I have heard people saying that you must place the signal injection resistor (10 to 100 R) between a high impedance point and a low impedance point. The usual place for voltage loop is between the output (low impedance) and the feedback resistor divider (high impedance).

Could someone explain the reasons behind this ? It seems logical to think that as long as the disturbance is injected into the loop, you should be able to measure the loop gain no matter where the injection takes place. I have had mixed success with this, sometimes it would work and sometimes not. So now I try to always inject directly after a buffer or the output.

Breaking the control loop in a place where there is high impedance looking forward and low impedance looking backward insures that the injection signal will be forced into the high impedance and the measurement taken on the low impedance will be the response from the injection. The closer the two impedances are together, the more difficult and suspect the control loop measurement will be. If the injection signal doesn't has low impedance for its reference, then the measurement could be partially injection signal. Injecting in a spot where there is high impedance looking forward and low impedance looking backwards insures that the signal measured on the high impedance is injected signal or input and that the signal measured on the low impedance is control loop response or output.

Thank you Paul for your explanation. It is now much clearer. This explains why sometimes my measurements looked "wrong".

When you inject a signal into a closed loop between a source impedance Z1 and a load impedance Z2 and measure the ratio of the upstream and downstream voltages, the result is Z1/Z2 -A. Where A is the gain function. If Z1 is much less than Z2, then the result approaches -A. If there is not a convenient injection point where Z1<<Z2, then A can be derived from this formula but don't forget this is complex arithmetic.

Oops! I got my signs twisted. The correct result is A-Z1/Z2 and as Z1 becomes small compared to Z2 the result approaches A.

Other considerations regarding the injection location include S/N ratio and injection drive amplitudes...ie, you don't want to saturate the control loop amplifier just to stay above the noise floor.. You also don't want too low of a signal amplitude, then your BODE plots will be noisy all mangled about... The location you mention right at the top of the sampling divider is popular location for all the many reasons, provided you buffer has the proper iso... I typically include a 50 Ohm resistor in my schematics with Test points by default at that same location...

There are times when you can inject your signal in series at the output of the error amplifier that is at ground level ...for example: when an output of the PWM may be at 5kV.... If you use a buffer, make sure to consider the Bandwidth of the buffer, ie it's contribution of phase shift to your measurement..