STM32 Bit-Banged 32-bit Communication: Multi-Slave Protocol Issues

[Puparia]

Hello,

I’m building a custom communication protocol using STM32 (F1 series) as a relay-only device, sitting between an ESP32F103C6T6A and multiple slave ESP32F103C6T6As.

The ESP handles all frame dispatching and reception. The STM32 "master" is just an intermediary to ensure matching internal clock timing between STM32s (since ESP32's timing is too loose). I can’t use external crystals or change pin assignments.

The system uses bit-banged GPIO for communication:

• Master TX (to Slaves): PA11 (HIGH idle)

• Master RX (from Slaves): PB10 (LOW idle)

• Slave RX: PA9 / Slave TX: PA10

• Timings: 20ms LOW front, 40ms HIGH sync, 7ms per bit (32 bits), 40ms end

The ESP sends 32-bit frames (e.g. 0xF1F1F1F1, 0xABABABAB) to the STM32 master via UART. The STM then dispatches them bit-banged to the slaves and listens for a bit-banged 32-bit response. That response is passed back to the ESP.

Each slave listens on PA9, and only responds if the 24 LSBs match its own ID. The ESP performs a manual DISCOVER loop, incrementing the ID one by one until no slave replies. There is no automatic discovery based on line state.

Issue: Slaves always receive the data correctly, but the master sometimes gets the wrong reply (previous frame) or times out. All delays (pre-bit sampling, post-reception delays) are implemented.

E.g: 
ESP sent: 

0xF1F1F1F1

Master recv/Sent to esclaves: 

0xF1F1F1F1

Slaves recv/sent: 

0xF1F1F1F1
RAW: 00001110000011100000111000001110

Master recv: 

RAW: 00001110000011100000111000001110
0x70707070

Finally, ESP recv: 

RAW: 00001110000011100000111000001110
Recv: 0x0E0E0E0E

Constraints:

• No external clock

• GPIO-only communication

• Cannot change pin mapping

Questions:

• How to ensure proper TX/RX timing between STM32s without a shared clock?

• Is there a better method than delay-based bit-bang for stable GPIO communication?

• Have you implemented similar "bit-forwarding relays" with STM32s or other MCUs?

Any insights from others who built custom GPIO protocols would be amazing. Thanks!

Master code: 

#include "main.h"
#include <string.h>
#include <stdio.h>

UART_HandleTypeDef huart1;
UART_HandleTypeDef huart2;
TIM_HandleTypeDef htim2;

void delay_us(uint16_t us);

#define MASTER_TX_PIN GPIO_PIN_11
#define MASTER_TX_PORT GPIOA
#define MASTER_RX_PIN GPIO_PIN_10
#define MASTER_RX_PORT GPIOB

void send_32bit_frame(uint32_t frame);
uint32_t receive_32bit_frame(void);

void MX_GPIO_Init(void);
void MX_USART1_UART_Init(void);
void MX_USART2_UART_Init(void);
void MX_TIM2_Init(void);

uint8_t uart_rx[4];

int main(void)
{
  HAL_Init();
  MX_GPIO_Init();
  MX_USART1_UART_Init();
  MX_USART2_UART_Init();
  MX_TIM2_Init();

  char *msg = "MASTER READY\r\n";
  HAL_UART_Transmit(&huart2, (uint8_t*)msg, strlen(msg), HAL_MAX_DELAY);

  while (1)
  {
    HAL_UART_Receive(&huart1, uart_rx, 4, HAL_MAX_DELAY);
    __HAL_UART_FLUSH_DRREGISTER(&huart1);

    uint32_t tx = (uart_rx[0] << 24) | (uart_rx[1] << 16) | (uart_rx[2] << 8) | uart_rx[3];

    send_32bit_frame(tx);
    uint32_t rx = receive_32bit_frame();

    char dbg[64];
    sprintf(dbg, "Sent: 0x%08lX | Recv: 0x%08lX\r\n", tx, rx);
    HAL_UART_Transmit(&huart2, (uint8_t*)dbg, strlen(dbg), HAL_MAX_DELAY);

    uint8_t resp_buf[4];
    resp_buf[0] = (rx >> 24) & 0xFF;
    resp_buf[1] = (rx >> 16) & 0xFF;
    resp_buf[2] = (rx >> 8) & 0xFF;
    resp_buf[3] = rx & 0xFF;

    HAL_UART_Transmit(&huart1, resp_buf, 4, HAL_MAX_DELAY);
    HAL_Delay(5);
  }
}

void send_32bit_frame(uint32_t frame)
{
  HAL_GPIO_WritePin(MASTER_TX_PORT, MASTER_TX_PIN, GPIO_PIN_RESET);
  delay_us(20000);
  HAL_GPIO_WritePin(MASTER_TX_PORT, MASTER_TX_PIN, GPIO_PIN_SET);
  delay_us(40000);

  for (int i = 31; i >= 0; i--) {
    HAL_GPIO_WritePin(MASTER_TX_PORT, MASTER_TX_PIN, (frame >> i) & 1 ? GPIO_PIN_SET : GPIO_PIN_RESET);
    delay_us(7000);
  }

