Embedded Developer

So far in The Way of the Register series we have only looked at SPI from a slave device point of view as we have been working towards creating a Netduino GO! module. For every slave device there must be a master, here we will look at configuring the STM8S to operate in SPI master mode.

The project will look at controlling a TLC5940 in order to emulate the work described in the post TLC5940 16 Channel PWM Driver. We could simply bit-bang the data out to the chip but instead we will use the SPI interface to achieve this.

The project breaks down into the following steps:

1. Generate the grey scale clock and blank signals
2. Bit-Bang data out over GPIO pins to create an operational circuit
3. Convert the data transmission to SPI

See the quoted post for a description of how this chip works and for an explanation of the terminology used.

Generating the Grey Scale Clock and Blank Signals

The TLC5940 generated 4,096 grey scale values by using a PWM counter. Once the counter reaches 4096 pulses it stops until it is told to restart. The Blank pulse acts as a restart signal. This project will be controlling LEDs and so will want to continuously keep the counter running. If we did not keep the counter in the TLC5940 running then the LEDs would light for a short while and then simply turn off and remain off.

The greyscale clock is generated by using the Configurable Clock Output (CCO) pin on the STM8S. This pin simply outputs the clock pulses used to drive the STM8S. Reviewing the data sheet we find that the maximum value for the grey scale clock is 30MHz. Using out standard clock initialisation generates a clock with a frequency of 16MHz (approximately). This is well within the tolerances of the TLC5940. To output this we need to make a simple modification to our standard code, name change the line:

CLK_CCOR = 0;   //  Turn off CCO.

to:

CLK_CCOR = 1;   //  Turn on CCO.

The starting point for our application becomes:

#if defined DISCOVERY
    #include <iostm8S105c6.h>
#elif defined PROTOMODULE
    #include <iostm8s103k3.h>
#else
    #include <iostm8s103f3.h>
#endif
#include <intrinsics.h>

//
//  Setup the system clock to run at 16MHz using the internal oscillator.
//
void InitialiseSystemClock()
{
    CLK_ICKR = 0;                       //  Reset the Internal Clock Register.
    CLK_ICKR_HSIEN = 1;                 //  Enable the HSI.
    CLK_ECKR = 0;                       //  Disable the external clock.
    while (CLK_ICKR_HSIRDY == 0);       //  Wait for the HSI to be ready for use.
    CLK_CKDIVR = 0;                     //  Ensure the clocks are running at full speed.
    CLK_PCKENR1 = 0xff;                 //  Enable all peripheral clocks.
    CLK_PCKENR2 = 0xff;                 //  Ditto.
    CLK_CCOR = 1;                       //  Turn on CCO.
    CLK_HSITRIMR = 0;                   //  Turn off any HSIU trimming.
    CLK_SWIMCCR = 0;                    //  Set SWIM to run at clock / 2.
    CLK_SWR = 0xe1;                     //  Use HSI as the clock source.
    CLK_SWCR = 0;                       //  Reset the clock switch control register.
    CLK_SWCR_SWEN = 1;                  //  Enable switching.
    while (CLK_SWCR_SWBSY != 0);        //  Pause while the clock switch is busy.
}

//
//  Main program loop.
//
void main()
{
    //
    //  Initialise the system.
    //
    __disable_interrupt();
    InitialiseSystemClock();
    __enable_interrupt();
    while (1)
    {
        __wait_for_interrupt();
    }
}

Wiring up the STM8S and connecting the scope to PC4 (CCO output pin) gives the following trace on the scope:

CCO On Scope

The trace on the scope has a minimum value of around 680mV and a maximum of 2.48V. In an ideal world this signal should range from 0 to 3.3V (based upon a 3.3V supply). Adding an inverter from a 74HC04 and feeding the signal through one of the gates gives the following trace:

Inverter output on the scope

This is starting to look a lot better. The next task is to create the Blank signal. There are several ways of doing this. The most automatic way of doing this is to generate a very short PWM pulse using one of the timers in the STM8S. One drawback of this method is that it is more difficult to generate a Blank pulse on demand. Instead we will use the interrupt method described in the same article. Whilst not automatic it is still a trivial task to complete. We simply modify the code from the method to load the counters with 4,096. The code for the GPIO port, timer and interrupt becomes:

//
//  Timer 2 Overflow handler.
//
#pragma vector = TIM2_OVR_UIF_vector
__interrupt void TIM2_UPD_OVF_IRQHandler(void)
{
    PD_ODR_ODR4 = 1;
    PD_ODR_ODR4 = 0;
    TIM2_SR1_UIF = 0;       //  Reset the interrupt otherwise it will fire again straight away.
}

//
//  Setup Timer 2 to generate an interrupt every 4096 clock ticks.
//
void SetupTimer2()
{
    TIM2_PSCR = 0x00;       //  Prescaler = 1.
    TIM2_ARRH = 0x10;       //  High byte of 4096.
    TIM2_ARRL = 0x00;       //  Low byte of 4096.
    TIM2_IER_UIE = 1;       //  Turn on the interrupts.
    TIM2_CR1_CEN = 1;       //  Finally enable the timer.
}

//
//  Setup the output ports used to control the TLC5940.
//
void SetupOutputPorts()
{
    PD_ODR = 0;             //  All pins are turned off.
    PD_DDR_DDR4 = 1;        //  Port D, pin 4 is used for the Blank signal.
    PD_CR1_C14 = 1;         //  Port D, pin 4 is Push-Pull
    PD_CR2_C24 = 1;         //  Port D, Pin 4 is generating a pulse under 2 MHz.
}

Hooking up the scope to PD4 gives the following trace:

Blanking pulses

The single pulses are being generated at a frequency of approximately 3.9kHz. A little mental arithmetic dividing the 16MHz clock by 4,096 comes out to about 3,900.

Zooming in on the signal we see:

Single Blanking Pulse

This shows that the signal is 125nS wide. This is acceptable as the minimum pulse width given in the data sheet is 20nS.

So at this point we have the 16MHz grey scale clock signal and the Blank pulse being generated every 4,096 clock pulses.

Connecting the TLC5940

The next task is to connect the STM8S to the TLC5940. You should refer to the article TLC5940 16 Channel PWM Driver for more information on the pins and their meaning. For this exercise we will use the following mapping:

You will note that the serial data and clock are currently connected to PD5 and PD6 respectively. Whilst the eventual aim is to communicate with the TLC5940 via SPI, the initial communication will be using Bit-Banging. We will move on to using SPI once the operation of the circuit has been proven using tested technology.

The first changes we will have to make create some #define statements to make the code a little more readable. We also add some storage space for the grey scale and dot correction data.

//
//  Define which pins on Port D will be used as control signals for the TLC5940.
//
//  BLANK - A pulse from low to high causes the TLC5940 to restart the counter
//  XLAT - A high pulse causes the data to be transferred into the DC or GS registers.
//  VPRG - Determines which registers are being programmed, High = DC, Low = GS.
//
#define PIN_BLANK                   PD_ODR_ODR4
#define PIN_XLAT                    PD_ODR_ODR3
#define PIN_VPRG                    PD_ODR_ODR2
//
//  Bit bang pins.
//
#define PIN_BB_DATA                 PD_ODR_ODR5
#define PIN_BB_CLK                  PD_ODR_ODR6
//
//  Values representing the modes for the VPRG pin.
//
#define PROGRAMME_DC                1
#define PROGRAMME_GS                0

//
//  TLC5940 related definitions.
//
#define TLC_NUMBER                  1
#define TLC_DC_BYTES_PER_CHIP       12
#define TLC_DC_BYTES                TLC_NUMBER * TLC_DC_BYTES_PER_CHIP
#define TLC_GS_BYTES_PER_CHIP       24
#define TLC_GS_BYTES                TLC_NUMBER * TLC_GS_BYTES_PER_CHIP

//
//  Next we need somewhere to hold the data.
//
unsigned char _greyScaleData[TLC_GS_BYTES];
unsigned char _dotCorrectionData[TLC_DC_BYTES];

The Bit-Banging methods should look familiar to anyone who has been reading any of the posts in The Way of the Register series.

