AWG Sequence Programming#
The Arbitrary Waveform Generator functionality is realized using field-programmable gate array (FPGA) technology and is available on multiple instruments types, like the HDAWG or the SHFSG. Users can operate the AWG through a program called sequencer code, it defines which waveforms are played and in which order. The syntax of the LabOne AWG Sequencer programming language is based on C, but with a few simplifications.
The user manuals for each device, that has a AWG available, provides detailed explanation of all available commands and syntax.
LabOne provides the AWG module to compile and upload a sequencer program to the device. In zhinst-toolkit this module is also available. In addition the devices with AWG support also have helper function directly implemented in the node tree.
The functionality explained in the following example are valid for all devices and are available behind the awg node.
Connect devices and access the awg node.
[1]:
# Load the LabOne API and other necessary packages
from zhinst.toolkit import Session
session = Session("localhost")
device = session.connect_device("DEVXXXX")
AWG_CORE = 0
[2]:
if device.device_type.startswith('HDAWG'):
awg_node = device.awgs[AWG_CORE]
elif device.device_type.startswith('SHFSG') or device.device_type.startswith('SHFQC'):
awg_node = device.sgchannels[AWG_CORE].awg
Compile and Upload Sequencer Program#
The Sequencer program can be uploaded as string.
[3]:
SEQUENCER_CODE = """\
// Waveform paramaters
const WFM_LEN = 1008;
const GAUSS_CENTER = WFM_LEN/2;
const SIGMA = WFM_LEN/8;
// Define waveforms
wave w0_1 = gauss(WFM_LEN, 1.0, GAUSS_CENTER, SIGMA) + marker(128, 1);
wave w0_2 = drag(WFM_LEN, 1.0, GAUSS_CENTER, SIGMA);
wave w1_1 = gauss(WFM_LEN, 0.5, GAUSS_CENTER, SIGMA) + marker(128, 1);
wave w1_2 = drag(WFM_LEN, 0.5, GAUSS_CENTER, SIGMA);
// Assign waveforms to an index in the waveform memory
assignWaveIndex(1,2, w0_1, 1,2, w0_2, 0);
assignWaveIndex(1,2, w1_1, 1,2, w1_2, 1);
// Play wave 1
playWave(1,2, w0_1, 1,2, w0_2);
// Play wave 2
playWave(1,2, w1_1, 1,2, w1_2);
"""
awg_node.load_sequencer_program(SEQUENCER_CODE)
[3]:
{'messages': '', 'maxelfsize': 2147483648}
[4]:
awg_node.enable_sequencer(single=True)
awg_node.wait_done()
Offline compilation#
When uploading the sequencer code with load_sequencer_program as a string, zhinst-toolkit first compiles the the code into a binary elf format. After that it uploads the byte code to the device. One can also use the offline compiler directly or zhinst-toolkit to compile the sequence without an active device connection.
[5]:
# Using the buildin command provided by zhinst-toolkit
elf, info = awg_node.compile_sequencer_program(SEQUENCER_CODE)
info
[5]:
{'messages': '', 'maxelfsize': 2147483648}
[6]:
# Using the zhinst.core offline compiler directly
from zhinst.core import compile_seqc
# Fetch device properties
device_type = device.device_type
device_options = device.device_options
samplerate = device.system.clocks.sampleclock.freq() if device_type.startswith('HDAWG') else None
sequencer_type = 'SG' if device.device_type.startswith('SHFSG') or device.device_type.startswith('SHFQC') else 'auto'
# Offline compilation
elf, info = compile_seqc(
SEQUENCER_CODE, device_type, device_options, index=AWG_CORE, samplerate=samplerate, sequencer=sequencer_type
)
info
[6]:
{'messages': '', 'maxelfsize': 2147483648}
[7]:
# Uploading the binary elf to the device
awg_node.elf.data(elf)
Warning
zhinst-toolkit uses by default the offline compiler provided by zhinst-core. If a feature or setting is not supported by the offline compiler (e.g. channel grouping on the HDAWG) the best way is to fallback to the awg module from Labone.
