July - August 2021: NI-DAQmx integration and PyPhoto21 GUI

[john-judge]

Continuation of @https://github.com/john-judge/PhotoZ_upgrades/issues/21

7/26/21

Resolved errors by keeping sampling rate non-zero and finite, and resolving a name conflict for the DO buffer outputs Next: plot the fp_data we are getting back and see if it's reasonable.

[john-judge]

7/27/21

1)

• Issue with Analog Input (FP data) -- refer to TurboSM code. Multi-threaded readout memory management needed? See SM's NI_pt pointer.
• Is the sampling fixed to read 1221 (size of outputs) per each of the 4 FP channels? Might make sense since DO and AI use the same sampling rate/clock
• Is the FP readout from NI in dimension 4x(trace length) or (trace length)x4, or has some interleave pattern from the multithreaded readout in the multiplexer? Looking thru TurboSM code first for any hints: see saveNI_data in SM_take_tb-Chun.cpp -- excerpt:

      for (m = 0; m < num_frames*BNC_ratio; m++) {
          *NI_dpt++ = *NI_spt;
          NI_spt += numBNC_chan;
      }

  where NI_dpt appears to be the destination pointer and NI_spt appears to be the source pointer. It appears that the NI data is indeed interleaved, but in a very simple way.

2)

• Issue with LED task handle -- rewrite so that NI-DAQmx stops tripping over itself.

[john-judge]

7/29/21

Frame and Trace viewers connected. Next: stimulation settings, and use this to test whether NI-DAQmx is working properly

11:05 AM

Finished connected GUI for TTL Output config to hardware module. This should pass the num_bursts, etc data on to the C++ Lib, which will use that in fillPDout to populate the DO data. We should see a change in the analog input FP data as we vary these settings. That'll be evidence of whether NI-DAQmx is set up properly.

[john-judge]

8/3/21

Successfully enabled NI-DAQmx: stimulation administered with configurable number of pulses and pulse-interval, and properly read back in via analog input.

[john-judge]

8/4/21

Reply from Chun regarding sync'ing Little Dave via NI clock:

For BNC ratio = 1, we set the camera to trigger mode (@ TXC 1) and make DO0 (black cable, resting high, low to trigger) to go low (to zero) to trigger the camera. Blue cable has the camera’s clock, which triggers the NI input.

For BNC ratio > 1, again @ TXC 1, but we use NI clock (the output and input is supposed to be synched), and DO0 (50% duty cycle per camera frame) to trigger camera (low to trigger), and NI input uses the same clock, but you can go at multiple frame rate of the camera’s.

DO0 should be "Dev1/port0/line0"

From TurboSM:

    if (NI_flag == 1) {
        resting_v = 1;
        trigger_v = 0;
    }
    else {
        resting_v = 0;
        trigger_v = 1;
    }

    for (i = 0; i < delay; ++i)
        output_data[i] = resting_v + 2;

The "+2" is done because line1 is the LED shutter -- SM uses DIgital Write U32 instead of Write Lines.

if (BNC_ratio > 1) {
        int m;
        for (; i < output_sample / BNC_ratio; ++i) {
            for (m = 0; m < BNC_ratio / 2; m++)
                output_data[i*BNC_ratio + m] = trigger_v + 2;
            for (; m < BNC_ratio; m++)
                output_data[i*BNC_ratio + m] = resting_v + 2;
        }
        for (m = (output_sample / BNC_ratio)*BNC_ratio; m < output_sample; m++)
            output_data[m] = trigger_v + 2;
        output_data[output_sample - 1] = trigger_v + 2;
    }
    else {
        for (; i < output_sample; ++i)
            output_data[i] = trigger_v + 2;
    }

[john-judge]

8/5/21

Talked to Meyer:

• For now I will focus on BNC ratio 1, but keep BNC ratio > 1 in mind for later -- it may simplify the current hardware solution to have shorter pulse durations.
• For now we will have the camera be untriggered and have the camera trigger the NI acqui/write tasks (and patch-clamping), but we will ask Masoud if he would prefer to have the patch-clamp trigger NI-USB writing (which would then trigger camera, which would then trigger analog input sampling).

[john-judge]

8/6/21

• [x] Trial-based loop and trial iteration in GUI.
• [x] Enable configuring pulse duration (attribute originally in DapChannel class)
• [ ] Replace stimulator with newer hardware to shorten pulses -- later

8/7 - 8/8

• [x] Show current slice, location, run and let user increment
• [x] Allow save directory choice

8/9/21

• [x] Filtering
• [x] Baseline correction
• [x] Point line

8/10/21

To-do:

• [x] Camera reset button similar to TurboSM's. And equivalent functionality for NI?
• [x] RLI computation and division
• [x] Point line annotations

Syncing threads during Live Feed

Mainly we want to avoid the overhead of starting/closing true parallelism -- 2.08 s for single image is high. Evidence that reassembly takes only < 0.5 % of time spent in DLL code during live feed (current state, which is too slow)

Strategy

Use two sync flags (a numpy bool array) that is shared between the plotter Python thread and the acquisition thread that calls into the C++ DLL. This establishes a producer-consumer relationship:

• The "produced_image" flag is True if the hardware has deposited a new image in the shared frame memory. The hardware will block before acquiring the next image until this flag is cleared by the plotting side. Note that this blocking can be done without synchronizing the omp threads.
• Only the omp thread with id 0 will do the reassembly. The time is negligible compared to acquisition time.
• The "stop_livefeed" flag communicates to the hardware that the infinite parallelized loop should stop. The plotting thread sets this flag when it receives the instruction from the GUI. The acqui DLL thread checks this regularly.
• Main idea: the while loop in continue_livefeed_in_background will be moved to C++ in Controller::continueLiveFeed

[john-judge]

8/11/21 - 8/12/21

• [x] Live Feed (finally!)
• [x] More trace annotations: RLI value, MaxAmp, MaxAmp/SD, latency
• [x] Computation for adjusted stim times after filtering traces: ms -> frames -> offset after shortening traces when filtering. -- via refactoring trace processing to new Trace class, with management for window cropping.

8/13/21

• [x] Visualization for stim and acqui settings
• [x] Live Feed fixes
• [x] ZDA compatibility and other small but high-priority fixes