This document is part of the hoverboard firmware documentation.
One way to measure current is by measuring voltage over the lower branch transistor, utilizing its parasitic resistance Rds,on.
Both motors have voltage measurement over the lower leg transistor of two phases. The voltage measurement is biased towards 3.3 V supply (resistor R22 in figure below), has a gain of somewhere around 2.5 and measurement range, depending on the op-amp type, probably around -500 ... 500 mV. This converts to about 0.5 ... 2.8 V for the ADC. If the measured phase voltage goes higher, there is a diode clamping circuit to protect the op-amp. On negative side the measurement voltage is not clamped.
Since the MOSFET in on-state is effectively an resistor with a value of Rds,on, it is straightforward to calculate current from the voltage measurement, if the resistance value is known. However, the resistance is quite nonlinear, especially temperature has a large effect on the value. Rds,on measurement is often used to estimate the MOSFET temperature when the current is known, now we'd need to work around the temperature effect.
The figure below shows one phase of the output inverter together with the measurement circuit and its equivalent circuit when the bottom transistor is conducting (and top one is open). Note that on the left side the component values somewhat match what is on the board (at least on my version), the right side markings are not related to real components.
Because the MOSFETs can also conduct in reverse direction when it is turned on, it is possible to measure both positive and negative current. The measurement offset when the current is zero must be measured and compensated for in the firmware.
The voltage measurements are connected to following pins.
| Voltage | Pin | ADC channel |
|---|---|---|
| Left motor B phase | PC4 | 14 |
| Left motor C phase | PC5 | 15 |
| Right motor A phase | PA0 | 0 |
| Right motor B phase | PC3 | 13 |
ADC1 is used to measure all 4 phase voltages. ADC1 was selected since it is the only one which maps all the pins to channels and has DMA capability. DMA 1 channel 1 is used for ADC1.
To make sure that the voltages are measured when bottom transistors are conducting, the measurement is done during a 000 zero vector (all lower transistors on). With the selected modulation schemes using the center-aligned mode (up/down counting) in the timers, the center of 000 zero vector is when the timer transitions from downcounting to upcounting. The timer also can generate an interrupt at that point, which then triggers the ADC.
Sampling at the center of the zero vector has an additional benefit that it is (hopefully) far from the switching instants which could cause noise in the measurement. And as the timers from PWM of both motors are synchronized, it is possible to measure all voltages almost simultaneously with single trigger.
The ADC scans all 4 voltages in series and transfers the values to a buffer. When the timer transitions from upcounting to downcounting and triggers an interrupt again, the values are transferred to another variables. At this point also DC (zero) offsets are subtracted from the measurement. The offsets are measured during power-up.
Operation of the current measurement with shunt resistor is described here.
Note that this method is not implemented at the moment, since the required minimum pulse lengths would degrade the voltage quality quite much, and the Rds,on method is much simpler and gives good enough results.
In the hoverboard electronics, both motors drivers contain current measurement. The measurement is done by a shunt resistor in the bottom branch, connected such that the sources of the lower leg transistors are connected together to one terminal of the resistor and the other terminal of the resistor is connected to the battery negative.
The voltage over the shunt resistor is then fed to an operational amplifier. The signal has a small positive offset applied to it (approximately 140 mV), which allows for bipolar current measurement. The operational amplifier has voltage gain of 11, and because the current shunt resistor is two 7 mOhm resistors in parallel (= 3.5 mOhm), the total gain is 0.0385 V/A.
The small offset applied to the input signal centers the output to about 1.54 V and if the rail-to-rail measurement range is 0 ... 3.3 V, the measurement range is then -38.46 ... 45.71 A. In reality we can probably expect around +-35 A.
Because only one shunt resistor is used per motor, the phase current measurement becomes difficult. However, with space vector modulation it is possible if done carefully. The method is described in for example this document from Texas Instruments.
First, we assume that the system is symmetric, i.e. the sum of all phase currents is zero. Then, for different switch combinations (phase A, B and C) we can derive that the shunt current is
| A | B | C | Current |
|---|---|---|---|
| 0 | 0 | 0 | - |
| 1 | 0 | 0 | Ib+Ic=-Ia |
| 0 | 1 | 0 | Ia+Ic=-Ib |
| 1 | 1 | 0 | Ic |
| 0 | 0 | 1 | Ia+Ib=-Ic |
| 1 | 0 | 1 | Ib |
| 0 | 1 | 1 | Ia |
| 1 | 1 | 1 | - |
It can be seen that current cannot be measured during zero vectors and if two different active vectors are used (which are separated only by one turn of a switch) then two different currents can be measured. The third current can then be calculated as the sum is zero.
For this method to work, three things must be guaranteed:
- Active vectors are long enough to allow stable measurement
- Two different active vectors are used
- Current is sampled at a precise instant after change of vector
The ADCs can sample following channels:
ADC1
| IN | Signal |
|---|---|
| 0 | Right A voltage |
| 1 | Power SW voltage |
| 2 | Left UART TX |
| 3 | Left UART RX |
| 10 | Right current |
| 11 | Left current |
| 12 | Battery voltage |
| 13 | Right B voltage |
| 14 | Left B voltage |
| 15 | Left C voltage |
ADC2
| IN | Signal |
|---|---|
| 0 | Right A voltage |
| 1 | Power SW voltage |
| 2 | Left UART TX |
| 3 | Left UART RX |
| 10 | Right current |
| 11 | Left current |
| 12 | Battery voltage |
| 13 | Right B voltage |
| 14 | Left B voltage |
| 15 | Left C voltage |
ADC3
| IN | Signal |
|---|---|
| 0 | Right A voltage |
| 1 | Power SW voltage |
| 2 | Left UART TX |
| 3 | Left UART RX |
| 10 | Right current |
| 11 | Left current |
| 12 | Battery voltage |
| 13 | Right B voltage |
Because ADCs 1 and 2 can sample all channels and ADC 3 everything but left motor phase voltage, it was decided to use ADC1 for samplign all non-motor related signals and ADCs 2 and 3 motor currents and phase voltages.
However, it turns out only ADC1 and ADC3 can generate DMA request, so it is only possible to sample single value from ADC2. As a result, both ADC2 and ADC3 sample only the current and do not use DMA.
The method described above is implemented in the firmware in following fashion.
ADC2 is used for left motor and ADC3for right motor.
ADCs are set in scan mode and channels are defined according to following table.
| Position | Sampling time | Source |
|---|---|---|
| 1 | xx | Motor current |
ADCs are set to generate an interrupt at end of conversion. In the interrupt servicing routine, the current sign is corrected and value is copied to correct phase current variable, depending on which voltage vector was active during sampling.
About the three issues, first issue can be handled when calculating the vector lengths and when converting those to individual phase controls. It is easy to check the resulting vector lengths and limit them.
The second issue is normal operation in almost all the time. However, during sector changes there would only be need to use one active vector. It is necessary to force usage of the neighboring vector at the sector borders to allow for current measurement, even though this distorts the output voltage.
Third one can be handled by triggering the analog to digital conversion after the timers doing the PWM change state. This could be done automatically for some timers in STM32F1, but to make sure all channels trigger the A/D conversion, a manual trigger must be used.
The manual trigger can be launched from the compare interrupt of the timer. Dead time must be taken
into account in the sampling position, i.e. Tsample = Tdead + Tsettling, which means that the ADC
needs to wait some time before sampling. Easiest method to do this is to sample some other channel(s)
first with long sampling time, and then sample the current.
Copyright (C) 2019 Lauri Peltonen