I2c frequency

[fary99]

Board

esp32

Device Description

esp32

Hardware Configuration

not have

Version

v2.0.1

IDE Name

Arduino IDE

Operating System

windows10

Flash frequency

80

PSRAM enabled

no

Upload speed

921600

Description

I want to change the I2C clock, but I find that if the frequency exceeds 100K, the IIC frequency will not change to the size I set. For example, if I set 400K, it can only reach about 350K. Is there any way to optimize it?

Sketch

#include "Wire.h"

void setup() {
  Serial.begin(115200);
  Wire.begin(SDA,SCL,(uint32_t)409600);
}

void loop() {
  byte error, address;
  int nDevices = 0;

  Serial.println("Scanning for I2C devices ...");
  for(address = 0x01; address < 0x7f; address++){
    Wire.beginTransmission(address);
    error = Wire.endTransmission();
  }
}

Debug Message

not have

Other Steps to Reproduce

not have

I have checked existing issues, online documentation and the Troubleshooting Guide

• I confirm I have checked existing issues, online documentation and Troubleshooting guide.

[Jason2866]

Please try latest Arduino git version. Many changes since version 2.0.1
No support anymore for v2.0.1. Release of new version 2.0.3 will be in a few days

[VojtechBartoska]

Hi @fary99, did you try it with v2.0.2?

[fary99]

Tried it. Same thing

[VojtechBartoska]

Hello @fary99, thanks for your test.

@PilnyTomas Can you please test this issue on the latest master? In the meantime.

[PilnyTomas]

Hi, what exactly does it means that the frequency does not change - are you checking it on an oscilloscope, or how?
Right now I can check it only via software and it seems ok. The maximum frequency is 1 million - anything above that will set that 1 million.
Please try to call Wire.getClock() to see what the software has to say about the requested clock frequency.
It is also good practice to check the result of functions. Wire.begin returns true if started successfully.

[PilnyTomas]

Hi again, we already checked the issue on the oscilloscope and indeed the real output CLK frequency is off by a lot. We will investigate...

[fary99]

Hi again, we already checked the issue on the oscilloscope and indeed the real output CLK frequency is off by a lot. We will investigate...

Yes, the IIC frequency can not exceed 100K set in the library file, but I have seen the introduction of ESP-IDF, the IIC frequency can exceed 100K. So I modified the source code so that it could be set to be greater than 100K. Then I used the logic analyzer to capture the frequency of IIC, but it was different from the actual. Looking forward to your solution. Here is the code I modified
esp_err_t i2cInit(uint8_t i2c_num, int8_t sda, int8_t scl, uint32_t frequency){
if(i2c_num >= SOC_I2C_NUM){
return ESP_ERR_INVALID_ARG;
}
#if !CONFIG_DISABLE_HAL_LOCKS
if(bus[i2c_num].lock == NULL){
bus[i2c_num].lock = xSemaphoreCreateMutex();
if(bus[i2c_num].lock == NULL){
log_e("xSemaphoreCreateMutex failed");
return ESP_ERR_NO_MEM;
}
}
//acquire lock
if(xSemaphoreTake(bus[i2c_num].lock, portMAX_DELAY) != pdTRUE){
log_e("could not acquire lock");
return ESP_FAIL;
}
#endif
if(bus[i2c_num].initialized){
log_e("bus is already initialized");
return ESP_FAIL;
}
//-----------------------------This is modified-----------------------//
if(!frequency){
frequency = 100000UL;
}
//--------------------------------------------------------------------//

[dalbert2]

Confirmed on an oscilloscope that the I2C bus frequency is significantly lower than configured. For example, when Wire() is initialized at 400kbps, SCL is clocked around 360kHz. When Wire.begin() is initialized at 450kbps, SCL is around 384kHz.

I'm using IO21, IO22, 2K pull-ups on SDA and SCL, and current Arduino Core:
SDK:v4.4-dev-3569-g6a7d83af19-dirty
IDF:v4.4-dev-3569-g6a7d83af19-dirty

I was also surprised to see the ESP32 not assert SDA synchronously with SCL for the first bit after an ACK. It looks like glitching at 400kbps and is clearer at 100kbps (both shown below). For the bit after an ACK, SDA floats briefly before the ESP32 drives it low. Is it possible that this is contributing to ESP32 I2C instability at higher clock speeds? This happens to the next data bit after each ACK. Is there a reason the ESP32 doesn't drive SDA low (for a 0 bit) concurrently with falling SCL?

[dalbert2]

I tried upstream (4.4-beta1) and I2C reliability is dramatically better.
SDK:v4.4-beta1-189-ga79dc75f0a
IDF:v4.4-beta1-189-ga79dc75f0a

When I initialize wire at 400kbps: Wire.begin(IO_SDA, IO_SCL, (uint32_t)400000);
SCL is clocked around 333kbps, but transactions are reliable (which they are not in 4.4-dev)

[PilnyTomas]

Hi, we are investigating this issue with ESP-IDF team.
IDF-4960

[VojtechBartoska]

Status update: Issue is solved in ESP-IDF, we will let you know when it's reflected to Arduino core.

[dalbert2]

Thank you @VojtechBartoska, I'm actually working on switching to ESP-IDF, so I'll try it there.

[VojtechBartoska]

@PilnyTomas Can you please just quickly retest this? It's supposed to be solved.

[PilnyTomas]

The IDF fix has improved the situation, but it is still not exactly matching the requested frequency.
For example:
Requested 500 000Hz
Before the fix, with 10k Ohm external pullup: 430 000Hz
After the fix, with 10k Ohm external pullup: 460 000Hz

[dpharris]

This appears to be pertinent: espressif/esp-idf#4300 David

…

On Mon, Aug 1, 2022 at 1:05 AM Tomáš Pilný ***@***.***> wrote: The IDF fix has improved the situation, but it is still not exactly matching the requested frequency. For example: Requested 500 000Hz Before the fix, with 10k Ohm external pullup: 430 000Hz After the fix, with 10k Ohm external pullup: 460 000Hz [image: image] <https://user-images.githubusercontent.com/34927466/182102339-462b4379-c8f7-432c-aa8b-31519f353c2e.png> — Reply to this email directly, view it on GitHub <#6444 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEDQSVMOROO6CZ4C66WMGDVW6AN3ANCNFSM5RBJDR5Q> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[PilnyTomas]

Hi everyone,
The IDF fix is available for Arduino users on a branch idf-release/v4.4 git checkout idf-release/v4.4 and it should be merged in next version (?) @VojtechBartoska

I have retested the frequency response to the external pull-up value and it seems that lower values correspond better to the requester frequency - try using values less than 5kOhm - 1k should be pretty good.

Also according to Technical Reference Manual the supported speed is up to 400kbit/s (see page 278)

[VojtechBartoska]

Yes, this will be included for 2.0.5 milestone.