Posts with «hackaday columns» label

Ask Hackaday Answered: The Tale of the Top-Octave Generator

We got a question from [DC Darsen], who apparently has a broken electronic organ from the mid-70s that needs a new top-octave generator. A top-octave generator is essentially an IC with twelve or thirteen logic counters or dividers on-board that produces an octave’s worth of notes for the cheesy organ in question, and then a string of divide-by-two logic counters divide these down to cover the rest of the keyboard. With the sound board making every pitch all the time, the keyboard is just a simple set of switches that let the sound through or not. Easy-peasy, as long as you have a working TOG.

I bravely, and/or naïvely, said that I could whip one up on an AVR-based Arduino, tried, and failed. The timing requirements were just too tight for the obvious approach, so I turned it over to the Hackaday community because I had this nagging feeling that surely someone could rise to the challenge.

The community delivered! Or, particularly, [Ag Prismatic]. With a clever approach to the problem, some assembly language programming, and an optional Arduino crystalectomy, [AP]’s solution is rock-solid and glitch-free, and you could build one right now if you wanted to. We expect a proliferation of cheesy synth sounds will result. This is some tight code. Hat tip!

Squeezing Cycles Out of a Microcontroller

Let’s take a look at [AP]’s code. The approach that [AP] used is tremendously useful whenever you have a microcontroller that has to do many things at once, on a rigid schedule, and there’s not enough CPU time between the smallest time increments to do much. Maybe you’d like to control twelve servo motors with no glitching? Or drive many LEDs with binary code modulation instead of primitive pulse-width modulation? Then you’re going to want to read on.

There are two additional tricks that [AP] uses: one to fake cycles with a non-integer number of counts, and one to make the AVR’s ISR timing absolutely jitter-free. Finally, [Ag] ended up writing everything in AVR assembly language to make the timing work out, but was nice enough to also include a C listing. So if you’d like to get your feet wet with assembly, this is a good start.

In short, if you’re doing anything with hard timing requirements on limited microcontroller resources, especially an AVR, read on!

Taking Time to Think

The goal of the top-octave generator is to take an input clock and divide it down into twelve simultaneous sub-clocks that all run independently of each other. Just to be clear, this means updating between zero and twelve GPIO pins at a frequency of 1 MHz or so — updating every twenty clock cycles at the AVR’s maximum CPU speed. If you thought you could loop through twelve counters and decide which pins to flip in twenty cycles, you’d be mistaken.

But recognizing the problem is the first step to solving it. Although the tightest schedule might require flipping one pin exactly twenty clocks after flipping another, most of the time there are more cycles between pin updates — hundreds up to a few thousand. So the solution is to recognize when there is time to think, and use this time to pre-calculate a buffer full of next states.

[Ag]’s solution uses a few different loops that run exactly 20, 40, and 60 cycles each — the longer versions being just the 20-cycle one padded out with NOPs. These loops run inside an interrupt-service routine (ISR). When there are 80 or more cycles of thinking time until the next scheduled pin change, control is returned to the main loop and the next interrupt is set to re-enter the tight loops at the next necessary update time.

All the fast loop has to do is read two bytes, write them out to the GPIO pins, increment the pointer to the next row of data, and figure out if it needs to stall for 20 or 40 additional cycles, or set the ISR timer for longer delays and return to calculations. And this it can do in just twelve of the twenty cycles! Slick.

Buffers

Taking a step back from the particulars of the top-octave generator, this is a classic problem and a classic solution. It’s worth your time to internalize it, because you’ll run into this situation any time you have real-time constraints. The problem is that on average there’s more than enough time to complete the calculations, but that in the worst cases it’s impossible. So you split the problem in two parts: one that runs as fast as possible, and one that does the calculations that the fast section will need. And connecting together fast and slow processes is exactly why computer science gave us the buffer.

In [AP]’s code, this buffer is a table where each entry has two bytes for the state of the twelve GPIO pins, and one byte to store the number of clock cycles to delay until the next update. One other byte is left empty, yielding 64 entries or 256 bytes for the whole table. Why 256 bytes? Because the AVR has an 8-bit unsigned integer, wrapping around from the end of the table back to the beginning is automatic, saving a few cycles of wasteful if statements.

But even with this fast/slow division of labor, there is not much time left over for doing the pre-calculation. Sounding the highest C on a piano keyboard (4186 Hz) with a 20 MHz CPU clock requires toggling a GPIO pin every 2,390 cycles, so that’s the most time that the CPU will ever see. When the virtual oscillators are out of phase, this can be a lot shorter. By running the AVR at its full 20 MHz, and coding everything in assembly, [AP] can run the calculations fast enough to support twelve oscillators. At 16 MHz, there’s only time for ten, so every small optimization counts.

Some Optimization Required

Perhaps one of the cleverest optimizations that [AP] made is the one that makes this possible at all. The original top-octave chips divide down a 2 MHz square wave by a set of carefully chosen integer divisors. Running the AVR equivalent at 2 MHz resolution would mean just ten clocks per update and [AP]’s fast routine needed twelve, so the update rate would have to be halved. But that means that some odd divisors on the original IC would end up non-integral in the AVR code. For example, the highest C is reproduced in silicon as 2 MHz / 239, so to pull this off at 1 MHz requires counting up to 119.5 on an integer CPU. How to cope?

You could imagine counting to 119 half of the time, and 120 the other. Nobody will notice the tiny difference in duty cycle, and the pitch will still be spot on. The C programmer in me would want to code something like this:

uint8_t counter[12] = { 0, ... };
uint8_t counter_top[12] = { 119, ... };
uint8_t is_counter_fractional[12] = { 1, 0, ... };
uint8_t is_low_this_time[12] = { 0, ... };

// and then loop
for ( i=0 ; i<12; ++i){
  if ( counter[i] == 0 ){
    if ( is_counter_fractional[i] ){
      if ( is_low_this_time[i] ){
        counter[i] = counter_top[i];
    is_low_this_time = 0;
      else {
        counter[i] = counter_top[i] + 1;
    is_low_this_time = 1;
      }
    }
  }
}

That will work, but the ifs costs evaluation time. Instead, [AP] did the equivalent of this:

uint8_t counter[12] = { 0, ... };
uint8_t counter_top[12] = { 119, ... };
uint8_t phase[12] = { 1, 0, ... };

for ( i=0 ; i<12; ++i){
  if ( counter[i] == 0 ){
    counter[i] = counter_top[i] + phase[i];
    counter_top[i] = counter[i];
    phase[i] = -phase[i];
  }
}

What’s particularly clever about this construction is that it doesn’t need to distinguish between the integer and non-integer delays in code. Alternately adding and subtracting one from the non-integer values gets us around the “half” problem, while setting the phase variable to 0 means that the integer-valued divisors run unmodified, with no ifs.

The final optimization shows just how far [AP] went to make this AVR top-octave generator work like the real IC. When setting the timer to re-enter the fast loop in the ISR, there’s the possibility for one cycle’s worth of jitter. Because AVR instructions run in either one or two clock cycles, it’s possible that a two-cycle instruction could be running when the ISR timer comes due. Depending on luck, then, the interrupt will run four or five clocks later: see the section “Interrupt Response Time” in the AVR data sheet for details.

In a prologue to the ISR, [AP]’s code double-checks the hardware timer to see if it has entered on a one-cycle instruction, and adds in an extra NOP to compensate. This makes the resulting oscillator nearly jitter free, pushing out a possible source of 50 ns (one cycle at 20 MHz) slop. I don’t think you’d be able to hear this jitter, but the result surely looks pretty on the oscilloscope, and this might be a useful trick to know if you’re ever doing ultra-precise timing with ISRs.

The Proof of the Pudding

Naturally, I had to test out this code on real hardware. The first step was to pull a random Arduino-clone out of the closet and flash it in. Because “Arduinos” run at 16 MHz with the stock crystal, the result is that a nominal 440 Hz concert A plays more like a slightly sharp F, a musical third down. It sounds fine on its own, but you won’t be able to play along with any other instruments that are tuned to 440 Hz.

[AP]’s preferred solution is to run the AVR chip at 20 MHz. Since the hardware requirements are very modest, you could use a $0.50 ATTiny816 coupled with a 20 MHz crystal and you’d have a top-octave generator for literal pocket change — certainly cheaper than buying even an Arduino clone. I tested it out with an ATMega48P and a 20 MHz crystal on a breadboard because it’s what I had. Or you could perform crystalectomy on your Arduino to get it running at full speed.

We went back and forth via e-mail about all the other (firmware) options. [AP] had tried them all. You could trim the ISR down to 16 cycles and run at 16 MHz, but then there’s only enough CPU time in the main loop to support ten notes, two shy of a full octave. You could try other update rates than 1 MHz, but the divisors end up being pretty wonky. To quote [AP] from our e-mail discussion on the topic:

“After playing with the divider values from the original top octave generator IC and trying different base frequencies, it appears that the 2 MHz update rate is a “sweet spot” for getting reasonable integer divisors with < +/-2 cents of error. The original designers of the chip must have done the same calculations.”

