Interactive SPI

The SPI protocol can be tricky enough to get working especially if you are unsure of the MCU you are using and/or the peripheral.  Logic analyzers can help but can also be expensive.  With the help of the following Energia MSP430G2553 code and a dumb terminal serial application program (on your PC) you can interact live with an SPI peripheral and hopefully come to grips with its operation.

The peripheral can be wired as follows:
LaunchPads-MSP430G2-—-Pins-Maps-13-42

Launchpad                   Peripheral
MOSI------------------------MOSI
MISO------------------------MISO
P1_0------------------------SS (slave select or CE)
GND-------------------------GND
Vcc-------------------------Vcc

Check the peripheral power requirements first and don’t connect a 5V peripheral directly to a 3.3V MSP430

The program presents the user with a simple menu:

Please select from one of the following:
0: SS Low
1: SS High
2: Write byte
3: Read byte

If you choose 0 or 1, SS is raised or lowered as appropriate and the menu recycles. If you choose 2 you see this (I entered the value ‘a9’ (not case sensitive))

Please select from one of the following:
0: SS Low
1: SS High
2: Write byte
3: Read byte
Enter a 2 character hex value: a9
Out : A9
In : 0

If you choose 3 you will see something like this:

Please select from one of the following:
0: SS Low
1: SS High
2: Write byte
3: Read byte
In : 0

The code is shown below. You will probably need to check out which SPI modes and byte ordering suit you. Also, the SPI interface is running at the very low speed of 125kHz. This was deliberate as it reduces the risk of data errors on shaky test leads and may help debugging. You can of course change this. The divider is divided into 16MHz to give an SPI data rate. This is very definitely version 0.1 and changes are likely in the future when I do some real testing.

/*
 * SPI protocol tester using the MSP430G2553 G2 Launchpad
 * This program allows you manage an SPI bus,write and read data
 * using a serial dumb terminal application
 * The program makes use of the Energia Serial and SPI libraries
 * Serial interface : 9600,n,8,1
 * SPI library reference : http://energia.nu/reference/spi/
 * 
 */
#include <SPI.h>

// Will use P1_0 as SS pin as there is a handy LED there on the launchpad
#define SS_Pin  P1_0


int getUserCommand();
int getInteger(String Prompt);
void setup() {
  // put your setup code here, to run once:
  // Default SPI configuration : feel free to change!
  // Set up the SS Pin and make it HIGH initially (low wakes up a slave)
  pinMode(SS_Pin,OUTPUT);  
  digitalWrite(SS_Pin,HIGH);
  SPI.begin();
  SPI.setDataMode(SPI_MODE0); // can choose modes 0,1,2,3
  SPI.setBitOrder(MSBFIRST);  // can be MSBFIRST or LSBFIRST
  SPI.setClockDivider(128);   // assuming a system clock of 16MHz this gives an 
                              // SPI speed of 125kHz - deliberately slow to be more forgiving and
                              // to make signals easier to see with a scope of logic analyser
  // Serial communications to host setup
  Serial.begin(9600);
}
int TXByte,RXByte;
void loop() {
  
  // put your main code here, to run repeatedly: 
  switch (getUserCommand())
  {
    case 0 : {
      // Command 0 : drop SS pin down
      digitalWrite(SS_Pin,LOW);    
      break;
    }
    case 1 : {
      // Command 1 : raise SS pin up
      digitalWrite(SS_Pin,HIGH);
      break;
    }
    case 2 : {
      // Command 2 : send a byte
      TXByte = getInteger("Enter a value to transmit: ");
      RXByte = SPI.transfer(TXByte);
      Serial.print("Out : ");
      Serial.println(TXByte,HEX);
      Serial.print("In : ");
      Serial.println(RXByte,HEX);
      break;
    }
    case 3 : {
      // Command 3 : read a byte (send a dummy byte out)
      RXByte = SPI.transfer(0x00);
      Serial.print("In : ");
      Serial.println(RXByte,HEX);
      break;      
    }
    default : {
      Serial.println("Invalid choice");
    }
  }
  delay(100);
}
int showMenu(String Menu[],int MenuItemCount)
{
  Serial.flush();
  Serial.println("Please select from one of the following:");
  for (int item=0; item < MenuItemCount; item++)
  {
    Serial.print(item);
    Serial.print(": ");
    Serial.println(Menu[item]);
  }
  while(Serial.available()==0);
    
  return Serial.read() - '0'; // assuming a numeric choice is made - convert to decimal from ascii
}

int getUserCommand()
{
  String Menu[4];
  Menu[0]="SS Low";
  Menu[1]="SS High";
  Menu[2]="Write byte";
  Menu[3]="Read byte";
  return showMenu(Menu,4);
  