//--------------------------------------------------------------------------------
//
//  Bit bang data.
//
//  TLC5940 expects the data to be shifted MSB first.  The data
//  is shifted in on the rising edge of the clock.
//
void BitBang(unsigned char byte)
{
    for (short bit = 7; bit >= 0; bit--)
    {
        if (byte &amp; (1 << bit))
        {
            PIN_BB_DATA = 1;
        }
        else
        {
            PIN_BB_DATA = 0;
        }
        PIN_BB_CLK = 1;
        PIN_BB_CLK = 0;
    }
    PIN_BB_DATA = 0;
}

//--------------------------------------------------------------------------------
//
//  Bit bang a buffer of data.
//
void BitBangBuffer(unsigned char *buffer, int size)
{
    for (int index = 0; index < size; index++)
    {
        BitBang(buffer[index]);
    }
}

Related to the Bit-Banging methods are the two methods which will send the grey scale and dot correction data:

//--------------------------------------------------------------------------------
//
//  Send the grey scale data to the TLC5940.
//
void SendGreyScaleData(unsigned char *buffer, int length)
{
    PIN_VPRG = PROGRAMME_GS;
    BitBangBuffer(buffer, length);
    PulseXLAT();
    PulseBlank();
}

//--------------------------------------------------------------------------------
//
//  Send the dot correction buffer to the TLC5940.
//
void SendDotCorrectionData(unsigned char *buffer, int length)
{
    PIN_VPRG = PROGRAMME_DC;
    BitBangBuffer(buffer, length);
    PulseXLAT();
    PulseBlank();
}

We also need a few methods to make the TLC5940 latch the data and also restart the counters:

//--------------------------------------------------------------------------------
//
//  Pulse the Blank pin in order to make the TLC5940 reload the counters and
//  restart timer.
//
void PulseBlank()
{
    PIN_BLANK = 1;
    PIN_BLANK = 0;
}

//--------------------------------------------------------------------------------
//
//  Pulse the XLAT pin in order to make the TLC5940 transfer the
//  data from the latches into the appropriate registers.
//
void PulseXLAT()
{
    PIN_XLAT = 1;
    PIN_XLAT = 0;
}

Next we need to set the initial condition. For this we set the TLC dot correction off and also turn all of the LEDs off:

//--------------------------------------------------------------------------------
//
//  Initialise the TLC5940.
//
void InitialiseTLC5940()
{
    for (int index = 0; index < TLC_DC_BYTES; index++)
    {
        _dotCorrectionData[index] = 0xff;
    }
    for (int index = 0; index < TLC_GS_BYTES; index++)
    {
        _greyScaleData[index] = 0;
    }
    SendDotCorrectionData(_dotCorrectionData, TLC_DC_BYTES);
    SendGreyScaleData(_greyScaleData, TLC_GS_BYTES);
}

The final support method we need to add is the method which sets the brightness of an LED. The brightness is a 12-bit value (0-4095). This means each LED uses 1.5 bytes for the brightness value. The following methods breaks down the value and ensures that the correct bits are set in the grey scale buffer depending upon which LED is being changed:

//--------------------------------------------------------------------------------
//
//  Set the brightness of an LED.
//
void SetLEDBrightness(int ledNumber, unsigned short brightness)
{
    int offset = (ledNumber >> 1) * 3;
    if (ledNumber & 0x01)
    {
        _greyScaleData[offset + 1] = (unsigned char) (_greyScaleData[offset + 1] & 0xf0) | ((brightness & 0x0f00) >> 8);
        _greyScaleData[offset + 2] = (unsigned char) (brightness & 0xff);
    }
    else
    {
        _greyScaleData[offset] = (unsigned char) ((brightness & 0x0ff0) >> 4) & 0xff;
        _greyScaleData[offset + 1] = (unsigned char) ((brightness & 0x0f) >> 4) | (_greyScaleData[offset + 1] & 0x0f);
    }
}

We should also create a similar method for changing the dot correction value for an LED. This is left as an exercise for the reader as we will not be changing this value in this code.

Proving the concept

If we have connected the TLC5940 correctly and our code works we should be able to connect up some LEDs (common anode) to the TLC5940 and change the brightness under program control.

This main program loop slowly increases the brightness of the LEDs. When they are at full brightness they are turned off and the process starts again:

//--------------------------------------------------------------------------------
//
//  Main program loop.
//
void main()
{
    //
    //  Initialise the system.
    //
    __disable_interrupt();
    InitialiseSystemClock();
    SetupOutputPorts();
    SetupTimer2();
    
    InitialiseTLC5940();
    __enable_interrupt();
    //
    //  Main program loop.
    //
    int brightness = 0;
    int counter = 0;
    while (1)
    {
        __wait_for_interrupt();
        counter++;
        if (counter == 20)
        {
            TIM2_CR1_CEN = 0;
            counter = 0;
            for (int index = 0; index < 16; index++)
            {
                SetLEDBrightness(index, brightness);
            }
            SendGreyScaleData(_greyScaleData, TLC_GS_BYTES);
            brightness++;
            if (brightness == 4096)
            {
                brightness = 0;
            }
            TIM2_CR1_CEN = 1;       //  Finally re-enable the timer.
        }
    }
}

If you connect a scope to the cathode of one of the LEDs you will see that the wave form slowly changes over time. At the start, the LED is fully on and the trace on the scope shows a horizontal line, i.e. a constant voltage. As time moves on and the value in the dot correction buffer changes you start to see a PWM signal similar to the following:

PWM Output On Scope 1

This trace shows the signal when the LEDs are a little brighter:

PWM Output On Scope 2

Having arrived here we now know that the circuit has been connected correctly and that the control logic in the main method works. We can now move on to considering what we need to do in order to use SPI in master mode. The aim will be to simply remove the Bit-Banging methods and replace these with an interrupt driven SPI master algorithm.

SPI Master

So now we have a working circuit we need to look at SPI on the STM8S. Firstly let’s remind ourselves of the serial communication parameters for the TLC5940. This chip reads the data on the leading clock edge (CPHA = 1). We have also set the clock idle state to low (CPOL = 0).

It is also advisable to start off using the lowest clock speed for SPI in order to confirm correct operation of the software and circuit. Lower speed are less likely to be subject to interference.

SPI Registers

You should review the previous articles on SPI communication if you are not already familiar with the SPI registers we have used so far. In this post we will only discuss the new settings required to switch from being a SPI slave device to a SPI master device.

SPI_CR1_BR – Baud Rate Control

The SPI baud rate is determined by the master clock frequency and the value in this register. The divisor used to set the baud rate according to the following table:

The SPI baud rate is calculated as f_master / divisor. So for our master clock speed of 16MHz we get the lowest clock speed of 16,000,000 / 256, or 62,500Hz.

SPI_CR1_MSTR – Master Selection

Setting this bit switches SPI into master mode (see also SPI_CR1_SPE).

Note that the reference for the STM8S also states that this bit (and SPI_CR1_SPE) will only remain set whilst NSS is high. It this therefore essential to connect NSS to V_cc if this device is not being used as a slave device.

Implementing SPI

Using SPI presents us with a small problem, namely the program will have to start to operate in a more asynchronous way. The code presented so far has only one interrupt to be concerned with, namely the timer used to control the Blank signal. Adding SPI to the mix means that we will have to also consider the SPI interrupt as well. It also adds the complication of sending dot correction data followed by grey scale data. This last problem will not be covered here and is left as an exercise for the reader.

The initialisation code merely sets things up for us:

//--------------------------------------------------------------------------------
//
//  Initialise SPI to be SPI master.
//
void SetupSPIAsMaster()
{
    SPI_CR1_SPE = 0;                    //  Disable SPI.
    SPI_CR1_CPOL = 0;                   //  Clock is low when idle.
    SPI_CR1_CPHA = 0;                   //  Capture MSB on first edge.
    SPI_ICR_TXIE = 1;                   //  Enable the SPI TXE interrupt.
    SPI_CR1_BR = 7;                     //  fmaster / 256 (62,500 baud).
    SPI_CR1_MSTR = 1;                   //  Master device.
}

Much of the code should be familiar as it has been used in previous posts discussing SPI slave devices. Not however that we do not enable SPI at this point. We simply set the scene for us to use SPI later.

The SPI data transfers will be controlled by using an interrupt service routine:

//--------------------------------------------------------------------------------
//
//  SPI Interrupt service routine.
//
#pragma vector = SPI_TXE_vector
__interrupt void SPI_IRQHandler(void)
{
    //
    //  Check for an overflow error.
    //
    if (SPI_SR_OVR)
    {
        (void) SPI_DR;                      // These two reads clear the overflow
        (void) SPI_SR;                      // error.
        return;
    }
    if (SPI_SR_TXE)
    {
        //
        //  Check if we have more data to send.
        //
        if (_txBufferIndex == _txBufferSize)
        {
            while (SPI_SR_BSY);
            SPI_CR1_SPE = 0;
            _txBuffer = 0;
            PulseXLAT();
            PulseBlank();
            TIM2_CR1_CEN = 1;
        }
        else
        {
            SPI_DR = _txBuffer[_txBufferIndex++];
        }
    }
}

The main works starts when we have established that the transmit buffer is empty (SPI_SR_TXE is set). If we have more data then we put the byte into the data register (SPI_DR). If we have transmitted all the data we have then we wait for the last byte to complete transmission (SPI_SR_BSY becomes false) before we start to terminate the end of the SPI communication.

In order to send some data we really just need to setup the pointers and counters correctly and then enable SPI. So the SendGreyScaleData method becomes:

//--------------------------------------------------------------------------------
//
//  Send the grey scale data to the TLC5940.
//
void SendGreyScaleData(unsigned char *buffer, int length)
{
    PIN_VPRG = PROGRAMME_GS;
    _txBuffer = buffer;
    _txBufferIndex = 0;
    _txBufferSize = length;
    TIM2_CR1_CEN = 0;
    SPI_CR1_SPE = 1;
}

We also need to have a look at the main program loop as we use the __wait_for_interrupt() method in order to determine when we should start to process the next LED brightness value. We now need to ignore the interrupts when SPI is enabled otherwise the brightness will increase each time the transmit buffer is empty. A crude implementation eliminating the SPI interrupts is:

int brightness = 0;
int counter = 0;
while (1)
{
    __wait_for_interrupt();
    if (!SPI_CR1_SPE)
    {
        counter++;
        if (counter == 20)
        {
            TIM2_CR1_CEN = 0;
            counter = 0;
            for (int index = 0; index < 16; index++)
            {
                SetLEDBrightness(index, brightness);
            }
            SendGreyScaleData(_greyScaleData, TLC_GS_BYTES);
            brightness++;
            if (brightness == 4096)
            {
                brightness = 0;
            }
            TIM2_CR1_CEN = 1;       //  Finally re-enable the timer.
        }
    }
}

Making these changes and running the code shows that the system operated as before.