To make a full organ out of this setup, you’ll also need twelve binary counter chips to divide down each note to fill up the lower registers of the keyboard, but these are easy to design with and cost only a few tens of cents apiece. All in all, thanks to [AP]’s extremely clever coding, you can build a fully-polyphonic noisemaker that spits out 96 simultaneous pitches, all in tune, for under $10. That’s pretty amazing.

And of course, I’ve already built a small device based on this code, but that’s a topic for another post. To be continued.

Hands-On with New Arduino FPGA Board: MKR Vidor 4000

Hackaday brought you a first look the Arduino MKR Vidor 4000 when it announced. Arduino sent over one of the first boards so now we finally have our hands on one! It’s early and the documentation is still a bit sparse, but we did get it up and running to take the board through some hello world exercises. This article will go over what we’ve been able to figure out about the FPGA system so far to help get you up and running with the new hardware.

Just to refresh your memory, here’s what is on the Vidor board:

  • 8 MB SRAM
  • A 2 MB QSPI Flash chip — 1 MB allocated for user applications
  • A Micro HDMI connector
  • An MIPI camera connector
  • Wi-Fi and BLE powered by a U-BLOX NINA W10 Series device
  • MKR interface on which all pins are driven both by SAMD21 (32-bit ARM CPU) and FPGA
  • Mini PCI Express connector with up to 25 user programmable pins
  • The FPGA (an Intel/Altera Cyclone 10CL016) contains 16K Logic Elements, 504 KB of embedded RAM, and 56 18×18 bit HW multipliers

Sounds good. You can get more gory technical details over at Arduino and there’s even a schematic (.zip).

Documentation

Documentation is — so far — very hard to come by but the team is working to change that by the day. Here are the resources we’ve used so far (in addition to the schematic):

In addition, Arduino just released an example FPGA project for Quartus. I’ll explain what that means in a bit.

Get Up and Running with the Arduino Desktop IDE

Despite the getting started guide, it doesn’t appear the libraries are usable from the cloud-based IDE, so we followed the instructions to load the beta board support for the MKR 4000 into our desktop IDE. Be aware that the instructions show the “normal” SAMD board package, but you actually want the beta which says it is for the MKR 4000. If you search for SAMD in the Boards Manager dialog, you’ll find it (see the second entry in the image below).

 

The libraries we grabbed as ZIP files from GitHub and used the install library from ZIP file option with no problems.

What’s the Code Look Like?

The most interesting part of this board is of course the inclusion of the FPGA which left us wondering what the code for the device would look like. Browsing the code, we were a bit dismayed at the lack of comments in all but the JTAG code. We decided to focus first on the VidorPeripherals repository and dug into the header file for some clues on how everything works.

Looking at VidorPeripherals.h, you can see that there’s a few interesting I/O devices include SPI, I2C, UART, reading a quadrature encoder, and NeoPixel. There’s also a few headers that don’t exist (and presumably won’t get the define to turn them on) so don’t get too excited by some of the header file names until you make sure they are really there.

Then we decided to try the example test code. The library provides a global FPGA object that you need to set up:

// Let's start by initializing the FPGA
if (!FPGA.begin()) {
    Serial.println("Initialization failed!");
    while (1) {}
}

// Let's discover which version we are running
int version = FPGA.version();
Serial.print("Vidor bitstream version: ");
Serial.println(version, HEX);

// Let's also ask which IPs are included in this bitstream
FPGA.printConfig();

The output of this bit of code looks like this:

Vidor bitstream version: 1020107
number of devices 9
1 01000000 MB_DEV_SF
1 02000000 MB_DEV_GPIO
4 04000000 MB_DEV_I2C
6 05000000 MB_DEV_SPI
8 06000000 MB_DEV_UART
1 08000000 MB_DEV_SDRAM
4 09000000 MB_DEV_NP
11 0A000000 MB_DEV_ENC
0 0B000000 MB_DEV_REG

In many cases, the devices provided by the FPGA are pretty transparent. For example, here’s another snip from the example code:

// Ok, so we know now that the FPGA contains the extended GPIO IP
// The GPIO pins controlled by the FPGA start from 100
// Please refer to the online documentation for the actual pin assignment
// Let's configure pin A0 to be an output, controlled by the FPGA
FPGA.pinMode(33, OUTPUT);
FPGA.digitalWrite(33, HIGH);

// The same pin can be read by the SAMD processor :)
pinMode(A0, INPUT);
Serial.print("Pin A0 is ");
Serial.println(digitalRead(A0) == LOW ? "LOW" : "HIGH");

FPGA.digitalWrite(33, LOW);
Serial.print("Pin A0 is ");
Serial.println(digitalRead(A0) == LOW ? "LOW" : "HIGH");

That’s easy enough and it is nice that the pins are usable from the CPU and FPGA. We couldn’t find the documentation mapping the pins, but we assume it is coming.

Using, say, an extra serial interface is easy, too:

SerialFPGA1.begin(115200);
while (!SerialFPGA1);
SerialFPGA1.println("test");

Bitstream

So where’s the FPGA code? As far as you can tell, this is just a new Arduino with a lot of extra devices that connect through this mysterious FPGA object. The trick is that the FPGA code is in the library. To see how it works, let’s talk a little about how an FPGA operates.

When you write a program in C, that’s not really what the computer looks at, right? The compiler converts it into a bunch of numbers that tell the CPU to do things. An FPGA is both the same and different from that. You write your program — usually in a hardware design language like Verilog or VHDL. You compile it to numbers, but those numbers don’t get executed like a CPU does.

The best analogy I’ve been able to think of is that an FPGA is like one of those old Radio Shack 100-in-1 electronic kits. There are a bunch of parts on a board and some way to connect them with wires. Put the wires one way and you have a radio. Put them another way and you have a burglar alarm. Rewire it again and you have a metal detector. The numbers correspond to wires. They make connections and configure options in the FPGA’s circuitry. Unless you’ve built a CPU, there’s nothing in there examining and acting on the numbers like there would be with a CPU.

The numbers that come out of an FPGA tool is usually called a bitstream. Someone has to send that bitstream to an FPGA like the Cyclone onboard the Arduino every time it powers up. That someone is usually a memory device on the board, although the CPU can do it, too.

So that leads to two questions: Where is the bitstream? How does it get to the FPGA?

The answer to the first question is easy. If you look on Github, you’ll see in the library there is a file called VidorBase.cpp. It has the following lines:

__attribute__ ((used, section(".fpga_bitstream")))
const unsigned char bitstream[] = {
    #include "app.ttf"
};

What this means if there is an array called bitstream that the linker will put it in a specially marked section of memory. That array gets initialized with app.ttf which is just an ASCII file full of numbers. Despite the name, it is not a TrueType font. What do the numbers mean? Hard to say, although, in theory, you could reverse engineer it just like you can disassemble binary code for a CPU. However, it is the configuration required to make all the library calls we just talked about work.

The second question about how it gets to the FPGA configuration is a bit of a mystery. As far as we can tell, the bootloader understands that data in that section should get copied over to the FPGA configuration memory and does the copying for you. It isn’t clear if there’s a copy in the main flash and a copy in the configuration flash but it seems to work transparently in any event.

There’s a checksum defined in the code but we changed it and everything still worked. Presumably, at some point, the IDE or the bootloader will complain if you have the wrong checksum, but that doesn’t appear to be the case now.

By the way, according to the Arduino forum, there are actually two bitstreams. One that loads on power-up that you would rarely (if ever) change. Then there is another that is the one included with the library. You can double-click the reset button to enter bootloader mode and we suspect that leaves the FPGA initialized with the first bitstream, but we don’t know that for sure. In bootloader mode, though, the red LED onboard has a breathing effect so you can tell the double click works.

What about my FPGA Code?

This isn’t great news if you were hoping for an easy Arduino-like way to do your own FPGA development in Verilog or VHDL. Intel will give you a copy of Quartus Prime which will generate bitstreams all day for you. We think — but we aren’t sure — that the ASCII format is just a raw conversion from binary of the bitstream files.

Very recently, Arduino provided a Quartus project that would create a bitstream. This provides a few key pieces of the puzzle, like the constraint file that lets the FPGA compiler find the different parts on the board.

However, even with that project, you still have some reverse engineering to do if you want to get started. Why? Here’s what Arduino says about loading your own FPGA code (we added the emphasis):

Quartus will produce a set of files under the output_files directory in the project folder. In order to incorporate the FPGA in the Arduino code you need to create a library and preprocess the ttf file generated by Quartus so that it contains the appropriate headers required by the software infrastructure. Details of this process will be disclosed as soon as the flow is stable.

