stimgrid

Your stimulator cannot produce every pulse you write down. It sets current in fixed steps and timing on a fixed clock, so every value gets rounded. For an asymmetric pulse (for example a short strong first phase and a long weak recharge phase) that rounding quietly breaks the charge balance you designed. A rate of 130 Hz may also become 129.87 Hz.

stimgrid shows what your device will actually deliver, and finds the nearest pulse on the device's grid whose two phases carry exactly equal charge. Pure Python, no dependencies.

For electrical neurostimulation and microstimulation on multi-electrode arrays (MEAs), microelectrodes and neural interfaces, with presets for the Intan RHS2116, Multi Channel Systems STG4000 and STG5, Tucker-Davis Technologies IZ2 and Cortical Labs CL, and support for any other current-controlled stimulator from its current and timing steps.

Research use only. stimgrid works from the resolution you give it (or a preset's published figures). Check the result on your own hardware and settings.

Install

Requires Python 3.9 or newer, and nothing else.

pip install git+https://github.com/sanjeevmuthusamy/stimgrid

Or from a downloaded copy, inside the stimgrid folder:

pip install .

This gives you the stimgrid command. You can also run it without installing, from the folder, as PYTHONPATH=src python -m stimgrid ....

Quick start

1. Find your device. stimgrid devices lists the built-in presets and where their figures come from. If yours is not there, give its own steps instead (see Devices).

2. Check your pulse. Give the first phase, the second phase and the rate:

   stimgrid quick --device mcs-stg4000 --amplitude-ua 7.3 --width-us 117 \
     --second-amplitude-ua 1.9 --second-width-us 450 --frequency-hz 130
   

3. Read the result (see Reading the output). Use the balanced pulse, or one of the options, in your stimulation software, or export it as a file that Intan RHX or MCS MC_Stimulus can load. Already have such a file? stimgrid check tells you whether it is balanced.

4. Save it for later or for a lab mate as a request file, and rerun it with stimgrid fit my_pulse.json.

Leave out --second-amplitude-ua and --second-width-us for a symmetric pulse; the second phase then copies the first.

Reading the output

For the quick start command above:

device: mcs-stg4000 (1.6 mA version): 20 µs time steps, 0.2 µA steps, max 1600 µA
  plan      7.3 µA x 117 µs | gap 0 µs | 1.9 µA x 450 µs
  rounded   7.4 µA x 120 µs | gap 0 µs | 2 µA x 460 µs
            charge 0.888 nC (+3.97% vs plan), net -0.032 nC, imbalance 3.4783%
  balanced  7.2 µA x 120 µs | gap 0 µs | 1.8 µA x 480 µs
            charge 0.864 nC (+1.16% vs plan), net 0 nC, imbalance 0.0000%
  option 1  7 µA x 120 µs | gap 0 µs | 2 µA x 420 µs
            charge 0.84 nC (-1.65% vs plan), net 0 nC, imbalance 0.0000%
  ...
  rate      frequency_hz 130 Hz -> 129.87 Hz (385 steps per period, -0.100%)

WARNING SG201 simple rounding leaves the pulse unbalanced (imbalance 3.4783%, net -0.032 nC per pulse)

best pulse: {"amplitude_ua": 7.2, "width_us": 120, "second_amplitude_ua": 1.8, "second_width_us": 480}
result: OK

Here the planned pulse would leave 0.032 nC on the electrode with every pulse; the balanced one leaves none and keeps charge per phase within 1.2% of the plan.

Another example: Intan RHS2116

An asymmetric pulse on an Intan RHS2116 with 0.1 µA steps, driven at 30,000 samples/s:

stimgrid quick --device intan-rhs2116 --step-ua 0.1 --sample-rate-hz 30000 \
  --amplitude-ua 7.3 --width-us 117 --second-amplitude-ua 1.9 --second-width-us 450 --frequency-hz 40

