TPS40304: Remote Sensing: Feasibility, Implementation, and Feedback Compensation

Yahya Farag said:

Is it possible to use the single-ended feedback on this regulator as a remote sense line, and if so what consideration must be made in terms of noise performance, feedback delay, and/or compensation network tuning.

Absolutely.  Single-ended remote sense has been very commonly used for decades.  This compensates for voltage drop on the supply side, but does not compensate for ground to ground differential (rise in the ground voltage from the controller to the load) like true remote sense does, and the design of the TPS40304 does not allow the ground of the controller to be referenced to the load-ground.

Yahya Farag said:

 I am expecting a feedback trace between 20-40mm from the output of the regulator to the center of the SoC.

The length of the remote sense feedback path is much less important to the loop performance and stability than the length, and potential inductance, of the power-path from the inductor to the remote sense point.  Parasitic trace inductance, along with capacitance at the load, can form a second L-C filter, adding significant phase-lag to the control loop, which may force the loop to a lower bandwidth than could be achieved without this additional lag, or when that lag was outside of the control loop.

Yahya Farag said:

what consideration must be made in terms of noise performance, feedback delay, and/or compensation network tuning.

For noise, it is important to keep the remote sense path away from high-frequency noise injectors, like the switching node, to minimize coupling.  It is also generally best to keep the remote sense line low impedance until it is close to the controller's feedback.  In practice, this means minimizing resistance between the remote sense point and the feedback line to the top resistor in the resistor divider.  A small R-C filter with a  <80ns (2MHz) time-constant, such as 10Ω + 6800pF close to the TPS40304 and before the resistor divider, can help remove high-frequency noise that might couple into the feedback without adding additional phase delay to the sensing.

If there is additional phase shift due to the trace inductance and load-side bypass capacitance, it may be necessary to reduce the loop bandwidth to 1/2 of the L-C corner frequency to maintain stability, or provide capacitive feedback from the local output voltage near the inductor, into the remote sense path after a series remote sense resistor and before the high-side resistor of the feedback divider (R78 in your schematic) 

Dear Peter

Thank you very much, your in-depth analysis is very valuable! I will move forward with this implementation.

Best,

Yahya