Programming the FPGA is possible in various ways:

  • Flashing the image along with Arduino code creating a library which incorporates the ttf file
  • Programming the image in RAM through USB Blaster (this requires mounting the FPGA JTAG header). this can be done safely only when SAM D21 is in bootloader mode as in other conditions it may access JTAG and cause a contention
  • Programming the image in RAM through the emulated USB Blaster via SAM D21 (this component is pending release)

In addition, the repository itself says that some key pieces are missing until they can work out licensing or clean up the code. So this gets us closer, but you’d still need to reverse engineer the header from the examples and/or figure out how to force the processor off the JTAG bus. The good news is it sounds like this information is coming, it just isn’t here yet.

Of course, you are going to need to understand a lot more to do anything significant. We know the FPGA is set in the AS configuration mode. We also asked Arduino about the clock architecture of the board and they told us:

[The CPU] has its own clock which is used to generate a 48 MHz reference clock that is fed to the FPGA (and that can be removed at any time to “freeze” fpga). In addition to this reference clock, [the] FPGA has an internal RC oscillator which can’t be used as [a] precise timing reference for tolerance issues but can be used in case you don’t want [the CPU] to produce the reference clock.

Of course, the FPGA has a number of PLLs onboard that can take any valid clock and produce other frequencies. For example, in the vision application, Arduino demonstrated, the 48 MHz clock is converted into 24 MHz, 60 MHz, 100 MHz, and 120 MHz clocks by PLLs.

Mix and Match?

One thing that is disappointing is that — at least for now — you won’t be able to mix and match different FPGA libraries. There is exactly one bitstream and you can’t just jam them together.  Although FPGAs can often be partially configured, that’s a difficult technique. But we were a little surprised that the IDE didn’t understand how to take libraries with, for example, EDIF design files for IP that would all get compiled together. That way I could pick the Arduino UART and mix it with the Hackaday PWM output module along with my own Verilog or VHDL.

The way things are structured now you will have one bitstream that is precompiled by another tool (probably Quartus for the foreseeable future). It will match up with a particular C++ library. And that’s it. Doesn’t matter how much of the FPGA is left over or how much of it you really use, you will use it all for the one library.

Of course, you can load another library but it is going to replace the first one. So you only get one set of functions at a time and someone else gets to decide what’s in that set. If you roll your own, you are going to have to roll your own all the way.

What’s Next?

It is still early for the Arduino Vidor. We are hopeful we’ll get the tools and procedures necessary to drop our own FPGA configurations in. It would be great, too, if the stock libraries were available in source format including the Verilog HDL. The recent GitHub release shows quite a bit, although it isn’t all of the examples, it is probably enough if we get the rest of the information.

As for a more intuitive interface, we don’t know if that’s in the cards or not. We don’t see much evidence of it, although posts on the Arduino forum indicate they will eventually supply an “IP Assembler” that will let you compose different modules into one bitstream. We don’t know if that will only work with “official” modules or not. However, we know the Arduino community is very resourceful so if we don’t get a good ecosystem it will not surprise us if someone else makes it happen. Eventually.

For now, we will continue to play with the existing bitstreams that become available. There are some neat new features on the CPU, too. For example, you can map two of the unused serial modules.  There’s a hardware-based cooperative multitasking capability. As more details on the FPGA emerge, we’ll keep you posted and if you learn something, be sure to leave word in the comments so everyone can benefit.

Entropy and The Arduino: When Clock Jitter is Useful

What do you do, when you need a random number in your programming? The chances are that you reach for your environment’s function to do the job, usually something like rand() or similar. This returns the required number, and you go happily on your way.

A shift register configured as a pseudo-random
number generator. [by KCAuXy4p CC0 1.0]
Except of course the reality isn’t quite that simple, and as many of you will know it all comes down to the level of randomness that you require. The simplest way to generate a random number in software is through a pseudo-random number generator, or PRNG. If you prefer to think in hardware terms, the most elementary PRNG is a shift register with a feedback loop from two of its cells through an XOR gate. While it provides a steady stream of bits it suffers from the fatal flaw that the stream is an endlessly repeating sequence rather than truly random. A PRNG is random enough to provide a level of chance in a computer game, but that predictability would make it entirely unsuitable to be used in cryptographic security for a financial transaction.

There is a handy way to deal with the PRNG predictability problem, and it lies in ensuring that its random number generation starts at a random point. Imagine the  shift register in the previous paragraph being initialised with a random number rather than a string of zeros. This random point is referred to as the seed, and if a PRNG algorithm can be started with a seed derived from a truly unpredictable source, then its output becomes no longer predictable.

Selecting Unpredictable Seeds

Computer systems that use a PRNG will therefore often have some form of seed() function alongside their rand() function. Sometimes this will take a number as an argument allowing the user to provide their own random number, at other times they will take a random number from some source of their own. The Sinclair 8-bit home computers for example took their seed from a count of the number of TV frames since switch-on.

The not-very-random result of a thousand analogRead() calls.

The Arduino Uno has a random() function that returns a random number from a PRNG, and as you might expect it also has a randomSeed() function to ensure that the PRNG is seeded with something that will underpin its randomness. All well and good, you might think, but sadly the Atmel processor on which it depends has no hardware entropy source from which to derive that seed. The user is left to search for a random number of their own, and sadly as we were alerted by a Twitter conversation between @scanlime and @cybergibbons, this is the point at which matters start to go awry. The documentation for randomSeed() suggests reading the random noise on an unused pin via analogRead(), and using that figure does not return anything like the required level of entropy. A very quick test using the Arduino Graph example yields a stream of readings from a pin, and aggregating several thousand of them into a spreadsheet shows an extremely narrow distribution. Clearly a better source is called for.

Noisy Hardware or a Jittery Clock

As a slightly old-school electronic engineer, my thoughts turn straight to a piece of hardware. Source a nice and noisy germanium diode, give it a couple of op-amps to amplify and filter the noise before feeding it to that Arduino pin. Maybe you were thinking about radioactive decay and Geiger counters at that point, or even bouncing balls. Unfortunately though, even if they scratch the urge to make an interesting piece of engineering, these pieces of hardware run the risk of becoming overcomplex and perhaps a bit messy.

The significantly more random result of a thousand Arduino Entropy Library calls.

The best of the suggestions in the Twitter thread brings us to the Arduino Entropy Library, which uses jitter in the microcontroller clock to generate truly random numbers that can be used as seeds. Lifting code from the library’s random number example gave us a continuous stream of numbers, and taking a thousand of them for the same spreadsheet treatment shows a much more even distribution. The library performs as it should, though it should be noted that it’s not a particularly fast way to generate a random number.

So should you ever need a truly random number in your Arduino sketch rather than one that appears random enough for some purposes, you now know that you can safely disregard the documentation for a random seed and use the entropy library instead. Of course this comes at the expense of adding an extra library to the overhead of your sketch, but if space is at a premium you still have the option of some form of hardware noise generator. Meanwhile perhaps it is time for the Arduino folks to re-appraise their documentation.

The subject of entropy and generating random numbers is one that has appeared on these pages many times. [Voja Antonic] made a in-depth study using uninitialized RAM as an entropy source for microcontrollers. If you have an insatiable appetite for understanding Linux entropy, we point you at [Elliot Williams]’ comprehensive examination of the subject.

[Arduino image: DustyDingo Public domain]


Filed under: Arduino Hacks, Hackaday Columns, Microcontrollers, Skills

KIM-1 to COSMAC Elf Conversion — Sort Of

In the mid-1970s, if you had your own computer, you probably built it. If you had a lot of money and considerable building skill, you could make an Altair 8800 for about $395 — better than the $650 to have it built. However, cheaper alternatives were not far behind.

In 1976, Popular Electronics published plans for a computer called the COSMAC Elf which you could build for under $100, and much less if you had a good junk box. The design was simple enough that you could build it on a piece of perf board or using wire wrap. We featured the online archive of the entire Popular Electronics collection, but hit up page 33 of this PDF if you want to jump right to the article that started it all. The COSMAC Elf is a great little machine built around a 40-pin RCA 1802 processor, and for many was the first computer they owned. I lost my original 1802 computer in a storm and my recent rebuild in another completely different kind of storm. But there is a way to reclaim those glory days without starting from scratch.  I’m going to repurpose another retro-computing recreation; the KIM-1.

I’ll admit it, Rewiring a real KIM-1 to take an 1802 CPU would be difficult and unnecessary and that’s not what this article is about. However, I did have a KIM UNO — [Oscar’s] respin of the classic computer using an Arduino mini pro. Looking at the keyboard, it occurred to me that the Arduino could just as easily simulate an 1802 as it could a 6502. Heck, that’s only two digits different, right?

