Bonsai Realtime stimulation graph

[soyyso]

Hello,
I’m planning an optogenetics experiment. I’ve already built the nodes to output pulses, but after switching to a Harp node, I can no longer see the graph when I use Show Visualizer.

I’d like to have a real-time time axis running, and have each pulse stimulation appear on the graph as it happens. Has anyone implemented something like this in Bonsai, or does anyone have suggestions for how to achieve it?

Thank you.

First screenshot: When I press KeyDown, the stimulation is triggered and the timestamp and Train timestamp are written to a CSV. Pressing the other KeyDown again immediately stops the stimulation.
Second screenshot: This is the node setup used to generate/output the pulse.

[banchan86]

Hi @soyyso, here is one way to do it!

First initialize a BehaviorSubject called StimValue with a Boolean.

For each time that you trigger a Boolean, you can add a MulticastSubject to update StimValue. As a Sink, it will not modify the input/output value.

Then, you can use a Timer with a WithLatestFrom to continously poll StimValue for a RollingGraph from the Bonsai.Gui package. You can modify the Period in Timer to adjust the update speed, and Capacity in RollingGraph to determine the number of points to display.

Uploaded the toy example here:
stim-plot.zip

[soyyso]

Hi!
I really really appreciate your generous, kind, and detailed explanation. It was extremely helpful. you're my Hero!!

It works! but I have a small issue with my Bonsai workflow.. on the graph, sometimes the pulses don’t appear as a TTL-like square waveform but instead look like sharp peaks, and the intervals also appear irregular. Could this be an issue with the node configuration, or is it more likely a problem with the graph, visualization output?

I will also upload the Bonsai file that contains that issue.

Thank you again for your support!

GRAPH.zip

[banchan86]

@soyyso Aww thank you for the compliment! Thank you for being an active user of Bonsai!

Now that I have the full workflow on hand, I can see some other issues. Sorry I didn't catch them earlier! We will put aside the visualization first and discuss it later:

1. Using Delay for pulse width:

I see that you are trying to set up a pulse train, where the pulse width is 20ms, and the pulse frequency is 20 Hz (50ms period for the timer). Unfortunately, since Windows is a non-real-time operating system, there is always some jitter and latency when executing commands and sending messages. At these timescales, that level of precision for the pulse train will not be achievable by sending a delayed Clear command from Bonsai, and you can see anywhere from 10-15ms jitter/latency, so your pulse width, for example, may end up being >30ms.

Ideally, for that level of precision, you want to set up a stimulation protocol in hardware, like an Arduino or a National Instruments board. The Harp Behavior board has a dedicated pulse mode, where you can set a pulse duration in milliseconds, but unfortunately, as far as I know, there is no corresponding way to set a dedicated pulse frequency (happy to be corrected here!). We can set up the pulse mode for that digital output, so that at least the pulse width will be more accurate.

We first set up a device pattern using subjects to enable us to send commands and receive messages from anywhere else in the workflow.

We then send commands to enable pulse mode for DO1 and set the pulse width. We use a SubscribeWhen to make sure the device is ready to receive messages before sending those out.

For pulse mode in the Train node, we only need to send a Behavior.OutputSetPayload via DeviceCommands to trigger the pulse. There is no need to clear the payload. Technically, we don't need to send a Boolean for Behavior.OutputSetPayload, but we are keeping the Boolean, Delay, and StimValue in here to simulate the pulse train for the visualizer (more on that later), as there is currently no way to record the falling edge of the pulse.

2. Visualizer

So at these timescales, the polling method doesn't work that well, which is why you see the sharp peaks and irregular intervals. Ideally, you want to mirror the digital output into a digital input line so that you can get a more accurate readout, and the overall workflow can then be simplified, as you don't need the bits that simulate the pulse train for the visualizer. If you do want to keep it, though, then you can simply decrease the period in Timer to poll faster, but there is a limit to how fast you can go. Keep in mind that it is also just for visualization, and the actual pulse width and pulse onset will be different in hardware.

3. Logging

For the reasons I mentioned, using the standard Timestamp operator is inaccurate. With the Harp behavior board, you actually have access to hardware timestamps for those events, which you should record, which I cover more in this comment (https://github.com/orgs/bonsai-rx/discussions/2504#discussioncomment-15573659).

I attached the modified workflow; there's more stuff in there that I didn't have the time to cover. I hope it makes sense, let me know if you have any questions!

GRAPH_modified.zip

[soyyso]

Hi, thank you so much for kindly creating and sharing the workflow file! I truly appreciate it.

I’m sorry for the late reply, I took some time to carefully read through the workflow you sent and the information on the Harp website. Your guidance helped me a lot, and I want to sincerely thank you for pointing things out so clearly.

If I have any further questions, I’ll reach out again.
I’d honestly love to treat you to a meal as a small thank-you. If there’s ever an opportunity, I would really like to return the favor.

Thank you again for all your help!^0^