Episode 3 — Field-Oriented Control of PMSMs

"How to Make a Three-Phase Motor Think It's a DC Motor"

Pillar: Embedded & Control Systems · Runtime ~14 min · ~2,050 words · Elena & Marcus

MARCUS: Episode three. We've done a power-systems lesson and a communications lesson, and both came down to the same move — find the right coordinate frame and the hard problem goes diagonal. Today, embedded control, and Elena promises the exact same trick shows up in motors. Set us up.

ELENA: The objective today: understand field-oriented control — vector control — of a permanent-magnet synchronous motor well enough to explain why it works, not just that it works. We'll re-derive why a three-phase machine is hard to control directly, then build the Clarke and Park transforms that fix it, then connect it to why your electric car drives the way it does. This is the control scheme behind essentially every modern high-performance motor drive.

MARCUS: And the hook?

ELENA: We're going to take a screaming three-phase AC machine, full of time-varying sinusoidal currents, and through nothing but a change of coordinates, make it as simple to command as a brushed DC motor — two knobs, one for torque, one for flux, fully decoupled.

The Refresher: Why Motors Are Hard and DC Motors Aren't

ELENA: Let's start by appreciating the old brushed DC motor, because it's the gold standard of controllability, and understanding why tells us everything. In a brushed DC machine, the commutator mechanically guarantees that the armature current always flows at right angles to the field flux. Torque is the cross product of those two, and because they're held orthogonal by construction, torque comes out simply proportional to armature current. You want more torque, you push more current. The field and the torque-producing current are physically decoupled. One knob.

MARCUS: And that's exactly what you lose going to a three-phase machine.

ELENA: Right. A permanent-magnet synchronous motor is wonderful hardware — no brushes to wear, high power density, the rotor flux supplied by permanent magnets. But to control it you face three stator windings carrying three sinusoidal currents, 120 degrees apart, and a rotor that's spinning. The torque depends on the instantaneous angle between the stator's combined magnetic field and the rotor's magnet flux, and that angle is changing constantly as the thing rotates. In the native abc frame — the three phase currents as functions of time — everything is a moving sinusoid, everything is coupled to everything else, and there's no single knob for torque. Try to run a simple PI controller on a quantity that's sinusoidally varying at the electrical frequency and it's always chasing, never settling.

MARCUS: So the machine is great but the coordinate system is hostile.

ELENA: Beautifully put — and that's the entire setup. The hardware is fine. The frame is the problem. And just like OFDM and just like the power-systems episode, the fix is to stop fighting in the bad frame and transform into a good one.

What Changed: Vector Control Went Mainstream

ELENA: Field-oriented control as theory dates back to the early 1970s — Blaschke and Hasse. So why is it a catch-up topic? Because what changed since your 2013 baseline isn't the theory, it's that the implementation went from exotic to universal and then got smarter.

MARCUS: What made it universal?

ELENA: Three things converging. One, cheap microcontrollers fast enough to run the transforms and current loops at tens of kilohertz — a modern Cortex-M with a fast ADC and sometimes hardware math does this on a chip that costs a couple dollars. Two, wide-bandgap inverters — silicon carbide and gallium nitride — switching fast and efficiently, which gives the control loop clean, high-bandwidth actuation. Three, and this is the big modern shift: sensorless control matured. Classic field-oriented control needs to know the rotor angle precisely, which historically meant an encoder or resolver bolted to the shaft — cost, wiring, a failure point. Modern drives increasingly estimate the rotor angle from the electrical signals themselves, using observers, and delete the sensor.

MARCUS: So the lesson is the same algorithm, now running everywhere, often without the position sensor it used to require.

ELENA: Exactly. It's the control scheme in your EV traction motor, your e-bike, your drone, your washing machine, your HVAC compressor. Understanding it is understanding how modern electromechanical energy conversion is actually commanded.

The Mechanism: Clarke and Park

ELENA: Here's the heart of the episode — two transforms, in sequence, and I'll teach them as a journey from a hard frame to an easy one.

MARCUS: Start with Clarke.

ELENA: The Clarke transform. You have three phase currents — i_a, i_b, i_c — but they're not independent; in a balanced machine they sum to zero. So three numbers, but really only two degrees of freedom. The Clarke transform collapses the three-axis abc representation into an equivalent two-axis representation, called alpha-beta. It's a fixed, stationary frame — think of it as describing the stator's total magnetic field as a single vector with two components instead of three overlapping windings. You've gone from three coupled sinusoids to a single rotating vector in a plane. That's progress, but that vector is still spinning at the electrical frequency.

MARCUS: So alpha-beta is still time-varying.

ELENA: Still spinning, yes. Which is where Park does the magic. The Park transform rotates the alpha-beta frame so that it spins along with the rotor. You take the stationary two-axis frame and continuously rotate it by the rotor's electrical angle, into what we call the d-q frame — direct and quadrature axes. The d-axis is aligned with the rotor's magnet flux; the q-axis is ninety degrees ahead of it.

MARCUS: And what happens to the currents when you view them from a frame that's spinning with the rotor?

ELENA: This is the payoff — the whole reason we're here. From a frame rotating in lockstep with the rotor, the currents that looked like fast sinusoids in the abc frame become constants at steady state. DC values. The vector that was whirling around now sits still relative to your rotating viewpoint. And once your control quantities are DC, a plain PI controller can drive them to a reference with zero steady-state error — no chasing a moving target. You've transformed a sinusoid-tracking problem into a DC-regulation problem, exactly the way you'd want.

MARCUS: And the two axes mean two knobs.