[8]:
import time
awg = session.modules.awg
awg.device(device.serial)
awg.index(AWG_CORE)
awg.sequencertype('sg' if device.device_type.startswith('SHFSG') or device.device_type.startswith('SHFQC') else 'auto-detect')
awg.execute()
awg.compiler.sourcestring(SEQUENCER_CODE)
# The following lines are not mandatory but only to ensure that everything was compiled and uploaded correctly.
timeout = 100.0 # seconds
compiler_status = awg.compiler.status()
start = time.time()
while compiler_status == -1:
if time.time() - start >= timeout:
raise TimeoutError("Program compilation timed out")
time.sleep(0.1)
compiler_status = awg.compiler.status()
if compiler_status == 1:
raise RuntimeError(
"Error during sequencer compilation. Check the log for detailed information"
)
if compiler_status == 2:
print(f"Warning during sequencer compilation {awg.compiler.statusstring()}")
# Check and wait until the elf upload to the device was successful
progress = awg.progress()
while progress < 1.0 or awg.elf.status() == 2 or awg_node.ready() == 0:
if time.time() - start >= timeout:
raise TimeoutError(f"Program upload timed out")
time.sleep(0.1)
progress = awg.progress()
if awg.elf.status() or not awg_node.ready():
raise RuntimeError(
"Error during upload of ELF file. Check the log for detailed information"
)
Sequencer Class#
zhinst-toolkit also offers a class Sequence representing a LabOne Sequence. This class enables a compact representation of a sequence for a Zurich Instruments device. Although a sequencer code can be represented by a simple string this class offers the following advantages:
* Define a constants dictionary. The constants will be added
automatically to the top of the resulting sequencer code and helps
to prevent the use of fstrings (which require the escaping of {})
* Link Waveforms to the sequence. This adds the waveform placeholder
definitions to the top of the resulting sequencer code.
(see the Waveform section below)
Note:
This class is only for convenience. The same functionality can be achieved with a simple string.
[9]:
from zhinst.toolkit import Sequence
seq = Sequence()
seq.code = """\
// Hello World
repeat(5)
...
"""
seq.constants["PULSE_WIDTH"] = 10e-9 #ns
print(seq)
// Constants
const PULSE_WIDTH = 1e-08;
// Hello World
repeat(5)
...
Waveform#
The waveform must be first defined in the sequence, either as placeholder (like in the sequencer code above) or as completely defined waveform with valid samples. In the first case, the compiler will only allocate the required memory and the waveform content is loaded later. The waveform definition must specify the length and eventual presence of markers. This should be respected later when the actual waveform is loaded.
In our case 2 waveforms have been defined. One has been assigned to index 0 and the other one two index 2.
The waveform data can be uploaded directly in the native AWG waveform format through its respective node (/.../awgs/n/wavforms/wave/m) which is also accessible through zhinst-toolkit. For more information on the native AWG waveform format take a look at the awg section in the labone user manuals. zhinst-toolkit itself offers a second, more user friendly approach through the class called Waveforms.
The Waveform class is a mutable mapping, meaning it behaves similar to a Python dictionary. The key defines the waveform index and the value is the waveform data provided as a tuple. The waveform data consists of the following elements:
wave 1 (numpy array between -1 and 1)
wave 2 (numpy array between -1 and 1)
markers (optional numpy array)
(wave 1 can be a complex array in which case the imaginary part will be treated as wave 2)
The conversion to the native AWG waveform format (interleaved waves and markers as uint16) is handled by the Waveform class directly.
[10]:
import numpy as np
from zhinst.toolkit import Waveforms
waveforms = Waveforms()
# Waveform at index 0 with markers
waveforms[0] = (0.5*np.ones(1008), -0.2*np.ones(1008), np.ones(1008))
# Waveform at index 2 without markers
waveforms[2] = (np.random.rand(1008)), np.random.rand(1008)
awg_node.write_to_waveform_memory(waveforms)
Note:
The waveform data can also be assigned via the helper function
assign_waveformwhich converts the data into the same tuple as used above. Similarassign_native_awg_waveformcan be used to assign already to a single native AWG format array converted waveform data to an index.