Increasing the Baud Rate

As noted earlier, the baud rate has been set low in order to reduce the chance of any problems being experienced due to interference. Now we have established that using SPI communication is possible and the circuit works as before we can start to increase the baud rate. Using our 16MHz clock we find we have the following baud rates which are theoretically possible:

A little experimentation is called for. Being ambitious I started with a clock frequency of 1MHz. This resulted in a flickering effect on the LED display. So, 1MHz is too ambitious, let’s start to reduce the frequency. I finally settled on 250 KHz as this allowed the circuit to function as before.

Conclusion

Using SPI master for data transmission was not as difficult as I originally thought. To make this application complete there are a few tasks to follow up on, namely:

1. Receiving data over SPI
2. Create the method to allow setting the dot correction values
3. Transmitting buffers from a queue
4. Minor tidying up of the timer control

The use of SPI here actually increased the time taken (597uS Bit-Banging c.f. 795uS for 250 KHz SPI) to reliably send the grey scale data to the TLC5940. I suspect that the time can be decreased if the circuit was taken from breadboard and put onto a PCB manufactured for the purpose. The breadboard for this circuit currently looks like this:

Bread Board And Flying Leads

As you can see, there is a lot of opportunity for interference with all those flying leads.

While the time taken might have increased, the load on the microcontroller will have decreased as the SPI method is interrupt driven. The actual transmission is off-loaded to the microcontrollers dedicated circuitry.

As usual, the source code for this project is available to download.

Sometimes you have to do something just because you can. Today was no exception. Take one Raspberry Pi board (Linux) and give it a dose of .NET loving…

I can hear people screaming NO! but it’s too late, I’ve done it.

To be honest, I’m not the first person to use .NET on the Raspberry Pi. In fact I used this installation guide to install Mono on the Raspberry Pi.

So what interested me? Simply put, hardware interfacing. Could the .NET hardware classes (i.e. SerialPort) be used on the Raspberry Pi? There was only one way to find out.

Raspberry Pi Setup

The first thing to do was to set up the Raspberry Pi using the instructions in the blog mentioned above. This was not difficult although it was time consuming as it took a while to compile.

When this is complete a simple program should demonstrate if the serial ports are available to us. The following code should show this:

namespace ConsoleApplication1
{
	class Program
	{
		public static void Main(string[] args)
		{
			SerialPort sp = new SerialPort("/dev/ttyAMA0", 115200, Parity.None, 8, StopBits.One);
			sp.Open();
			for (int index = 0; index < 10; index++)
			{
				string result = string.Format("{0} Testing", index);
				sp.Write(result);
			}
			sp.Close();
		}
	}
}

As you can see, this code is not too complex. It merely opens a serial port and then starts to output data to the port. The only tricky part of the job was to locate the name to be used for the serial port. It’s been too long since I used Unix in anger.

Testing the Application

Testing is not too complex as the test application is simply outputting a text message. We can see the output by hooking up PuTTY to a serial port on a computer.

The hardware is also simple to hook up. The FTDI cable I have allows me to connect a 3.3V TTL serial device to my PC over USB. This USB device then appears as a standard serial device on the PC. Three connections are required:

Strictly speaking, we do not have to connect the T_x on the FTDI cable to the R_x on the Raspberry Pi as we will not be transmitting anything from the PC to the Raspberry Pi in this example.

Once the hardware is connected you will need to fire up PuTTY and connect it to the serial port on your PC. Ensure that the settings used match those in the application code above.

Assuming that you now have Mono installed and the application typed in ready to go, type the following commands into a shell on the Raspberry Pi:

mcs serial.cs
mono serial.exe

The first command invokes the Mono C# compiler. The second tells mono to execute the application which has been compiled.

If all has gone well then you will see something like this in the PuTTY window running on the PC:

Serial Output in Putty

Conclusion

I was rather interested to see that the C# skills acquired on the PC could also be used on the Raspberry Pi. The only thing missing on the Pi was a good development environment.

Making the OutputExpander module has been an interesting journey. The original drawings started in August 2012 and then sat on the hard drive for about eight months. Much of the time following the original drawings were spent working out how the STM8S worked. You can find out more in The Way of the Register series (something I will pick up again soon, to my mind there are a few missing topics).

For those who do not know, I’m a software engineer and electronics is a hobby. The prospect of designing a board and using SMD components would have been unthinkable to me two years ago. Today I sit here with my first prototype PCB connected to a commercial board and the output looks reasonably professional – well I’ll let you decide.

Completed Board

So let’s look at what I have learned and also how long the project took.

Lessons Learned

With all projects we should look back and learn from the experience, both good and bad. So here are a few things I have learned over the past few months.

Designing the Board

The original design started life in August 2012. I probably should have taken the plunge and developed the board a little quicker than I did although in truth, I did not get the major requirement of the board, namely GoBus 1.0 really sorted out until late November 2012.

Prototyping

This was probably the simplest bit of the project. I have all of the standard components in my toolbox already and I also have the tools required. This was really a case of getting the system working. The hardest part was getting to grips with the STM8S, a story I have documented in The Way of The Register series of posts.

Schematic

During this part of the design phase I tried several different packages. All of them had strengths and weaknesses. I finally settled on DesignSpark. For me this package had three major strengths:

• It’s free

In fairness it does have a few weaknesses. The most obvious for me was the lack of the ability to add images of any kind to the design. Come on, at version 5 you should have this one!

Nets – I discovered these when producing the final draft of the schematic. These allowed the separation of the nets into logical groups/functions/areas. It made the schematic a lot cleaner.

Schematic to Manufacture

For me this was the where I learned the most. The first thing I learned was that auto-routers are dreadful. They are slow and produce some interesting board layouts. This board is a simple board and yet the auto-router still took a long time to make a half-hearted attempt at routing the board. In the end I did this manually. This was not too much of a problem as the board was simple.

Next, you have to learn to think in three dimensions. You have two layers so use them.

The Netduino modules produced by Secret Labs have nice rounded corners – these are a devil to produce in DesignSpark. I think that the module I produced has one corner which is slightly different from the others.

The cost of prototyping is a lot lower than I thought. Ten boards including shipping costs about £18 and only took 10 days.

I now know what 0403 means. The ‘0’ stands for Ohhh my goodness that’s small. Seriously, the four digits should be split into two and thy give the dimensions of the component. So for a metric component an 0403 part is 0.4 x 0.3 mm – that’s small.

The STM8S part selected has a 0.65mm pitch for the pins. I originally found this a little worrying. Don’t be afraid – they are not that bad.

Get a USB microscope when soldering SMD components. This tools is cheap and allows the examination of joints for shorts. The quality will never be great, mine only runs at 640×480, but a 400x zoom means you can be sure that you have no problems.

Add test points. There came a point when I was making the board and I needed to see the data going through to the 74HC595’s. I did not have a suitable connection and so I had to solder a piece of wire to the board:

Improvised Test Point

A good test point would have made this easier.

How Long Did it Take

The original drawing started in August 2012 and the final board was put together and tested only yesterday. So in elapsed time that’s about nine months. In real working time this broke down as follows:

Something to bear in mind is that no production evaluation or component selection has been conducted as part of this project. It was supposed to be the final item on the list. I am still not sure if this should be taken through to manufacture – time will tell.

Another item to be considered is to achieve the Netduino GO! logo approval. At the time of writing this required the approval of the board by Secret Labs – this activity has not been completed.

Conclusion

Well, that was a hectic few weeks.

Did I enjoy it – YES!

Would I recommend that you try it – YES!

As for me, I’ll be taking a few days off of hardware development and blogging. Love doing it but it can take it’s toll.

I suppose you may be interested in some downloads…

• STM8S Source Code
• Netduino Source Code
• Schematic
• Gerber files – for iTead

If you use any of the code or techniques discussed in this series of posts then please let me know as I’m interested in what other people are doing with this work.

A few days ago I received a package from China, namely my Output Expander prototype boards:

Bubble Wrapped Boards From China

Could not wait to unwrap them:

OutputExpander Bare Boards

Only one thing left to do, start assembling them. As with all projects this will be broken down into steps:

• Add the STM8S microcontroller and test
• Add one 74HC595 shift register and test
• Complete the board and add connectors and of course, test

By using a modular approach it should be easy to detect a problem with the design or assembly.

Component List

The board requires the following components:

Component	Value	Quantity
STM8S103F3	NA	1
IDC socket	1.27″ pitch	1
Sr1, SR2, Sr3, SR4	SOL / SOP16	4
C2	1uF	0403
C1, C3, C4, C5, C6, C7	100nF – 0403	6
Connectors	0.1″	Misc

When these arrive be prepared, they are small!

Adding the Microcontroller

The board will need a microcontroller and some way of programming it. The logical first task is to add the controller, socket and the supporting passive components. Doing this will allow us to programme the controller with the firmware. As a test we can connect the programmed board to the Netduino Go!. If the connections between the board and the Netduino GO! are correct then the blue LED on the socket on the Netduino Go! should light.

If you are attempting to follow this series and you are making your own board then I recommend you browse the net and have a look for videos on soldering SMD components. I found the tutorials on drag soldering really useful.

Out with the soldering iron, a magnifier (it was needed). One thing I noticed was the difference between the 74HC595 pads and the pads for the STM8S. The 74HC595 component used was a built in component whilst the STM8S was a component I had created. The most noticeable difference between the two parts was the size of the pads on the PCB compared to the size of the component. The 74HC595 pads were elongated. These make soldering easier.

