Slow python DAQmx reading & writing

[e1010511@u.nus.edu]

Hi,

when I was using nidaqmx lib in Python 3.10 with USB6363(maximum 2MS/s), but the largest sample rate is only 500Hz (I was using 10000 sample rate, 25 samples in one time). My code only contains a simple PID controller. Could you help me find if anything is wrong?

sample_rate = 10000
sample_per_channel = 25

# initialize PID & device
PID = PID()
task_AI = nidaqmx.Task()
task_AI.ai_channels.add_ai_voltage_chan("Dev1/ai0:1")
task_AI.timing.cfg_samp_clk_timing(sample_rate, sample_mode=AcquisitionType.CONTINUOUS)

task_AO = nidaqmx.Task()
task_AO.ao_channels.add_ao_voltage_chan("Dev1/ao0", min_val= -3, max_val=3)
# task_AO.timing.cfg_samp_clk_timing(sample_rate, sample_mode=AcquisitionType.CONTINUOUS)

# There is also a problem here. When I try to configure timing for AO using the above line, there is an error: nidaqmx.errors.DaqError: Non-buffered hardware-timed operations are not supported for this device and Channel Type. 
# Set the Buffer Size to greater than 0, do not configure Sample Clock timing, or set Sample Timing Type to On Demand.

print("Running task. Press Ctrl+C to stop.")

def update():
    global cnt, sum_deflV, ZVolt, last_output
    av = task_AI.in_stream.avail_samp_per_chan

    row_data = task_AI.read(sample_per_channel)
    length = np.shape(row_data)
    # print(f"available data:{av}, data read: {length}", end="\r")
    PID.setDeltaTime(1/sample_rate*sample_per_channel)
    cnt += 1
    if len(row_data[0]) != 0:
        task_AO.write(ZVolt)
        data = np.array(row_data).mean(axis=1)
        data_DeflV = data[0]
        data_ZVolt = data[1]
        setpoint = 0.5
        PID.error = setpoint - data_DeflV
        PID.update()
        output = PID.output
        last_output = output
        ZVolt += output
        print(f"write:{ZVolt:+.6f}V, ZVolt:{data_ZVolt:.6f}V", end="\r")
        sum_deflV += data_DeflV
    return

Thank you so much!

[ZYOng · Accepted solution]

Using a USB device and Python is the worst combination for control.

Firstly, USB has a communication latency of 100 times slower than PCI/PXI.

Reference: Instrument Bus Performance – Making Sense of Competing Bus Technologies for Instrument Control

Python is an interpreter language and can run up to 100 times slower than C. C++ VS Python benchmarks, Which programming language or compiler is faster

For PID control, you should use a PCI(e)/PXI(e) DAQ device via LabVIEW/C/C#, preferably via NI-DAQmx Hardware-Timed Single Point Mode

[santo_13]

And if you need a very fast PID, you need to implement it on FPGA like with cRIO.

[e1010511@u.nus.edu]

Thank you so much!

And if there is anyway to change USB-6363 to a PXI/PCI connector?

[e1010511@u.nus.edu]

Thank you so much!

But it's not possible to use python on cDAQ devices, while I need to use machine learning through python on signal control. Is there any methods to deal with this problem?

[ZYOng]

And if there is anyway to change USB-6363 to a PXI/PCI connector?

It would require a change in the PCB so it is not possible. You would need to get a separate module.

But it's not possible to use python on cDAQ devices, while I need to use machine learning through python on signal control. Is there any methods to deal with this problem?

I believe that you are referring to cRIO, as mentioned by santo_13. All cDAQ (except the cDAQ-913x) do not have their own controllers and suffer from the same high latency issue. You can install and run Python on a cRIO. Just select Other programming languages when installing the software when NI MAX.

[e1010511@u.nus.edu]

Thank you so much! Now I plan buying a new PCIe device (maybe PCIe-6321 with 2 AO). I have tested the USB sample rate, which is 10kS/s for python reading, and 500S/s for python writing. So how fast would you suppose for a PCIe-6321 sample rate in python? Would it exceed 5kS/s? Thank you again for your kindly help!

[santo_13]

---

e1010511@u.nus.edu wrote:

Thank you so much! Now I plan buying a new PCIe device (maybe PCIe-6321 with 2 AO). I have tested the USB sample rate, which is 10kS/s for python reading, and 500S/s for python writing. So how fast would you suppose for a PCIe-6321 sample rate in python? Would it exceed 5kS/s? Thank you again for your kindly help!

---

The secret to achieve the highest hardware performance is to write/read in large chunks, this allows the hardware to create/use the chunk of data while your application generates or acquires a new chunk of data.