Amplitude Errors in the Summed Response of Audio Crossover Filters

[johnc124]

In order for the full advantage of feedback to be realized, the feedback signal and the input signal must follow the same path through the circuit, otherwise errors will be introduced.

If non-differential input is used, feedback is applied to emitter of input transistor, introducing a B-E discrepancy. When using differential in (circuit) non-inverting mode, there is a 2 x B-E discrepancy.
With the exception of DiFETs (OPA627/637), all opamps will have an additional source of error due to reverse bias diode-substrate leakage that exists for all components on the die (Substrate is connected to V-), and input capacitance-to-substrate modulation.

Substrate leakage artifacts are the root cause of the "Opamp Sound".

The solution is differential input where input node and feedback node are identical. This will also ensure the application of feedback does not introduce any new distortions.

Makes no difference if op amp, discrete transistor, or tube implementation.

Dielectrically isolated processes solve the problem of input capacitance variation with input signal. In fact I have measured this variation down to the 10s of femto-farads while writing an article ( http://www.ti.com/lit/an/slyt595/slyt595.pdf ) The DiFET process is not the only dielectrically isolated process though. In fact many modern complementary bipolar processes are dielectrically isolated.

The dirty little secret though is that the input capacitance of a cheap CMOS op amp can but just as linear as an expensive op amp on a DI process. And you can't completely eliminate diodes to the substrate unless you also eliminate ESD protection on the input of the op amp. Some op amps do this in the pursuit of extremely low input bias current, but they require careful handling!

 

[rayma]

In order for the full advantage of feedback to be realized, the feedback signal and the input signal
must follow the same path through the circuit, Makes no difference if op amp, discrete transistor, or tube implementation.

Yes, though few are willing to admit this. JLH discussed this many years ago.