STM8S	75HC595

Although the pins on the STM8S are only 0.65mm pitch, soldering is not as difficult as it first appears. A quick first attempt gave the following:



STM8S Mounted On Board

There is only one item of concern and that is the whisker of solder between the fourth and fifth pins down on the right hand side of the image. This was quickly tidied up by dragging the soldering iron between the two pins.

Next task was to add the passives which supported the STM8S leaving the passives for the shift registers for later. This is where you get some idea of the difference between the size of the components vs the size of the tools you are using:



Capacitor and Tools

At this point I realised that an 0403 (metric sizing) component is 0.4mm x 0.3mm and the smallest soldering iron bit I has was about 1.5mm. Not to worry, the pads on the board are a reasonable size, simply tin the pads and then slide the capacitor into the molten solder.

The next job was to add the socket for the GoBus. The sockets are surface mounted 1.27″ pitch IDC sockets. I found the easiest way to add these was to tin one pad and then slide the socket into place. The remaining pads could be soldered by placing the solder at the end of the connector and then applying heat and letting the solder run under the socket. It’s not as difficult as it sounds.

At this point, the microcontroller should be in place with enough supporting hardware to allow it to be programmed. This was achieved by connecting the ST-Link/V2 programmer to the prototype board using the Komodex Breakout Board. The firmware developed in the previous posts was loaded into the development environment and deployed top the microcontroller.



Programming the STM8S

No deployment errors!

A good indication that the microcontroller and the supporting hardware are functioning correctly.

Add a Shift Register

Next step is to add a single shift register and see if we get some output. Soldering the shift registers was a lot simpler than the STM8S as the pin pitch was greater. These could be soldered more conventionally although the pitch was finer than you may be used to if you have only worked with PTH components.

Connecting the module to the Netduino GO! acts as a quick check:



Netduino Go! Connected to OutputExpander

The blue LED lights up – the Netduino GO! recognises the OutputExpander as a valid module.

Adding the single register worked and so the next task is to add the remaining registers and connectors.

But All Was Not Well…

During the assembly and testing process I had managed to accidentally short a few pins on the shift registers. This resulted in no output from the OutputExpander module. Breaking out the scope and the logic analyser proved that something was very wrong. The following trace shows the problem:



Original Output From the OutputExpander

It appears that the latch and clear lines were being triggered at the same time. I was able to establish by disconnecting the module from the circuit that there was not short between the two lines. Something else must be going on. Some further digging into the output from the logic analyser showed that the clear signal was being triggered slightly before the latch signal and that the latch was being released slightly after the clear signal. As a result I would expect no output from the shift registers – this is what I was seeing.

Not wanting to waste money on components I continued to check the circuit but could not find anything else obviously wrong with the soldering or the software.

Only one remaining option. Try a putting together a new board. Back to step one.

A New Board

Building the new board was a lot quicker than the first. Following the same procedure (one step at a time and test all the way) produced a new board:

Before testing the board there was a final modification to make. This time to the software. The board has the outputs labelled from left to right with the lower bits being to the right of the board. The prototype module had the shift registers ordered from right to left. A quick change to the C code on the STM8S soon resolved this problem:

//--------------------------------------------------------------------------------
//
//  GO! function 2 - Output the specified bytes to the shift registers.
//  Tx buffer.
//
void SetShiftRegisters()
{
    for (int index = 0; index < _numberOfShiftRegisters; index++)
    {
        _registers[index] = _rxBuffer[5 - index];
    }
    OutputData();
    NotifyGOBoard();
}

A small bug had also been noticed in the clock configuration method. The code stated that CCO was turned off but the code actually turned it on. The code should read:

//--------------------------------------------------------------------------------
//
//  Setup the system clock to run at 16MHz using the internal oscillator.
//
void InitialiseSystemClock()
{
    CLK_ICKR = 0;                       //  Reset the Internal Clock Register.
    CLK_ICKR_HSIEN = 1;                 //  Enable the HSI.
    CLK_ECKR = 0;                       //  Disable the external clock.
    while (CLK_ICKR_HSIRDY == 0);       //  Wait for the HSI to be ready for use.
    CLK_CKDIVR = 0;                     //  Ensure the clocks are running at full speed.
    CLK_PCKENR1 = 0xff;                 //  Enable all peripheral clocks.
    CLK_PCKENR2 = 0xff;                 //  Ditto.
    CLK_CCOR = 0;                       //  Turn off CCO.
    CLK_HSITRIMR = 0;                   //  Turn off any HSIU trimming.
    CLK_SWIMCCR = 0;                    //  Set SWIM to run at clock / 2.
    CLK_SWR = 0xe1;                     //  Use HSI as the clock source.
    CLK_SWCR = 0;                       //  Reset the clock switch control register.
    CLK_SWCR_SWEN = 1;                  //  Enable switching.
    while (CLK_SWCR_SWBSY != 0);        //  Pause while the clock switch is busy.
}

Testing

The step by step testing process had shown that a single shift register worked, now to prove that four worked. Now it was time to add some more and connect some LEDs:



Netduino Go OutputExpander and Some LEDs

And here’s a video of it working:

Conclusion

Assembly was not as difficult as it first appears even considering the small size of the components. In fact the STM8S was programmed first time.

One piece of equipment I did find invaluable was a cheap USB microscope. These don’t give a high resolution image but they do allow you to zoom in on the board and check for problems.

One final post left – time to reflect on the process.

This series of posts follows the steps required in order to make your own Netduino GO! module. To do this we are using a simple idea, an Output Expander (yes, it seems to have a name now) for the Netduino GO!. Using only a hand full of simple components, the Output Expander will add 32 digital outputs to the Netduino Go!.

The Story So Far…

The module has so far completed the following steps:

• Concept and prototype on breadboard
• Basic drivers
• PCB design and layout
• Generation of PCB manufacturing files

The first batch of 10 boards were ordered on 20th April 2013 using iTeads prototyping service. I am expecting the manufacturing process to take about one week with a further two weeks for shipping as I used the cheap, slow courier service.

In the meantime we still have the breadboard prototype to work with. This may only have two shift registers attached to it but by carefully parametrising the code we should be able to develop a driver which only needs to be recompiled when the final hardware becomes available.

Features

The driver for this module should be relatively simple as we are merely setting outputs from the shift registers to be either on or off. We shall start with the following list of desired features:

• Default initialisation to set up for one board
• Set the bytes for the digital outputs
• Array like access for the bits
• Configurable latching mode (automatic or manual)
• Clear the output
• Turn outputs from the registers on or off

The only remaining task is to decide where the features are to be implemented. We have two options:

• On the Netduino GO! in C#
• On the STM8S in C

Some of the code is better left on the Netduino GO! with the base features (setting all of the register values etc) being implemented on the STM8S.

Initialisation

Initialisation should allow for the use of cascaded boards. If you look back at the schematic you will notice a connector called CascadeConn. This connector allows the addition of a simpler board as an expansion module. This board only need to supply additional shift registers leaving the first, main board supplying the logic to communicate with the Netduino GO!. The concept is that if you want 64 outputs then you would have a single Output Expander module with a single, cheaper daughter board.

In order to support the addition of the daughter board the initialisation will need to support the specification of the number of shift registers in the cascade.

In supporting the cascading of these boards we will also need to provide some sensible default values. The basic case is a single board which contains four shift registers.

We should also consider a maximum value for the number of shift registers. In this case I am going to set this to 12 for two reasons:

• Power considerations – all of the power for the shift registers is being provided by the Netduino GO!
• Data packet size – the data packets used in the GoBus are fixed size. Keeping the number of shift registers to a value where only one data packet is required simplifies the communication between the Netduino GO! and the module as a single packet can be used for all messages.

In order to facilitate this we will need code on both the Netduino GO! and the module.

Module Code

The code on the module should allow for the number of shift registers to be set up (assuming a default value of four registers) and then clear the registers. The default code should be called when the module is re-initialised.

//--------------------------------------------------------------------------------
//
//  GO! function 3 - Set up the module.
//
void SetUp()
{
    U8 n = _rxBuffer[2];
    if ((n < 0) && (n < MAX_REGISTERS))
    {
        _numberOfShiftRegisters = n;
        free(_registers);
        _registers = (U8 *) malloc(n);
        ClearRegisters();
        OutputData();
        NotifyGOBoard();
    }
}

Netduino GO! Code

The initialisation code on the Netduino GO! will assume that the startup code on the module will initialise itself to four shift registers. This will reduce the communications overhead between the Netduino GO! and the module. The following when added to the Initialise method should set up the module driver on the Netduino GO! for a variable number of shift registers in the sequence:

//
//  Next, set up the space to store the data.
//
_shiftRegisterData = new byte[numberOfShiftRegisters];
for (int index = 0; index < numberOfShiftRegisters; index++)
{
    _shiftRegisterData[index] = 0;
}
LatchMode = LatchingMode.Automatic;
if (numberOfShiftRegisters != 4)
{
    _writeFrameBuffer[0] = GO_BUS10_COMMAND_RESPONSE;
    _writeFrameBuffer[1] = CMD_SETUP;
    _writeFrameBuffer[2] = (byte) (numberOfShiftRegisters & 0xff);
    WriteDataToModule("Initialise: cannot setup the OutputExpander module");
}

Set Outputs

Setting the outputs is simply a case of sending a number of bytes, one for each shift register to the module.