device: intan-rhs2116 (step 0.1 µA, 30000 samples/s): 33.3333 µs time steps, 0.1 µA steps, max 25.5 µA
  plan      7.3 µA x 117 µs | gap 0 µs | 1.9 µA x 450 µs
  rounded   7.3 µA x 133.333 µs | gap 0 µs | 1.9 µA x 466.667 µs
            charge 0.97333 nC (+13.96% vs plan), net 0.08667 nC, imbalance 8.9041%
  balanced  7.2 µA x 100 µs | gap 0 µs | 1.8 µA x 400 µs
            charge 0.72 nC (-15.70% vs plan), net 0 nC, imbalance 0.0000%
  option 1  7.2 µA x 133.333 µs | gap 0 µs | 1.8 µA x 533.333 µs
            charge 0.96 nC (+12.40% vs plan), net 0 nC, imbalance 0.0000%
  ...
WARNING SG201 simple rounding leaves the pulse unbalanced (imbalance 8.9041%, net 0.08667 nC per pulse)
WARNING SG203 the balanced pulse changes charge per phase by -15.7%

Simple rounding would leave 0.087 nC behind on every pulse. stimgrid lists exactly balanced pulses instead, closest to your plan first, and warns when the grid forces charge per phase to move.

Checked against Intan RHX

The intan-rhs2116 preset was compared with the source code of Intan RHX 3.5.2, Intan's own acquisition and stimulation software. RHX uses a time step of 1 / sample rate and a current step equal to the chosen step size, and it sends each phase as the nearest whole number of samples and current steps, with halves rounded up. That is the same rule stimgrid uses.

The one difference is in RHX's stimulation dialog, which rounds typed values from a step shortened to three decimals (33.333 µs at 30,000 samples/s). Across 300,000 test values, 42 landed one sample apart, all at 30,000 samples/s and all within about 0.01 µs of a half-sample. For example, 383.33 µs gives 11 samples (366.67 µs) in stimgrid and 12 samples (400 µs) in RHX. At 20,000 and 25,000 samples/s the two agreed on every value. If you type widths into RHX, use the rounded or balanced widths stimgrid prints; they are whole samples, so RHX keeps them as they are.

This is a comparison with RHX's source code, not a measurement on hardware.

How it works

1. Rounded. Each amplitude, phase width and gap is rounded (halves up) to the device's steps and clamped to its ranges. This is what you get if you type the plan in and let the device round it.
2. Balanced. stimgrid searches small changes to both phases, mostly the second phase's width and amplitude, the usual balancing knobs. On a gridded device, balance is checked exactly, as whole current steps times whole time steps, so "balanced" means equal to the last bit. On a device with continuous amplitude it also tries the first-phase amplitude that keeps your planned charge exactly.
3. Ranking. Exactly balanced candidates are ranked by closeness to your plan, with first-phase charge counted double because it sets the effect. The best and three alternatives are shown.
4. Rates. Each rate is rounded to a whole number of time steps per period, and the delivered rate is reported.

Command options

stimgrid quick takes the pulse as flags. Amplitudes are in µA, times in µs, rates in Hz.

Both quick and fit also take:

stimgrid --help and stimgrid quick --help list them too.

Export to your stimulation software

Instead of typing the fitted pulse into your stimulation software, --export writes it as a file that software can load. It works with both quick and fit:

stimgrid fit examples/intan_asymmetric.json --export intan-rhx --out stim.xml --channels A-000,A-001 --pulses 10
stimgrid fit examples/mcs_stg4000_asymmetric.json --export mcs-ascii --out pulse.txt --pulses 100

The report still prints as usual, and the file is written next to it. With --out - only the file contents are printed. Nothing is exported if the fit has errors.

Intan RHX. Load the file with File > Load Stimulation Settings (not Load Settings, which expects a full RHX settings file and warns that no signal groups were found). It sets, for each listed channel: stimulation on, the shape (biphasic, or biphasic with interphase delay when there is a gap), the polarity, both phase durations and amplitudes, and for a train the number of pulses and the train period. Trigger, amp settle and charge recovery are left out, so the values already set in RHX stay. Other channels are not touched.

• The file names the sample rate and step size it was fitted for. RHX ignores every stimulation setting in the file, with only a warning, if these differ from the running system, so use the same --sample-rate-hz (20,000, 25,000 or 30,000) and --step-ua as in RHX.
• Durations are written in µs to six decimals. RHX converts them to the nearest whole sample, which is the sample count stimgrid fitted.
• stimgrid will not write what RHX would reject: phases or gaps over 5,000 µs, more than 256 pulses, or a train period shorter than the pulse.

