EMF #

Brief #

emf measures the back EMF of a coasting motor on the F3 (HV board): with the bridge off it demodulates every PWM period’s unfiltered phase voltage sample against the rotor angle and sums the flux linkage psi, its 5th and 7th harmonic back EMF, and psi per magnet pole, to tell a demagnetised magnet from a weak rotor. It is loaded by stm32f303/src/main.c as emf0 (rt_prio 7, the last in the chain). It is controlled from the F4 through config words and read back one result at a time through a state word: idpmsm’s psi test drives it (conf/template/id_pmsm.txt links hv0.emf_run/emf_sel/emf_pp to idpmsm0 and idpmsm0.emf_val = hv0.emf_val). It runs on the F3 because the F4 only sees each phase voltage every ~2 ms and through io0’s filter.

Component Explanation #

1. Wiring on the F3 (fixed in main.c):

• emf0.u/v/w = io0.ur/vr/wr, emf0.pos = ls0.pos, emf0.si/co = dq0.si/co, emf0.en = ls0.en.
• emf0.run/sel/pp = ls0.emf_run/emf_sel/emf_pp (F4 config words), ls0.emf_val = emf0.val (F4 state word).

2. Speed and turns (rt, always):

• vel is the electrical speed from the change of pos per tick (the angle steps once per F4 packet, every PWM_TICKS_PER_PACKET ticks, 3 at the default 15 kHz PWM).
• A wrap of pos from +pi to -pi counts one electrical cycle forward, the other way one back.

3. When it sums (rt):

• Only with the bridge off: en must have been 0 for 30 ms (the winding current has to die out first), and |vel| must be above 20 rad/s electrical.
• run < 0 clears the sums every tick while it stays negative. rt_start clears them too.
• The live psi pin is updated whenever the conditions above hold, also while run is 0.

4. Demodulation (rt):

• The phase voltages are Clarke transformed (amplitude invariant, so the common mode drops out) into e = a + j b. With the rotor flux psi(th) = sum psi_k exp(j k th) over the harmonics k = 1, -5, 7, the back EMF is e = d/dt psi, so

e * exp(-j k th) / w  averages to  j k psi_k   // w = electrical speed

• This holds whatever the direction of rotation, so psi comes out in Vs (peak, per electrical rad) at any speed.
• Every sample adds e exp(-j th) / w, e exp(+j 5 th) / w and e exp(-j 7 th) / w to three sums, and n counts them.
• Per pole bins: the fundamental is also summed per half electrical cycle, bins = 2 * pp bins over a mechanical turn (pp rounded, 1..8, so up to 16 bins), which is one bin per magnet pole. Bin 0 is wherever the count started, not a fixed rotor position, so compare the bins with each other: a weak pole reads low in its bin on every run.

5. DC offset removal:

• A DC offset between the phase voltage channels (divider or ADC offsets that differ per phase) averages out over a whole electrical cycle but not over half of one, so it would make the bins alternate high and low.
• The offset d is estimated as the mean Clarke vector over whole electrical cycles only (the back EMF integrates to 0 over each one): summing starts at the first wrap of pos after the first sample, and a snapshot is taken at every later wrap. Each bin also keeps the sum of exp(-j th) / w, so its share of d is subtracted when the bin is read. Until one whole cycle has been seen, d reads 0.

6. Results (rt, one per tick):

• sel picks what val returns:
• 0 (or below): the live psi (low pass filtered |psi_1| with a 20 ms time constant).
• 1: number of samples summed.
• 2, 3: psi_1 real and imaginary part, averaged over the whole coast.
• 4, 5: 5th harmonic back EMF (-5 psi_-5), real and imaginary, on the same scale as psi_1.
• 6, 7: 7th harmonic back EMF (7 psi_7), real and imaginary.
• 8: number of bins in use (2 * pp).
• 9 + 3b, 10 + 3b, 11 + 3b: bin b (0..15), psi real, psi imaginary (DC offset removed, averaged over the bin’s samples), sample count.
• 57, 58: the DC offset d, alpha and beta (V).
• Anything above returns 0.
• On the F4 the value arrives through the rotating state words of the link, so after changing sel the reader must wait a few ms before emf_val holds the new result (idpmsm does).

• vel comes from differencing pos, which on the F3 is the F4’s angle extrapolated between packets, so the F4 must keep sending the rotor angle while the bridge is off.
• Bins past 2 * pp stay 0, and changing pp between clears mixes bin indices.
• The bin count is limited to 16 (8 pole pairs) by the F3’s HAL memory (HAL_MAX_CTX); a motor with more pole pairs gets bins that each span several poles.

Pins #

References #

Source: .c File