The result is pretty pleasing. A “real” Elf had 8 toggle switches, but there were several variations that did have keypads, so it isn’t that far off. Most Elf computers had 256 bytes of memory (without an upgrade) but the 1802 UNO (as I’m calling it) has 1K. There’s also a host of other features, including a ROM and a monitor for loading and debugging programs that doesn’t require any space in the emulated 1802.

Repurpose

The KIM UNO has 24 switches. There are 16 for the hex digits, of course. The top two rows mimic functions from the original KIM-1. A real Elf had a way to input a byte (usually 8 toggle switches), a load switch, a run switch, a memory protect switch, and a push button wired to a CPU pin. That means the hardware has more than enough switches.

On the display side, a normal Elf had a single-byte hex display although some clones had more. There was also the Q LED that a program could light or extinguish. The KIM UNO hardware has many 7-segment displays so it is possible to put those digits to use like an Elf clone. There isn’t an LED, however, except for the Arduino’s built in LED which is not normally visible in operation. However, the digital displays have decimal points and they are connected to the Arduino. So if you don’t mind using those, you have plenty of LEDs, too.

The hardware is open source and easy to duplicate. [Oscar] sometimes has kits as well and they are very inexpensive (about $20).

The KIM UNO software is open source, so I started there. I first stripped all the code out of the main file other than the parts that drove the display and the keyboard, then built up everything need to suppot 1802 emulation. You can find all the code in my 1802UNO GitHub repository.

Inside the 1802

The 1802 instruction set is very regular and quite simple. Most instructions use the top 4 bits as an op code and the bottom 4 bits to select one of sixteen 16-bit registers. So 0x12 increments register 2 and 0x15 increments register 5. There are only a handful of op codes that don’t follow this pattern. There’s also an 8-bit accumulator called “D” (not to be confused with register D).

One unique feature in the 1802 architecture is the program counter. There isn’t one. Well, more precisely, there are up to 16. Any of the registers can be the program counter and a subroutine call can be as simple as switching the program counter. Unfortunately, that isn’t very reentrant (or good for recursion). If you want a proper subroutine call, you had to code it yourself. RCA provided the “standard call and return technique” that had the unfortunate downside of destroying the accumulator.

With so few instructions, the emulator turns out to be a few switch statements and some pretty simple code. Although it is made to run with the KIM UNO hardware, like the KIM UNO, you should be able to use it with just about any Arduino via the serial port. It isn’t quite as fun as having the real hardware, but it is simpler.

Unreal

The emulator is reasonably accurate except it doesn’t simulate interrupts (since there is no source of them). However, it doesn’t faithfully reproduce the 1802’s load mode which used DMA. Instead, load mode is just completely custom code that enters data into memory. It does not simulate the cycle and register manipulations that go on in a real 1802 using DMA in load mode.

In addition to loading a program with the ersatz load mode, you can also move RAM back and forth to EEPROM or a PC via the serial port.

Serial and Push Buttons

The serial port is just the usual Arduino serial port set for 9600 baud. By default, the serial input will mimic the hardware keys. However, you can use the pipe character (‘|’) to shift the serial port into terminal mode. Then the 1802 code can read data from the serial port. You lose the front panel functions and there’s no way to go back until you cycle the power unless you make the 1802 code release the port.

A few of the push buttons have special functions if you hold them down for more than one second. For example, the AD button writes the EEPROM data into RAM. This is useful for storing a self-contained demo, for example.

You can find a summary of the keyboard and serial commands on the GitHub site. The serial port can do things you can’t do from the front panel, like set a trace mode, dump the CPU registers, and more.

Building

The hardware doesn’t require any changes to the stock KIM UNO kit. There’s a lot to solder and once you solder the displays on, it would be hard to get the Arduino back off the board.

You could probably build the software using the Arduino IDE, but I used Platform IO. That lets me use the editor of my choice, but you ought to be able to get the code to work in the IDE, as well. There is enough memory to make the RAM slightly bigger, but I didn’t do it. Since one way to save and load the RAM is to EEPROM, I didn’t want the RAM to be larger than the EEPROM. In addition, the RAM “maps” like a real Elf (that is, RAM at location 0x0 also appears at 0x4000, 0x8000, etc). This would be more difficult if you added a little bit more than 1K of RAM.

There are a few other options at the top of 1802config.h. You can select how often the screen and keyboard refresh. Higher values are slower to refresh but faster to execute code. You can change the I/O ports associated with the keyboard, displays, and serial port. You can also change the serial escape character.

Examples

There are some examples provided that blink the LEDs and manipulate the serial port. If you look around, there’s a lot of 1802 code on the web. However, be aware that most 1802s don’t have a hardware UART. They emulate serial ports using the Q output and one of the EF inputs. That’s fine for a real device even though it takes lots of code, but for this virtual device, it isn’t practical. You’ll need to rip out any code that does serial I/O and replace it with single I/O instructions.

If you have a binary file (or a format you can convert to binary) I have a converter written in C included on GitHub. You can compile it on nearly any platform and use it to convert. It always assumes address. If that’s not right, you can always open the output in a text editor and adjust.

In addition, there are three ROMs included that you can try. By default, there is a simple high-low game. There are also two monitors, one for use with the built-in keyboard and another for use with a serial port. To select a ROM, edit 1802rom.h and change the comments so the ROM you want is not commented and the others are.

Practical?

Emulators are fun, but as the song goes, there’s nothing like the real thing. If that’s not authentic enough for you, it is possible to build a very authentic looking Elf, even today. The reason real 1802s are still around is they had several desirable characteristics, namely low power consumption and resistance to radiation.

The Arduino simulation has neither of those features. However, it is a fun retrocomputing toy, inexpensive, and a great learning tool. The CPU is simple enough to program directly in machine code and the portability is better than most other old school computers.

If you want to learn more about the 1802 there are several sites dedicated to it and a very helpful Yahoo group. One site has a very prolific software author, but most of the code won’t fit in the 1802 UNO’s 1K RAM. Maybe a version with more memory is in the future.


Filed under: Arduino Hacks, classic hacks, computer hacks, Hackaday Columns

First Look at ABC: Basic Connections

[Alberto Piganti], aka [pighixxx] has been making circuit diagram art for a few years now, and has just come out with a book that’s available on Kickstarter. He sent us a copy to review, and we spent an hour or so with a refreshing beverage and a binder full of beautiful circuit diagrams. It doesn’t get better than that!

[pighixxx] started out making very pretty and functional pinout diagrams for a number of microcontrollers, and then branched out to modules and development boards like the Arduino and ESP8266. They’re great, and we’ll admit to having a printout of his SMD ATMega328 and the ESP-12 on our wall. His graphical style has been widely copied, which truly is the sincerest form of flattery.

But after pinouts, what’s next? Fully elaborated circuit diagrams, done in the same style, of course. “ABC: Basic Connections” started out life as a compendium of frequently used sub-circuits in Arduino projects. But you can take “Arduino” with a grain of salt — these are all useful for generic microcontroller-based projects. So whether you want to drive a 12 V solenoid from a low-voltage microcontroller, drive many LEDs with shift registers, or decode a rotary encoder, there is a circuit snippet here for you.

One of the things that we like most about the graphics in “ABC” is that they’re not dumbed down — they’re fundamentally just well-done circuit diagrams, but with graphic touches and extra detail where it actually helps to clarify things. This is a middle ground between the kind of schematic you use in a PCB layout program and the kind of diagram you get from Fritzing. In the former, every part has a symbol but multifunction parts like microcontrollers are just represented as squares bristling with pin numbers. In the latter, wiring up an IC is easy because the parts and pins are represented graphically, but you quickly run out of colors for the different wires, and the “breadboard” turns into a rat’s nest with a circuit of any complexity.

“ABC” takes the middle road, using standard circuit diagram style overall, but also the nice graphic representations of the ICs and modules that [pighixxx] is good at. Is a 2N2222 pinned EBC or BCE? You don’t have to look that up, because it’s sketched out for you here. We’d guess that this attractive, but information-rich, style is a great fit for the target audience — people with some electronics experience who do not yet have their favorite transistor symbol tattooed on their forearm. [pighixxx]’s diagrams are simple, easy to understand, easy to use, and pretty to boot.

There is a planned online counterpart to the book, with further elaborations of all of the circuit setups. They’re not finished yet, but they have a lot more of the flavor of the Fritzing-style, this-wire-goes-to-that-hole diagrams. This style does work better in an online format than in a physical book, because you can build up the rat’s nest in bite-sized steps, none of which are too overwhelming. But honestly, for an advanced beginner or intermediate electronics hacker, the book can be treated as stand-alone. The web content may help the rank newbie when they get stuck.

Tee-hee.

