 0x00A0 No signal
Channel: 3 DAC: 0x00B0 No signal
Channel: 3 DAC: 0x00C0 No signal
Channel: 3 DAC: 0x00D0 No signal
Channel: 3 DAC: 0x00E0 No signal
Channel: 3 DAC: 0x00F0 No signal
Channel: 3 DAC: 0x0100 39% high, low 60%
Channel: 3 DAC: 0x0110 40% high, low 59%
Channel: 3 DAC: 0x0120 45% high, low 54%
Channel: 3 DAC: 0x0130 50% high, low 49%
Set!
Writing to EEPROM...
finished!

This gives me a very good signal distribution:

The source code of the calibration application can be found on GitHub.

Summary

The new encoders are a big improvement for the robot:  higher precision, less costs, but requires some signal processing. Right now one robot is equipped with the new system, so the next days will be filled with more testing. It started with manually tuning the encoders, and now I have a self-calibrating in place. I will need more tight sampling, but I will get a higher resolution. The extra workload will not be an issue for the ARM Cortex-M0+ I have as brain in my robot. The new encoders are a big quadrature step forward. And for the next iteration we plan to integrate everything into the robot base board, including the microcontroller :-).

PS: Thanks to Christian Jost and Andreas Albisser for their help and contributions in this project!

Happy Encoding 🙂