Opamp phase shift - open and closed loop.

[Circlotron]

As we all know, any opamp has a certain amount of phase shift that gets greater with frequency. Manufacturers publish these figures for open loop. What happens to the phsae shift when the loop is closed though? Obviously the phase shift still exists inside the feedback loop, but what if you measure it outside the loop? Does it decrease in proportion to gain reduction like distortion does? Does it stay the same?

[peranders]

If you have a flat response and you are not near the "drop" the phase shift is zero, but inside the amp and the feedback loop the phaseshift is probably 90 degrees up to 120-140 degrees at - 3 dB.

[Steve Eddy]

Originally posted by peranders
If you have a flat response and you are not near the "drop" the phase shift is zero, but inside the amp and the feedback loop the phaseshift is probably 90 degrees up to 120-140 degrees at - 3 dB.

And it's worth noting that unless you're considering amplifier stability, phase shift is meaningless in itself. A constant delay (i.e. one which delays all frequencies equally) results in phase shift, yet all frequencies arrive at the same time. And this will be the case as long as the phase shift is linear with frequency.

That's why group delay is really the only meaningful figure of merit when it comes to accurate reproduction.

se

[Pjotr]

Originally posted by Steve Eddy
That's why group delay is really the only meaningful figure of merit when it comes to accurate reproduction.

se

That's why constant group delay is really the only meaningful figure of merit when it comes to accurate reproduction.

But is it the only meaningful figure? Dropping open loop gain with frequency results in reducing distortion reduction with frequency.

[Pedja]

Hi Circlotron and others,

Open and closed loop phase shifts are related. Then, both are related to the opamp’s bandwidth. The higher the bandwidth, the lower the phase shifts at given frequency. Also, opamp set to lower gain has higher bandwidth and hence lower phase shift.

Here are the numbers. With gain of 20, LM6172’s (100MHz at unity gain opamp) phase shift at 20kHz is less than 0.2deg, and group delay (difference between 1kHz and 20kHz) is notably less than 1picosecond. In the same circuit OP27 (8MHz at unity gain) has more than 3degs phase shift and difference in group delay is more than 1ns. The same OP27 when works with gain of 4 has phase shift less than 0.5deg and group delay of a few picoseconds.

If you check these numbers calculating these shifts from corresponding group delays you’ll se, as Steve mentioned, the bulk of these phase shift numbers are caused by the propagation delay and not by the non-linear group-delay. But again, propagation delay is in relation with the non-linear group delay.

Pedja

[Steve Eddy]

Originally posted by Pjotr
That's why constant group delay is really the only meaningful figure of merit when it comes to accurate reproduction.

Well yes, a constant group delay would be the ideal. My point though was that phase shift in and of itself tells you nothing particularly meaningful.

But is it the only meaningful figure? Dropping open loop gain with frequency results in reducing distortion reduction with frequency.

No, it's not the only meaningful figure of merit. The context of my comments was limited to phase versus group delay. They weren't meant to imply that group delay is the only meaningful figure of merit to the exclusion of all else. Only phase.

se

[Steve Eddy]

Originally posted by Pedja
If you check these numbers calculating these shifts from corresponding group delays you’ll se, as Steve mentioned, the bulk of these phase shift numbers are caused by the propagation delay and not by the non-linear group-delay.

Yeah, the notion that ANY delay, even a constant delay which delays all frequencies equally, results in phase shift is not a terribly intuitive one.

The phase shift for a given frequency and a given delay is a simple calculation of the time delay times the frequency times 360 degrees.

For example, with a constant delay of 12.5 microseconds, the phase at 20 Hz will be:

0.0000125 x 20 x 360 = 0.09 degrees.

At 200Hz it will be:

0.0000125 x 200 x 360 = 0.9 degrees.

At 2kHz it will be:

0.0000125 x 2,000 x 360 = 9.0 degrees.

And finally at 20 kHz, it will be:

0.0000125 x 20,000 x 360 = 90 degrees.

Yet in spite of a 90 degree phase shift at 20 kHz, all frequencies will arrive at the same time, so there's no smearing or phase distortion, etc.

se

[ebrewste]

From a theory point of view, in a unity feedback system (buffer) the frequency response of the closed loop system is G/(G+1) where G is the complex open loop response. I made a plot showing the relationship for an arbitrary system with a delay.

The simple way of looking at it is that closing the loop reduces phase lag in the 0dB gain neighborhood, but not by much.

[ebrewste]

If anyone cares, here it the spreadsheet I threw together to make that plot.
http://members.cox.net/quotes/closed%20open.xls

Also, you can see some wierdish mechanical systems measured by a program I make here using the same principles:
http://support.motioneng.com/Utiliti...e/bode_07.html
http://support.motioneng.com/Utiliti...e/bode_08.html