Poor man's PWM generator

In the signal generator application packaged with USBee SX, every waveform has to be manually 'designed'. I wanted to implement a led fading "chase" effect using Pulse Width Modulation (PWM). It is far too much work to manually design the waveforms in that case. Also, as I wanted a solution that also works on Linux, I did not use the DLL API, but instead what I've learnt up to now about the USBee protocol and put it to use with Python and libusb, a convenient way to interface to USB devices without writing kernel modules or device drivers. This is the kind of effect I wanted to achieve (works browsers supporting HTML5 canvas only):

(js source)

First, I implemented the PWM algorithm. The most basic algorithm, Time Proportioning, uses a counter that increments periodically and is reset at the end of every period of the PWM. When the counter value is more than the reference value, the PWM output changes state from high to low (or low to high). To simulate this I wrote the following piece of Python to go from an array of 8 intensity levels to 256 samples in binary format efficiently:

def pwm(intensities):
    """In: 8 intensity values 0..256, Out: 256 8-bit samples"""
    import numpy
    from numpy import newaxis
    intensity = numpy.array(intensities)
    # For a 256 sample PWM, generate array with numbers 0..255
    counter = numpy.arange(256)

    # Replicate 8 times to make 8 by 256 array, so we have one counter value for each output bit in each sample
    counters = counter[:, newaxis].repeat(8, 1)

    # Compare each value in the two-dimensional array to the intensity of the led it refers to
    # This results in a 8 by 256 array of booleans, which is cast to an integer array
    # (as numpy.packbits does not accept bool array)
    bits = numpy.array(counters < intensity, "I")

    # Pack to binary string
    data = numpy.packbits(bits, 1).tostring()
    return data

This allows setting the intensity of each of the 8 leds separately. Now, we just have to set up the hardware to start sending samples (see link at bottom of this post for the source code of the USBee class).

# Get handle to USBee SX device 0
dev = USBee(0)

# Put device in reset mode, send "signal generator" firmware, then take it out of
# reset. Wait for it to become ready.
f = open("firmware/firmware2.bin", "rb")
data = f.read()

# Send at 12Mhz
    0x01, 0x01, 0x01, 0x01, 0x02, 0x03, 0x3c, 0x07,
    0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x07, 0x00,
    0xe3, 0xe3, 0xe3, 0xe3, 0xe3, 0xe0, 0xe2, 0xe3,
    0x3f, 0x47, 0x00, 0x00, 0x00, 0x00, 0x36, 0x3f]))

Then we can perform the actual effect:

intensity = [0] * 8
delta = [x*0.33 for x in xrange(8)] # each output "lags" a bit behind the previous one
start_time = time.time()
period = 1
while True:
    t = time.time() - start_time
    t = t/period
    # Compute the eight intensities from current time using a triangle wave "bounce"
    # x=0.0   0
    # x=1.0   255
    # x=2.0   0
    # ...
    for i in xrange(8):
        x = (t - delta[i]) % 2.0
        if x >= 1.0:
            val = 256 - (x-1.0) * 256
            val = x * 256
        intensity[i] = val 
    # generate PWM samples
    wave = pwm(intensity)
    # Replicate the 256 samples 256 times, to get 64kB of data (max packet size),
    # and send to device.
    dev.write(wave * 256)

The full source (including setup code) can be found on my bitbucket page, just click on "get source" or use Mercurial. The signal generator test is test.py. I call this "Poor man's PWM" as the ideal solution would be to implement it on the micro-controller itself, instead of computing the waveform on the PC and sending all the data over USB. Then again, this is a proof of concept.

Written on May 27, 2010
Filed under