failure with 1MHz I2C update rate

[christianrauch]

Version

1.0.9

Describe the bug

When setting the I2C rate on a Raspberry Pi 5 to 1MHz (dtparam=i2c_arm_baudrate=1000000), the mpu9250 fails with:

$ ./build/mpu9250 -e read
iic: read failed.
: Resource temporarily unavailable
mpu9250: mag read who am i failed.
mpu9250: mag init failed.
mpu9250: run failed.

It works with 400kHz (dtparam=i2c_arm_baudrate=400000).

Reproduce

In /boot/firmware/config.txt, set:

[all]
# I2C 1MHz
dtparam=i2c_arm_baudrate=1000000

Expected behavior

The library should still work.

Additional context

No response

[libdriver]

Hello, according to the MPU9250 datasheet(datasheet/mpu9250_product_specification.pdf), the I2C interface supports max 400kHz.

[christianrauch]

I was able to read from the gyroscope and accelerometer at 1MHz with an alternative C/C++ implementation.

In Chapter 3 (Electrical Characteristics), the Gyroscope and Accelerometer MAX "Output Data Rate" is advertised as 8000 and 4000 Hz, respectively. I am not sure how the "i2c_arm_baudrate" and the "SCL Clock Frequency" are related. With dtparam=i2c_arm_baudrate=400000, In "only" get a rate of 800Hz. The example also uses MPU9250_BASIC_DEFAULT_RATE of 50Hz. But changing this seems to have no effect.

How can I read the gyroscope and accelerometer at the advertised 8kHz and 4kHz? Do you have an example configuration?

[libdriver]

400KHz is the I2C interface rate, and 8kHz for gyroscope, 4KHz for accelerometer are sensor adc sample rates.You can try this:

[christianrauch]

In the end, I tried the following configuration:

-#define MPU9250_BASIC_DEFAULT_RATE                           50                                        /**< 50Hz */
-#define MPU9250_BASIC_DEFAULT_LOW_PASS_FILTER                MPU9250_LOW_PASS_FILTER_3                 /**< low pass filter 3 */
+#define MPU9250_BASIC_DEFAULT_RATE                           500                                       /**< 500Hz */
+#define MPU9250_BASIC_DEFAULT_LOW_PASS_FILTER                MPU9250_LOW_PASS_FILTER_0                 /**< low pass filter 0 */
-#define MPU9250_BASIC_DEFAULT_GYROSCOPE_CHOICE               0                                         /**< 0 */
-#define MPU9250_BASIC_DEFAULT_ACCELEROMETER_CHOICE           0                                         /**< 0 */
-#define MPU9250_BASIC_DEFAULT_ACCELEROMETER_LOW_PASS_FILTER  MPU9250_ACCELEROMETER_LOW_PASS_FILTER_3   /**< low pass filter 3 */
+#define MPU9250_BASIC_DEFAULT_GYROSCOPE_CHOICE               3                                         /**< 3 */
+#define MPU9250_BASIC_DEFAULT_ACCELEROMETER_CHOICE           1                                         /**< 1 */
+#define MPU9250_BASIC_DEFAULT_ACCELEROMETER_LOW_PASS_FILTER  MPU9250_ACCELEROMETER_LOW_PASS_FILTER_0   /**< low pass filter 0 */

but without a noticeable effect.

When I only time the call of mpu9250_basic_read in main.c with:

            struct timespec start, end;
            clock_gettime(CLOCK_MONOTONIC, &start);

            /* loop */
            for (i = 0; i < times; i++)
            {
                /* read data */
                if (mpu9250_basic_read(g, dps, ut) != 0)
                {
                    (void)mpu9250_basic_deinit();

                    return 1;
                }
            }

            clock_gettime(CLOCK_MONOTONIC, &end);

I get a slight improvement to 914.82Hz.

Within mpu9250_basic_read -> mpu9250_read there are further calls to a_mpu9250_read. As far as I can see, this always reads "read config", "read accel config", and "read gyro config", with 1 byte each, 14 bytes of acceleration and gyro data, and 8 bytes of magnetometer data (25 bytes in total).

At 400kHz (bit/s) that should give 400000 bit/s /(25 byte * 8) = 2000 Hz, but I am only getting half of that.

If I remove the a_mpu9250_mag_read from mpu9250_read I get 1238.48Hz, which matches roughtly the relation 914.82Hz * 25 bytes / 17 bytes = 1344.

If I further remove the 3 config reads via a_mpu9250_read, I can actually achieve 2379.42Hz measured. For the 14 bytes accel+gyro, this should be theoretically 400000/(14*8) = 3571.

With the magnetometer and without the 3 config reads, I can still get 1398.74Hz. So reading the config clearly has more time overhead than just transmitting the 3 bytes.

I assume it is possible to achieve the advertised rates when disabling the individual components (accel, gyro, magn).

But reading the 3 configs in every mpu9250_read seems to be quite some overhead that could be avoided by caching the config after it has been set. This already increases the accel+gyro+magn read way above 1kHz. Is there a particular reason why the config is read every time?

[libdriver]

For batch data collection, you can refer to FIFO mode data collection example(/example/driver_mpu9250_fifo.c), maybe this mode is more suitable for your need.