How can I increase the PSRR of Bandgap Reference ?

[devop]

what's the usage of C1 here??

How can I increase the PSRR of Bandgap Reference ????
What's the spec of OPAMP here?can I use a 2 stage OPA here since the C1 may decrease the stability here.

 

[safwatonline]

a bandgap question

Lower the OTA rejection

 

[DenisMark]

a bandgap question

C1 servers for dominant-pole frequency compensation. For increase PSRR it's better use it instead pure miller compensation. Capacitance between Vout and grond also increase PSRR.
OTA may be with any number of stages But, M13 and R1 form last amlification stage in loop. If OTA is single stage then the loop will now consist 2 stage and this can lead to stability issue. And C1 with proper size helps with this.

Generally OTA must have high gain, low offset voltage (this increase PSRR too), low 1/f and white noise. This requirements are related with requered accuracy of reference.

 

[rfsystem]

Re: a bandgap question

As a first step you can match the drain/source voltages of the matched PMOS triple. Insert the resistor R3 into the drain of the kT bias loop. That improves up to the point of matching. I guess 20x with reasonable PMOS devices sizes. So please verify with matching! There are other topologies based on NMOS source followers but better would be the use of a preregulator. There two versions. The first using a cascade. The cascade stage is the bandgap cell with a rationed buffer driving the second bandgap. The second version is a rationed buffer self driving his own bandgap. It need more detailed startup analysis and good startup shutoff circuit but removes the power/area overhead.

If the power is low R2 could be fairly large. So a cascaded or selfbuffered version could be more area efficient than pure improvements by matching of opamp gain.

 

[devop]

Re: a bandgap question

C1 servers for dominant-pole frequency compensation. For increase PSRR it's better use it instead pure miller compensation. Capacitance between Vout and grond also increase PSRR.
.

Thanks safwatonline,DenisMark and rfsystem
I have one questions about C1 here.
it's better to connect C1 to vdd or gnd?why?

 

[safwatonline]

a bandgap question

Vdd

Added after 2 minutes:

to couple vdd smalls signal to the gate of the PMOS so that its Vgs=Vg-Vs=vdd-vdd=0 >> better rejection
if gnd, then Vgs=0-vdd>> worse rejection ;
VOUT=id*Rout=(gm* Vgs)*Rout

 

[gpwu]

a bandgap question

Vdd
C1 need to close to M12,M13 on layout.

 

[surianova]

Re: a bandgap question

Vdd

Added after 2 minutes:

to couple vdd smalls signal to the gate of the PMOS so that its Vgs=Vg-Vs=vdd-vdd=0 >> better rejection
if gnd, then Vgs=0-vdd>> worse rejection ;
VOUT=id*Rout=(gm* Vgs)*Rout

y

i agree, C1 connect to VDD give better psrr at high frequency. How about conpensation for stability? any different for phase margin when u connect C1 to VDD or GND?

for me, i think shd be not different, bcoz VDD and GND also AC GND. What do u all think?

Will it affect transient?

Thank you

 

[wjlzhx]

a bandgap question

I think VDD is better

IN my idea,C1 let M13's Vgs and the OTA stable to let Im13's current stabel.then increate psrr

 

[gpwu]

Re: a bandgap question

Fro stability, C1 connects to VDD or GND has the same phase margin.
But traisient response will be different, but usually given enough time it will settle down.

 

[surianova]

Re: a bandgap question

Fro stability, C1 connects to VDD or GND has the same phase margin.
But traisient response will be different, but usually given enough time it will settle down.

for transient c1 connect to vdd or gnd better? why got difference if PM the same?

 

[gpwu]

Re: a bandgap question

Yes, transient should be the same for the same PM.
The only difference is PSRR to VDD or to GND.

 

[safwatonline]

a bandgap question

actually they are not the same transient response

 

[mists]

Re: a bandgap question

There are other topologies based on NMOS source followers but better would be the use of a preregulator. There two versions. The first using a cascade. The cascade stage is the bandgap cell with a rationed buffer driving the second bandgap. The second version is a rationed buffer self driving his own bandgap. It need more detailed startup analysis and good startup shutoff circuit but removes the power/area overhead.

If the power is low R2 could be fairly large. So a cascaded or selfbuffered version could be more area efficient than pure improvements by matching of opamp gain.

rfsystem, may you explain in detail about two version of NMOS source followers, thanks!