Today I looked at the voltages at the output of the feedback loop op-amps and at the COMP input of the R7731A. Since there are no significant load changes, these voltages should be pretty stable, so I measured them with a voltmeter. I also measured the 8.4V raw output because it is used as the bias voltage for the PH2 opto-coupler. The results were as follows:
PH2-2 PH2-4 8.4V raw (DC) 8.4Vraw (AC+DC)
------------ ------------ ------------------ ----------------------
100VAC 7.192V 1.620V 8.41V 8.356V
125VAC 7.192V 1.565V 8.41V 8.356V
140VAC 7.193V 1.535V 8.41V 8.356V
From the above I noticed that the only significant variations were at PH2-4, which is connected to the COMP input of the R7731A. This voltage controls the duty cycle, the lower the voltage, the lower the duty cycle. This is supposed to happen when the load is light and the output voltage rises. A lower duty cycles delivers less energy and brings the output voltage back to its nominal value. However, the COMP input is supposed to be controlled by the feedback loop op-amps via the opto-coupler. Neither one of the voltages driving the opto-coupler's LED is varying, so the COMP input shouldn't vary either, in theory. Of course, the input AC voltage is varying, and as we know from previous observations, this has caused other surprises, like causing the R7731's osc frequency to vary. So somehow, it seems to be causing this effect too.
Note that it's normal for the duty cycle to vary when the AC input voltage varies, but on current mode PWM controllers like the R7731A, this is accomplished via the effect of the current sense (CS) input. A higher AC input voltage causes more current to flow through the primary, which in turn causes the CS threshold to be reached sooner, and as a result causes Q1 to be turned off sooner.
Hoping that viewing the voltages with a scope rather than just the voltmeter may reveal something that I may have been missing, I decided to capture waveforms for the previously measured voltages. PH2-2 and the 8.4V raw output are on the secondary side and can be easily captured using GND-C as the reference. PH2-2 is a different story because it is on the primary side and is referenced to GND-A. I did try to capture it anyway, but the results are not trustworthy because you again have to use the math function and the signals are low level. So I'm not going to post that capture. The captures for PH2-2 were done using average mode to eliminate some of the random noise from the results, the 8.4V raw output was captured using sample mode. Again, I couldn't see any significant variations that could justify the voltage variations on PH2-4. So why the voltage at PH2-4 varies is still a mystery to me!