ELENA: And now we recover the brushed DC motor's beautiful property. In the d-q frame, the q-axis current — i_q — is the one aligned to produce torque against the rotor flux. For a surface-magnet machine, torque is essentially proportional to i_q. That's your torque knob, clean and linear, just like armature current in the DC motor. The d-axis current — i_d — points along the rotor flux. In the simplest control you hold i_d at zero, because it produces no torque and you don't want to waste current. So you run two independent PI loops: one drives i_q to whatever torque you command, the other holds i_d at zero. Two decoupled knobs. You have made the three-phase synchronous machine behave like a DC motor, purely by choosing the right rotating coordinate frame.

MARCUS: And then you transform back to drive the actual inverter.

ELENA: Right — the loop closes by running the transforms in reverse. Your PI loops produce voltage commands in the d-q frame; inverse Park rotates them back to stationary alpha-beta; and space-vector modulation turns those into the switching pattern for the six transistors of the inverter. Every control cycle: measure the phase currents, Clarke and Park into d-q, regulate with PI, inverse-transform, modulate. Tens of thousands of times a second. That's field-oriented control.

Where It Gets Modern: Field Weakening, MTPA, and Sensorless

ELENA: Now let's push past the textbook version into what makes a real EV drive, because this is the catch-up that matters.

MARCUS: You said hold i_d at zero. When is that wrong?

ELENA: Two big situations, and they're the soul of modern drive engineering. First, field weakening. The spinning rotor magnets induce a back-EMF that grows with speed. Eventually that back-EMF approaches the voltage your inverter can supply, and you can't push any more current — you've hit a speed ceiling. The fix: deliberately command a negative i_d. That d-axis current creates a stator flux that opposes the magnet flux, weakening the net field, lowering the back-EMF, and letting the motor spin faster — past base speed — at the cost of some torque. That's how an EV keeps accelerating at highway speed instead of slamming into a wall at base speed. Field weakening is i_d earning its keep.

MARCUS: You said two situations.

ELENA: The second is the interior permanent-magnet machine — the kind used in most EV traction motors. There the magnets are buried inside the rotor iron, which makes the rotor magnetically asymmetric — the d-axis and q-axis have different inductances. That asymmetry produces a second torque component, reluctance torque, on top of the magnet torque. And reluctance torque depends on the product of i_d and i_q. So now i_d is no longer something you zero out — there's an optimal nonzero i_d, and because the q-axis inductance is the larger one in these machines, that optimal i_d is negative. Combined with i_q, it produces a given torque with the minimum total current. That optimization is called maximum torque per amp, MTPA, and modern drives compute that operating point continuously to squeeze the most torque out of every amp — which means the most range out of every electron in the battery. MTPA plus field weakening together is the operating map of a real traction inverter.

MARCUS: And the sensorless piece you flagged earlier.

ELENA: Right, the last modern pillar. Every transform we did needs the rotor's electrical angle to do the Park rotation. Sensorless control estimates that angle without a shaft sensor. At medium and high speed you use the back-EMF: it carries the rotor position, and a state observer — often a sliding-mode observer or a flux observer — reconstructs the angle from the measured currents and voltages. At very low speed and standstill the back-EMF vanishes and that fails, so you use high-frequency signal injection — inject a small probing signal and exploit the rotor's magnetic saliency, that same d-q asymmetry, to find the angle. Modern firmware blends the two: injection down low, observer up high. The result is full field-oriented control with no position sensor at all — cheaper, more reliable, one less thing to wire.

Industry Standard Today

MARCUS: Where does this live in real products?

ELENA: Field-oriented control is the default for any application that cares about efficiency or dynamic performance. EV traction drives run interior-PMSM field-oriented control with MTPA and field weakening, increasingly on silicon-carbide inverters for the efficiency. Industrial servo drives — robotics, CNC, automation — run it for the precision. Even commodity appliances: high-efficiency HVAC compressors and washing-machine motors moved to sensorless field-oriented control to hit energy-efficiency regulations. On the implementation side, vendors ship it as turnkey libraries and reference designs — the classic example is Texas Instruments' InstaSPIN, and every major MCU vendor has a motor-control library and often dedicated hardware accelerators for the transforms. The point for you: this is not niche. It is the standard way electromechanical power gets controlled in 2026.

Self-Test

MARCUS: Make it skill. What should the listener work?

ELENA: Two exercises. First, the relationship that trips people up: electrical versus mechanical frequency. A PMSM has multiple pole pairs, and the electrical frequency is the mechanical rotation frequency times the number of pole pairs. So work this: a motor with, say, four pole pairs spinning at three thousand RPM — what's the electrical frequency your field-oriented control loop and your Park transform are actually tracking? You'll get a number in the hundreds of hertz, and you'll appreciate why the current loop has to run at tens of kilohertz to stay way above it. That ratio — loop rate over electrical frequency — is a real design constraint.

MARCUS: And the second?

ELENA: The conceptual one that proves understanding. Explain why holding i_d at zero is the right call for a surface-magnet motor but leaves performance on the table for an interior permanent-magnet motor. If you can articulate that the buried magnets create d-q inductance asymmetry, that the asymmetry produces reluctance torque proportional to the product of i_d and i_q, and that this means an optimal nonzero i_d exists — maximum torque per amp — then you understand why your electric car's motor controller is doing something more sophisticated than the textbook version. That's the catch-up nailed.

MARCUS: Next episode we go down into the silicon itself — the move from FinFETs to gate-all-around nanosheet transistors, and why the gate had to wrap further and further around the channel just to keep Moore's law breathing.

ELENA: And you'll notice the theme one more time. Today we controlled a motor by changing frames. Next time we control a channel — an electron channel — by changing geometry. It's control all the way down.