MCS MC_Stimulus. Import the file in MC_Stimulus II or III. Each pulse is written as rows of current (µA) and duration (µs): the first phase, the gap if there is one, the second phase, then zero current for the rest of the period when a rate is given. Every pulse in the train is written out, because the import format has no repeat count.

• Importing overwrites the open worksheet, and channels that are not in the file come out empty. Save your current setup first.
• The import takes times in multiples of 20 µs on both the STG4000 and the STG5, although the STG5 itself has a 5 µs time step. If an STG5 pulse is not on 20 µs steps, stimgrid says so; fit it again with --time-step-us 20.
• The manuals do not say which sign is cathodic on your electrode. --first-phase sets the sign of the first phase; check the waveform preview in MC_Stimulus.

Check the loaded values in the vendor software before stimulating. The file formats were written from the vendors' published descriptions. An intan-rhx file for the Intan example (10 pulses at 40 Hz) was loaded in the RHX 3.5.2 RHS demo at 30,000 samples/s and 0.1 µA steps, and RHX showed the exported values exactly (7.2 µA × 100 µs, 1.8 µA × 400 µs, cathodic first, 10 pulses, 25 ms period) and reported the charge as balanced. No mcs-ascii file has been loaded in MC_Stimulus yet.

Check an existing file

stimgrid check reads a stimulation file you already have and reports, channel by channel, what the stimulator will deliver and whether each pulse is charge-balanced. It reads the same two formats that export writes: Intan RHX stimulation settings (.xml, saved with File > Save Stimulation Settings) and MC_Stimulus II / III ASCII files (format 4, current mode).

stimgrid check stim.xml

file: stim.xml: Intan RHX, 30000 samples/s, 0.1 µA steps
A-000:
  in file   7.3 µA x 117 µs | gap 0 µs | 1.9 µA x 450 µs  (100 pulses every 25000 µs)
  delivered 7.3 µA x 133.333 µs | gap 0 µs | 1.9 µA x 466.667 µs
            charge 0.97333 nC (+13.96% vs file), net 0.08667 nC, imbalance 8.9041%
  balanced  7.2 µA x 100 µs | gap 0 µs | 1.8 µA x 400 µs
            charge 0.72 nC (-15.70% vs file), net 0 nC, imbalance 0.0000%
  WARNING SG106 width_us 117 becomes 133.333 after rounding (+14.0%)
  WARNING SG203 the balanced pulse changes charge per phase by -15.7%
  net charge over the whole channel: 8.667 nC (NOT balanced)
A-001: off: stimulation is off

summary: 2 channel(s): 1 checked (1 not balanced), 1 off, 0 skipped
result: NOT OK

• in file is what the file says; delivered is that pulse rounded to the device's steps, the way RHX and MC_Stimulus do it (nearest step, halves up). Here RHX turns 117 µs into 4 samples at 30,000 samples/s, and the pulse leaves 0.087 nC behind every time, 8.7 nC over the train.
• balanced is the nearest exactly balanced pulse, shown only when the delivered pulse is not balanced.
• Channels that are off, use a shape stimgrid does not check (such as triphasic), or leave values out of the file (RHX then keeps what it already has) are listed but not checked.
• It also flags what would stop the file working: in RHX, a phase over 5,000 µs, more than 256 pulses, or a train period shorter than the pulse (RHX then skips the channel with only a warning); in MC_Stimulus, times that are not multiples of 20 µs.

The exit code is 0 when every checked channel is balanced and nothing would stop the file working, and 1 otherwise, so stimgrid check can run in a script before each experiment.

In an MC_Stimulus file, stimgrid splits each channel into pulses: a phase, then an opposite-sign phase straight after it or after one zero-current row (the gap). Identical pulses are counted together. A phase with no opposite phase after it is reported as monophasic, and so not balanced.

Whole protocols

A request file can describe a whole protocol instead of one pulse: a list of steps, each a single pulse, a train, a set of bursts or a rest. stimgrid fit fits every step's pulse and adds what only shows across steps: how long each step really takes once its rates fall on whole time steps, how many pulses each channel gets, and the charge per phase and net charge over the whole protocol.

