The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> RF Design >> Tuning cirtuit for a Gm-C filter
https://designers-guide.org/forum/YaBB.pl?num=1278817059

Message started by spring on Jul 10^th, 2010, 7:57pm

Title: Tuning cirtuit for a Gm-C filter
Post by spring on Jul 10^th, 2010, 7:57pm

Hi, everyone
Now I am doing a Gm-C RF tracking band pass filter whose working freq is from 48M to 300M by using Tow-Thomas architecture; the tuning circuit is confused me a lot. i try to use PLL as tuning circuit, i use the same Gm-C cell to building a vco, but the oscillate freq is not the same as the BPF, why? anybody can help me, thank you:) or can other proper method be used as tuning circuit?
Spring

Title: Re: Tuning cirtuit for a Gm-C filter
Post by rfcooltools.com on Jul 10^th, 2010, 11:00pm

that the large signal oscillation biasing is not the same as the small signal one (it may start out similar but in oscillation it probably changes). The large signal oscillation may include current slewing that will make the time const different.

Oscillating at the Bandpass frequency is not important. adjusting the right amount of R or C is what is needed to remove process variations from the filter is. this does not have to happen at frequency instead
Instead make a current proportional to R
charge C with that current to a voltage V
compare to a time reference and make a correction
dump C an repeat. If your circuit has a parasitic capacitance that is large the include devices that make up that parasitic in the charging of C.
http://rfcooltools.com

Title: Re: Tuning cirtuit for a Gm-C filter
Post by spring on Jul 11^th, 2010, 12:02am

rfcooltools.com wrote on Jul 10^th, 2010, 11:00pm:

thank you for you reponse:)
i know for active RC the method you mentioned is effective3, but for Gm-C circuit, the Gm is not easy to measure.

Title: Re: Tuning cirtuit for a Gm-C filter
Post by rfcooltools.com on Jul 11^th, 2010, 10:10am

I was not paying attention sorry..
Gm-C tuning here is a way
make a Gm=1/R circuit where R can either be an external resistor or using a NMOS in triode and an opamp to control the the gate create a resistor copy
This is done by using a constant current as the drain current for the NMOS and the gate is adjusted mimic a Bandgap related voltage. This gate voltage in conjunction with another equivalent size NMOS will be the R in the Gm=1/R

From the Gm=1/R circuit mirror the current from this block and use it to charge the capacitor then repeat the steps above my previous post.

http://rfcooltools.com

Title: Re: Tuning cirtuit for a Gm-C filter
Post by loose-electron on Jul 11^th, 2010, 12:19pm

Good grief no!

Tune the -6dB point of the filter.

Create an input frequency signal at the -6dB point of the filter.
Put input signal into filter.
Put input signal resistor divider
Compare amplitudes of the 2 signals (build an amplituide detector)

Adjust filter until amplitudes match.
Done.

Title: Re: Tuning cirtuit for a Gm-C filter
Post by spring on Jul 11^th, 2010, 10:40pm

rfcooltools.com wrote on Jul 11^th, 2010, 10:10am:

Thank you for you attention:) I have got you mean, but the difficult is how to build a Gm=1/R circuit .

Title: Re: Tuning cirtuit for a Gm-C filter
Post by spring on Jul 11^th, 2010, 10:49pm

loose-electron wrote on Jul 11^th, 2010, 12:19pm:

Thank you for you response:)

Why choose -6dB point of the filter, i see same papers choose 0 dB point of the filter, but the accuracy of the amplitude detecor limit the tuning accuracy of the filter.

Spring

Title: Re: Tuning cirtuit for a Gm-C filter
Post by rfcooltools.com on Jul 12^th, 2010, 11:01am

GM=1/R for longer geometries of fets the ratio is around 1: 4 but with smaller geometries 0.18uM or less you will have to play around with the ratios to get Gm proportional to 1/R

_____________
|current mirror|
------------------
| |
|___ |
| | |
|---| | |---
4 |o--o| 1
|---| |---|
\ |
R / |
\ |
---------------

http://rfcooltools.com

Title: Re: Tuning cirtuit for a Gm-C filter
Post by aaron_do on Jul 13^th, 2010, 2:03am

Hi Loose_electron,

you mentioned tuning to the -6 dB frequency. I think you may have misread the post since this is a bandpass filter. Otherwise could you please explain a bit further. Thanks.