The same way one can upload the waveforms through a simple function one can also download the waveforms from the device
[11]:
waveforms_device = awg_node.read_from_waveform_memory()
waveforms_device[0]
[11]:
(array([0.49995422, 0.49995422, 0.49995422, ..., 0.49995422, 0.49995422,
0.49995422]),
array([-0.20001831, -0.20001831, -0.20001831, ..., -0.20001831,
-0.20001831, -0.20001831]),
array([1, 1, 1, ..., 1, 1, 1], dtype=int16))
Automatic sequencer code generation#
As already discussed the waveforms must be defined in the sequencer program before they can be uploaded to the device. In addition to convert the assigned waveforms to the native AWG format the Waveform class can also generate a sequencer code snippet that defineds the waveforms present in the Waveform object.
[12]:
print(waveforms.get_sequence_snippet())
assignWaveIndex(placeholder(1008, true, false), placeholder(1008, false, false), 0);
assignWaveIndex(placeholder(1008, false, false), placeholder(1008, false, false), 2);
Additional meta information can be added to each waveform to customize the snippet output. The following meta information are supported: * name: The name of the waveform. If specified, the placeholder will be assigned to a variable that can be in the sequencer program. * output: Output configuration for the waveform.
[13]:
from zhinst.toolkit.waveform import Wave, OutputType
waveforms = Waveforms()
waveforms[1] = (
Wave(0.5 * np.ones(1008), name= "w1", output= OutputType.OUT1 | OutputType.OUT2),
Wave(-0.5 * np.ones(1008), name= "w2", output= OutputType.OUT1 | OutputType.OUT2),
(1 << 0 | 1 << 1 | 1 << 2 | 1 << 3) * np.ones(1008),
)
waveforms.assign_waveform(2,
Wave(np.ones(1008), name= "w3", output= OutputType.OUT2),
Wave(-np.ones(1008), name= "w4", output= OutputType.OUT2),
(1 << 1 | 1 << 3) * np.ones(1008),
)
waveforms[0] = (0.2 * np.ones(1008), -0.2 * np.ones(1008))
waveforms[0][0].name = "test1"
print(waveforms.get_sequence_snippet())
wave test1 = placeholder(1008, false, false);
assignWaveIndex(test1, placeholder(1008, false, false), 0);
wave w1 = placeholder(1008, true, true);
wave w2 = placeholder(1008, true, true);
assignWaveIndex(1, 2, w1, 1, 2, w2, 1);
wave w3 = placeholder(1008, false, true);
wave w4 = placeholder(1008, false, true);
assignWaveIndex(2, w3, 2, w4, 2);
The Waveforms object can also be added to a Sequence object. This allows a a more structured code and prevents uploading the wrong waveforms. It also allows an easy declaration of the waveforms in the sequencer code since the above explained code snippet is automatically added to the sequencer code (can be disabled)
[14]:
seq = Sequence("""\
// Play wave 1
playWave(1,2, w0_1, 1,2, w0_2);
// Play wave 2
playWave(1,2, w1_1, 1,2, w1_2);
""")
seq.waveforms = Waveforms()
seq.waveforms[0] = (
Wave(0.5 * np.ones(1008), name= "w0_1", output= OutputType.OUT1 | OutputType.OUT2),
Wave(-0.5 * np.ones(1008), name= "w0_2", output= OutputType.OUT1 | OutputType.OUT2),
(1 << 0 | 1 << 1 | 1 << 2 | 1 << 3) * np.ones(1008),
)
seq.waveforms.assign_waveform(2,
Wave(np.ones(1008), name= "w1_1", output= OutputType.OUT1 | OutputType.OUT2),
Wave(-np.ones(1008), name= "w1_2", output= OutputType.OUT1 | OutputType.OUT2),
(1 << 1 | 1 << 3) * np.ones(1008),
)
print(seq)
// Waveforms declaration
wave w0_1 = placeholder(1008, true, true);
wave w0_2 = placeholder(1008, true, true);
assignWaveIndex(1, 2, w0_1, 1, 2, w0_2, 0);
wave w1_1 = placeholder(1008, false, true);
wave w1_2 = placeholder(1008, false, true);
assignWaveIndex(1, 2, w1_1, 1, 2, w1_2, 2);
// Play wave 1
playWave(1,2, w0_1, 1,2, w0_2);
// Play wave 2
playWave(1,2, w1_1, 1,2, w1_2);
Waveform validation#
The waveform definitions must match the assignment in the sequencer code. To validate if a waveform matches a sequencer code the waveform class has a validation function.
