[SOLVED] The impacts of using two ota's in parallel within the feedback network?

[prestonee]

I have been trying to find some papers on this but am coming up short.

I have heard that if you place a very broad band ota in parallel with a narrow band, the narrow band helps keep the loop stable.
But what is the exact impact? If OTA1 has BW of x and OTA has bw of 3x, is the bandwidth of the system, 4x/2=2x? or is it 4x? I also expect the open loop gain to increase , is this true?
I am also curious about impacts as far as noise, i would expect this to go up? Im starting to wonder, would this share same concerns and impacts as a full rail to rail amp, since those are actually 2 amps in parallel, 1 being nfet pair while the other is a pfet pair...

-Pb

 

[rakshitdatta]

When you have two parallel paths between the input and the output, the actual transfer function will be the sum of the two. Let us say in the first path, we have a narrow band OTA(usually mulitple stages for large DC gain) and the broadband OTA(single-stage usually)is in the second parallel path. Let us assume that the unity gain frequency of the first OTA is "w". Now, for frequencies beyond "w", the net transfer function is mainly dominated by second path. Hence you have a single pole roll-off again upto the unity gain frequency of the second OTA. If you draw bode plots for this configuration, you will come to know that there is a zero being formed in the left half plane which is any way expected when you have multiple paths from input to output. The exact bandwidth of the composite OTA will depend on the bandwidth and the dominant pole locations of each of the individual OTAs. This kind of compensation is called "feedforward compensation". There are some papers on this by J. Harrison from Australia.
It is slightly different from a rail to rail input OTA. You can use nMOS input pair OTAs in both the parallel paths. This will ensure that you dont have the input dependent offset problem that you usually have with a rail to rail input OTA.
Regarding noise, compared to a miller compensated OTA, feedforward compensated OTA will not have extra noise.

 

[prestonee]

So according to this, I could use an unstable Broadband amplifier, in parallel with a stable narrow band, and put my sc network around them, and itd be the best of both worlds? stable and broad band?

 

[rakshitdatta]

With two parallel paths from input to output, effectively you are creating a zero. From phase margin point of view, you can at the maximum compensate for a pole. You cannot definitely use an "unstable broadband amplifier" in parallel with a "stable narrow band" as the effective OTA will still be unstable. I would put it like this. A "small gain(at dc) broadband amplifier" can be used in parallel with a "large gain(at dc) narrow band amplifier" to get a "large gain(at dc) and broadband amplifier"

 

[prestonee]

Ok Thank you, I was able to verify the A + B result, it was hidden at first it seemed A + B =A, then I realized that it was the input loading of the other amp that was causing my B term to change significantly reducing my overall gain to still be around A. kind of sucks, but when I place it in open loop its obvious to see A + B, and when I use in closed loop and sim the loop gain while using vcvs as input buffers i still get Beta(A+B). so I just need to discover why the input loading is having a such a significant impact on my Beta term.

-Pb

 

[prestonee]

I discovered the cause of my confusion, I had assumed that all nodes in a folded cascode were low impedance (that is the characteristic of a folded cascode for correct operation). However my "low impedance at the drains of my input pair in the folded cascode were around rds/2, which isnt that low compared to a 1/gm term that i was assuming. but is relatively low compared to the output impedance which is a rds^2 type of value. hence my folded cap impedance was a factor of 10 larger then previously thought and did in fact substantially affect my beta. when I took this value into consideration and recalculated my AB values for amp A, amp B, ampA+B, it did correlate.

thank you for the mention of feedforward.
I havent used that compensation before but its theories do apply to my amplifier architecture.
-Pb

- - - Updated - - -

But I will say that I do believe you can use the parallel stable amp to stabilize the unstable high gain amp. thats the original purpose of the feedforward compensation technique. (it basically extends the non dom pole to allow stability without reducing the bw that is nomally done via cc cap.