Python driver for the Trinamic SIXpack 2 six-axis stepper controller over RS232/RS485, using its binary 9-byte protocol, plus a Bliss motor controller.
Written from the SIXpack 2 Manual (Trinamic, V1.01, May 2006), which is not included here. The newer TMCL modules use a different protocol, and this driver does not support them.
pip install -e .The package has no runtime dependency. The examples use pyserial.
import serial
from trinamic_driver import SixPack
with serial.Serial("/dev/ttyS9", 19200, timeout=0.05) as ser:
pack = SixPack(ser, address=0) # unit address = rotary switches
pack.unit_info() # firmware, reset flag, temperature, serial
pack.position(2) # position, action, stop flag of motor 2
pack.start_ramp(2, 5400) # trapezoidal ramp to 5400 microsteps
pack.stop(1 << 2) # $2A: target := actual position- Frames. Every command is 9 bytes: address, command, P0..P6. Values are sent least significant byte first, and there is no checksum.
- No acknowledgements. Set and motion commands get no reply, and the unit silently ignores a bad one. Only queries answer: $20, $21, $28, $30 and $43.
- Reply format is ambiguous in the manual. The $20 example reads 8 bytes
(command, then P0..P6). The command table suggests 9 bytes, with an address
first. Both are accepted by default and the first reply fixes the format
(
SixPack(reply_length=...)forces one).examples/check_readonly.pyreports which one the unit uses. - Settings are write-only and volatile. They cannot be read back and are lost on a power cycle. $43 has a reset flag: it reads 1 on the first read after a reset, then 0.
- Read side effects. Reading $43 clears the reset flag, and reading $20 clears the stop flag. The driver keeps both.
- Speed. f = 20 MHz / (clkdiv+1) · v / 2^(14+div), with v ≤ 511. clkdiv is global, div is per motor. During acceleration v grows by amax/64 every 2 ms, and amax ≤ 64·vstart.
- Never sent. The commands that change the unit itself are refused, also as raw commands: baud rate, timeout, address, power-down mode, hardware reset, flash and EEPROM ($40–$42, $45, $CC, $F2–$F7).
trinamic_driver.bliss_trinamic.Trinamic is one controller per unit, with
one axis per motor channel. The XRD-3003TT config is
bliss_config/xrd3003tt_trinamic.yml,
translated from spec.
How it behaves:
-
Serial port. A local device (
/dev/...) is opened with pyserial. bliss.comm's Serial lost bytes on this instrument's ViX line, and the SIXpack has no acknowledgements to catch a damaged frame. Remote URLs use bliss.comm. -
Programming. When an axis initialises, the controller programs it; all motors must be stopped, otherwise it refuses. By default (
program_parameters: true) that is the currents, speeds, $15 flags (home_method, spec's 16-bit encoding), reference-search parameters and null offset. Withprogram_parameters: falseit sends only what spec sends: the clock divider ($12), the stop limits ($31) and the speeds ($13/$14), and the unit keeps its own power-on settings. The controller programs the axes again after detecting a unit reset, checking the reset flag at most everycheck_intervalseconds, like spec. -
Lost replies. A query that gets no reply or a garbled one is asked again up to twice, on an emptied input buffer, and logged. Commands are never repeated, so a move is never sent twice. $43 is not simply asked again, because reading it clears the reset flag: after a lost $43 reply the controller asks again up to
unit_info_retriestimes (default 5),unit_info_pauseseconds apart (default 1), and also checks the positions, since a reset puts every motor at 0. A motor that was elsewhere and now reads 0 is treated as reset (positions lost). The pause is there because the unit sometimes falls silent for a moment (on 2026-11-02 it missed two $43 queries 0.6 s apart and answered normally a few minutes later), which used to stop a long scan series. -
Lost positions. After a reset the axes report
READY | POSLOST. Moves and jogs are refused until the axis is homed (axis.home(), the unit's reference search) or its dial is set. The state is not FAULT because Bliss refuseshome()on a FAULT axis. The lost flag is kept in Redis, so a Bliss restart cannot forget it. -
Velocity and acceleration are converted to v/amax. div is chosen per axis from the configured velocity, for the best resolution. A velocity outside that divider's range is refused (set
divto go faster). Accelerations above 64·vstart are limited, with a warning. -
Stop. $2A sets the target to the current position. A fast motor overshoots and ramps back to that point. Jogs (rotation, $25) decelerate with amax and are then ended with $23.
-
Limits. The unit's stop condition (StopNull or the analogue limits) is reported as LIMNEG/LIMPOS in the direction of the move, so the move raises
AxisOnLimitError. Withignore_limits: true(spec'sdisable_limit_checks) it is only logged. -
Homing uses the unit's reference search. Its direction comes from the
home_methodflags, not fromhome(±1).stop_after_search: true(the driver's default) stays at the reference point; the unit's default (false) drives back to the old position number. Withprogram_parameters: falsethe unit's own reference settings apply. -
Not supported:
hw_limit()(the unit has no limit search),axis.on()/off(), the PI controller, and trajectories.
Custom commands: axis.trinamic_inputs() and axis.trinamic_supply().
raw_write_read("20 02 00") takes hex bytes (command, P0..) and returns P0..P6.
Since October 2026 the controller runs the monochromator m1 and the two
absorber wheels a1 and a2 (one SIXpack 2, firmware 1.66, RS232 at
19200 baud). Confirmed there:
- Replies are 9 bytes, with the response address first.
- Positions are the unit's microstep counter; dial = steps / steps_per_unit matched spec's dial values.
- spec sends very little. A debug capture of spec showed only $40 (baud
rate) and $31 (stop limits 0/1024) at start-up, $12 (clock divider 23) and,
per move, $13/$14 (speeds) and $23. The unit otherwise runs on its power-on
settings, $15 flags 0x0AA0. spec's
motor_par dumpshows 0x60D0 instead, which is not what the unit uses: programming those flags madea2stop between wheel positions, in Bliss and afterwards in spec. The XRD config therefore usesprogram_parameters: falseandclkdiv: 23. - Absorber backlash. A backlash move to wheel position 1 overshoots below 1,
so the low limit of
a1anda2is 1 minus the backlash (0.984375).
examples/check_readonly.py /dev/ttyS9 (with spec and Bliss stopped) shows
the reply format, the supply voltage and the positions as dial values, without
moving anything.
The tests need no hardware:
python -m unittest discover -s tests- Unit tests check frames against the manual's worked examples, the
conversions, and the unit layer against a simulator (
tests/fake_sixpack.py).tests/test_bliss_trinamic.py(needs Bliss, no Beacon) checks the controller's $43 retries and reset detection. - End-to-end:
tests/bliss_e2e.pyruns Bliss axes from the XRD config against the simulator, served on a pseudo-terminal bytests/pty_sim.py(setup in the docstring). It covers programming, moves, stop, jog, the stop condition, a power cycle with recovery by homing or by setting the dial, and raw commands. It passes with both 8- and 9-byte replies.
These come from the manual's text and are tested only against the simulator:
- homing (the reference search) and jogging: the deceleration of $25 with velocity 0, and $23 ending a rotation;
- a stop condition during a move;
- detecting a real power cycle of the unit, and the retries after lost or garbled replies.