Good perception. Lack of attention in the lab runs rampent in these times, and I'm glad to see someone paying attention.
Chances are you have one or both of the following:
1 - The impedance of your test circuit varies with frequency and interacts with the source. After all, the signal generator has some impedance (typically 50 ohms) and in some cases, poorly designed equipment can interact with reactive loads.
2 - The generator's output naturally varies with frequency. I have an old piece of equipment. I love it dearly, but it has the same problem.
The answer is, and always will be - take more data! Keep a column for your excitation as well as your output. If it's a linear system, you can correct it in math later. Otherwise, you can adjust the source before each reading. But, take the reading anyway and make a note that you had to adjust the output and why.
Keeping track of things that seem unimportant, and teaching your staff to do the same, will likely aid you in your career. In many cases the ambient temperature, time of day, season of the year, applied torque during assembly, etc.. ended up being linked to experiment results.
Temperature is pretty obvious. Time of day tends to be related to what people are doing, how much vibration is present, and what the heating and air conditioning are doing. Season of the year - humidity and heat load. Torque - leaky fittings and poor electrical connections.
At one time, these tenants were held self evident, but now, most companies and labs take for granted that calibration stickers mean that everything in the system is working properly. It was the old school technicians, turned out at such places as GE, that embarassed me a few times and led me to understand the value of careful measurement, recording, and compensation.
I wish you well,
Mike