The breadth of circuits in “ABC” is fairly wide, covering most of the microcontroller-interfacing problems that we’ve ever encountered. None of the circuits are revolutionary — they’re the tried-and-true, correct solutions to the various problems, rather than anything too hacky or clever. We weren’t surprised by any of the circuits, but we didn’t find anything that we wouldn’t use ourselves either. These are basic connections after all, and a darn solid collection of them.

To sum up, “ABC” is an attractive book in a handy binder format that would make a great collection of solutions for anyone who’s just getting started in the whole “Arduino” scene but who gets hung up on interfacing the chips with the real world. It’s a handy reference for the pinouts of a number of frequently used parts, combined with the resistors, flyback diodes, level-shifting circuits, and whatever else that you’d need to make them work. It’s what we wish our simple circuit diagrams looked like. We like it.


Filed under: Hackaday Columns, reviews

Friday Hack Chat: Tenaya Hurst From Arduino

Join us this Friday at noon PDT for a Hack Chat with Tenaya Hurst of Arduino. If you’ve been one of the big Maker Faires over the last few years (or innumerable other live events) and stopped by the Arduino area you’ve probably met Tenaya. She is the Education Accounts Manager for Arduino and loves working with wearable electronics.

Come and discuss maker education and the role Arduino is playing in getting our students excited about electronics, and STEAM education in general. Tenaya will also be discussing a new wearable tech kit she’s been working on. We hope to see the gear in person at Bay Area Maker Faire next week.

Here’s How To Take Part:

Our Hack Chats are live community events on the Hackaday.io Hack Chat group messaging.

Log into Hackaday.io, visit that page, and look for the ‘Join this Project’ Button. Once you’re part of the project, the button will change to ‘Team Messaging’, which takes you directly to the Hack Chat.

You don’t have to wait until Friday; join whenever you want and you can see what the community is talking about.


Filed under: Arduino Hacks, Hackaday Columns, wearable hacks

PlatformIO and Visual Studio Take over the World

In a recent post, I talked about using the “Blue Pill” STM32 module with the Arduino IDE. I’m not a big fan of the Arduino IDE, but I will admit it is simple to use which makes it good for simple things.

I’m not a big fan of integrated development environments (IDE), in general. I’ve used plenty of them, especially when they are tightly tied to the tool I’m trying to use at the time. But when I’m not doing anything special, I tend to just write my code in emacs. Thinking about it, I suppose I really don’t mind an IDE if it has tools that actually help me. But if it is just a text editor and launches a few commands, I can do that from emacs or another editor of my choice. The chances that your favorite IDE is going to have as much editing capability and customization as emacs are close to zero. Even if you don’t like emacs, why learn another editor if there isn’t a clear benefit in doing so?

There are ways, of course, to use other tools with the Arduino and other frameworks and I decided to start looking at them. After all, how hard can it be to build Arduino code? If you want to jump straight to the punch line, you can check out the video, below.

Turns Out…

It turns out, the Arduino IDE does a lot more than providing a bare-bones editor and launching a few command line tools. It also manages a very convoluted build process. The build process joins a lot of your files together, adds headers based on what it thinks you are doing, and generally compiles one big file, unless you’ve expressly included .cpp or .c files in your build.

That means just copying your normal Arduino code (I hate to say sketch) doesn’t give you anything you can build with a normal compiler. While there are plenty of makefile-based solutions, there’s also a tool called PlatformIO that purports to be a general-purpose solution for building on lots of embedded platforms, including Arduino.

About PlatformIO

Although PlatformIO claims to be an IDE, it really is a plugin for the open source Atom editor. However, it also has plugins for a lot of other IDEs. Interestingly enough, it even supports emacs. I know not everyone appreciates emacs, so I decided to investigate some of the other options. I’m not talking about VIM, either.

I wound up experimenting with two IDEs: Atom and Microsoft Visual Studio Code. Since PlatformIO has their 2.0 version in preview, I decided to try it. You might be surprised that I’m using Microsoft’s Code tool. Surprisingly, it runs on Linux and supports many things through plugins, including an Arduino module and, of course, PlatformIO. It is even available as source under an MIT license. The two editors actually look a lot alike, as you can see.

PlatformIO supports a staggering number of boards ranging from Arduino to ESP82666 to mBed boards to Raspberry Pi. It also supports different frameworks and IDEs. If you are like me and just like to be at the command line, you can use PlatformIO Core which is command line-driven.

In fact, that’s one of the things you first notice about PlatformIO is that it can’t decide if it is a GUI tool or a command line tool. I suspect some of that is in the IDE choice, too. For example, with Code, you have to run the projection initialization tool in a shell prompt. Granted, you can open a shell inside Code, but it is still a command line. Even on the PlatformIO IDE (actually, Atom), changing the Blue Pill framework from Arduino to mBed requires opening an INI file and changing it. Setting the upload path for an FRDM-KL46 required the same sort of change.

Is it Easy?

Don’t get me wrong. I personally don’t mind editing a file or issuing a command from a prompt. However, it seems like this kind of tool will mostly appeal to someone who does. I like that the command line tools exist. But it does make it seem odd when some changes are done in a GUI and some are done from the command line.

That’s fixable, of course. However, I do have another complaint that I feel bad for voicing because I don’t have a better solution. PlatformIO does too much. In theory, that’s the strength of it. I can write my code and not care how the mBed libraries or written or the Arduino tools munge my source code. I don’t even have to set up a tool chain because PlatformIO downloads everything I need the first time I use it.

When that works it is really great. The problem is when it doesn’t. For example, on the older version of PlatformIO, I had trouble getting the mBed libraries to build for a different target. I dug around and found the issue but it wasn’t easy. Had I built the toolchain and been in control of the process, I would have known better how to troubleshoot.

In the end, too, you will have to troubleshoot. PlatformIO aims at moving targets. Every time the Arduino IDE or the mBed frameworks or anything else changes, there is a good chance it will break something. When it does, you are going to have to work to fix it until the developers fix it for you. If you can do that, it is a cost in time. But I suspect the people who will be most interested in PlatformIO will be least able to fix it when it breaks.

Bottom Line

If you want to experiment with a different way of building programs — and more importantly, a single way to create and build — you should give PlatformIO a spin. When it works, it works well. Here are a few links to get you started:

Bottom line, when it works, it works great. When it doesn’t it is painful. Should you use it? It is handy, there’s no doubt about that. The integration with Code is pretty minimal. The Atom integration — while not perfect — is much more seamless. However, if you learn to use the command line tools, it almost doesn’t matter. Use whatever editor you like, and I do like that. If you do use it, just hope it doesn’t break and maybe have a backup plan if it does.


Filed under: Arduino Hacks, ARM, Hackaday Columns, reviews, Skills

My DIY BB-8: Problems, Solutions, Lessons Learned

Imagine trying to make a ball-shaped robot that rolls in any direction but with a head that stays on. When I saw the BB-8 droid doing just that in the first Star Wars: The Force Awakens trailer, it was an interesting engineering challenge that I couldn’t resist. All the details for how I made it would fill a book, so here are the highlights: the problems I ran into, how I solved them and what I learned.

The Design Criteria: Portable and Inexpensive

Carrying BB-8

I had two design criteria in mind. The first was to keep it low-cost. Some spend $1000 to $1500 on their BB-8s. I wanted to spend as little as I could, so as many parts as possible had to come from my existing stock and from online classifieds and thrift stores. Not counting the parts that were discarded along the way, the cost came in just shy of $300.

The second design criteria was to make it portable. It had to be something I could take on the bus or carry while walking a reasonable distance (I once carried it twenty-five minutes to a nearby school).

Both of these criteria meant that it had to be smaller than full size. A full size ball is 20″ in diameter. Mine has a 12″ ball which makes it 3/5 scale. Also, the larger it is, the more powerful and costly the motors, batteries, motor controllers, magnets, and so on.

Version One: Quick And Easy

First I tried a minimalist approach. For the ball, I found my 12″ cardboard globe on kijiji.ca. I bought an RC toy truck at a yard sale and attached a pole to it for holding magnets near the top of the globe. I then made a head with corresponding magnets under it. The head magnets attract to the pole magnets, keeping the head on. Meanwhile, the truck rolling inside the ball makes the ball move.

Making the ball roll around was easy. Making it roll around while keeping the head on was very hard. The magnets at the top of the pole attract the magnets under the head, pulling them down hard onto the surface of the globe. That essentially glues the truck to the top of the globe.

To overcome that the truck needs sufficient traction. That also means the truck needs to be heavy. And lastly, the truck’s motor needs to be powerful enough to overcome its own weight and the grip of the magnets on the ball. The alternative is to make the magnetic attraction weaker, but if it’s too weak the head falls off. It’s a tricky balancing act, in both senses of the word.

But the most dome-like head I could make stay on was just a cardboard skeleton. Anything more filled out would be heavier and require a stronger magnetic attraction. The toy truck’s motor would not be up to it.