  HAL_GPIO_WritePin(MASTER_TX_PORT, MASTER_TX_PIN, GPIO_PIN_RESET);
  delay_us(40000);
  HAL_GPIO_WritePin(MASTER_TX_PORT, MASTER_TX_PIN, GPIO_PIN_SET);
}

uint8_t reverse_byte(uint8_t b) {
  b = (b & 0xF0) >> 4 | (b & 0x0F) << 4;
  b = (b & 0xCC) >> 2 | (b & 0x33) << 2;
  b = (b & 0xAA) >> 1 | (b & 0x55) << 1;
  return b;
}

uint32_t reverse_bits_per_byte(uint32_t val) {
  return (reverse_byte((val >> 24) & 0xFF) << 24) |
         (reverse_byte((val >> 16) & 0xFF) << 16) |
         (reverse_byte((val >> 8)  & 0xFF) << 8)  |
         (reverse_byte(val & 0xFF));
}

uint32_t receive_32bit_frame(void)
{
  char dbg[33];

  while (HAL_GPIO_ReadPin(MASTER_RX_PORT, MASTER_RX_PIN) == GPIO_PIN_SET);
  delay_us(20000);
  while (HAL_GPIO_ReadPin(MASTER_RX_PORT, MASTER_RX_PIN) == GPIO_PIN_RESET);
  delay_us(40000);
  delay_us(3500);

  uint32_t result = 0;

  for (int i = 0; i < 32; i++) {
    result <<= 1; 
    GPIO_PinState bit = HAL_GPIO_ReadPin(MASTER_RX_PORT, MASTER_RX_PIN);
    result |= (bit == GPIO_PIN_SET ? 1 : 0);
    dbg[i] = (bit == GPIO_PIN_SET ? '1' : '0');
    delay_us(7000);
  }

  dbg[32] = '\0';
  HAL_UART_Transmit(&huart2, (uint8_t*)"MASTER RX BITS: ", 17, HAL_MAX_DELAY);
  HAL_UART_Transmit(&huart2, (uint8_t*)dbg, 32, HAL_MAX_DELAY);
  HAL_UART_Transmit(&huart2, (uint8_t*)"\r\n", 2, HAL_MAX_DELAY);

  char hex[32];
  sprintf(hex, "Master value: 0x%08lX\r\n", result);
  HAL_UART_Transmit(&huart2, (uint8_t*)hex, strlen(hex), HAL_MAX_DELAY);

  while (HAL_GPIO_ReadPin(MASTER_RX_PORT, MASTER_RX_PIN) == GPIO_PIN_SET);
  delay_us(40000);
  return result;
}

void delay_us(uint16_t us)
{
  __HAL_TIM_SET_COUNTER(&htim2, 0);
  HAL_TIM_Base_Start(&htim2);
  while (__HAL_TIM_GET_COUNTER(&htim2) < us);
  HAL_TIM_Base_Stop(&htim2);
}

 
Slave's code:

#include "main.h"
#include <string.h>
#include <stdio.h>

UART_HandleTypeDef huart2;
TIM_HandleTypeDef htim2;

void delay_us(uint16_t us);

#define SLAVE_RX_PIN GPIO_PIN_9
#define SLAVE_RX_PORT GPIOA
#define SLAVE_TX_PIN GPIO_PIN_10
#define SLAVE_TX_PORT GPIOA

uint32_t wait_and_read_frame(void);
void send_32bit_frame(uint32_t frame);
void MX_GPIO_Init(void);
void MX_USART2_UART_Init(void);
void MX_TIM2_Init(void);

int main(void)
{
  HAL_Init();
  MX_GPIO_Init();
  MX_USART2_UART_Init();
  MX_TIM2_Init();

  char *msg = "SLAVE READY\r\n";
  HAL_UART_Transmit(&huart2, (uint8_t*)msg, strlen(msg), HAL_MAX_DELAY);

  while (1)
  {
    uint32_t received = wait_and_read_frame();

    char dbg[64];
    sprintf(dbg, "Recv: 0x%08lX\r\n", received);
    HAL_UART_Transmit(&huart2, (uint8_t*)dbg, strlen(dbg), HAL_MAX_DELAY);
    delay_us(10000);
    send_32bit_frame(~received);
  }
}

uint32_t wait_and_read_frame(void)
{
  while (HAL_GPIO_ReadPin(SLAVE_RX_PORT, SLAVE_RX_PIN) == GPIO_PIN_SET);
  delay_us(20000);

  while (HAL_GPIO_ReadPin(SLAVE_RX_PORT, SLAVE_RX_PIN) == GPIO_PIN_RESET);
  delay_us(40000);

  delay_us(3500);

  uint32_t frame = 0;
  for (int i = 0; i < 32; i++)
  {
    frame <<= 1;
    frame |= (HAL_GPIO_ReadPin(SLAVE_RX_PORT, SLAVE_RX_PIN) == GPIO_PIN_SET) ? 1 : 0;
    delay_us(7000);
  }

  while (HAL_GPIO_ReadPin(SLAVE_RX_PORT, SLAVE_RX_PIN) == GPIO_PIN_SET);
  delay_us(40000);

  return frame;
}

void send_32bit_frame(uint32_t frame)
{
  char dbg[33];

