Calibrate Wire.setClock so it sets correct frequency

[LonerDan]

This is more a observation then issue report since you have already written in description that Wire.setClock gives weird frequency. I played with it a while and found that at 400 it gives frequency around 1khz. At 300000 and more it doesnt work at all since the scl is all distorted and non consistent (more like some noise). Maximum frequency I got was ~307kHz with set values around 180000-200000.

[igrr]

Thanks for investigating.

The code that sets the frequency is here:

void i2c_freq(int freq_hz)
{
    s_i2c_delay = 1000000 / freq_hz / 4 - 1;
    if (s_i2c_delay < 0)
        s_i2c_delay = 0;
}

So this function needs a tweak to calculate the correct delay, maybe by using a static lookup table.
@ManiackCZ Since you have started experimenting with this, would you mind hacking this a bit to measure the frequencies we get for each value s_i2c_delay in range from 1 to 16? Then we can convert that data into a LUT and fix this.

[LonerDan]

Oh, havent found that function when I was looking for it. Ok, I have few free days this week so I will try to poke around and let you know.

[LonerDan]

It seems little bit rough to use delayMicroseconds(). It has big gaps between individual delay values due to operation with integers. What frequency are the ESP8266 modules usually running at? Are there any other sources to generate delays? (timers, way to use NOPs...)

[igrr]

80MHz or 160MHz most of the time.
Maybe we can use a calibrated delay loop?
There's a CCOUNT special register which counts the CPU clock cycles. It can
be read in asm using RSR instruction.

On Thu, Apr 2, 2015 at 4:22 PM, ManiackCZ notifications@github.com wrote:

Seems little bit rough to use delayMicroseconds(). It has big gaps between
individual delay values due to operation with integers. What frequency are
the ESP8266 modules usually running at? Are there any other sources to
generate delays? (timers, way to use NOPs...)

Reply to this email directly or view it on GitHub
#15 (comment).

[LonerDan]

That seems promising. I will have a look at that.

[LonerDan]

I thought I got it, but it was bahaving absolutely without any logic so I looked at my conversion algorithm itself and it seems to give 0 all the time.
UPDATE: I investigated more and it seems to be something wrong with float in general. I opened new issue report.
UPDATE: Wait, maybe I am geting somewhere...

[LonerDan]

UPDATE: I have my final leaving exams coming in few weeks, so I will have to (unforunately) pause my work on the calibration but I plan to get right back into it after a month or so.
Thank you for understanding

[igrr]

Fixed with the new i2c library.