{
  "stimgrid": 1,
  "device": {"preset": "mcs-stg4000"},
  "protocol": [
    {"name": "train", "pulse": {"amplitude_ua": 7.3, "width_us": 117, "second_amplitude_ua": 1.9, "second_width_us": 450},
     "pulses": 1300, "frequency_hz": 130, "channels": [1, 2]},
    {"name": "pause", "rest_s": 10},
    {"name": "theta", "pulse": {"amplitude_ua": 20, "width_us": 100},
     "bursts": 10, "pulses": 4, "intra_burst_hz": 100, "burst_rate_hz": 5, "channels": [1]}
  ],
  "repeat": 3
}

stimgrid fit examples/mcs_stg4000_protocol.json

device: mcs-stg4000 (1.6 mA version): 20 µs time steps, 0.2 µA steps
protocol: 3 step(s), run 3 times, using the best pulse of each step

step 1 train: 1,300 pulses at 130 Hz on channels 1, 2
  plan      7.3 µA x 117 µs | gap 0 µs | 1.9 µA x 450 µs
  using     7.2 µA x 120 µs | gap 0 µs | 1.8 µA x 480 µs
            charge 0.864 nC (+1.16% vs plan), net 0 nC, imbalance 0.0000%
  rate      frequency_hz 130 Hz -> 129.87 Hz (-0.100%)
  time      10 s planned, 10.01 s delivered (+0.100%)
  charge    1.123 µC per phase over 1,300 pulses, net 0 nC (balanced)
...

time: 1 min 6 s planned, 1 min 6.03 s delivered (+0.045%)
channel 1: 4,020 pulses, 3.61 µC per phase, net 0 nC (balanced)
channel 2: 3,900 pulses, 3.37 µC per phase, net 0 nC (balanced)
result: OK

Here 130 Hz on the STG4000's 20 µs grid becomes 129.87 Hz, so the 10 s train takes 10.01 s, and the whole protocol runs 30 ms longer than planned.

At the top level, repeat runs the whole list more than once, and use (or --use) picks the pulse each step uses: best (the balanced pulse, the default) or rounded (the planned pulse simply rounded, to see what it would leave behind). max_imbalance, keep_first, search_steps and electrode work as in a single-pulse request.

• A train's time is pulses × period, its last period included; a burst step's is bursts × burst period. Rests are taken as given.
• A burst that does not fit in one burst period is an error (SG601).
• The exit code is 0 when no step has an error and every pulse used is balanced, so stimgrid fit protocol.json --use rounded fails if simple rounding would leave charge behind.
• --format json gives every step and total. --format pulse and --export work on one pulse only, so they are not available for a protocol.

Charge density and Shannon k

Give the electrode's size and fit, quick and check also report the charge density of the larger phase and its Shannon k value, k = log10(Q) + log10(Q / A), with Q the charge per phase in µC and A the electrode's geometric area in cm² (Shannon, 1992).

stimgrid quick --device mcs-stg4000 --amplitude-ua 100 --width-us 200 --electrode-diameter-um 30

  electrode 706.858 µm²: 20 nC per phase, 2829 µC/cm², Shannon k 1.753 (limit 1.75 allows 19.94 nC)

WARNING SG501 Shannon k 1.753 is above 1.75 (20 nC per phase, 2829 µC/cm² on 706.858 µm²); at most 19.94 nC per phase stays within it

• fit and quick report the best pulse; check reports the pulse the file delivers.
• Going over the limit is a warning (SG501), so the exit code stays 0 unless you add --strict.
• A request file can carry it too: "electrode": {"diameter_um": 30, "k_limit": 1.75}. Flags on stimgrid fit replace the file's values.
• Shannon's line was fitted to damage data from macroelectrodes. For small microelectrodes it is not a reliable guide, and the limits that apply depend on the electrode material and the tissue. Treat k as one input to your own judgement, not as a safety limit.

Devices

All presets are current mode. The MCS range is the model version you bought (STG4000) or the range you select (STG5). The STG4000 160 µA version is left out because its official datasheet could not be confirmed.

Not included yet, because the vendor does not publish a current step or a timing step:

If you have one of these, describe it with your own numbers (below). To have it added as a preset, request a device with the values from its manual or datasheet; the same form works for any stimulator not listed.

stimgrid devices prints each preset's source. The RHS2116 has no pulse timers; the host sets every edge, so its timing grid depends on your host software (Intan RHX matches it; see Checked against Intan RHX). Any other stimulator can be described by its own numbers:

stimgrid quick --time-step-us 10 --amplitude-step-ua 0.5 --max-amplitude-ua 100 \
  --amplitude-ua 20 --width-us 105 --second-amplitude-ua 4 --second-width-us 520

Fields: time_step_us (required), amplitude_step_ua (leave out for continuous), max_amplitude_ua, min_phase_us, max_phase_us, max_charge_per_phase_nc, max_frequency_hz.