}
int HexDigitToDecimal(char Digit)
{
  if ( (Digit >= '0') && (Digit <= '9') )
  {
    return Digit - '0';
  }
  Digit = Digit | 32; // enforce lower case
  if ( (Digit >= 'a') && (Digit <= 'f') )
  {
    return Digit - 'a' + 10;
  }
  return 0;
}
int getInteger(String Prompt)
{
  char HexString[3];
  int ReturnValue = 0;
  HexString[2]=0;
  Serial.flush();
  Serial.print("Enter a 2 character hex value: ");
  while(Serial.available()==0);  
  HexString[0]=Serial.read();Serial.print(HexString[0]);
  while(Serial.available()==0);  
  HexString[1]=Serial.read();Serial.println(HexString[1]);
  ReturnValue = HexDigitToDecimal(HexString[0]);
  ReturnValue = ReturnValue << 4;
  ReturnValue += HexDigitToDecimal(HexString[1]);
  return ReturnValue;
}


Let your code do the wiring

SimpleVsComplex

The image above shows two implementations of the same gaming system (ArcadeSlam). The display has a parallel data interface and the left hand version maps this interface to the 8 bits of a single I/O port on the MSP430 MCU.  A simple write to the port data register is sufficient to write a byte to the display.  From a programming perspective this is easy but, as you can see, the wiring is a little complex.

The version on the right makes use of the connections within the breadboard to connect the display data interface to the MSP430.  This greatly simplifies the wiring however it pushes this complexity back into the code.  The wiring looks like this:

Wiring

The display data interface is spread across Ports 1 and 2.  Not only that, the bits are in reverse order.  Reversing bits at run-time represents a performance hit so, a lookup table was generated and the writing of the data bytes goes from this (for the left hand version)


P2OUT = data;

to

P1OUT &=0xc1;
P2OUT &=0xf8;
if (b)
{ // only write out bits if b is non zero to save time
P1OUT |= reverse_bits[ (b >> 3) ] >> 2;
P2OUT |= reverse_bits[ (b & 0x7) ] >> 5;
}

The lookup table reverse_bits was produced using the following python script (only a 5 bit table was necessary)


#!/usr/bin/python
# Output the specified range of numbers with their
# bits reversed as a lookup table suitable for C
print "const uint8_t reverse_bits[]={ \\"
for n in range(32):
print int('{:08b}'.format(n)[::-1], 2),",\\"
print "};"

Performance is not obviously affected by this wiring change and full code can be downloaded over here on github.

More information can also be found on roboslam.com

Low pass filtering using the STML432 Nucleo

This example uses a 4th order Butterworth low pass filter that was designed in GNU Octave.  The sampling rate was set to 200kHz and the cut-off frequency was set to 20kHz.  The filter output at 20kHz is shown below and, as expected, shows an attenuation of 0.7 (approx the square root of 2).

FilterOutput1

Various attempts were made to optimize the performance of the filter.  The execution time was measured by flipping an output bit either side of the filter code.  An oscilloscope trace of this output is below.

FilterTiming1

As can be seen, the execution time is 1.78 microseconds. This is pretty quick given that floating point numbers are being used.  I found that my attempts to manually improve the performance made no significant difference compared to what the compiler’s optimizer could do.  I also found that gcc’s -O2 optimization setting produced a faster filter than -O3.  The filter shuffles data in the input and output delay lines.  This may be considered less  than optimal but, given that the order of the filter is low, it probably would make little difference to use circular buffers (and manage buffer state etc).

Code can be downloaded here on Github and should be easily compiled on Linux/Windows/Mac

Analogue pass-through at 1MHz on the STM32L432 Nucleo board

Update: I previously had measured (incorrectly) a conversion rate of 4MHz – on moving to better instrumentation this proved to be incorrect.  The maximum stable conversion rate comes out just below 2MHz.  This example runs the system at 1MHz.20kHz_at_4MHz

The STM32L432KC Nucleo board is a low cost board (approx €13) in the same form factor as an Arduino Nano.  The onboard CPU is based on an ARM Cortex M4F running at 80MHz.  It features a very fast ADC and 2 DAC output as well as a number of timers, serial interfaces and so on.