Version Two: Drill Motors And Drill Batteries

Batteries and motors in Blender

For more powerful motors and more mass I figured I could kill two birds with one stone by going with drill motors and drill batteries in a hamster drive configuration. Using drill parts kept costs down as the batteries and one drill came from yard sales while the other drill was free through freecycle.org. Meanwhile, both are heavy.

To make sure it all fit, I drew up a 3D model in Blender, the free 3D modelling and animation software that I use a lot. In fact, finding out how to make the batteries and motors fit was the first step. They had to be as low as possible. Their large mass low down is what keeps the droid vertical, with the much lighter head at the highest point.

Batteries Velcroed and a connector

The drill batteries had to be easily removable for recharging. To hold them under the drive plate I simply used Velcro. Meanwhile, the drill battery stems went up through a hole in the drive plate. I made a connector to electrically connect to the battery terminals. It is a plastic rectangle with thin copper sheet metal for the contacts. Once the battery was Velcroed in place, this plastic and metal piece was lowered down onto the stem, the copper metal making contact with the battery terminals.

The Electronics

For the brains I used an Arduino UNO. To drive the motors I had all the parts for making two H-bridge driver boards, with the exception of 4 MOSFETs and some fuses. The Arduino does pulse width modulation (PWM) to the driver boards for speed control, as well as playing sounds at certain times when the motors are turned on.

For remote control, I hacked the RC receiver from the toy truck and added an extra set of AA batteries in parallel for more runtime. A problem I ran into right away though, was that the RC receiver put out voltages of both polarities based on which direction the motors should rotate, whereas the Arduino’s pins take only positive voltages. To solve that I came up with a converter board to go between them.

Getting all that to work reliably took a while. Before I added fuses, I burned a few MOSFETs. At one point I’d put an N-type MOSFET where a P-type should have gone and vice versa. That resulting problem alone took a few days of spare time to figure out.

The wheels were old Rollerblade wheels — I keep a small bucket of these in my shop. I decided I wanted the ball to roll at around 1 foot per second and doing the math, that meant the wheels would have to rotate at around 2 rotations per second or 120 RPM. I found a PWM value that would give something close to that and started blowing fuses. I started with 1 amp fuses, then 2, 5, and finally settled on 10 amp fuses.

My final hurdle was that the motors would behave oddly when the motors were told to turn in opposite directions but were fine when they were told to turn in the same direction. This turned out to be a bad assumption on my part about how the RC receiver was wired internally — none of the output wires were common inside. After some changes to the circuit, I now had stable electronics.

I had basically been treating the RC receiver as a black box, but when I asked for help about my converter board here on Hackaday, it was pointed out that the receiver likely contained H bridges. Opening it up, that’s exactly what I found. The converter board works fine for now, but in the near further I’ll use one of the suggestions from that Hackaday post to eliminate the board altogether. I might even try all of the suggestions, just for fun.

The Drive System

The drive system

The motors were too long to fit in line between the wheels and so had to be mounted off to the sides. To transfer rotation to the wheels, I drew up some gears in Blender and 3D printed them at our local University of Ottawa Makerspace. In the print settings I used 2 shells and only 50% infill. The gears are held firmly onto the shafts solely using nuts and washers on either side. They’ve held up amazingly well, even with slipping and grinding during development.

For the bearings for the center gears and the wheels I used an old trick of making bearing blocks from hardwood.

Putting Loctite and screwing gear to motor shaft

I wanted to keep as much room as possible on the drive plate available for adding things later and so initially I’d mounted the motors at only three points. But this allowed the motors to move a little causing the gears to slip. To fix that I later added a fourth mounting point and haven’t had any slipping since.

At the end of the shaft for the drill motors is a hole that a screw goes into. That’s part of how the chuck is kept on a drill motor, and that’s how one gear was kept on. However, this screw had a tendency to get loose. Putting a little Loctite on the threads fixed that.

Stability

Rearranging BB-8’s internals

Given that I was trying to fit a lot in a small droid, I had to mount some things higher than I’d have liked to. When the droid stops, mass high up causes the droid to wobble. In the BB-8 droid used for promotional events they’ve gone to a great extent to keep the majority of the mass as low as possible. Originally I had the Arduino batteries and the RC receiver with its extra batteries fairly high up. I later mounted them much lower. When holding the internals in my hands I could tell the difference but it didn’t make a noticeable difference with the wobble.

Instead, for that I added Adafruit’s BNO055 inertial measurement unit (IMU) board. With it I could tell what angle the droid was at when stopping and experimented with PID loops and other algorithms of my own to minimize wobble. That helped.

The Ball

As I said, I used a 12″ cardboard globe. To increase the traction of the wheels inside the globe I sprayed the globe’s interior with an anti-slip spray from a hardware store. This made a huge difference. However, over time the anti-slip coating vanished, and so I’m looking for another, more permanent coating, perhaps urethane or something. If anyone has any suggestions, please let me know in the comments. It also has to stop off-gassing eventually. The anti-slip spray had an odor for a long time.

Sealing globes – cardboard and fiberglass

But the main problem with the cardboard globe was its thinness. With the huge weight of the internals pushing down on it where the wheels are, it warped significantly and required a lot of effort and a copious amount of tape to keep the two hemispheres together. This large amount of tape also made an uneven ridge for the head to slide over, making it get stuck. At that point either the drive system continued to move while the head fell off or the drive system couldn’t move at all.

The solution was to carefully coat a new globe in three layers of fiberglass. I took my time doing this, over a month and a half, applying one piece at a time and then sanding before putting the next piece. The result was a fantastic improvement. It no longer deformed and now it takes a minimal effort to attach the two hemispheres with only eight narrow strips of transparent duct tape.

The Never-ending Saga

My DIY BB-8 in action

My BB-8 is now at the point that I can call it finished. At least it’s finished as far as all the engineering is concerned, and that’s usually where I call it quits.

While the paint job worked out well, up close you can see that the details are painted on. It’d be nice to have at least the lines on the head be actual grooves. Also, these drill motors are brushed motors and I’ve since learned that doing high frequency PWM to brushed motors damages them over time. I’d like to replace them with brushless motors. And as anyone who’s use cordless drills a lot knows, drill batteries don’t last long, and so it’d be great to switch to LiPos.

But for now, the reactions I get from both kids and adults is beyond my wildest expectations. Kids threat it like a friend while adults have petted it and called to it like it was a baby or a dog wagging its tail. I’d call that a success.


Filed under: Hackaday Columns, robots hacks

My Life in the Connector Zoo

“The great thing about standards is that there are so many to choose from.” Truer words were never spoken, and this goes double for the hobbyist world of hardware hacking. It seems that every module, every company, and every individual hacker has a favorite way of putting the same pins in a row.

We have an entire drawer full of adapters that just go from one pinout to another, or one programmer to many different target boards. We’ll be the first to admit that it’s often our own darn fault — we decided to swap the reset and ground lines because it was convenient for one design, and now we have two adapters. But imagine a world where there was only a handful of distinct pinouts — that drawer would be only half full and many projects would simply snap together. “You may say I’m a dreamer…”

This article is about connectors and standards. We’ll try not to whine and complain, although we will editorialize. We’re going to work through some of the design tradeoffs and requirements, and maybe you’ll even find that there’s already a standard pinout that’s “close enough” for your next project. And if you’ve got a frequently used pinout or use case that we’ve missed, we encourage you to share the connector pinouts in the comments, along with its pros and cons. Let’s see if we can’t make sense of this mess.

FTDI TTL Serial

The de-facto standard for a hacker’s TTL serial pinout is definitely the layout that FTDI uses for their USB/TTL serial cables. Said cable is just so handy to have on hand that you’d be silly to use any other pinout for the job. And the good news is that the rest of the world has basically joined in. From the Chinese “Pro Mini” cloneduinos to the Hackaday Edition Huzzah ESP8266 board, and from Adafruit’s FTDI Friend to Modern Device’s USB-BUB, almost everyone uses this pinout. A victory for the common man!

There is one slight point of contention, however, and that’s whether pin 6 is DTR or RTS#. We never use either, so we couldn’t care less, but if you’re counting on your programmer sending the DTR signal to enter programming mode on the device (we’re looking at you Arduino!) then you’ll want DTR on pin 6, and the original FTDI cable, ironically, has the “wrong” pinout. Perhaps that’s why Sparkfun avoided the whole debacle and went their own way, breaking out every signal off the FTDI chip into their own unique configuration.

If you’re only going to break out TTL serial lines, you’d be a fool to use any other pinout.

Modules and Other Communications

Unlike the case with simple serial, connecting various kinds of modules to mainboards is a difficult problem. Creating a single pinout or connector specification for many potential protocols or arbitrary signals is a Herculean task. Surprisingly, it’s been done a few times over. Here are some notables.