Request files

{
  "stimgrid": 1,
  "device": {"preset": "intan-rhs2116", "step_ua": 0.1, "sample_rate_hz": 30000},
  "pulse": {"amplitude_ua": 7.3, "width_us": 117, "second_amplitude_ua": 1.9, "second_width_us": 450},
  "rates": {"frequency_hz": 40},
  "max_imbalance": 0,
  "keep_first": false,
  "electrode": {"diameter_um": 15}
}

electrode is optional (see Charge density and Shannon k).

stimgrid fit examples/intan_asymmetric.json
stimgrid fit examples/intan_asymmetric.json --format json
cat request.json | stimgrid fit - --format pulse

A preset takes its options (step_ua, sample_rate_hz or range_ua, as in the table) and any device field to override, for example {"preset": "intan-rhs2116", "step_ua": 0.1, "max_amplitude_ua": 10} or {"preset": "mcs-stg5", "range_ua": 160}. Invalid input names the exact place, for example pulse: width_us must be a positive number.

A request file with protocol instead of pulse describes a whole protocol; see Whole protocols.

--format pulse prints only the best pulse, in the common shape (amplitude_ua, width_us, second_amplitude_ua, second_width_us, gap_us), ready to paste into a protocol file or another tool.

Findings

Exit codes

From Python

from stimgrid import Pulse, Rates, fit, intan_rhs2116

report = fit(Pulse(7.3, 117, second_amplitude_ua=1.9, second_width_us=450),
             intan_rhs2116(step_ua=0.1, sample_rate_hz=30000), Rates(frequency_hz=40))
print(report.balanced.as_pulse(), report.balanced.imbalance)

To export from Python:

from stimgrid import export

device = intan_rhs2116(step_ua=0.1, sample_rate_hz=30000)
report = fit(Pulse(7.3, 117, second_amplitude_ua=1.9, second_width_us=450), device, Rates(frequency_hz=40))
with open("stim.xml", "w") as fh:
    fh.write(export(report, device, "intan-rhx", channels=["A-000"], pulses=10))

To check a file from Python:

from stimgrid import check_file

result = check_file("stim.xml")
for channel in result.unbalanced:
    print(channel.channel, channel.net_charge_nc)

With an electrode:

from stimgrid import Electrode, mcs_stg4000

report = fit(Pulse(100, 200), mcs_stg4000(), electrode=Electrode(diameter_um=30))
print(report.density.density_uc_cm2, report.density.shannon_k, report.density.max_charge_nc)
result = check_file("stim.xml", electrode=Electrode(area_um2=2000))

A protocol:

from stimgrid import load_protocol, fit_protocol

protocol, device, options = load_protocol("examples/mcs_stg4000_protocol.json")
report = fit_protocol(protocol, device, **options)
print(report.delivered_us / 1e6, [(c.channel, c.pulses, c.net_charge_nc) for c in report.channels()])

Limits of this tool

• It models one pulse shape: rectangular phases with an optional gap. It does not model DAC settling, electrode impedance, channel-to-channel mismatch or the RHS2116's trim settings.
• Rates are assumed to fall on the same time grid as phase widths.
• Continuous amplitude means the device's real amplitude resolution is unknown; the reported amplitude may need rounding again on the hardware.

Citing

If you use stimgrid in published work, please cite it. GitHub's "Cite this repository" button gives the reference from CITATION.cff:

Muthusamy S. stimgrid (version 0.4.0) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.23694989

This DOI always resolves to the latest version; each release also has its own DOI on Zenodo.

Tests

python -m unittest discover -s tests

Licence

MIT. See LICENSE.

Trademarks

Intan, RHX, Multi Channel Systems, MC_Stimulus, Tucker-Davis Technologies, Synapse, Cortical Labs and other product names are trademarks of their owners. They are used only to say which devices and software stimgrid works with. stimgrid is an independent project, not affiliated with or endorsed by any of these companies.