/// <summary>
/// Set the shift registers using the values in the byte array.
/// "/summary>
/// "param name="registers">Bytes containing the shift register values.</param>
public void Set(byte[] registers)
{
    if (registers.Length != _shiftRegisterData.Length)
    {
        throw new ArgumentException("registers: length mismatch");
    }
    for (int index = 0; index < registers.Length; index++)
    {
        _shiftRegisterData[index] = registers[index];
    }
    Latch();
}

The module code needs a slight adjustment to transfer the correct number of incoming bytes to the register store:

//--------------------------------------------------------------------------------
//
//  GO! function 2 - Output the specified bytes to the shift registers.
//  Tx buffer.
//
void SetShiftRegisters()
{
    for (int index = 0; index < _numberOfShiftRegisters; index++)
    {
        _registers[index] = _rxBuffer[2 + index];
    }
    OutputData();
    NotifyGOBoard();
}

Array of Bits

From a software point of view, a shift register is nothing more than an array of boolean values. Internally it makes sense for the driver to allow this abstraction and use the indexing operator to set a single bit at a time. The code for this operator looks something like this:

/// <summary>
/// Overload the index operator to allow the user to get/set a particular 
/// bit in the shift register.
/// </summary>
/// <param name="bit">Bit number to get/set.</param>
/// <returns>Value in the specified bit.</returns>
public bool this[int bit]
{
    get
    {
        if ((bit >= 0) && (bit < (_shiftRegisterData.Length * 8)))
        {
            int register = bit >> 3;
            byte mask = (byte) (bit & 0x07);
            return ((_shiftRegisterData[register] & mask) == 1);
        }
        throw new IndexOutOfRangeException("OutputExpander: Bit index out of range.");
    }
    set
    {
        if ((bit >= 0) && (bit < (_shiftRegisterData.Length * 8)))
        {
            int register = bit >> 3;
            byte mask = (byte) ((1 << (bit & 0x07)) & 0xff);
            if (value)
            {
                _shiftRegisterData[register] |= mask;
            }
            else
            {
                mask = (byte) ~mask;
                _shiftRegisterData[register] &= mask;
            }
            if (LatchMode == LatchingMode.Automatic)
            {
                Latch();
            }
        }
        else
        {
            throw new IndexOutOfRangeException("OutputExpander: Bit index out of range.");
        }
    }
}

Adding the above code allows the programmer to use constructs such as:

OutputExpander outputs = new OutputExpander();
outputs[2] = true;

instead of the more obscure:

OutputExpander outputs = new OutputExpander();
byte[] data = new byte[4];
data[0] = 0x04;
SetOutputs(data);

Not only is the former example more elegant but it is also more concise.

Clear All Registers

This method will simply clear the shift registers and set the outputs to 0;

/// <summary>
/// This method calls the ClearRegister method on the GO! module and then waits for the
/// module to indicate that it has received and executed the command.
/// </summary>
public void Clear()
{
    _writeFrameBuffer[0] = GO_BUS10_COMMAND_RESPONSE;
    _writeFrameBuffer[1] = CMD_CLEAR_REGISTERS;
    for (int index = 0; index < _shiftRegisterData.Length; index++)
    {
        _shiftRegisterData[index] = 0;
    }
    WriteDataToModule("Clear cannot communicate with the Output Expander module");
}

Latch Mode

The introduction of the array indexing operators does introduce on complication, namely that we cannot set all of the outputs to a specified value at the same time without delaying the latching of the registers. Consider the following case:

OutputExpander outputs = new OutputExpander();
outputs[2] = true;
outputs[3] = true;

In this case we would set bit 2 of the lower shift register followed by bit 3 of the same shift register. Because of the speed of .NETMF there would be a slight delay between the two outputs of the shift register being set high. In order to allow for this we introduce the ability to delay the latching of the data from the internal shift register into the output register.

/// <summary>
/// Determine when the data should be sent to the module.
/// </summary>
public enum LatchingMode
{
    /// <summary>
    /// Automtically send the data to the module as soon as there are any changes.
    /// </summary>
    Automatic,
    /// <summary>
    /// Manually latch the data.
    /// </summary>
    Manual
}

/// <summary>
/// Backing variable for the LatchMode property.
/// </summary>
private LatchingMode _latchMode;

/// <summary>
/// Determine how the data will be send to the module.  The default is to 
/// automatically send data as soon as there are any changes.
/// </summary>
public LatchingMode LatchMode
{
    get { return (_latchMode); }
    set { _latchMode = value; }
}

The initialisation of the class would also need to be modified in order to set the mode to automatic:

LatchMode = LatchingMode.Automatic;

The most lightweight method of using the LatchMode is to simply not send the data to the shift registers until the mode is either reset or until the controlling program explicitly latches the data. The Set method will therefore need some adjustment to take into account the two modes:

/// <summary>
/// Set the shift registers using the values in the byte array.
/// </summary>
/// <param name="registers">Bytes containing the shift register values.</param>
public void Set(byte[] registers)
{
    if (registers.Length != _shiftRegisterData.Length)
    {
        throw new ArgumentException("registers: length mismatch");
    }
    for (int index = 0; index < registers.Length; index++)
    {
        _shiftRegisterData[index] = registers[index];
    }
    if (LatchMode == LatchingMode.Automatic)
    {
        Latch();
    }
}

Latch Operation

The introduction of the LatchMode means that we also need to allow for the data to be latched into the shift registers.

/// <summary>
/// Call the Set command on the module to set the outputs of the shift registers.
/// </summary>
private void Set()
{
    _writeFrameBuffer[0] = GO_BUS10_COMMAND_RESPONSE;
    _writeFrameBuffer[1] = CMD_SET_REGISTERS;
    for (int index = 0; index < _shiftRegisterData.Length; index++)
    {
        _writeFrameBuffer[2 + index] = _shiftRegisterData[index];
    }
    WriteDataToModule("Latch cannot communicate with the Output Expander module");
}

The above method simply sends the data to the module.

Testing

We can perform some simple testing of the software while the prototype boards are being made by using the breadboard test environment build in the first post. This board only has two shift registers but it should be possible to test the majority of functionality using this board.

In the previous posts we have cycled through the bits one at a time either from 0 to 15 or down from 15 to 0. In this example we will perform some counting and use the LEDs to display the number in binary. Our test code becomes:

output = new OutputExpander(2);
short cycleCount = 0;
byte[] registers = new byte[2];
output.LatchMode = OutputExpander.LatchingMode.Manual;
while (true)
{
    Debug.Print("Cycle: " + ++cycleCount);
    short mask = 1;
    output.Clear();
    for (int index = 0; index <= 15; index++)
    {
        if ((cycleCount & mask) != 0)
        {
            output[index] = true;
        }
        mask <<= 1;
    }
    output.Latch();
    Thread.Sleep(200);
}

Deploying this application should result in the Netduino GO! counting up from 0 and the binary representation of cycleCount being output on the shift registers. The following video shows this in action:

Conclusion

The above minor modifications to the STM8S module code and the Netduino GO! driver has added the following functionality:

• Default initialisation to set up for one board
• Set the bytes for the digital outputs
• Array like access for the bits
• Configurable latching mode (automatic or manual)
• Clear the output
• Turn outputs from the registers on or off

The code has been carefully written so that we should only need to change two parameters when the final PCBs arrive in order to change the drivers from two shift registers to four shift registers.

A quick test has shown that the main functionality appears to be working on the breadboard prototype as demonstrated by the above video. The prototype PCBs have completed manufacture and are currently with the Hong Kong postal service (as of 27th April 2013). Delivery should take another 7-10 days so there is plenty of time to complete the test suite.

In the previous posts a prototype output expander module was put together on breadboard and connected to the Netduino Go!. A small module driver was developed for the Netduino GO! and the STM8S. This video shows the basic module working:

The next stage in the process is to make the final design decisions and produce a prototype PCB.

Design Criteria

The criteria as defined in the first post stated that the hardware should provide at least 16 digital outputs using low cost components. The breadboard prototype produced shows that we can certainly control 16 digital outputs using the common shift register.

One possible extension is to use more shift registers than were used in the breadboard prototype. This will produce a module with more outputs. A quick look at a local suppliers web site shows that these can obtained for a relatively modest cost. Adding a further two shift registers will take the outputs from 16 to 32.

The prototype shows which pin is being activated by a LED. The LED circuit includes a resistor and MOSFET/Transistor to allow the power for the LED to be provided by an external power supply rather than the 74HC595. Doing this allows the use of LEDs which would exceed the maximum power rating of the 74HC595. Although it should be possible to include these components (resistor and MOSFET) in the final design, they will be omitted in order to reduce the production costs.

In summary:

1. Use the STM8S as it is low cost and powerful enough for this task
2. 32 digital output (4 shift registers)
3. Use 0.1" connectors to the board can be used in breadboard prototyping

PCB Design Software

Over the past few years I have spent a fair amount of time looking for the right PCB design software. I have looked at the following:

• Eagle PCB
• KiCad
• DesignSpark

There are also several other packages available but these seem to be the three main packages offering free version of their software. Being a hobbyist price is a critical factor in the choice of software to use.

Eagle PCB is available as a free personal edition. The software and licence are both restricted in some way. I have also found the interface to be exceedingly difficult to use. As a long time user of Windows applications the interface in Eagle PCB is counter-intuitive.

KiCad is free and looks to be well supported but I found the interface difficult to use.

DesignSpark looks to be well supported and has gone through several revisions, one major revision in the time in which I have been using the software.