I was curious to see how fast the ADC could be read using a timer as a trigger so I put together a simple program that reads an analogue input and writes this value back out to the DAC.  The graph above shows two traces:  the output is green and is overlaid on top of the input (yellow).  The input signal is a 20kHz sine wave (DC shifted to 1.5V).  The system is reading the input signal and updating the output at 1MHz.  An interrupt service routine (ISR) is called at each ADC conversion which consists of the following code:


void ADC_ISR()
{
  // The green LED output is used to measure the execution time of the ISR
  GPIOB_ODR |= BIT3;   // Turn on green led
  ADC1_ISR = BIT3;     // clear ADC interrupt flag
  GPIOB_ODR |= BIT3;   // Toggle green led
  ADCValue = ADC1_DR;  // Read latest value from ADC conversion
  writeDAC(ADCValue);  // Write new output to DAC
  GPIOB_ODR &= ~BIT3;  // Turn off green led
}

The onboard LED is driven high at the beginning of the ISR and low again on exit.  This allows a measurement to be made of CPU usage inside the ISR.  I used an oscilloscope to monitor the behaviour of the LED pin and this is shown in the trace below

1MHzSamplingCPUUsage

As can be seen, the CPU is loaded to around 25%

Source code for this example and others is available over here on Github

Compiling should be pretty straightforward:

(1) Run the build script (batch file) on Linux/Windows/Mac.

(2) Plug the nucleo board in to your computer and it should appear as disk

(3) Copy “main.bin” to this new “disk”

This should program the board and start the program running.

 

Multi-threading on the Tiva C Launchpad

Threads and processes

A process is a running program. Multitasking operating systems (e.g Linux, Windows etc.) run a number of processes simultaneously. Each process has a global (or static) memory area, a stack and code. Processes in multitasking OS’s are protected from one another using a hardware based memory management unit. A Scheduler allocates CPU time to each process. The simplest scheduler is a “round-robin” scheduler which allows each process run for a short time before switching to the next allowing each process a turn on the CPU.

multitasking

Threads are similar to processes in some ways however they share the same global/static data as well as the same code but have separate stacks.

threads1

Threads can be scheduled just like processes and so appear to operate in parallel – this is multi-threading.

threads2

Context switching

Each process or thread switch involves a context change: the current processor state (all of its register contents) must be saved and the processor state for the next thread or process loaded.  The image below illustrates a context change from Thread 1 to Thread 2

context_change

The context change is triggered by a timer interrupt and the ARM Cortex processors have a special timer aimed at just this role : the SysTick timer. In the following example the SysTick timer is configured to interrupt the CPU every millisecond which triggers a context change.

ARM Cortex M0 Exception handling

The following registers are placed on the interrupted thread stack (Process Stack) automatically following an interrupt (such as SysTick)

Address Contents
SP Prior to interrupt ????????
SP + 0x0000001C xPSR
SP + 0x00000018 PC
SP + 0x00000014 LR
SP + 0x00000010 R12
SP + 0x0000000C R3
SP + 0x00000008 R2
SP + 0x00000004 R1
SP + 0x00000000 R0

Why not save all of the registers? It is too slow (your ISR may not be changing all registers).

Why just these ones? R0-R3 typically are used for argument passing and should always be preserved by ISR’s. R12 is used by some compilers in their inner function call glue. The LR may hold a function return address. PC must be remembered so we know where to go back to and xPSR must be remembered for the flags.

For a full context switch, the remaining registers must be placed on the Process Stack also.

Address Contents
SP Prior to interrupt ????????
SP + 0x0000001C xPSR
SP + 0x00000018 PC
SP + 0x00000014 LR
SP + 0x00000010 R12
SP + 0x0000000C R3
SP + 0x00000008 R2
SP + 0x00000004 R1
SP + 0x00000000 R0
SP – 0x00000004 R11
SP – 0x00000008 R10
SP – 0x0000000C R9
SP – 0x00000010 R8
SP – 0x00000014 R7
SP – 0x00000018 R6
SP – 0x0000001C R5
SP – 0x00000020 R4

It is not possible carry this out in the C language so a little inline assembler is needed here to complete the context change.