  HAL_GPIO_WritePin(SLAVE_TX_PORT, SLAVE_TX_PIN, GPIO_PIN_RESET);
  delay_us(20000);
  HAL_GPIO_WritePin(SLAVE_TX_PORT, SLAVE_TX_PIN, GPIO_PIN_SET);
  delay_us(40000);

  for (int i = 31; i >= 0; i--) // MSB first
  {
    GPIO_PinState bit = (frame >> i) & 1 ? GPIO_PIN_SET : GPIO_PIN_RESET;
    HAL_GPIO_WritePin(SLAVE_TX_PORT, SLAVE_TX_PIN, bit);
    dbg[31 - i] = (bit == GPIO_PIN_SET ? '1' : '0');
    delay_us(7000);
  }

  dbg[32] = '\0';
  HAL_UART_Transmit(&huart2, (uint8_t*)dbg, 32, HAL_MAX_DELAY);
  HAL_UART_Transmit(&huart2, (uint8_t*)"\n", 1, HAL_MAX_DELAY);

  HAL_GPIO_WritePin(SLAVE_TX_PORT, SLAVE_TX_PIN, GPIO_PIN_RESET);
  delay_us(40000);
}

void delay_us(uint16_t us)
{
  __HAL_TIM_SET_COUNTER(&htim2, 0);
  HAL_TIM_Base_Start(&htim2);
  while (__HAL_TIM_GET_COUNTER(&htim2) < us);
  HAL_TIM_Base_Stop(&htim2);
}

[Tesla DeLorean]

Probably want to consider something that samples on reception interval, and you send some preamble or synchronization pattern, so the response isn't missed and not tied to hard to specific timing.

Your presentation of the data/raw is a bit confusing and the pattern look prone to shift in the time domain.

Perhaps use a scope or logic analyzer to visualize what's happening vs desired.

Have start bit different than idle state.

Be able to resync if shifted.

Perhaps pick patterns which are self clocking or have frequent transitions. Say 8b10b type encodings.

Perhaps time edge receptions rather then use delays.

Perhaps use 1KHz timing interrupt to drive transmission state machine.

[TDK]

This sounds a lot like UART with extra steps.

How accurate is delay_us? How much clock drift is allowed before things start to break? Can't see a way of developing a new protocol without a logic analyzer. Slave could be waiting for pin states and verifying delay is acceptable rather than doing a simple delay and hoping.

[Puparia]

Oh thank you! Well, I’ll try all of this tomorrow, because honestly I’m starting to get really fed up with it.

At times I managed to get something working, but there were always small errors. I wanted to switch to 64 bits before going further, but in the meantime I rebooted my PC and I have no idea what I had written to make it work in 32 bits. And unsurprisingly, with 64 bits, the delays and errors were massive, so I wanted to go back to 32 bits.

It’s frustrating because I’m limited by my physical design, so the GPIOs and using the internal oscillator is mandatory.

And yeah, because the slave PCBs have already been manufactured, and there’s no going back.

They’re wired with PA9 and PA10 for RX/TX, and no external crystals are planned.

I can still make changes on the master side, but overall it wouldn’t really help. And I can’t afford to order new PCBs for the slaves — this project has already cost me way too much in components, MCUs, prototypes of all

kinds...

[Puparia]

Well, the thing is, I’m physically constrained… the PCBs have already been made, and surprise surprise, I messed something up in the wiring. The master and slaves use a physical Ethernet connection (for the twisted pairs and the robustness of the connector).

But I wired the TX line into the 5V pair and… the RX into the 12V pair!

And on top of that, my slaves can be several meters apart from each other, so UART is basically a no-go...

Especially since I don’t have external crystals on the slaves, so the clock frequency is really low and way too imprecise for USART — especially over long distances and with such crappy physical connections between them and the master.

That’s why I moved toward bit-banging, which gives me more flexibility and better control over voltage modulation reliability, even across long distances.

And yeah, I don’t have a logic analyzer — which is a huge problem for debugging this kind of prototype. I plan on buying one eventually, but for now I just don’t have the budget, so I have to do without.

But right now I’m really tearing my hair out — sometimes I manage to get something working, but the moment I add a new feature to make the system more complete, it breaks everything and I spend all my time fixing stuff all over the place.

Especially since I’m really a beginner in C. I’m solid in a bunch of other languages, but C really isn’t my strong suit.