Please note that it is not mandatory to call the validation function before uploading. But especially when debugging or playing around with the waveforms it can be helpful.
The validation either takes the compiled elf file or the waveform informations from the device (only if the sequencer code is already uploaded).
[15]:
waveforms = awg_node.read_from_waveform_memory()
waveforms.validate(elf)
waveforms.validate(awg_node.waveform.descriptors())
Command Table#
The command table allows the sequencer to group waveform playback instructions with other timing-critical phase and amplitude setting commands in a single instruction within one clock cycle of 3.33 ns. The command table is a unit separate from the sequencer and waveform memory. Both the phase and the amplitude can be set in absolute and in incremental mode. Even when not using digital modulation or amplitude settings, working with the command table has the advantage of being more efficient in sequencer instruction memory compared to standard sequencing. Starting a waveform playback with the command table always requires just a single clock cycle, as opposed to 2 or 3 when using a playWave instruction.
For more information on the usage and advantages of the command table reference to the user manuals. Note that the command table is not supported on all devices that have an AWG.
The command tables is specified in the JSON format and need to conform to a device specific schema. The user manuals explain in detail how the command table is structured and used. Similar to the Waveforms zhinst-toolkit offers a helper class for the command table usage called CommandTable.
Note:
Since the command table structure is defined in a JSON schema the
CommandTableclass requires this json schema as well. Either one stores a copy of it locally or it can be accessed in zhinst-toolkit through the functionload_validation_schema
[16]:
# Create a CommandTable instance by using the schema
from zhinst.toolkit import CommandTable
ct_schema = awg_node.commandtable.load_validation_schema()
ct = CommandTable(ct_schema)
[17]:
# Alternately, load the existing command table from the device
ct = awg_node.commandtable.load_from_device()
The CommandTable class creates a pythonic approach of creating a command table that is similar to the node tree usage in zhinst-toolkit. Elements can be accessed either by value or by attribute.
Autocompletion is also available as well as on the fly validation.
[18]:
dir(ct.table[0])
[18]:
['amplitude0',
'amplitude1',
'as_dict',
'clear',
'info',
'is_empty',
'phase0',
'phase1',
'waveform']
[19]:
ct.table[0].waveform.info()
[19]:
{'type': 'object',
'properties': {'index': {'description': 'Index of the waveform to play as defined with the assignWaveIndex sequencer instruction',
'type': 'integer',
'minimum': 0,
'maximum': 15999},
'length': {'description': 'The length of the waveform in samples',
'type': 'integer',
'multipleOf': 16,
'minimum': 32},
'samplingRateDivider': {'descpription': 'Integer exponent n of the sample rate divider: SampleRate / 2^n, n in range 0 ... 13',
'type': 'integer',
'minimum': 0,
'maximum': 13},
'awgChannel0': {'description': 'Assign the given AWG channel to signal output 0 & 1',
'type': 'array',
'minItems': 1,
'maxItems': 2,
'uniqueItems': True,
'items': [{'type': 'string', 'enum': ['sigout0', 'sigout1']}]},
'awgChannel1': {'description': 'Assign the given AWG channel to signal output 0 & 1',
'type': 'array',
'minItems': 1,
'maxItems': 2,
'uniqueItems': True,
'items': [{'type': 'string', 'enum': ['sigout0', 'sigout1']}]},
'precompClear': {'description': 'Set to true to clear the precompensation filters',
'type': 'boolean',
'default': False},
'playZero': {'description': 'Play a zero-valued waveform for specified length of waveform, equivalent to the playZero sequencer instruction',
'type': 'boolean',
'default': 'false'},
'playHold': {'description': 'Hold the last played value for the specified number of samples, equivalent to the playHold sequencer instruction',
'type': 'boolean',
'default': 'false'}},
'additionalProperties': False,
'oneOf': [{'required': ['index']},
{'required': ['playZero', 'length']},
{'required': ['playHold', 'length']}]}
[20]:
ct.table[0].waveform.info("index")
[20]:
{'description': 'Index of the waveform to play as defined with the assignWaveIndex sequencer instruction',
'type': 'integer',
'minimum': 0,
'maximum': 15999}
[21]:
ct.table[0].waveform.index = 0
ct.table[0].waveform.index
[21]:
0
[22]:
from zhinst.toolkit.exceptions import ValidationError
try:
ct.table[0].waveform.index = -1
except ValidationError as err:
print(err)
-1 is less than the minimum of 0
Failed validating 'minimum' in schema:
{'description': 'Index of the waveform to play as defined with the '
'assignWaveIndex sequencer instruction',
'maximum': 15999,
'minimum': 0,
'type': 'integer'}
On instance:
-1
Each change to the command table will be validated on the fly. In addition the moment it gets converted the complete structure is validated.
