Balanced OTA-C cascode IC design

[sherif96]

If the filter's response is as follow what could possibly be the problem? the ideal response is to be flat between 1 and 3MHz with a 0dB gain and -3dB at both 1 and 3MHz, I understand the non ideal filter would differ slightly than the ideal one but it seems like there is a problem.

 

[FvM]

Lack of information to answer the question. You don't show the designed gm and capacitor values, thus we can't see if the measured filter gain is expectable or not.

Secondly you would determine the actual gm and ro of differential OTAs.

Finally, when paralleling differential OTA outputs, either G2 and G3 or G2a, G2b and G3, you must take care that the CMFB circuits don't saturate, either provide limited gain or use a common CMFB for all OTAs in parallel.

 

[sherif96]

Lack of information to answer the question. You don't show the designed gm and capacitor values, thus we can't see if the measured filter gain is expectable or not.

Secondly you would determine the actual gm and ro of differential OTAs.

Finally, when paralleling differential OTA outputs, either G2 and G3 or G2a, G2b and G3, you must take care that the CMFB circuits don't saturate, either provide limited gain or use a common CMFB for all OTAs in parallel.

Lack of information to answer the question. You don't show the designed gm and capacitor values, thus we can't see if the measured filter gain is expectable or not.

Secondly you would determine the actual gm and ro of differential OTAs.

Finally, when paralleling differential OTA outputs, either G2 and G3 or G2a, G2b and G3, you must take care that the CMFB circuits don't saturate, either provide limited gain or use a common CMFB for all OTAs in parallel.

-The ideal circuit with the ideal gm values and the capacitor values are in the first attachment, the second attachment is the response of the ideal circuit using the same architecture but with vccs instead of the OTAs the response is as designed in the ideal circuit as you can see, so I reckon the problem won't be with the gm values or capacitance but something to do with the non ideal circuit.
- The OTA gm values and routs are as follows:
Ideal gm value Non ideal value
47.08u 47.087u rout equals to 26.45K
62.882u 62.91u rout equals to 25.73K
127.23u, 127.26u rout equals to 18.8K
45.9u 45.93u rout equals to 26.4K
17.42u 17.423u rout equals to 179.24K
16.99u 17.004u rout equals to 179.27K

-what do you mean by the CMFB circuits do not saturate? shouldn't they saturate in order to have a vout of my reference voltage ? as if they do not saturate the voltage won't be equal to the reference voltage and biasing of the second stage transistors in the differential OTA would be affected

 

[sherif96]

Well you were absolutely right about connecting the parallel OTAs with same cmfb, the response developed vastly, however there is a small issue I wanted to ask you about, the response has a gain of -6.54 dB at 1MHz and 3.012dB at 3MHz, is ther a way to make the 1MHz+ frequencies have better gain to be close to the designed gain value ?

 

[sherif96]

Lack of information to answer the question. You don't show the designed gm and capacitor values, thus we can't see if the measured filter gain is expectable or not.

Secondly you would determine the actual gm and ro of differential OTAs.

Finally, when paralleling differential OTA outputs, either G2 and G3 or G2a, G2b and G3, you must take care that the CMFB circuits don't saturate, either provide limited gain or use a common CMFB for all OTAs in parallel.

Would you mind explaining why the parallel OTAs + the last OTA should be connected with a common cmfb? and in the photo you uploaded in a previous post of how to connect the whole circuit, why was the output of the parallel ota connected to the output of the second ota not the input, your advice made it work, however i still do not understand those 2 points if you may explain to me briefly.

 

[FvM]

CMFB is required for true differential OP and OTA to set the output common mode voltage to a defined level, usually mid supply. Without CMFB, the CM voltage can be expected to drift towards a rail due to unavoidable offsets.

In case of multiple OTA driving the same differential node, each CMFB amp might see a slightly different CM voltage and thus try to fight each other. With sufficient gain, the CMFB amps can saturate and cause an unsuitable OTA bias point.

why was the output of the parallel ota connected to the output of the second ota not the input

consider the identity i2 = gm2*(vo1-vi2) = gm2a*vo1 - gm2b*vi2

You obviously need to sum the outputs.

 

[sherif96]

CMFB is required for true differential OP and OTA to set the output common mode voltage to a defined level, usually mid supply. Without CMFB, the CM voltage can be expected to drift towards a rail due to unavoidable offsets.

In case of multiple OTA driving the same differential node, each CMFB amp might see a slightly different CM voltage and thus try to fight each other. With sufficient gain, the CMFB amps can saturate and cause an unsuitable OTA bias point.



consider the identity i2 = gm2*(vo1-vi2) = gm2a*vo1 - gm2b*vi2

You obviously need to sum the outputs.

I understand the part for having a common cmfb for u2a and u2b as they are driving the same differential node, but why u3 also should be connected to the same cmfb? shouldnt that be a different differential node with its own cmfb ?

 

[FvM]

The common CMFB amplifier is just a suggestion, reducing CMFB gain is another solution.

Anyway, the three OTA are driving the same differential node, so their CM output currents are fighting each other.

 

[sherif96]

The common CMFB amplifier is just a suggestion, reducing CMFB gain is another solution.

Anyway, the three OTA are driving the same differential node, so their CM output currents are fighting each other.

yes actually they are, i totally forgot about the feedback line coming from the third ota, if you do not mind I have one last question to end my design, the current noise figure for my filter is 44dB, if I want to reduce the NF about 9-10 dB, how can I decrease it with such a large value ?

 

[FvM]

Did you determine the dominant noise sources in a noise analysis? I guess it's the differential pair(s) in OTAs. If so, you can try to operate the OTA at different bias currents (= different gm level) and adjust the capacitors respectively.

 

[sherif96]

Did you determine the dominant noise sources in a noise analysis? I guess it's the differential pair(s) in OTAs. If so, you can try to operate the OTA at different bias currents (= different gm level) and adjust the capacitors respectively.

would it be worth it if i calculate nf,gain for each ota and rearrange stages based on the most optimized total nf ?