Yes, I know CDM324 is a doppler radar module, however, Signal Path posted a nice video taking a deep dive into this module and observed that when the VCC applied is varied between 4 and 5 volts there is a swing in the oscillation frequency of about 6MHz (going from 23.901GHz to 23.907GHz). For those unfamiliar with CDM324, it has a Colpitts(?) oscillator feeding an antenna, and then the transmitted and reflected signals are mixed using a rat-race coupler.
My hypothesis was that if I apply a relatively low-frequency triangle wave modulation to VCC, I may be able to observe some distance-correlated component coming out of the mixer on CDM324.
So, I set up a quick test. I had a 10-100kHz sawtooth wave generated on AWG of my oscilloscope, that was fed into an opamp to add a DC bias via a pot on the non-inverting input (CDM324 consumes about 30 mA at 5V, so it's within the opamp output spec). The result was a sawtooth waveform fluctuating from 4V to 5V (ballpark, voltages changed and I had to tune the circuit once the load was included). The IF output was fed into an active bandpass with gain of 100 and high cutoff of about 10kHz.
Given the modulation of VCC at 10kHz and the oscillator swing of about 6MHz per 1V between 4V and 5V at VCC, I was expecting to see some component in the IF output giving me 800Hz per meter distance to target. That assumption is based on that for every meter distance to target, the reflection should be delayed by about 6.67 ns. With 10kHz sawtooth waveform going from Vmin to Vmax in 50 us, I would estimate the frequency shift of the reflected wave to be about (6.67 ns / 50 us) * 6MHz, or 800 Hz. I was expecting 10kHz to be a low enough of a frequency to not get filtered out by any RC elements in the circuit and be slow enough for the 24GHz oscillator to adjust to.
I ran the experiment and saw nothing at all. The radar still responded to moving objects as expected but there was nothing but noise showing up on the FFT in the 400-1600 Hz range. Any thoughts from RF engineers as to why this doesn't work?