Why use 4K gyro task if readings are taken from it at 2K frequency?

[and-sh]

No filtration or accumulation are used in gyro task. So half of readings just discarded.
It's too wasteful.

[Jetrell]

That's correct. 4K looptime adds no performance difference over 2K for INAV. And whether it could or even should is debatable.

There is a place for higher looptime, when it comes to freestyle and race performance... But when it comes to navigation. It would be interesting to see how betaflight is managing to use higher looptimes, now that they have more automated flight. I can only assume they are limiting it.
Higher looptimes load up the processor more, leaving less time for other high priority tasks. Which is why projects like Ardupilot only run between 272Hz and 400Hz, for processing stability.

[and-sh]

Thank you. I have a small improvement. I'll test how it works. Do you know some good comparison different pid loop rates?

[and-sh]

The difference between picking one number out of a stream (top image) and averaging all the data (bottom image). This has been done on the accelerometer for a long time, so there is no difference there.

[and-sh]

Realization is quite simple. First function is in gyro loop. Second in pid loop.

void FAST_CODE NOINLINE icm45600IMURead(gyroDev_t *gyro)
{
STATIC_FASTRAM uint8_t data[12];

const bool ack = busReadBuf(gyro->busDev, ICM45600_RA_ACCEL_DATA_X1, data, 12);
if (!ack) {
    return;
}
GYRO_A[0]+= (float) int16_val_little_endian(data, 3);
GYRO_A[1]+= (float) int16_val_little_endian(data, 4);
GYRO_A[2]+= (float) int16_val_little_endian(data, 5);
GYRO_cntr++;

ACC_A[0]+=(float) int16_val_little_endian(data, 0);
ACC_A[1]+=(float) int16_val_little_endian(data, 1);
ACC_A[2]+=(float) int16_val_little_endian(data, 2);
ACC_cntr++;
return;

}

static bool FAST_CODE NOINLINE icm45600GyroRead(gyroDev_t *gyro)
{

if (GYRO_cntr==0) {
    return false;
}

gyro->gyroADCRaw[X] = GYRO_A[0]/GYRO_cntr;
gyro->gyroADCRaw[Y] = GYRO_A[1]/GYRO_cntr;
gyro->gyroADCRaw[Z] = GYRO_A[2]/GYRO_cntr;

GYRO_A[0]=0;
GYRO_A[1]=0;
GYRO_A[2]=0;
GYRO_cntr=0;    

return true;

}

[DzikuVx]

Actually, the original statement of

No filtration or accumulation are used in gyro task. So half of readings just discarded.

is false. The gyro sampling routine running at 4kHz read gyro values and then runs it through a antialiasing filter so it can be safely ingested by the main PID loop without any artifacts.

There is no need to run PID loop at 4kHz or 8kHz as there are no frequencies coming from flight that require such fast processing rate.

[and-sh]

Thank you. Now I see it. Pt1 250Hz. Actually mooving avereging on 2 points.

I notised a long time ago, in oder to lower noice you need to process as many data as you can.