Of the three packages I have settled on using DesignSpark for the following reasons:

• The interface is the most "standard Windows" like of the three packages
• Free licence with no restrictions
• Eagle parts can be imported into the library
• Additional DesignSpark libraries are also available including one for Sparkfun’s components and boards

DesignSpark is not restricted to producing just the PCB design file but can also render a 3D image of the final PCB. This allows you to visualise how the final board will look. The interface allows the board to be rotated and viewed from an infinite number of points of view.

Schematic

DesignSpark is fairly intuitive to use and there are a number of tutorials and how to guides available and so I will not go into too much detail regarding using the package. There is one tip I would like to give and that is use nets.

Using nets allows the design file to be simplified greatly. You can break the design down into a number of logical components. Consider our design, this breaks down into the following three sections:

1. STM8S and connector for the GoBus
2. Shift registers
3. Connectors for the digital outputs

Now before I discovered how the nets feature works I would have dropped the components for the STM8S and connector on the schematic and then wired them together.

Next I would have dropped the first of the shift registers onto the schematic and started to connect this to the STM8S. I would have then repeated this with the next shift register (connecting it to the first) and so on. The end result would have been a schematic which was difficult to read due to the number of connecting wires.

This can be simplified by the use of nets and input/output connectors/pins. So lets have a look at how this appears for the first of out logical blocks, the STM8S and the GoBus connector:






STM8S Showing Pin Labels

Looking at the above image you should note that the STM8S has two different connections, an un-named wire which goes off to another part of the circuit (i.e. pins PD5, PD6 etc.) and connections which are terminated with a name (i.e. PD3 is connected to something called SRClockOut).

The connections which go off of the image are standard interconnects between the various components on the board. These interconnects have been restricted to connecting components in our logical function block (STM8S and GoBus connector). This includes any components required to support the STM8S such as capacitors etc.

The second group of pins which go to named connections go to either an input or output pin. The names represent the function of the signal; so SRClockOut is the Shift Register Clock Output from the STM8S. If we look at the first shift register you will see that it has an input pin SR1ClockIn (Shift Register 1 Clock Input):



Shift Register Showing Net Names

We have two pins which have two different names but if we were to look at their properties we would find that they have both been connected to the same net, SRClock Shift Register Clock. By doing this the software knows that the two pins are in fact connected.

This process has been repeated and a number of nets have been created, some have only two pins on them (the data output from the STM8S – SRData – is only connected to the input of shift register 1), others have several (the SRClock net connected the STM8S clock output pin to the clock input of all of the shift registers).

Using nets simplifies the schematic and makes it easier to read. Using a standard naming convention for the input and output pins means that you should always be clear on which pins should be connected. It does add a new task to the design process. With a number of the pins named rather than connected you will need to verify that the pins are connected correctly.

For a small schematic like this one the process of checking the connections is relatively simple. The software allows a component (or net) to be selected from a list of all of the components/nets in the circuit. A part/net is then highlighted on the schematic once it has been selected:






Net Selection

The above diagram shows that the net SR4DataOut has been selected; see the blue highlighted name in the list at the right of the image. The green connections on the schematic show which pins are connected to the SR4DataOut net. Checking is then a case of repeating the process for each of the nets and noting which connections are highlighted.

The final schematic for the module looks like this:



Schematic

A PDF version is also available as this may be difficult to read.

Conclusion

Now that we have the schematic we can translate this board into a PCB and from there we can get a 3D view of the board. Here is a sneak preview of the board in 3D:



Output Expander – 3D View

In the next post we shall have a look at taking the schematic and laying out the PCB.

In the previous post a prototype for the multiple digital outputs was developed and implemented on breadboard. The STM8S code was independent and demonstrated that two chained 74HC595 shift registers could be controlled by bit-banging data through GPIO pins on the STM8S.

In this post we will continue the development of the module by merging the code from the previous post with the STM8S module code from the post STM8S SPI Slave (Part 3) – Making a Go Module. We will develop both the STM8S code and the NETMF driver to implement the following functionality:

• Clear the shift registers
• Set the outputs of the two shift registers

These two functions are already in the standalone version of the STM8S code and it should be a relatively simple exercise to merge this functionality with the Netduino Go Bus communication protocol.

STM8S Application

The first task to be completed for the STM8S code is to create a new project and add the code from the post STM8S SPI Slave (Part 3) – Making a Go Module. The two functions should then be removed and the two new ones we are implementing should be added.

Start a New Project

Start a new STM8S project and save the project into a new directory. Now take the source code in main.c from the project STM8S SPI Slave (Part 3) – Making a Go Module and paste over the code in you new project. Ensure that the project is targeting the correct microcontroller and save the project.

Alternatively, copy the project code from STM8S SPI Slave (Part 3) – Making a Go Module into a new directory.

Remove Dummy Functionality

The project code in the STM8S module post contained a couple of dummy functions illustrating the concept of module functionality and communications. This code should be removed and the application stripped down to the basics required for SPI communication with the Netduino Go!.

Add New Functionality

The final task is to add the code which implements the new functionality as defined for the basic module (i.e. clear the shift registers and set the outputs of the shift registers).

The first thing which is required is to add the #define statements to support the bit-banging:

//--------------------------------------------------------------------------------
//
//  Pins which we will be using for output the data to the shift registers.
//
#define SR_CLOCK            PD_ODR_ODR3
#define SR_DATA             PD_ODR_ODR2
#define SR_CLEAR            PC_ODR_ODR3
#define SR_OUTPUT_ENABLE    PC_ODR_ODR4
#define SR_LATCH            PD_ODR_ODR4

The InitialisePorts method will also need to be modified in order to ensure that the ports above are all setup correctly. We need these to be output ports capable of running at up to 10MHz. The code becomes:

//--------------------------------------------------------------------------------
//
//  Initialise the ports.
//
void InitialisePorts()
{
    //
    //  Initialise Port D.
    //
    PD_ODR = 0;             //  All pins are turned off.
    PD_DDR = 0xff;          //  All pins are outputs.
    PD_CR1 = 0xff;          //  Push-Pull outputs.
    PD_CR2 = 0xff;          //  Output speeds up to 10 MHz.
    //
    //  Initialise Port C.
    //
    PC_ODR = 0;             //  Turn port C outputs off.
    PC_DDR = 0x18;          //  PC3 & PC4 initialised for output.
    PC_CR1 = 0x18;          //  PC3 & PC4 Push-Pull outputs
    PC_CR2 = 0x18;          //  PC3 & PC4 can run up to 10MHz.
    //
    //  Initialise the CS port for input and set up the interrupt behaviour.
    //
#if defined(DISCOVERY)
    PB_ODR = 0;             //  Turn the outputs off.
    PB_DDR = 0;             //  All pins are inputs.
    PB_CR1 = 0xff;          //  All inputs have pull-ups enabled.
    PB_CR2 = 0xff;          //  Interrupts enabled on all pins.
    EXTI_CR1_PBIS = 2;      //  Port B interrupt on falling edge (initially).
#else
    PA_ODR = 0;             //  Turn the outputs off.
    PA_DDR = 0;             //  All pins are inputs.
    PA_CR1 = 0xff;          //  All inputs have pull-ups enabled.
    PA_CR2 = 0xff;          //  Interrupts enabled on all pins.
    EXTI_CR1_PAIS = 2;      //  Port A interrupt on falling edge (initially).
#endif
}

The application will also need some storage space for the data in the shift registers:

//--------------------------------------------------------------------------------
//
//  Number of registers in the chain.
//
#define NUMBER_OF_REGISTERS     2
//
//  Data area holding the values in the register.
//
U8 _registers[NUMBER_OF_REGISTERS];             //  Data in the shift registers.

Note that the variable used to store the data in the registers has been converted into a static array instead of a variable sized array using malloc.

The function table needs to be modified to hold the references to the methods which will implement the module functionality:

//
//  Forward function declarations for the function table.
//
void SetShiftRegisters();
void ClearShiftRegisters();
//
//  Table of pointers to functions which implement the specified commands.
//
FunctionTableEntry _functionTable[] = { { 0x01, ClearShiftRegisters }, { 0x02, SetShiftRegisters } };

The next step is to add the methods which will implement the functionality:

//--------------------------------------------------------------------------------
//
//  Clear the shift registers.
//
void ClearRegisters()
{
    for (U8 index = 0; index < NUMBER_OF_REGISTERS; index++)
    {
        _registers[index] = 0;
    }
}

//--------------------------------------------------------------------------------
//
//  GO! Function 1 - Clear the shift registers.
//
void ClearShiftRegisters()
{
    ClearRegisters();
    OutputData();
    NotifyGOBoard();
}

Note that the functionality has been implemented using two methods, the first ClearRegisters is the internal implementation. This is independent of the Netduino Go! communications and allows the functionality to be called in say the initialisation code of the module. The second method, ClearShiftRegisters is the method which is called by the code as determined by the Netduino Go! driver code. This has the additional output and notification methods which actually sets the data in the shift registers and sends a signal back to the Netduino Go! to let the board know that the request has been received and processed. Note that this split is not necessary but is a design decision taken for this particular module.

The code should be ready to compile and deploy to the STM8S.

Netduino Go! Driver

At this point we should have one half of the communications channel ready to test. The second stage is to implement the driver on the Netduino Go!. As our starting point, we will copy the code from the BasicGoModule class in the post STM8S SPI Slave (Part 3) – Making a Go Module into a new class OutputExpander. We will modify this to add create new functionality to clear and set the shift registers.

