Phase measuring experiments






This script is the cumulation of experiments toward learning how to measure the phase between two signals at the same frequency. I was at a loss of how to do this untill reading the excellent tutorials on costas loops by Eric Hagemann on this page. After reading about phase-lock-loops it became somewhat obvious: you mix the two real (not complex) signals and low-pass filter. If they're the same frequency, f, you end up with 2f plus a DC component indicative of their relative phase with some ambiguity (as we'll see shortly). One thing that is appearent is the magnitude of the signals matter, as simply changing the magnitude of either signal will change the filtered phase signal. So that was an opportunity to experiment with measuring signal strength and setting an amplifier gain to compensate.

The math of mixing two signals of equal frequency and differing phase:

sin(wct + φ) * sin(wct + θ) = ½ [ cos ( φ - θ) + cos( 2wct + φ + &theta) ]

where you can see there's a double frequency term cos( 2wct + φ + &theta) which gets filtered out, and the remaining term is simply the cosine of the phase difference - which matches the lab here when the two signals are in phase cos(0) = 1, and when 180 degrees out of phase cos(180) = -1. This also explains why it is difficult to accurately measure phase near 0 or 180 compared to 90 or 270 degrees - a small change of phase near 0 or 180 makes a smaller signal than the same change near 90 or 270.

This script no longer needs custom signal blocks. Instead, a sink called gr.probe_avg_mag_sqrd_X(1,.001) is used from the official gnuradio source. X=f for float or c for complex.

probe_avg_mag_sqrd_X takes an 'alpha' argument (plus an usused 'threshold' left over from squelch) for an iir filter, and reports the signal strength:

>>>meter = gr.probe_avg_mag_sqrd_f(1,.001)
>>>print meter.level()
12.509

The value 'm' returned by meter.level() is the magnitude squared divided by 2. Since the amplitude 's' of a signal from gr.sig_source_f() is PEAK, the relation between the two is: m=(.707s)**2, or s=sqrt(2m). That is, if a float source has an amplitude of 1, meter.level() returns .5 = (1**2)/2. If the source is 2, m=(2**2)/2 or 2. If s=8, m=(8**2)/2 or 32 etc. If we use meter = gr.probe_avg_mag_sqrd_c(1,.001) for a complex signal, the result returned by meter.level() is the peak squared, m=s**2.

That will be important in setting the gain of an amplifier to normalize the amplitude of the two signals before measuring their relative phase.

As far as I can tell the iir 'alpha' is a trade off between frequency and processor usage. Generally smaller is better, but more expensive, and it doesn't pay to use too small for a given frequency. All I know is to used trial and error so that the measured signal strength at the highest frequency measured doesn't bounce around too much. gr_signal_level2_ff.cc is for reading the DC phase shift signal and eleminates the iir filter.

The experiment flow graph:


sample freq = 1e6
signal freq = 1e4
signal ampl = .1

        +----(i_meter)
        |
        +----(oscope, 0)
        |
(src0)--+---------------------------------------------------+
        |                                                   |
        +----(src1)-----(mul2)---+---------------(X)--+----(X)---->A
             phase     amplitude |               pga  |    mul1
             shift      change   |                    |
                                 +----(oscope 1)      +---(oscope 2)
                                 |
                                 +----(o_meter)

                +----(meter)
                |
   A>---(lpf)---+----(oscope 3)
         x20


In the above, the signal source, 10Khz at 1Mhz sampling rate, is fed to our oscilloscope channel 0. It is also measured in amplitude at i_meter and fed to the phase multiplier. It is also altered in phase and amplitude (controllable by sliders in the gui) by src1 and mul2. The phase shift is controllable from 0 to 99 samples, which translates to 0-356 degree delay for a 10Khz signal sampled at 1Mhz. The altered signal is fed to scope channel 1, is measured by o_meter, then normalized by pga using a formula on the measured values from i_meter and o_meter. Specifically pga.set_k(math.sqrt(i_meter.level()/o_meter.level())) The normalized signal is display in scope channel 2. Then the normalized phase-shifted signal is mixed with the original signal in mul1, then low-pass filtered, measured and displayed in scope channel 3. The low-pass filter has a gain of 20 so the relatively small phase signal will show up on the same scale as the source signals. The measured values is updated in a text window in the gui every 100mSec, and the pga is also set every 100mSec to compensate for any changes to the amplitude slider.

The lab is more fun to play with using a Powermate knob to adjust the phase.

Here is the final script.

This is the display when first run:


The blue line is the original source signal, red is phase shifted or delayed 180 degress and amplified by 2 (the amplitude slider is zero to 100/25, or 0 to 4). The cyan curve is the normalized in amplitude signal, attenuated in pga to match the original signal. Finally the purple line at the bottom is the final phase measurement, -.1





In the above we have changed the delay from 180 to 135, so the measured phase signal is now -0.068





In the above we have increased the amplitude. As you can see, the i_meter, o_meter and pga have returned the cyan signal to .1 to match the original signal, and the measured phase signal is still -.068 for 135 degrees, even though the magnitude has changed.





Now the amplitude has been reduced to .02, and still the normalized signal matches the original, and the phase signal is still -.068





Now we've changed the phase to 225 degrees delay (symmetrical by 45 about 180) and the measured phase is still -.068 - that's the ambiquity mentioned above. To resolve the ambiguity in the final project, we will measure the phase shift of the device under test with two signals 90 degrees apart. That will tell us exactly what the phase actually is.





Lastly this is the signal delayed by 306 degrees, and the measured phase shift is 0.058