Expand description

National Instrument (NI) Integration with nicompiler_backend

National Instrument (NI) has long been a preferred choice for building experimental control systems, owing to the versatility, cost-effectiveness, extensibility, and robust documentation of its hardware. Their substantial documentation spans from system design (NI-DAQmx Documentation) to APIs for both ANSI C and Python.

While NI provides fine-grained control over its hardware, existing drivers present the following challenges:

Challenges with Existing Implementations

1. Streaming Deficiency

The NI driver, while versatile, demands that output signals be pre-sampled and relayed to the device’s output-buffer. Consider an experiment that runs for an extended duration (e.g., 10 minutes) and requires high time-resolution (e.g., 1MHz for 10 analogue f64 channels). Pre-sampling the entire waveform becomes both computationally demanding and memory-intensive (requiring around ~44.7Gb for storage). A more practical approach would be streaming the signal, where a fraction of the signal is sampled and relayed while the preceding chunk is executed. This approach reduces memory and computational overhead while maintaining signal integrity.

2. Device-Centric Abstraction

NI drivers typically interface at the device level, with software “task” entities corresponding to specific device channels. Modern experiments, however, often require capabilities that exceed a single NI card. Using a NI experimental control system consisting of multiple devices necessitates managing multiple device tasks concurrently, a problem fraught with complexity. Ideally, researchers should interface with the entire system holistically rather than grappling with individual devices and their concurrent tasks. See Device for more details on synchronization.

3. Trade-offs between High vs. Low-Level Implementation

Low-level system implementations promise versatility and performance but at the expense of development ease. Conversely, a Python-based solution encourages rapid development but may be marred by performance bottlenecks, especially when dealing with concurrent streaming across multiple devices.

Introducing nicompiler_backend

nicompiler_backend is designed to bridge these challenges. At its core, it leverages the performance and safety guarantees of Rust as well as its convenient interface with C and python. By interfacing seamlessly with the NI-DAQmx C driver library and providing a Python API via PyO3, nicompiler_backend offers the best of both worlds. Coupled with an optional high-level Python wrapper, researchers can design experiments in an expressive language, leaving the Rust backend to handle streaming and concurrency.

Currently, this crate supports analogue and digital output tasks, along with synchronization between NI devices through shared start-triggers, sampling clocks, or phase-locked reference clocks.

Example usage

Rust

use nicompiler_backend::*;
let mut exp = Experiment::new();
// Define devices and associated channels
exp.add_ao_device("PXI1Slot3", 1e6);
exp.add_ao_channel("PXI1Slot3", 0);

exp.add_ao_device("PXI1Slot4", 1e6);
exp.add_ao_channel("PXI1Slot4", 0);

exp.add_do_device("PXI1Slot6", 1e7);
exp.add_do_channel("PXI1Slot6", 0, 0);
exp.add_do_channel("PXI1Slot6", 0, 4);

// Define synchronization behavior:
exp.device_cfg_trig("PXI1Slot3", "PXI1_Trig0", true);
exp.device_cfg_ref_clk("PXI1Slot3", "PXI1_Trig7", 1e7, true);

exp.device_cfg_trig("PXI1Slot4", "PXI1_Trig0", false);
exp.device_cfg_ref_clk("PXI1Slot4", "PXI1_Trig7", 1e7, false);

exp.device_cfg_samp_clk_src("PXI1Slot6", "PXI1_Trig7");
exp.device_cfg_trig("PXI1Slot6", "PXI1_Trig0", false);

// PXI1Slot3/ao0 starts with a 1s-long 7Hz sine wave with offset 1
// and unit amplitude, zero phase. Does not keep its value.
exp.sine("PXI1Slot3", "ao0", 0., 1., false, 7., None, None, Some(1.));
// Ends with a half-second long 1V constant signal which returns to zero
exp.constant("PXI1Slot3", "ao0", 9., 0.5, 1., false);

// We can also leave a defined channel empty: the device / channel will simply not be compiled

// Both lines of PXI1Slot6 start with a one-second "high" at t=0 and a half-second high at t=9
exp.high("PXI1Slot6", "port0/line0", 0., 1.);
exp.high("PXI1Slot6", "port0/line0", 9., 0.5);
// Alternatively, we can also define the same behavior via go_high/go_low
exp.go_high("PXI1Slot6", "port0/line4", 0.);
exp.go_low("PXI1Slot6", "port0/line4", 1.);

exp.go_high("PXI1Slot6", "port0/line4", 9.);
exp.go_low("PXI1Slot6", "port0/line4", 9.5);

// Compile the experiment: this will stop the experiment at the last edit-time plus one tick
exp.compile();

// We can compile again with a specific stop_time (and add instructions in between)
exp.compile_with_stoptime(10.); // Experiment signal will stop at t=10 now
assert_eq!(exp.compiled_stop_time(), 10.);

Python

Functionally the same code, additionally samples and plots the signal for PXI1Slot6/port0/line4. The primary goal of the Experiment object is to expose a complete set of fast rust-implemented methods for interfacing with a NI experiment. One may easily customize syntactic sugar and higher-level abstractions by wrapping nicompiler_backend module in another layer of python code, see our project page for one such example.