[23]:
ct.clear()
try:
ct.table[0].waveform.index = 2
ct.as_dict()
except ValidationError as err:
print(err)
Once the command table is finished the upload to the device is taken care of by zhinst-toolkit
[24]:
ct.clear()
if device.device_type.startswith('HDAWG'):
ct.table[0].waveform.index = 0
ct.table[0].amplitude0.value = 0.0
ct.table[0].amplitude0.increment = False
ct.table[0].amplitude1.value = 0.0
ct.table[0].amplitude1.increment = False
ct.table[1].waveform.index = 0
ct.table[1].amplitude0.value = 0.007
ct.table[1].amplitude0.increment = True
ct.table[1].amplitude1.value = -0.007
ct.table[1].amplitude1.increment = True
elif device.device_type.startswith('SHFSG') or device.device_type.startswith('SHFQC'):
ct.table[0].waveform.index = 0
ct.table[0].amplitude00.value = 0.0
ct.table[0].amplitude01.value = 0.0
ct.table[0].amplitude10.value = 0.0
ct.table[0].amplitude11.value = 0.0
ct.table[0].amplitude00.increment = False
ct.table[0].amplitude01.increment = False
ct.table[0].amplitude10.increment = False
ct.table[0].amplitude11.increment = False
ct.table[1].waveform.index = 0
ct.table[1].amplitude00.value = 0.07
ct.table[1].amplitude01.value = -0.07
ct.table[1].amplitude10.value = 0.07
ct.table[1].amplitude11.value = 0.07
ct.table[1].amplitude00.increment = True
ct.table[1].amplitude01.increment = True
ct.table[1].amplitude10.increment = True
ct.table[1].amplitude11.increment = True
awg_node.commandtable.upload_to_device(ct)
Command table active validation#
Command table validates the given arguments on the fly by default. The feature has overhead and can be turned off to improve production code runtimes. Disabling it is good especially when creating a large command table with multiple table indexes. In any case, the command table is always validated upon upload and the instruments itself checks again errors.
On the fly validation can be disabled by:
ct.active_validation = False
Performance optimzation#
Often the limiting factor for an experiment is the delay of the device communication. If this is the case it is best trying to reduce the number of uploads. For the AWG core this means uploading everything in a single transaction.
Warning: The order is to some extend crutial. Meaning the sequencer code needs to be at the beginning and the enable call at the end.
Note: The bundling of the upload is not limited to a single awg core but can combine multiple cores.
Note: It is also possible to do the transaction on a session level so that it applies for multiple instruments.
[25]:
seq = Sequence()
seq.code = """\
// Simple playback loop
while(true) {
executeTableEntry(0);
}
"""
seq.waveforms = Waveforms()
seq.waveforms[10] = (np.zeros(1024), np.ones(1024))
ct = CommandTable(ct_schema)
ct.active_validation = False
ct.table[0].waveform.index = 10
with device.set_transaction():
awg_node.load_sequencer_program(seq)
awg_node.write_to_waveform_memory(seq.waveforms)
awg_node.commandtable.upload_to_device(ct)
awg_node.enable(True)
Optionally, the sequence can be compiled and the resulting ELF uploaded manually
[26]:
elf,_ = awg_node.compile_sequencer_program(seq)
with device.set_transaction():
awg_node.elf.data(elf)
awg_node.write_to_waveform_memory(seq.waveforms)
awg_node.commandtable.upload_to_device(ct)
awg_node.enable(True)
Multi-core programming#
So far, all the example referred to a single sequencer, but the signal generators have multiple output channels. To control all of them we must program all the relative sequencers.