// Preserve remaining registers on stack of thread that is being suspended (Thread A)
asm(" cpsid i "); // disable interrupts during thread switch
asm(" MRS R0,PSP "); // get Thread A stack pointer
asm(" SUB R0,#32"); // Make room for the other registers : R4-R11 = 8 x 4 = 32 bytes
asm(" STMIA R0! , { R4-R7 } "); // Can only do a multiple store on registers up to R7
asm(" MOV R4,R8 "); // Copy higher registers to lower ones
asm(" MOV R5,R9 ");
asm(" MOV R6,R10 ");
asm(" MOV R7,R11 ");
asm(" STMIA R0! , { R4-R7 } "); // and repeat the multiple register store
// Locate the Thread Control Block (TCB) for Thread A
asm(" LDR R0,=TCB_Size "); // get the size of each TCB
asm(" LDR R0,[R0] ");
asm(" LDR R1,=ThreadIndex "); // Which one is being used right now?
asm(" LDR R1,[R1] ");
asm(" MUL R1,R0,R1 "); // Calculate offset of Thread A TCB from start of TCB array
asm(" LDR R0,=Threads "); // point to start of TCB array
asm(" ADD R1,R0,R1 "); // add offset to get pointer to Thread A TCB
asm(" MRS R0,PSP "); // get Thread A stack pointer
// Save Thread A's stack pointer (adjusted for new registers being pushed
asm(" SUB R0,#32 "); // Adjust for the other registers : R4-R11 = 8 x 4 = 32 bytes
asm(" STR R0,[R1] "); // Save Thread A Stack pointer to the TCB (first entry = Saved stack pointer)

// Update the ThreadIndex
ThreadIndex++;
if (ThreadIndex >= ThreadCount)
  ThreadIndex = 0;

// Locate the Thread Control Block (TCB) for Thread B
asm(" LDR R0,=TCB_Size "); // get the size of each TCB
asm(" LDR R0,[R0] ");
asm(" LDR R1,=ThreadIndex "); // Which one is being used right now?
asm(" LDR R1,[R1] ");
asm(" MUL R1,R0,R1 "); // Calculate offset of Thread A TCB from start of TCB array
asm(" LDR R0,=Threads "); // point to start of TCB array
asm(" ADD R1,R0,R1 "); // add offset to get pointer to Thread B TCB
asm(" LDR R0,[R1] "); // read saved Thread B Stack pointer
asm(" ADD R0,#16 "); // Skip past saved low registers for the moment
asm(" LDMIA R0!,{R4-R7} "); // read saved registers
asm(" MOV R8,R4 "); // Copy higher registers to lower ones
asm(" MOV R9,R5 ");
asm(" MOV R10,R6 ");
asm(" MOV R11,R7 ");
asm(" LDR R0,[R1] "); // read saved Thread B Stack pointer
asm(" LDMIA R0!,{R4-R7} "); // read saved LOW registers
asm(" LDR R0,[R1] "); // read saved Thread B Stack pointer
asm(" ADD R0,#32 "); // re-adjust saved stack pointer
asm(" MSR PSP,R0 "); // write Thread B stack pointer

Threads are managed using a structure called a Thread Control Block which is defined as follows:


typedef struct {
uint32_t *ThreadStack;
void (*ThreadFn )();
uint32_t Attributes;
} ThreadControlBlock;

Implementation

A demonstrator application with three threads was developed for the Tiva C Launchpad.  Each thread flashes an LED on the board at a different rate.  The trickiest part to get right was the initial launching of the thread switching which involved a little bit of stack fiddling.  Code is available over here on Github

9 DOF on the STM32L476 Discovery board

The STM32L476 Discovery board has an LSM303 Accelerometer/Compass IC and an L3GD20 gyroscope attached to the MCU using an SPI bus and some chip select lines.  I wanted to experiment with them with a view to putting together a balancing robot.  Supporting code for the following was needed for this:

  • an SPI interface
  • the LSM303
  • the L3GD20
  • serial communications
  • periodic interrupts to pace data capture

Rather than build a complex Makefile I went with a simple shell script (or batch file if you prefer) with the following commands:

arm-none-eabi-gcc -static -mthumb -g -mcpu=cortex-m4 *.c -T linker_script.ld -o main.elf -nostartfiles
arm-none-eabi-objcopy -g -O binary main.elf main.bin

Note: your PATH environment variable must include the directory where arm-none-eabi-gcc is located.

The resulting main.bin file can then be copied to the virtual disk presented by the mbed interface on the STM32L476 discovery board.  (The program waits for you to press the centre joystick button before starting).`

Serial communications is carried out over the built-in ST-Link USB-Serial emulator so no additional hardware is needed (9600,n,8,1).

Code is available over here