Core concepts of qblox-scheduler#

Overview#

Experiments in qblox-scheduler run sequences of pulses and acquisitions, returning the results in a xarray.Dataset structure.

On the user side, the package introduces these core concepts:

In this document, we shall present the HardwareAgent class, which is the fundamental interface between users and the hardware backend of the Qblox control stack. We will also present the different type of data returned by the execution of experimental sequences (named Schedules in the qblox-scheduler framework).

Each of the other core concepts listed above has its own dedicated documentation page, which we encourage the interested reader to explore for more detail.

The HardwareAgent class#

Generalities#

The HardwareAgent class is the primary interface for controlling a Qblox-based experimental setup. It provides a software representation of the hardware stack, including the quantum hardware as well as the qblox clusters and modules.

Its main responsibilities are to:

  1. Describe the hardware setup: it loads the physical layout and static settings for the instruments.

  2. Manage experiment execution: it serves as the main interface for compiling and running experiments on the hardware and retrieving the resulting measurement data.

The HardwareAgent is configured using two configuration files:

  • the hardware configuration: a required file that defines the hardware components (clusters, modules) and their instrument settings.

  • device configuration: an optional file describing the device under test (DUT) and the logical operations that can be performed on it. In the context of quantum computing, that DUT will often be a quantum processing unit (QPU).

To get started, you instantiate the HardwareAgent with these configuration files:

hw_agent = HardwareAgent(hardware_configuration, quantum_device_configuration)

Configuration files#

Detailed documentation for the hardware configuration and the device element configuration can be found on their respective pages.

Hardware configuration and pulse-level schedules#

In a minimal setup, only the hardware configuration is required. Providing it to the HardwareAgentenables the description of an experimental sequence in terms of pulses and acquisition operations, without the introducing notion of device under test and gates.

You can provide the hardware configuration to the HardwareAgent in one of three ways:

  • As a Python dict.

  • As a JSON or YAML file representing that dictionary.

  • As a QbloxHardwareCompilationConfig instance.

Example of a hardware configuration dictionary

hardware_config = {
    "version": "0.2",
    "config_type": "QbloxHardwareCompilationConfig",
    "hardware_description": {
        "cluster0": {
            "instrument_type": "Cluster",
            "modules": {
                "2": {"instrument_type": "QCM"},
                "4": {"instrument_type": "QRM"},
                "6": {"instrument_type": "QCM_RF"},
                "8": {"instrument_type": "QRM_RF"},
                "10": {"instrument_type": "QTM"},
            },
            "ref": "internal",
            "ip": "<cluster_IPv4>",
        }
    },
    "hardware_options": {},
    "connectivity": {
        "graph": [
            ["cluster0.module4.complex_output_0", "q0:port"],
            ["cluster0.module4.complex_input_0", "q0:port"],
        ]
    },
}

For more complex examples, please refer to the hardware configuration. With a valid hardware configuration, the HardwareAgent understands which modules are available and which ports can be targeted by operations. This allows for the direct compilation and execution of pulse-level schedules, as shown below:

Pulse-level schedule example


# Create a schedule
# -----------------
schedule = Schedule("Minimal pulse-level schedule")

# Play a 1 µs square pulse on "q0:port".
schedule.add(
    SquarePulse(
        duration=1e-6
        amp=0.577_215
        port="q0:port"
        )
)

# Simultaneously, start a 3 µs trace acquisition on the same port.
# A Trace acquisition captures a waveform, similar to an oscilloscope.
schedule.add(
    Trace(duration=3e-6, port="q0:port", clock="cl0.baseband"),
    ref_pt="start",
)

# Execute the schedule
# --------------------
xr_trace_acquisition = hardware_agent.run(schedule)

Executing this schedule returns the following trace acquisition data:

alt

Device element configuration and gate-level representation of a schedule#

Optionally, you can provide a device configuration to enable a gate-level description of your experiment. This file defines a collection of elements (like qubits) and their associated logical operations (gates). Each gate corresponds to a pre-defined sequence of pulses and acquisitions. Using this higher-level of abstraction simplifies the process of writing complex quantum algorithms by abstracting away the underlying pulse details.