Pmod

Digilent makes a wide range of FPGA demo boards, and they needed an in-house standard pinout that they could use to plug into various add-on peripheral modules that they sell. Thus Pmod was born. It has since become a full-fledged, and trademarked, open standard that you can use in your designs. Here’s the PDF version of the specification for you to print out, so you know they mean business.

There are a few aspects of Pmod that we think are particularly clever. First, the number of pins involved is “just right” at six, and it’s easily expandable. They use standard 0.1″ pitch pins and headers. Two lines carry power and ground, leaving four free pins for SPI, UART, or whatever else. The specification is that all power and signal voltages are 3.3 V because they’re designed for FPGAs after all. You can mix and match if you know what you’re doing, but they won’t let you call it Pmod(tm).

Eric Brombaugh’s iceRadio FPGA SDR, plugged together with Pmods

If you need more than four signals, there’s a twelve-pin version which is just two six-pin Pmods stacked into a double-row header. The extra power and ground are redundant, but it makes a twelve-pin output very flexible, because nothing stops it from being used as two sixes. The standard also says that the twelve-pin headers are to be spaced at 0.9″ center-to-center, so you can even connect two of them together when you need sixteen board-to-board signal connections. We like the modularity and expandability.

Pmod connectors are multi-protocol, but for each protocol there is a single pinout. So there’s an SPI Pmod and an I2C Pmod, and the pins are always in the same place. There isn’t a Pmod standard for every conceivable application, of course, so there’s a GPIO pinout that gives you free rein over what goes where. We think that it would be nice if some additional notable protocols (I2S? one-wire? servos? analog stereo audio?) were included in the specs, but the community can also handle these lower-level details.

In our eyes, Pmod is nearly perfect. It uses cheap hardware, is easily expandable, and the smallest incarnations are small enough to fit on all four sides of a one-inch-square board. If you’re willing to pay the brand-name premium, Digilent makes an incredible range of modules. We want to see more hackers outside of the FPGA scene get on it.

mikroBUS

What Digilent is to development boards in the US, MikroElektronika is in Europe. While Pmod aims to be capable of doing anything, Mikro-E’s mikroBUS connector wants to do everything, which is to say it has I2C, SPI, UART, two voltages, and even a few extra signals all on the same pinout. Physically, it’s two single rows of eight pins, spaced 0.9″ apart side-to-side, which means it fits into a breadboard nicely. Here’s the spec in PDF.

The tradeoff here, relative to Pmod, is that a lot of pins go unused on any given design. With (only) one “analog” channel, you wouldn’t choose mikroBUS to send stereo audio, whereas nothing stops you from calling the Pmod’s GPIO lines analog and sending four channels of sound. But that mikroBUS gains is fool-proofness. (Well, they could have also made it asymmetric…) There’s no chance of a newbie plugging an SPI module in where an I2C module is expected and scratching their heads. With mikroBus, it should just work.

Microchip has added a mikroBus port to their Curiosity boards, and MikroElektronika makes a ton of modules. If your audience consists of beginners, and one footprint for all protocols, it’s worth considering.

Seeed’s Grove

Meanwhile in China, Seeed Studios makes open-source modules, and makes them cheap. Their Grove connector uses only four pins, with power and ground among them. The have standard pinouts for UART, I2C, and for servo motors. Sensors and other analog peripherals are allocated one “primary” pin and one “secondary” and it’s assumed that you know what you’re doing. The idea behind their system is that you add a shield to your microcontroller board, and they break out the relevant pins into these four-pin Grove headers.

This is great for small things and I2C devices, which is Seeed’s catalog, but there just aren’t enough signal pins to run SPI or an analog RGB LED, for instance. But because of the small number of pins, they use very inexpensive polarized cables and shrouds that you can’t plug in the wrong way, and that take up relatively little board space. That’s Grove’s design trade-off.

Servo Motor control

One of these things is not like the others…

Hobby servo motors need three wires: voltage, ground, and a signal to tell them where to point. There are three distinct ways to arrange these wires, but Futaba, HiTec, Tower, GWS, and JR servos all chose to put them in SVG (or GVS) order, and there’s no reason to buck the trend. (Airtronics, what were you thinking?!)

SVG is also a handy pinout to use for all sorts of one-signal sensors or actuators where space is a premium, and we’ve seen this in a few designs (here and here, for instance). But we’re torn. Relative to the Grove, for instance, you’re just saving one pin. Even the Pmod would work with only three pins’ overhead. Is that worth complicating your life with another pinout? If you need a lot of powered one-signal devices, or servos, it probably is, and you can hardly beat SVG or GVS, whichever way you look at it.

Arduino

Viewed in the light of any of the other module interconnection systems, the Arduino is the worst of all worlds. It’s monolithic like mikroBUS, but it’s gigantic — you have to build a 55×73 mm board and accommodate 30 pins and pass-throughs if you’d like it to be stackable. The pins have a funny spacing (that gap!), that doesn’t fit any normal protoboard. Nobody goes through the trouble of building a shield just for an I2C connection. No wonder most Arduino projects look like a breadboard hedgehog. About the only good thing we can say about it is that it’s hard to plug one in backwards.

There’s also the tiny little factor that there’s a million Arduino shields out there, a huge community built around them, and widespread support for them. Which turns out to trump all of the reasonable design concerns. (Shakes head.)

Miscellany

Of course, there are other very specific pinouts that one might encounter, like the old ESP-01 module, or the XBee, or the nRF24. Adapting modules to fit boards is always going to be a pain, because the manufacturers will pick whatever suits them on that day. Programmer pinouts for specific microcontrollers are a similar story, as is JTAG, which is a beautiful standard with a dogs’ breakfast of pinout possibilities. (We could do a whole column!)

Faced with this inevitability, and the need for many one-off adapters, what can you do? What we do is buy a lot of those cheap “Dupont” female-to-female cables, get the connections working and tested, and then tape them permanently together and label them. It’s not as pretty as a dedicated PCB adapter, but it’s quick and easy and gets you moving on to what you wanted to do in the first place.

Wrapup and Recommendations

The goal of connectors, and their standards, is putting parts together. If you’re designing a sensor module with more than a couple components, and you want it to be maximally easy for yourself and others to hook up to whatever mainboard they’ve got, this is no easy task. The end result is a proliferation of translators, adapter boards, hats, shields, capes, or whatever else. We have a drawer and a half full, and we bet you do too.

Yes, I do see what I’m suggesting here. [source: xkcd 927]
We’d be happy to see the world settle on Pmod for most needs, honestly, and we’d even throw away our beloved FTDI serial pinout in the name of standardization (or make an adapter). We can also see the need for exceptions like SVG / servo connectors when small sensors or multiples are in play. There will always be the need for dedicated on-board connectors as well, of course. Nobody said hardware was easy.

What’s your solution to the ultimate connector conundrum? Are there important connector systems that we’ve left out? What are their design tradeoffs? How stoked would you be if things could just plug together? Let us know!

Thumbnail image courtesy of [Raspberry Pi Controller].


Filed under: Engineering, Hackaday Columns, hardware, rants

3D Printering: Trinamic TMC2130 Stepper Motor Drivers

Adjust the phase current, crank up the microstepping, and forget about it — that’s what most people want out of a stepper motor driver IC. Although they power most of our CNC machines and 3D printers, as monolithic solutions to “make it spin”, we don’t often pay much attention to them.

In this article, I’ll be looking at the Trinamic TMC2130 stepper motor driver, one that comes with more bells and whistles than you might ever need. On the one hand, this driver can be configured through its SPI interface to suit virtually any application that employs a stepper motor. On the other hand, you can also write directly to the coil current registers and expand the scope of applicability far beyond motors.

The TMC2130 SilentStepStick’s top side with SPI headers and heatsink.

Last month, we took a closer look at microstepping on common stepper driver ICs, but left out the ones that we actually want to use: the smart ones. Trinamic provides some of the smartest stepper motor drivers on the market, and since the German hacker store Watterott released their SilentStepStick breakout boards for the TMC2100 and TMC2130, they are also setting a new standard for DIY 3D printers, mills and pick-and-place robots. I recently acquired a set of both of them for my Prusa i3 3D printer, and the TMC2130 with its SPI configuration interface really caught my attention.

The TMC2130 SilentStepStick should not be confused with the — far more popular — TMC2100 variant. As the name suggests, it comes as a StepStick-compatible breakout board, and just like it’s famous sibling, features a Trinamic IC on the bottom side of the little PCB. Several vias and copper spills conduct heat away from the IC’s center pad, allowing a heatsink on the top side to effectively cool the driver.

The bottom side with the stand-alone mode solder blob jumper next to the IC.