Hi Spring,

IMO if you're going with tuning using the method mentioned by loose_electron, then you could do a phase comparsion at the input and output of the filter since the phase shift should be zero at the center frequency.

cheers,
Aaron

Title: Re: Tuning cirtuit for a Gm-C filter
Post by aLittleKnowledge on Jul 13^th, 2010, 4:20am

Once you have acquired the signal, the phase-shift method proposed by aaron_do looks good. But unless it's a single pole design the required phase shift will not be zero.

Title: Re: Tuning cirtuit for a Gm-C filter
Post by wave on Jul 15^th, 2010, 3:06pm

Your architecture will depend on your system. Can this be a one time tune (with external reference signal), an occassional on chip tune between operations, or a continuous tuning to track temperature/VDD, etc. (ie, master/slave operation).

The question of how to make 1/R from a Gm is pretty fundamental.
A Gm can be a single xter or diff pair, with sometimes complex biasing (ie, CMFB, degeneration tuning).
The outputs are connected to the inputs of an appropriate polarity.

The ascii stick figure someone drew looked more like a delta-VGS bias than a Gm to me. Not like a trans-conductance cell.

I recall the Johns & Martin text has some tuning examples that might be useful to you .

Good luck,
Wave

;)

Title: Re: Tuning cirtuit for a Gm-C filter
Post by vp1953 on Jul 17^th, 2010, 11:14am

Hi Rfcooltools,

The stick figure looks very cool!!

The current in each leg is proportional to 1/R^2. And Gm is proportional to 1/R but why the ratio 4:1; the relationship seems to hold for many other ratios.

_____________
|current mirror|
------------------
| |
|___ |
| | |
|---| | |---
4 |o--o| 1
|---| |---|
\ |
R / |
\ |
---------------

Title: Re: Tuning cirtuit for a Gm-C filter
Post by loose-electron on Jul 17^th, 2010, 12:21pm

aaron_do wrote on Jul 13^th, 2010, 2:03am:

you mentioned tuning to the -6 dB frequency. I think you may have misread the post since this is a bandpass filter. Otherwise could you please explain a bit further. Thanks.

Ok, my bad, I was thinking a LPF.

For a BPF its a little different.

If the BPF is narrow in BW you tune for an amplitude peak using a band centered reference signal.

If the BPF is a bit wider, you generate two reference signals that sit on band edges (one above band center, one below), and you adjust your tune for amplitude balance between the two signals.

Another option is using a band centered test signal and adjusting the tuning up-down and find the -6dB points of the
tuning system, and then splitting the difference to center the BPF. Tuning system has to be linear.

Many possible options there.

The reason I suggested -6dB is because it's easily generated, V/2 is pretty easy to do right?

gmC filters will require updating due to temps and process corners. Will need to update settings periodically.

Oh, and I am a big proponent of digital tuning systems. Switched systems are just a bit more predictable than an analog FB system.

Title: Re: Tuning cirtuit for a Gm-C filter
Post by rfcooltools.com on Jul 17^th, 2010, 3:44pm

to get exactly gm=1/R 4:1 is required for a square-law device to get exactly that. see my attached derivation

otherwise gm=X/R where X is some other constant multiplier

http://rfcooltools.com

Title: Re: Tuning cirtuit for a Gm-C filter
Post by vp1953 on Jul 18^th, 2010, 2:30pm

Hi Rfcooltools,

Thanks for the detailed derivation. I had assumed proportionality and not equality. For Gm =1/R, agreed that it should be 4:1

Title: Re: Tuning cirtuit for a Gm-C filter
Post by rfcooltools.com on Jul 19^th, 2010, 1:17pm

vp1953,

your thanks made the effort worthwhile!
http://rfcooltools.com