Command Constants

The first this we need to do is to remove the old command constants and add two new ones:

/// <summary>
/// Command number for the ClearShiftRegister command.
/// </summary>
private const byte CMD_CLEAR_REGISTERS = 1;

/// <summary
/// Command number for the SetRegister command.
/// </summary>
private const byte CMD_SET_REGISTERS = 2;

Modify the Supporting Methods

The existing code in the module can be optimised for this module. To do this rename the WriteDataToModule method to WriteDataToSPIBus. Now add the following code:

/// <summary>
/// Write the data in the _writeFrameBuffer to the module.  Make several
/// attempts to write the data before throwing an exception.
/// </summary>
/// <param name="exceptionMessage">Exception message to the used in the constructor if the write attempt fails.</param>
private void WriteDataToModule(string exceptionMessage)
{
    int retriesLeft = 10;
    bool responseReceived = false;

    WriteDataToSPIBus();
    while (!responseReceived && (retriesLeft > 0))
    {
        //
        //  We have written the data to the module so wait for a maximum 
        //  of 5 milliseconds to see if the module responds with some 
        //  data for us.
        //
        responseReceived = _irqPortInterruptEvent.WaitOne(5, false);
        if ((responseReceived) && (_readFrameBuffer[1] == GO_BUS10_COMMAND_RESPONSE))
        {
            //
            //  Assume good result, it is up to the calling method to determine if
            //  the command has been executed correctly.
            //
            return;
        }
        else
        {
            //
            //  No response within the 5ms so lets make another attempt.
            //
            retriesLeft--;
            if (retriesLeft > 0)
            {
                WriteDataToSPIBus();
            }
        }
    }
    throw new Exception(exceptionMessage);
}

By making this change we are also making the assumption that the signal back from the module (via the interrupt) is always indicating success. This is a decision which is appropriate for this module but may not be appropriate for other applications/modules.

Add New Functionality

Before adding new functionality it is necessary to remove the existing AddFive functionality from the BasicGoModule.

Out OutputExpander module will provide the application with two basic functions:

• Clear – clear the shift registers setting the output to 0
• Set – Set the values in the shift registers to those specified by the parameters

This functionality is provided by the following code:

/// <summary>
/// This method calls the ClearRegister method on the GO! module and then waits for the
/// module to indicate that it has received and executed the command.
/// </summary>
public void Clear()
{
    _writeFrameBuffer[0] = GO_BUS10_COMMAND_RESPONSE;
    _writeFrameBuffer[1] = CMD_CLEAR_REGISTERS;
    WriteDataToModule("Clear cannot communicate with the Output Expander module");
}

/// <summary>
/// Set the shift registers using the values in the byte array.
/// </summary>
/// <param name="registers">Bytes containing the shift register values.</param>
public void Set(byte[] registers)
{
    if (registers.Length != 2)
    {
        throw new ArgumentException("registers: length should be 2 bytes");
    }
    _writeFrameBuffer[0] = GO_BUS10_COMMAND_RESPONSE;
    _writeFrameBuffer[1] = CMD_SET_REGISTERS;
    for (int index = 0; index < registers.Length; index++)
    {
        _writeFrameBuffer[2 + index] = registers[index];
    }
    WriteDataToModule("Clear cannot communicate with the Output Expander module");
}

Testing

At this point we should be able to add some code the Main method of the application to test the module. We will perform something similar to the code in the previous post, we will move a single LED but this time we will illuminate the LED starting a position 15 and working down to 0. This should ensure that we have deployed new code to both the module and the Netduino Go! board.

public static void Main()
{
    OutputExpander output = new OutputExpander();
    int cycleCount = 0;
    byte[] registers = new byte[2];
    while (true)
    {
        Debug.Print("Cycle: " + ++cycleCount);
        for (int index = 15; index >= 0; index--)
        {
            output.Clear();
            if (index < 8)
            {
                registers[0] = (byte) (1 << index);
                registers[1] = 0;
            }
            else
            {
                registers[0] = 0;
                registers[1] = (byte) (1 << (index - 8));
            }
            output.Set(registers);
            Thread.Sleep(200);
        }
    }
}

The test breadboard and Netduino Go! were connected through the Komodex GoBus Breakout Module. This allows the Netduino Go! and the ST-Link/2 programmer to be connected to the breadboard:

Output Expander Connected to Netdunio GO!

And here is a video of this working:

You can just see the Netduino Go! in the bottom left corner of the video.

Conclusion

At this point the basic concept has be proven and the application can perform the following:

1. STM8S code can control one or more LEDs
2. STM8S can be controlled using SPI
3. Netduino Go! can pass instructions to the STM8S

One thing that did become apparent during this part of the process was that both the STM8S code and the Netduino code can be improved by improved grouping the functionality. For instance, in order to add the methods to the function table in the STM8S code I had to jump around the source code a little.

The next step in the process is to look at the hardware and start to prepare for prototype PCB manufacture.

This series of posts will examine the activities required to take the concept of a module for the Netduino GO! through to production.

Project Definition

A simple project will allow the series of posts to concentrate on the principles required for the production of a Netduino GO! module without being too distracted by the functional aspects of the module. The following project definition should meet this brief:

• Provide 16 or more digital outputs
• Work with the Netduino GO!
• Low cost manufacture
• Use simple tried and tested components/techniques

Meeting the Objectives

As you can no doubt see, this is a reasonably simple project and with the exception of cost, we should have no major problems reaching the objectives. The most obvious solution to this problem is to look at using 74HC595 serial to parallel chips. These are cheap components and the techniques needed to solve this type of problem are tried and tested. The project definition looks like the counting example which is discussed in the Counting Using 74HC595 Shift Registers post with the addition of the Netduino Go! functionality.

Project Plan

Initial assessment of the project indicates that the following steps are required in order to take the project from concept to manufacture:

1. Build a hardware prototype using the STM8S103F3 to control two shift registers. This will have some form of visual output to prove that we can control the digital lines (probably some LEDs)
2. Write the software for the STM8S which will control the output of the 74HC595 chips
3. Generate the schematic for the board
4. Layout a prototype board and send to manufacture
5. Write the software for the Netduino GO! while waiting for the manufactured boards to turn up
6. Assemble a prototype on one of the prototype boards
7. Conclusion

The first post in the series will build the breadboard prototype and start to control the LEDs using the STM8S.

The STM32 contains several clocks each of which can be turned on or off when not used. In this post we will examine the clocks available and how they can be configured. To do this we will modify the simple GPIO example to change the clock parameters. Any changes to the system clock speed will increase or decrease the period of the delay. The results can be seen using an oscilloscope. We will look at the following topics:

• Clock Sources
• System and Peripheral Clocks
• Microcontroller Clock Output
• Configuring the Clock

At the end of this post you should be able to configure the system clock and introduce a known delay subject to the tolerances of the system clocks.

Clock Sources

The STM32 Reference Manual (RM0090) states that there are five clocks available:

• High Speed Internal (HSI) oscillator
• High Speed External (HSE) oscillator
• Phase Locked Loop (PLL)
• Low Speed Internal (LSI) clock
• Low Speed External (LSE) clock

As already noted, each clock can be turned on/off as required. Turning an unused clock off reduces the power consumption of the microcontroller.

The first three clocks are used to drive the system clock for the microcontroller. The final two are low speed clocks and are primarily used to drive the watchdogs.

HSI

After reset, the STM32 enables the HSI oscillator. This has a relatively low accuracy, only 1%, but is suitable for most applications. Using the HSI oscillator eliminates the need for an external clock in the final circuit design. On the STM32F4, the HSI oscillator has a clock speed of 16 MHz.

HSE

The STM32 can operate using an external clock circuit. It is possible to design an external clock to run with a greater accuracy than the internal HSI clock enabling finer control of the operating parameters of the final circuit. The exact specification of the external clock frequency varies but is typically 4-16 MHz.

The STM32F4 Discovery Board has a built in external oscillator circuit fitted with a 8 MHz crystal.

PLL

The PLL is used to multiply it’s input clock source by a factor varying between 2 to 16. The input of the PLL is one of HSI, HSE or HSE/2.

It is important to note that the configuration of the PLL cannot be changed once it has been enabled.

LSI

The LSI is a low power clock used for the watchdog timers.

LSE

The LSE is powered by an external 32.768 KHz clock. This provides a method of providing a low speed accurate clock for the real time clock.

System and Peripheral Clocks

Once the clock source has been selected it is necessary to configure the internal system and peripheral clocks. The internal clocks are:

• System Clock
• Advanced High Performance Bus (AHB)
• Low speed Advanced Peripheral Bus (APB1)
• High speed Advanced Peripheral Bus (APB2)

Each of these clocks can be scaled using prescalers. A fuller picture of the types of clocks and their relationships can be found in the STM32 Reference Manual (RM0090).

System Clock

The system clock is used to determine the speed at which instructions are executed and has a maximum speed of 168MHz.

Advanced High Performance Bus (AHB)

Derived from the system clock, this bus has a maximum speed of 168MHz.

Low speed Advanced Peripheral Bus (APB1)

Derived from AHB, this bus has a maximum speed of 42MHz.

High speed Advanced Peripheral Bus (APB2)

Derived from AHB, this clock has a maximum frequency of 84MHz.

Non-System Clock Peripherals

A number of the peripheral clocks are not derived from the system clock but have their own independent source.

USB OTG FS, Random Number Generator and SDIO Clock

These clocks are all driven by an independent output of the PLL. The maximum speed for these peripherals is 48MHz.

