National Instruments hardware over the bridge

National Instruments DAQ hardware cannot be reached from a web browser directly — there is no browser API for it. The web logger drives NI hardware through the local pydvma serve bridge: a small Python process on the lab PC that owns the nidaqmx driver and streams data to the browser over a WebSocket. See Acquisition and setup for how to start it, and Installation for the extras.

This page collects the practical NI details you need when acquiring through the bridge. It applies equally to the desktop Python interface (the same recorder code runs underneath) — see Data acquisition for the scripting API.

Requirements

• Windows or Linux (NI-DAQmx has no macOS driver).
• The NI-DAQmx driver installed from National Instruments.
• The Python bindings: pip install "pydvma[ni,serve]" (or pydvma[full]).

Start the bridge with the NI driver and open the app:

pydvma-serve --driver nidaq --open

When the bridge reports NI hardware, the app's Setup stage shows an extra NI group with the controls below, and the device picker lists your NI devices alongside any soundcards.

Devices and how they differ

pydvma has been verified end-to-end (over the bridge, on real hardware) with three representative devices. They behave differently in ways that affect your measurements:

Device	Sampling	AO timing	IEPE	Notes
USB-6003	Multiplexed (inter-channel skew)	Software-timed	No	Low-cost; single ADC scans the channel list, so channels are sampled slightly staggered. AO max ~5 kS/s.
USB-6212	Multiplexed (inter-channel skew)	Hardware-timed	No	M-series; supports shared-clock AI/AO sync.
cDAQ-9174 chassis	Simultaneous (DSA)	Hardware-timed	Yes (9234)	Delta-sigma modules with per-channel ADCs/DACs: every channel is sampled at the same instant, no skew. Example modules: NI 9234 (4-ch AI, IEPE), NI 9260 (2-ch BNC AO).

Multiplexed vs simultaneous matters for phase-sensitive work (transfer functions, mode shapes): on the multiplexed USB devices there is a small, fixed inter-channel time skew; on the DSA chassis there is none.

Addressing a cDAQ chassis

A CompactDAQ chassis is addressed as a single device, not one per module. The requested channel count is consumed across the chassis's AI modules in slot order, and AO-only modules in the middle are skipped when counting AI channels. So on a chassis with two 4-channel AI modules, asking for 8 channels spans both modules automatically, and an accelerometer on the second module's ai1 lands at capture column 5 — the same index used for its IEPE current, sensitivity and units. Full details and the channel-mapping table are in the Python acquisition guide.

Sample-rate ladders and coercion

Discrete rate ladders are the norm in measurement hardware, not an NI quirk: an audio interface runs 44.1 kHz and up in fixed steps, and delta-sigma modules (like the NI 9234) only run at rates on their internal divider ladder. The multiplexed USB devices (6212/6003) are the exception — their timebase divides finely enough to treat as continuous, though it still coerces slightly (the 6003's 80 MHz timebase turns a 48 kHz request into 48019.2 Hz).

What differs is what happens to an off-ladder request. A DSA module coerces it to the nearest legal value — requesting 8000 Hz on a 9234 can yield 8533.33 Hz, and 5000 Hz can yield 5120 Hz. pydvma adopts the true hardware rate: the recorder reads back the actual sample-clock rate and uses it for every time and frequency axis, so your data is correctly scaled even when it differs from what you asked for. In the app, Setup and Acquire show a note when the rate has been coerced, and the fs picker constrains to the selected device's ladder to reduce surprises. The output side is handled the same way: a DSA AO module (the NI 9260) coerces output_fs onto its own ladder too, so the stimulus — generated at the rate you asked for — is resampled onto the rate the AO really runs before it is played. A 30 s sweep requested at 8000 Hz on a 9234/9260 chassis therefore still sweeps its band over 30 s at 8533.33 Hz, rather than playing 6.7 % fast and stopping 2 s early. A sound card, whose ladder is published, is handled the other way round: pydvma captures at a rate the hardware really runs and resamples to the rate you asked for — see Capture rate and delivered rate.

Tip

Pick a rate from the device's ladder in the first place and the coercion never happens. If you must hit an exact arbitrary rate on NI hardware, a multiplexed device (USB-6212/6003) gets closest.

Oversampling: which rule each device follows

The digital low-pass captures above your fs and resamples down. How far above depends on whether the converter anti-aliases in silicon, which splits the NI devices:

• The 9234 is delta-sigma, so its anti-alias filter is inherent and locked to the converter rate. Content above the capture Nyquist is gone before the ADC and capturing faster rejects nothing extra, so it follows the same 'lowest' rule as an audio interface: the lowest available rate at or above 2.56 × fs.
• The USB-6003 and 6212 have no anti-alias filter at all, so a high capture rate is their only alias protection. They default to 'highest' — as fast as the device will go. On these multiplexed devices the published maximum is the aggregate rate across the channel list, so pydvma divides it by the channel count to get the per-channel ceiling.

MySettings(oversample='lowest'|'highest') overrides the default either way — 'highest' on a 9234 still buys roughly 10·log₁₀(M) dB of broadband-noise process gain, at the cost of the data volume.

Terminal configuration

The NI group exposes the input terminal configuration. The legal options depend on the module:

• Multiplexed devices support RSE (referenced single-ended, default), NRSE and differential.
• DSA modules (9234) are pseudo-differential only. If a configuration a module cannot support is requested, pydvma falls back to the module's supported configuration (with a note) rather than crashing — the app surfaces the effective mode.

Voltage ranges and output rails

• Input range (VmaxNI, default ±5 V) is the AI full-scale. Pick the smallest range that covers your signal for the best resolution; DAQmx rejects samples outside the range.
• Output rail (output_VmaxNI) is the AO full-scale. Some modules have a hard rail below the default — the NI 9260, for instance, is limited to ±4.24 V, which is less than the default 5 V. The bridge reports each device's real ai_vmax / ao_vmax, and the app clamps the input-range and output-amplitude controls to those rails, so you cannot ask for a voltage the hardware will refuse.

IEPE / ICP excitation (DSA only)

DSA modules with built-in excitation (the 9234) can power IEPE/ICP accelerometers directly. In the NI group you enable IEPE per channel; the 9234 accepts exactly 0 mA (off) or 2 mA.

Only enable IEPE where an ICP/IEPE sensor is wired

Forcing excitation current into a voltage/charge input (a force hammer, a signal generator, or a loopback to an AO output) can damage it. Leave every non-ICP channel off. Enabling a channel switches it to AC coupling and adds a ~2 s bias-settle on the first capture while the sensor's DC bias stabilises.

The scripting equivalent and a full worked recipe (IEPE accelerometers plus a force hammer on a cDAQ, with per-channel calibration) are in the Python acquisition guide.

Output stimulus and pretrigger

Through the bridge the app can play an output stimulus (sine sweep, white/uniform noise, or Gaussian noise) during a capture and can arm a pretrigger to catch transients — see Acquisition and setup. On the DSA chassis the sweep plays from the 9260 AO into the structure while the 9234 captures the response; a requested stimulus frequency above Nyquist (fs/2) is rejected with a clear message. Pretrigger crossings land at exactly the requested pre-trigger sample count on all verified devices.

What runs where

Everything on this page needs the pydvma serve bridge, because it needs the native NI driver. The no-install browser modes (the Pages app and JupyterLite) can analyse NI data you have already saved, but they cannot acquire from NI hardware — only a local Python process can. See the three modes for the full picture.