# Instantiate experiment, define devices and channels
from nicompiler_backend import Experiment
import matplotlib.pyplot as plt

exp = Experiment()
exp.add_ao_device(name="PXI1Slot3", samp_rate=1e6)
exp.add_ao_channel(name="PXI1Slot3", channel_id=0)

exp.add_ao_device(name="PXI1Slot4", samp_rate=1e6)
exp.add_ao_channel(name="PXI1Slot4", channel_id=0)

exp.add_do_device(name="PXI1Slot6", samp_rate=1e7)
exp.add_do_channel(name="PXI1Slot6", port_id=0, line_id=0)
exp.add_do_channel("PXI1Slot6", port_id=0, line_id=4)

# Define synchronization behavior
exp.device_cfg_trig(name="PXI1Slot3", trig_line="PXI1_Trig0", export_trig=True)
exp.device_cfg_ref_clk(name="PXI1Slot3", ref_clk_line="PXI1_Trig7",
                       ref_clk_rate=1e7, export_ref_clk=True)
exp.device_cfg_trig(name="PXI1Slot4", trig_line="PXI1_Trig0", export_trig=False)
exp.device_cfg_ref_clk(name="PXI1Slot4", ref_clk_line="PXI1_Trig7",
                       ref_clk_rate=1e7, export_ref_clk=False)
exp.device_cfg_samp_clk_src(name="PXI1Slot6", src="PXI1_Trig7")
exp.device_cfg_trig(name="PXI1Slot6", trig_line="PXI1_Trig0", export_trig=False)

# Define signal
# Arguments of "option" type in rust is converted to optional arguments in python
exp.sine(dev_name="PXI1Slot3", chan_name="ao0", t=0., duration=1., keep_val=False,
         freq=7., dc_offset=1.)
exp.constant(dev_name="PXI1Slot3", chan_name="ao0", t=9., duration=0.5, value=1., keep_val=False)

exp.high("PXI1Slot6", "port0/line0", t=0., duration=1.)
exp.high("PXI1Slot6", "port0/line0", t=9., duration=.5)

exp.go_high("PXI1Slot6", "port0/line4", t=0.)
exp.go_low("PXI1Slot6", "port0/line4", t=1.)
exp.go_high("PXI1Slot6", "port0/line4", t=9.)
exp.go_low("PXI1Slot6", "port0/line4", t=9.5)

exp.compile_with_stoptime(10.)
# Returns a 100-element vector of float
sig = exp.channel_calc_signal_nsamps("PXI1Slot6", "port0/line4", start_time=0., end_time=10., num_samps=100)
plt.plot(sig)

The nicompiler_backend crate is organized into primary modules - experiment, device, channel, and instruction. Each serves specific functionality within the crate. Here’s a quick guide to help you navigate:

experiment Module: Your Starting Point

If you’re a typical user, you’ll likely spend most of your time here.

  • Overview: An Experiment is viewed as a collection of devices, each identified by its name as recognized by the NI driver.
  • Usage: The Experiment object is the primary entity exposed to Python. It provides methods for experiment-wide, device-wide, and channel-wide operations.
  • Key Traits & Implementations: Refer to the BaseExperiment trait for Rust methods and usage examples. For Python method signatures, check the direct implementations in Experiment, which simply wrap BaseExperiment implementations.

device Module: Delving into Devices

If you’re keen on understanding or customizing device-specific details, this module is for you.

  • Overview: Each Device relates to a unique piece of NI hardware in the control system. It contains essential metadata such as physical names, sampling rates, and trigger behaviors.
  • Key Traits & Implementations: See the BaseDevice trait and the entire device module for more insights. Devices also hold a set of channels, each referred to by its physical name.

channel Module: Channel Instructions & Behaviors

Ideal for those wanting to understand how instructions are managed or need to design a new TaskType as well as TaskType-specific customized channel behavior.

  • Overview: A Channel signifies a specific physical channel on an NI device. It administers a series of non-overlapping InstrBook which, after compilation, can be sampled to render floating-point signals.

instruction Module: Deep Dive into Instructions

For those interested in the intricacies of how instructions are defined and executed.

  • Overview: Each InstrBook holds an Instruction coupled with edit-time metadata, like start_pos, end_pos, and keep_val. An Instruction is crafted from an instruction type (InstrType) and a set of parameters in key-value pairs.

We encourage users to explore each module to fully grasp the capabilities and structure of the crate. Whether you’re here for a quick setup or to contribute, the nicompiler_backend crate is designed to cater to both needs.

Re-exports

Modules

  • Struct and methods corresponding to NI-hardware channels. See BaseChannel for implementation details.
  • Implements struct and methods corresponding to NI devices. See BaseDevice for implementation details.
  • The experiment module provides the highest level of abstraction for managing NI experiments, and the single place by which methods are exposed to python.
  • Provides definitions and implementations for instruction-related functionalities.
  • Provides a collection of utility functions and structures primarily focused on time-tracking and extracting specific data from string patterns.

Macros

  • A macro to generate boilerplate implementations for structs representing experiments.