I2S

The I2S peripherals have their own internal clock (PLLI2S). Alternatively, this can be derived by an independent external clock source.

USB OTG HS and Ethernet Clock

These clocks are derived from an external source.

Microcontroller Clock Output

The STM32 provides the ability to output up to two clocks to the external circuit using the Microcontroller Clock Output (MCO) pins. Both of the outputs can have a prescaler applied. This can be in the range 1 to 5 inclusive.

Microcontroller Clock Output 1 (MCO1) is output on pin PA8. MCO1 output can be one of HSI, LSE, HSE or PLL scaled appropriately.

Microcontroller Clock Output 2 (MCO2) is output on pin PC9. MCO2 output can be one of HSE, PLL, System Clock or PLLI2S scaled appropriately.

The clock output is configured by selecting the appropriate alternative function for the GPIO port and is restricted to 100MHz.

Configuring the Clock

The number of features on the STM32F4 analysis and discussion of the registers and their function large. Luckily ST have come to the rescue and provided a configuration tool to help.

ST Clock Configuration Tool in Wizard Mode

This Excel spreadsheet can be run in both Expert and Wizard mode making it suitable for a wide range of users.

ST Clock Configuration Tool in Expert Mode

The grey number in the above illustration have been calculated whilst the numbers in black can be changed by the users.

The output from this spreadsheet is a source file (system_stm32f4xx.c) which can be added to your project. This file, system_stm32f4xx.c, contains three methods which can be called to setup the clocks on the microcontroller:

• SystemInit – Setup the system including the clocks
• SystemCoreClockUpdate – Setup the SystemCoreClock variable (clock frequency)
• SetSysClock – Setup the system clock (called from SystemInit)

We can see the effect of calling SystemInit by using the code from the simple GPIO example to the following:

#include "stm32f4xx_rcc.h"
#include "stm32f4xx_gpio.h"
#include "system_stm32f4xx.h"

int main()
{
    //
    //  Uncomment this line to see the effect of
    //  the clock change.
    //
	//SystemInit();
    //
    //  Initialise the peripheral clock.
    //
    RCC->AHB1ENR |= RCC_AHB1Periph_GPIOD;
    //
    //  Initialise the GPIO port.
    //
    GPIOD->MODER |= GPIO_Mode_OUT;
    GPIOD->OSPEEDR |= GPIO_Speed_100MHz;
    GPIOD->OTYPER |= GPIO_OType_PP;
    GPIOD->PUPDR |= GPIO_PuPd_NOPULL;
    //
    //  Toggle Port D, pin 0 indefinitely.
    //
    while (1)
    {
        GPIOD->BSRRL = GPIO_Pin_0;
        GPIOD->BSRRH = GPIO_Pin_0;
    }
}

If we run the code above with the SystemInit line commented out and hook an oscilloscope up to pin 0 of port D we get a square wave output with a frequency of about 1.5MHz.

STM32F4 Output (1.46MHz)

Now uncomment the SystemInit line of code and re-run the example. This time we see that the output signal has a frequency of nearly 4.9MHz.

STM32F4 Output (5.75MHz)

Turning the configuration tool into expert mode and tweaking a few of the parameters allows the frequency of the output signal to go to 8.77MHz.

STM32F4 Output (8.77MHz)

Timings are based upon the optimisation level being set to no optimisation. Turning the optimisation on to the maximum speed allows an output signal of nearly 20MHz to be generated.

Conclusion

The STM32 contains a variety of clock sources and configuration using the registers within the microcontroller would be a complex task. The Clock Configuration tool eases the configuration of the internal and peripheral clocks generating the source code for system_stm32f4xx.c automatically.

The Clock Configuration tool and the associated guide can be downloaded from ST’s web site. The tool and document are in the following files:

• stsm-stm32091 Clock Configuration Tool
• AN3988 Clock Configuration Tool User Guide

A few weeks ago I published two articles on development environments for the STM32F4 Doscovery Board. Since then I have become aware of CooCox IDE. This is a small development environment for a number of Microcontrollers including some of the STM32 range of processors.

The IDE promises to be free, unrestricted and simple offering the ability to develop high-quality software solutions in a timely and cost effective manner.

Too good to be true?

Let’s find out.

Web Site

I must admit, it took me a long time to take this tool seriously. I found the web site in the very early days in my search for a reasonable cost development environment for the STM32F4 Discovery board. The look and feel of the site left me a little cold. The design looks like it was put together in the early 1990s.

My only hope was that the tools were better.

Clicking through to the CoIDE page promised to show a video instead I found myself presented with a large amount of white space where the video should be. Maybe this works with IE or Firefox but not with Chrome. If you do see the white spave please take the time to scroll to the bottom of the page. There you will find tables detailing the supported chips/boards along with some download links.

On the whole this web site left me feeling nervous and I must admit that the first time I came across this tool I dismissed it purely based upon the look and feel of the web site.

So what made me come back, a comment made by a colleague, i.e. it worked and was simple to use.

Software

In order to use this IDE to develop applications for your board you also need to download the ARM GCC 4.7 toolchain. This is the same toolchain I used when setting us Eclipse to work with the STM32F4 Discovery board (see Setting up GCC for the STM32F4 Discovery Board). I was lucky enough to have this toolchain already installed making one less task to perform.

Downloading the IDE is simple although you do need to register on the site.

Installation was painless and the initial setup was also simple.

I’m starting to get over the feeling of unease I experienced when I first visited the web site.

Starting the IDE present you with a very familiar look and feel:

New Project

It feels like Eclipse.

Creating a Project

Let’s see how good the environment is by taking the code from STM32 – Simple GPIO and Some Flashing LEDs post.

Creating a new project is simple, just head over to the Project->New Project menu item. Doing this will present you with a wizard which will take you through the steps required to select your board and create an empty project.

Some hints:

• When creating a new project you need to untick the Use default path option if you want your code to be put anywhere other than under the installation directory.
• The select board or chip does not have an option for the STM32F4 Discovery Board (or any for the Discovery boards for that matter). To use this board select Chip and then locate the STM32F407VG chip.

Upon completion of the wizard you will be presented with the component selection tab in the IDE:

Component Selection

This tab is really powerful and makes it easy for the new comer to this board to get off the ground. In the previous example I used two sets of definitions; Reset and clock control and GPIO. Ticking the RCC option in the component tab automatically adds the headers and source files needed to use the RCC library to the project. It also adds any dependencies, in this case the CMSIS core system startup files. After selecting the RCC and GPIO components the IDE looks as follows:

Initial Component Selection

Opening file explorer on the PC showed that the directory structure had been modified and the files put in place. So what happens if you untick and option? Removing components removes the files from the project and it also removes them from file store as well.

Now this feature may seem simple but for the beginner it is really powerful. It ensures that all of the source and header files required are added to the project. This is something which whilst not difficult can trip up the beginner. It is also a little tedious to do with IAR and Eclipse.

One downside of the component selection tab is that it does not easily allow you to remove the source code for the components you have added without also removing the header files as well. I am following the principle of The Way of the Register and not using the Standard Peripheral Library. This requires the header files but not the source code for the library itself. We can only hope the linker is smart enough to remove unused code from the final object executable.

Building the project is as simple as clicking on the Build icon (or using the Project->Build menu item). A few seconds later I have the default applicaiton and the library files compiled.

Debugging

An IDE is really nothing without a reasonable debugger. I remember from the work I did setting up Eclipse (see Setting up GCC for the STM32F4 Discovery Board) that Eclipse needed the OpenOCD library in order to be able to debug an application using ST-Link. There is no mention of OpenOCD when using CooCox. The only additional software required for debugging appears to be the ST-Link USB driver. Again, I already have this installed but you may need to visit ST’s web site for the latest version.

In order to debug we need some code so let’s borrow the code from the Simple GPIO example:

#include "stm32f4xx_rcc.h"
#include "stm32f4xx_gpio.h"

int main()
{
    //
    //  Initialise the peripheral clock.
    //
    RCC->AHB1ENR |= RCC_AHB1Periph_GPIOD;
    //
    //  Initialise the GPIO port.
    //
    GPIOD->MODER |= GPIO_Mode_OUT;
    GPIOD->OSPEEDR |= GPIO_Speed_25MHz;
    GPIOD->OTYPER |= GPIO_OType_PP;
    GPIOD->PUPDR |= GPIO_PuPd_NOPULL;
    //
    //  Toggle Port D, pin 0 indefinitely.
    //
    while (1)
    {
        GPIOD->BSRRL = GPIO_Pin_0;
        GPIOD->BSRRH = GPIO_Pin_0;
    }
}

Starting the debugger (Ctrl-F5) changes the IDE context and presents the debugging interface with the application stopped at the first line of code:

CooCox debugging aspect

Warning – the above image is large

Pressing F5 runs the application to the next breakpoint. As I have not set any breakpoints this should run the application indefinitely and indeed it does. Hooking up the scope to pin 0 of port D shows the following output:

Scope output

Single stepping is achieved using F10 and F11. So re-running the application and pressing F10 should single step over the current line of code and indeed it does. The registers window also highlights the ARM registers that were changed by executing the last instruction:

Register View

One thing that I have found missing which I really appreciate with IAR is the ability to view the state of the peripheral registers (as opposed to the microcontroller registers). This is a really useful feature as it means you do not have to locate the registers in memory and decode them yourself.

Conclusion

CooCox IDE is a simple way to get into working with the STM32F4 Discovery board. The power and simplicity of the tool is only let down by the dated look of the web site.

A cracking tool; well worth looking at.