Example of device element configuration

device_config = {
    "name": "qpu_0",
    "elements": {
        "qubit_0": {
            "name": "q0",
            "element_type": "BasicTransmonElement",
            "reset": {
                "name": "reset",
                "duration": 100e-6,
            },
            "rxy": {
                "name": "rxy",
                "duration": 40e-9,
                "amp180": 0.1,
                "beta": 0.0,
            },
            "measure": {
                "name": "measure",
                "pulse_type": "SquarePulse",
                "pulse_duration": 5e-06,
                "pulse_amp": 0.05,
                "acq_channel": 0,
                "acq_delay": 200e-09,
                "integration_time": 5e-06,
                "reset_clock_phase": True,
                "acq_weights_a": {"data": "", "shape": [0], "dtype": "float64"},
                "acq_weights_b": {"data": "", "shape": [0], "dtype": "float64"},
                "acq_weights_sampling_rate": 1.0,
                "acq_weight_type": "SSB",
                "acq_rotation": 0,
                "acq_threshold": 0,
            },
            "ports": {
                "name": "ports",
                "microwave": "q0:mw",
                "flux": "q0:fl",
                "readout": "q0:res",
            },
            "clock_freqs": {
                "name": "clock_freqs",
                "f01": 4.349_984_473e9,
                "f12": 51e6,
                "readout": 7.393_717_793e9,
            },
        },
    },
}

In this example, the BasicTransmonElement implements \(\hat{R}_{xy}\) gates using DRAG pulses, with the modulation frequency taken from the clock_freqs.f01 entry.

This dictionary describes a single-qubit QPU that supports Reset, Measure, and rotation operations (\(\hat{R}_{xy} (\theta, \phi)\)). By defining the static parameters for these gates, the HardwareAgent can transpile them into the correct pulse sequences. The available and required fields of this configuration are depending on the qubit architecture, and therefore the subsequent implemented element_type class.

qbs-struct

General structure of the qblox-scheduler user-side concepts, and the way they interconnect.#

Dataset#

Generalities#

The execution of schedules by an instance of HardwareAgent is ensured by the HardwareAgent.run method, which returns a xarray.Dataset (xarray documentation).

More details about the structure of those datasets can be found in the tutorial page about acquisitions with qblox-scheduler. The important idea is that the acquired data are stored in a “Data variables” mapping under the form of xarray.DataArrays, and the values of those data series are indexed with “Coordinates” (typically a set of experimental parameters that was applied when the data point was taken). An example of such a dataset is given below, where the xarray.DataArray “data” has values depending on the (frequency, amplitude) coordinate system.

<xarray.Dataset> Size: 48kB
Dimensions:               (acq_index_data: 1000)
Coordinates:
  * acq_index_data        (acq_index_data) int64 8kB 0 1 2 3 ... 996 997 998 999
    amplitude             (acq_index_data) float64 8kB -1.0 -1.0 ... 1.0 1.0
    loop_repetition_data  (acq_index_data) int64 8kB 0 1 2 3 ... 996 997 998 999
    frequency             (acq_index_data) float64 8kB 8e+07 ... 8.25e+07
Data variables:
    data                  (acq_index_data) complex128 16kB (-0.05443004396678...
Attributes:
    tuid:     20250911-164958-178-63babd

By default, the acquired data are assumed to be potentially sparse, meaning that they do not densily populate the grid of available parameters listed in the Coordinate section: in practice the data and coordinate series are one dimensional, and all indexed by the acq_index_<acq_channel> array of integers. Nevertheless, when the data are dense (or “gridded”) this representation can be changed with the acq_coords_to_dims than can be loaded from the qblox_scheduler.analysis.helpers module. An example is provided in the dedicated acquisitions tutorial.

Data saving#

One can also notice the presence of a time-based unique id (tuid) attached to datasets. This tuid simplifies the retrieval of data saved on a filesystem. Indeed, by default data are saved by the HardwareAgent in the $HOME/qblox-data/<YYYYMMDD>/<tuid>/ directory, where $HOME is /home/username on Linux, /Users/username on macOS, and C:\Users\username on Windows.

  • xarray.Dataset are saved in a HDF5 formatted file, with default name dataset.hdf5.

  • The saving of data can be deactivated when the HardwareAgent.run method is called: using the save_to_experiment=False optional argument.

  • The default data directory can be overridden at the HardwareAgent instantiation, using the output_dir optional argument.

  • Loading datasets from hard drive is made easy with the AnalysisDataContainer that can be loaded with

    from qblox_scheduler.analysis.data_handling import AnalysisDataContainer

    this class implements many classmethods designed to facilitate data handling. The interested user can refer to the API documentation for an exhaustive description of the data_handling module of qblox-scheduler, but just to name a few:

    • load_dataset enables loading data from their tuid

    • load_dataset does the same but but providing a path rather than the tuid

    • get_latest_tuid returns the most recent tuid

Snapshot#

In addition to the raw data saved in datasets, qblox-scheduler also features the saving of low-level hardware description and configuration. This functionality is inherited from the QCoDeS data acquisition framework, and is documented here. These snapshot collecting information about the instruments in use, are saved by default in the same directory as the dataset files ($HOME/qblox-data/<YYYYMMDD>/<tuid>/), in a snapshot.json JSON file.