Need a setClock function in Wire Library

[mjs513]

To effectively use the i2c bus users need a way to programmatically change the I2C bus speed. Currently this is implemented for most of the Arduino boards like the mega as well as the teesnsy. Many sensors that would be connected to the bus would benefit from this ability.

[bigdinotech]

Hi, I added the functionality to change the I2C clock speed. The PR is under review.

Meanwhile, you can try it out here:
https://github.com/bigdinotech/corelibs-arduino101/tree/i2cclock

[mjs513]

Hi bigdinotech
I downloaded the updates per the instructions. I decided to run the test on the new Teensy Prop Shield that has two I2C sensors from NXP (FXAS21002C and the FXOS8700CQ). Running 1024 sensor reads using both sensors here is what I got (average of 1024 reads);

slow = 69 milliseconds
fast = 68 milliseconds

Not much of a difference the bus speed change and very slow for i2c. On the mega I get:

slow = 0.047 milliseconds and
fast = 0.032 milliseconds.

Big difference between the mega and curie. I think something else is wrong here.
Here is the sketch for your reference:
`#include <Wire.h>
#include <FXAS21002Cnxp.h>
#include <FXOS8700CQnxp.h>

FXOS8700CQ sensor = FXOS8700CQ();
FXAS21002C sensor2 = FXAS21002C();

unsigned long start, stop;
int ax, ay, az;
int gx, gy, gz;
int mx, my, mz;

void setup() {
Serial.begin(115200);
Wire.begin(I2C_SPEED_SLOW);

// Initialize the FXOS8700CQ

sensor.begin();

// Initialize the FXAS21002C
sensor2.begin();
delay(6000);
}

void loop() {
Serial.println("Testing raw reading speed (average on 1024 samples):");
start = micros();

for(int i=0; i<1024; i++) {
// Read the motion sensors
sensor.readMotionSensor(ax, ay, az, mx, my, mz);
sensor2.readGyro(gx, gy, gz);
}

stop = micros();
Serial.print("--> result: ");
Serial.print((stop - start) / 1024);
Serial.print(" microseconds .... ");
Serial.print(((stop - start) / 1024) / 1000);
Serial.println(" milliseconds");

Serial.println("Looping again..");
Serial.println("----");
}
`

[mjs513]

Hi. Hooked up a MPU-6050 sensor to the mega and 101 here are the results:
100hz 400hz
Mega 0.426 0.178 milliseconds
101 4.28 3.29 milliseconds.

all I can say is something does not make sense with the 101 and the I2C implementation
Hope this info helps
Mike.

[bigdinotech]

@mjs513 can you give a link to the library you are using?
I have hunch that Wire.begin() is being called inside that library and thus overriding the speed back to 100Khz

[mjs513]

@bigdinotech I did too so I looked at all libs used and deleted the wire.begin's in the FXOS8700CQ and the FXAS21002C libraries. There are calls to wire.begintransmission(addr) etc. I also looked at the I2Cdev lib and the MPU60xx lib that I am using from Jeff Rowberg and there are no wire.begin calls in the libs or anything strange to reset the TWBRs (but that would have a thrown an error).

I also just hooked up the FDRM-STBC-AGM01 shield from Freescale which have the same chip set as the prop shield with the Curie and got the same results. However, when I first hooked it up I had the sketch already loaded and it gave me a 2 ms reading for slow. When I changed the setting to fast it went bac to around 68 ms and when I put it back to slow it gave me 68ms. Not sure what is happening in this case.

Mike

PS. Did you guys change the loader script - don't see all the gibberish with DFU etc anymore.

[bigdinotech]

@mjs513 I just tried the example sketch from the FXAS21002C library.
I have a scoped attached to SCL and I can see the clock getting faster as expected.
Do you have access to a scope? What clock speed are you getting on the SCL pin?

[mjs513]

@bigdinotech I have access to a usb scope but now can't get a nice clean signal but from what I am seeing at I2CSLOW the freq varies from 0.058 kHz to 0.086 kHz using a 20ms time div. At I2CFast I get a pretty steady reading of 0.058 kHz using the same settings. Go figure. I am attaching the screenshots so you can interpret for me - I really have no idea what I am doing with the scope yet. This testing was done with hardware in the loop - using the Prop-Shield from Paul S.

Thanks for your help
Fast

Slow shot 1

Slow shot2

[bigdinotech]

@mjs513 I think if you zoom in more into the signal you would see that the clock speed is correct.

[mjs513]

@bigdinotech I believe you. Like I said I am not very good with a scope. Anyway so if the clock speed has increased (I am assuming slow is 100 and Fast is 400) why such a difference in the sensor read timing tests between the mega and the Curie? Why no change in the timings between slow and fast on the curie? Is this a separate issue or am I doing something wrong?

Thanks for your help
Mike

[mjs513]

@bigdinotech Would it be possible to add a function that does the same thing called Wire.setClock(xxxxxxx); - this would make it similar to existing calls to setClock in the Arduino wire library. The function would just do a wire.begin(I2C_SPEED_FAST) or wire.begin(I2C_SPEED_SLOW) if xxxxxx is 4000000 or 4000000L. It would avoid the user having to change existing code if they are moving a sketch from a mega or uno over to the Cuire.

[bigdinotech]

@mjs513. I'll take a look at it