However, unlike the TMC2100, this one won’t let your motors spin right away. You’ve got two options: Hard-wire it in stand-alone mode, which practically turns it into a TMC2100, or hook up to its SPI-interface and dial in if you want your stepper motor shaken or stirred. In fact, plentiful configuration registers make the TMC2130 an extremely hackable chip, so I’m not even thinking about bridging that solder jumper on the SilentStepStick’s bottom side that activates the stand-alone mode.

First Steps

Wiring the TMC2130 to a classic RAMPS 1.4.

As said, before the driver does anything, it wants to be configured, and it’s worth mentioning that all configuration registers are naturally volatile, so if I want to use them in my 3D printer, I need to configure them as part of the printers startup routine.

The RAMPS 1.4 on my 3D printer breaks out the hardware SPI interface of the underlying Arduino through its AUX3 pin header, along with two additional digital pins (D53 and D49), which I used for the cable select signals. After crimping a cable to connect two TMC2130’s to the AUX3 header, I could start digging into the software part.

Watterott provides an example sketch, which writes a basic configuration to the driver’s registers and spins an attached stepper motor. Great stuff, but the datasheet describes 23 configuration registers waiting to be finely tuned, and 8 more to read diagnosis and status data from. So, I wrote a little Arduino library that would make the numerous configuration parameters available in a more practical way. From there, I could just include my library into the Marlin-RC7 3D printer firmware I’m using. Luckily, the current Marlin release candidate already features support for TMC26X drivers, so I could reuse some of its code to put together a Marlin fork that includes 59 of the TMC2130’s parameters in its define-based configuration files. And then, I could take the little buddies out for a spin.

First steps on a RAMPS 1.4 on a somewhat-uino (sorry Massimo). The testing-contraption to the left is a NEMA 17 stepper motor attached to an encoder.

Taking Them For A Spin

With the hardware set up and the software working as supposed, I ran a few sanity tests: toggling parameters on and off and checking how the driver’s behavior changes during printing. Since the TMC2130 let’s you tune almost everything it’s doing, that’s a good first step that helps to eliminate some variables and picking others that are worth a deeper look. Most of the settings can be changed on the fly and mid-print, however, not all parameters can actually be safely changed while the motors are running.

The TMC’s in service. I’m using the SPI-configurable TMC2130’s (silver heatsink) for the X- and Y- axis. The Z-axis and the extruder feature the TMC2100 (black heatsink). All of them are sitting on additional free-runner diode protection shields.
An excerpt of Trinamic’s thorough quick start guide.

To actually tune the drivers for a certain application, Trinamic provides a quick start guide in the datasheet, as well as detailed information on each parameter, and on how they interact. Basically, the first step is adjusting the RMS coil current by using the onboard potentiometer on the SilentStepSticks. Then, we need to chose the analog input pin as a current scaling reference to actually make use of the potentiometer. The mentioned library lets me do this through a simple method:

myStepper.set_I_scale_analog(1); // 0: internal, 1: AIN

The running and holding current are the first real parameter that should be tuned, with the running current typically at the desired maximum current, and the holding current at 70% of this value. The delay between a stillstand and the transition from running current to holding current can be adjusted between 0 and 4 seconds, and for now, I set it to 4 seconds, practically disabling the current reduction while the 3D printer is running. The three values share one write-only register, so the corresponding method call looks like this:

myStepper.set_IHOLD_IRUN(22,31,5); // [0-31],[0-31],[0-5]

and sets the running current to 100% (≙ 31), the holding current to about 70% of this value (≙ 22), and the delay between the two to 4 seconds (≙ 5).

I want torque, so I can leave stealthChop disabled. The datasheet suggests some starting values for configuring the chopper’s off time and the comparator’s blank time settings, but since it’s a key tradeoff between switching noise and torque, it makes sense to iterate through other values as well. The library methods for the two values look like this:

myStepper.set_tbl(1); // [0-3]
myStepper.set_toff(8); // [0-15]

And finally, I need to pick a microstepping resolution and choose if I want to make use of the 256 microstep interpolation feature, covered later in this article:

myStepper.set_mres(32); // {1,2,4,8,16,32,64,128,256}
myStepper.set_intpol(1); // 0: off, 1: interpolate

I have yet to walk through the entire tuning procedure, which includes monitoring the coil current on the scope and eliminating distortions in the zero crossing, but I’m getting a clue of the driver’s potential.

Juice

It’s maximum continuous RMS current of about 1.2 A per coil (at least in the QFN package on the SilentStepSticks) lets it look like a low-current driver, inferior to the common A4988 and DRV8825. In practice, it outperforms both of them by making intelligent use of a 2.5 A peak current margin. This gives it more than enough torque for 3D printing. I wouldn’t recommend pushing them over 0.9 A RMS though since the IC will momentarily pull more current if it needs more. For SilentStepStick users, that’s a Vref of 0.88 V. Through the SPI-interface, you can choose how much current you want to send through the motor coils when it’s spinning, and when it’s idling. You can choose after how many seconds it will start to decrease the current to a lower holding current when the motor is in standstill, and then to an even lower idling current. And, of course, you can also set it to squeeze out the maximum juice for everything.

Shifting The Gears

Where it starts getting interesting are settings like the high-velocity mode. Above a configurable velocity threshold, the driver offers you to automatically switch the chopper to a faster decay time to squeeze out some extra speed. You can also literally shift the gears by letting the driver internally switch from microstepping to full-step mode once it’s up to speed.

Microstepping

Choosing a finer microstepping resolution smoothens the stepper’s movement, reduces vibrations and sometimes even increases the positioning accuracy. However, it also multiplies the load on the microcontroller, which has to churn out 16, 32 or 256 times more step pulses per second. The TMC2130 lets you pick an input resolution between 1 and 256 microsteps per full-step, and then gives you the option of interpolating the output resolution to 256 microsteps. This allows for smooth operation even on increasingly retro 8-bit AVR motion controllers, which cannot deliver high step frequencies. Also, by configuring the TMC2130’s interface to double-edge step pulses, you can at least double the step frequency at almost no cost. Given that the modern IC still features the classic step/direction interface and even an enable pin, those few additional features actually make it a sweet drop-in upgrade for less-recent CNC and 3D printer electronics.

Noise Reduction

The TMC2130’s datasheet promises undistorted output with stealthChop.

Just like the TMC2100, the TMC2130 features two efficient and silent drive modes: spreadCycle, and stealthChop. The former delivers high torque at relatively low noise emissions, the latter one is almost inaudible but delivers a dramatically reduced torque. The flexible IC also allows you to tweak the chopper yourself to find the right balance between torque, noise, and efficiency for your application. One of the more noteworthy options in this regard is the possibility of randomizing the chopper’s off time. Since most of the audible noise is released due to dubstep the chopper busily switching the stepper motor’s coils, this option spreads the noise over a wider frequency range to subjectively silence the stepper motor.

Stall Detection

The TMC2130 notices when the motor is stalled and losing steps by measuring the motor’s back EMF. Along the way, it counts missed steps, allowing the controller to compensate for otherwise irreversible step-loss. It’s also a great way to react to obstacles rather than running into them full-force and, of course, the feature can be used as an axis endstop. Trinamic calls this feature StallGuard, and just like anything else in this motor driver, it’s highly configurable.

Direct Mode

Instead of letting the motor driver handle everything for you, you can also choose the direct mode. This mode practically turns the driver into a two-channel, bipolar constant-current source with SPI interface. You can still use it as a motor driver, but the possibilities reach far beyond that. It’s worth mentioning that the datasheet might be a bit confusing here, and the corresponding XDIRECT register actually accepts two signed 9 bit integers (not 8 bit) for each coil and operates as expected within a numeric range of, naturally, ± 254 (not ± 255) to vary the current between ± Imax/RMS..

The Takeaway

About half a year after the release of Watterott’s breakout board, the potential of smarter stepper motor drivers piqued the curiosity of the 3D printing community, but not much has happened in terms of implementation. Admittedly, it takes some effort to get them running. If you’re still busy dialing in the temperature on your 3D printer, you surely don’t want to add a few dozen new variables, but if you’re keen on getting the best out of it, the TMC2130 has a lot to offer: low-noise printing, high-speed printing, print interrupt on failure and recovery from lost steps. Because the driver IC is so hackable, it’s clearly intended to be tuned in to accommodate specific applications. Throwing it on a general purpose test bench probably won’t yield meaningful, general purpose results.

I hope you enjoyed taking a look at a smarter-than-usual stepper motor driver, as one of the new frontiers of DIY 3D printing, and as an interesting component with many other applications. If you’re thinking about experimenting with this IC or breakout board in your 3D printer, feel free to try my Marlin fork to get started. If you’re building something entirely different, the underlying Arduino library will help you out. Who else is using this part? I’ll be glad to hear about your ideas, applications, and experiences in the comments!


Filed under: 3d Printer hacks, Hackaday Columns