This could be done just by sequentially loading the relative sequences:
[27]:
# Get a list of all the AWG cores on a device
if device.device_type.startswith('HDAWG'):
awg_nodes = list(device.awgs)
elif device.device_type.startswith('SHFSG') or device.device_type.startswith('SHFQC'):
awg_nodes = [sgchannel.awg for sgchannel in device.sgchannels]
[28]:
# Generate a list of test sequences
from textwrap import dedent
num_cores = len(awg_nodes)
sequences = [
dedent(f"""\
// Core {i:d} sequence
waitDigTrigger(1); //Wait for a trigger to syncronize all the cores
playWave(ramp(1024, 0, {(i+1)/num_cores:f}));
""")
for i in range(num_cores)
]
[29]:
# Sequentially compile and upload all the sequences
from zhinst.core.errors import CoreError
with session.set_transaction():
for index, (awg_node, sequence) in enumerate(zip(awg_nodes, sequences)):
try:
_ = awg_node.load_sequencer_program(sequence)
except CoreError as e:
print("Compilation error on core", index, e)
break
If the sequences are particularly long, it worth to compile them in parallel. The ThreadPollExecutor will use a number of threads depending on the CPU core count. Differently from generic Python code, the seqc compiler can use multiple cores at the same time, if used correctly.
Please note that in such case the set_transaction is mandatory. The LabOne API is not thread safe, and the transaction ensures that the sequences are sequentially uploaded to the device(s).
[30]:
from concurrent.futures import ThreadPoolExecutor, as_completed
with session.set_transaction(), ThreadPoolExecutor() as executor:
#Compile and upload all the sequences
# Get and store a future to verify the compilation once is done
futures = {}
for index, (awg_node, sequence) in enumerate(zip(awg_nodes, sequences)):
future = executor.submit(awg_node.load_sequencer_program, sequence)
futures[future] = index
#Checks the errors
# Iterate over all futures to verify the compilation
for future in as_completed(futures):
index = futures[future]
try:
_ = future.result()
except CoreError as e:
print("Compilation error on core", index, e)
Here a more complex example, with “heavy” sequences and associated waveforms and command tables.
As in the regular case, it’s important to respect the order of upload of sequence, waveforms and command table. This can be done by doing that after the compilations futures are done
[31]:
# Generate a more complex example, where every sequence has associated waveforms and command table
from numpy.random import default_rng
rng = default_rng()
WFM_NUM = 24 #Number of waveforms
CT_NUM = 15000 #Number of calls to `executeTableEntry`
sequences = []
command_tables = []
for i in range(num_cores):
#Generate a sequence of random `executeTableEntry`
seq = Sequence()
seq.code = dedent(f"""\
// Core {i:d} sequence
waitDigTrigger(1); //Wait for a trigger to syncronize all the cores
""")
seq.code += "\n".join([f"executeTableEntry({rng.integers(WFM_NUM):d});" for _ in range(CT_NUM)])
# Generate random waveform and associated command table to play them
seq.waveforms = Waveforms()
ct = CommandTable(ct_schema)
ct.active_validation = False
for j in range(WFM_NUM):
# Random waveform parameters, gaussian
wfm_len = rng.integers(2,100)*16
x = np.linspace(-1, 1, wfm_len)
sigma = rng.standard_normal()
ampl = rng.random()
seq.waveforms[j] = (ampl * np.exp( - x**2 / (2 * sigma**2) ))
ct.table[j].waveform.index = j
sequences.append(seq)
command_tables.append(ct)
[32]:
with session.set_transaction(), ThreadPoolExecutor() as executor:
#Compile and upload all the sequences
# Get and store a future to verify the compilation once is done
futures = {}
for index, (awg_node, sequence) in enumerate(zip(awg_nodes, sequences)):
future = executor.submit(awg_node.load_sequencer_program, sequence)
futures[future] = index
#Checks the errors
# Iterate over all futures to verify the compilation
# Then, upload waveforms and command table. This step is done here, to be sure
# such upload is done only after the sequence. Since the futures are available as soon as
# the compilation is done, their order is not predictable
for future in as_completed(futures):
index = futures[future]
try:
_ = future.result()
awg_nodes[index].write_to_waveform_memory(sequences[index].waveforms)
awg_nodes[index].commandtable.upload_to_device(command_tables[index])
except CoreError as e:
print("Compilation error on core", index, e)