i2c scanner not working correctly

[sanchosk]

I am trying to run the i2c detect script from Arduino from http://playground.arduino.cc/Main/I2cScanner
I modified the setup and started with
Wire.pins(2, 0);
Wire.setClock(100000);
Unfortunately, the code returns error == 0 for all addresses.
Am I doing something wrong? Did I not understood something?

[wizard23]

I also have a problem with I2C communication when I build form current source.

I have an I2C accelerometer which I can sucessfully read out with the precompiled linux release.

But it does not work anymore in the current version built from source even when I call Wire.begin instead of the depricated Wire.begin.

I will gladly assist in any debugging if that helps.

[sanchosk]

Well, after a short digging within the source code I found the problem.
Current endTransmission call returns number of sent bytes instead of the ACK status.
I modified a single line within the i2c.cpp file - removing
return size;
replacing it with
return ack;
within the i2c_master_write_to function.
All of the sudden, the i2c scanner works OK :)

[igrr]

@wizard23 can you capture the interaction using a logic analyzer or an oscilloscope?

[wizard23]

I measured with a oscilloscope and it seems that in the new version the data pin remains floating while the clock pin seems fine.

In the old version and with the same oscilloscope both cloc and data seem fine.

I'm not 100% sure if it's really floating or stays high all the time with a lot of noise.

[igrr]

Can you please share the picture you see on the scope? Do you have a pull-up resistor on the SDA line?

[wizard23]

i've reduced the code to a minimum and done some more research with an oscilloscope. The floating pin was a false alert. I forgot the ground connection so it was floating. Yes I have external 4.7k in addition to the internal pullups.

the first image is from the current (not working for me) version. Some notes: I had to use the i2c low level functions instead of the Wire lib because Wire checks the return codes and it chokes on some return value and the stops transmitting so you dont see much on the scope there. My guess is that it somehow generates one more clock low to high transition and that my MMA8452 accelerometer does not like that. Other i2c devices might accept it.:

the below image is from the precompiled 1.6.1 version. Here the last bits (that come from the accelerometer) have the correct value of 0x2A (in the top image its 0 which is wrong) :

I hacked together a version of the new i2c library that emulates the send bit style of the old lib in the new library and it works for me :) I should have made a picture of that too...I hope I find time tomorrow and test it again. I'm not 100% sure if the code is correct I2C and Wire still does not work for me. but it wortks for me so I copied it here. Sorry I am not yet experienced with pull requests but I'll try that tomorrow after more carefull testing.

changes in core_esp8266_si2c.c:

static bool twi_write_start(void) {
  SCL_HIGH();
  SDA_HIGH();
  if (SDA_READ() == 0) return false;
  twi_delay(twi_dcount);
  SDA_LOW();
  twi_delay(twi_dcount);

  // NEW: start condition
  SCL_LOW();
  twi_delay(twi_dcount);

  return true;
}

static bool twi_write_stop(void){
  unsigned int i = 0;

  // NEW: commented out next 2 lines
  //SCL_LOW();
  //SDA_LOW();

  twi_delay(twi_dcount);
  SCL_HIGH();
  while (SCL_READ() == 0 && (i++) < TWI_CLOCK_STRETCH);// Clock
stretching (up to 100us)
  twi_delay(twi_dcount);
  SDA_HIGH();
  twi_delay(twi_dcount);

  return true;
}

static bool twi_write_bit(bool bit) {
  unsigned int i = 0;

  // NEW: set clock low below
  //SCL_LOW();

  if (bit) SDA_HIGH();
  else SDA_LOW();
  twi_delay(twi_dcount+1);
  SCL_HIGH();
  while (SCL_READ() == 0 && (i++) < TWI_CLOCK_STRETCH);// Clock
stretching (up to 100us)

  // NEW: set clock low
  SCL_LOW();

  twi_delay(twi_dcount+1);
  return true;
}

static bool twi_read_bit(void) {
  unsigned int i = 0;

  // NEW: set cock low below
  // SCL_LOW();

  SDA_HIGH();
  twi_delay(twi_dcount+1);
  SCL_HIGH();
  while (SCL_READ() == 0 && (i++) < TWI_CLOCK_STRETCH);// Clock
stretching (up to 100us)
  bool bit = SDA_READ();
  twi_delay(twi_dcount);

  // NEW: set clock low
  SCL_LOW();

  return bit;
}

[tytower]

Have tried Wire.begin(sda,scl); using pins 4 and 5 and other combinations of begin functions including //Wire.setClock(40000); and many other speeds to no avail
Cant get I2c working at all in 1.6.4 IDE

have to check your edits with my I2C things here, because some lines you removed deal with shitty slaves that do not want to release the data line on time.
Scope image looks great BTW.
Can you post images of both with and without your changes so I understand what is actually different?

I'm currently running some I2C things on my ESP without any modifications to the Wire lib
tested 100000 and 400000 speeds

#303
the fix

I guess some devices need clock cycles to finish an operation. Had caught this in the read method, but never got it in write.
What your changes did was exactly that... add a low clock on repeated start. But messed up the clock timing much. And broke other parts of the protocol.

[tytower]

So I wish to update so I can get this working . How do I do that now with the json setup? I have update at startup ticked . Will that do it automatically or do I have to do something else?

I'm not positive... I think that @igrr needs to update the package as well so you get the latest code. Then you need an Arduino 1.6+ info on how to add the json is on the forum as well :)

[wizard23]

I just tested by building from current source and it works with Wire now :) I will make a scope picture in the next days. :)
ktnx!

I'm so jealous that you have a 70 MHz digi scope... any picture is like porn to me...