How to improve AD834 based VGA circuit

[dzseki]

I try to improve a variable gain amplifier based on the AD834 (500MHz) multiplier that I’ve made years ago, the thing is used for (CRT) home theater projectors as an improvement over original video chain design.
The source signal has a bandwidth of about 100MHz, and the idea would be to pass this signal with the least attenuation to the later circuit stages.
The design is based on a circuit found in the AN-212 application note of the AD834. Although for some reason I am failing to meet the performance that is presented in the application note. More precisely I am able to meet the 500MHz performance point  within -3dB, but my frequency response is not as flat as I would like.
The problem is that one of the deepest point in the frequency response waviness is located right around 100MHz, about -1.5dB, the response eventually comes back at higher frequency but there the signal has very little harmonic content, therefore having a very subtle performance lead compared to slower multipliers, especially the AD835 (250MHz), with which I also have first hand experience.
With this circuit I have made several iterations both with component values and circuit layout already, but this waviness remained essentially the same.
As for the changes:
-I have tried different common base transistors, those only made things worse if any (in line with the application note’s suggestion)
-I have varied R31-32 resistors
-I have tried several opamps on the output (HFA1100, CLC449, THS3201, LMH6703) along with different feedback resistors, with these I could control the amplitude of the last peak before the final roll off, but overall had little effect on the flatness.

I am loading you with some data:
-the AN-212 application note for reference
-snapshot of the latest PCB layout
-circuit diagram
-frequency response (VNA)

[David Hess]

The design is based on a circuit found in the AN-212 application note of the AD834.

Which circuit?  The 60 MHz resistor level shifted circuit or the 450 MHz bipolar level shifted circuit?

Although for some reason I am failing to meet the performance that is presented in the application note. More precisely I am able to meet the 500MHz performance point  within -3dB, but my frequency response is not as flat as I would like.

The problem is that one of the deepest point in the frequency response waviness is located right around 100MHz, about -1.5dB, the response eventually comes back at higher frequency but there the signal has very little harmonic content, therefore having a very subtle performance lead compared to slower multipliers, especially the AD835 (250MHz), with which I also have first hand experience.

With this circuit I have made several iterations both with component values and circuit layout already, but this waviness remained essentially the same.

I wonder if there could be a problem with the substrate you are using.  Things can get weird when maximal bandwidth flatness is a requirement.  Humidity might be used as a test to see if the substrate is absorbing water which is a common problem.

-I have tried different common base transistors, those only made things worse if any (in line with the application note’s suggestion)

What was their Ft?  There is a deplorable availability of fast PNP transistors now since NXP discontinued theirs.

-I have varied R31-32 resistors

Which are those?  The two schematics are not marked.

It seems like some type of equalization adjustment would be called for but the effect you are observing is pretty subtle.

You might consider using a more complex termination for the AD834 (or bipolar level shifter) involving a t-coil although construction can be difficult.

-I have tried several opamps on the output (HFA1100, CLC449, THS3201, LMH6703) along with different feedback resistors, with these I could control the amplitude of the last peak before the final roll off, but overall had little effect on the flatness.

Honestly I would be tempted to drop using the operational amplifier as a differential to single ended converter and replace the output stage with some sort of transconductance (series feedback) based bipolar transistor based version.

[dzseki]

Which circuit?  The 60 MHz resistor level shifted circuit or the 450 MHz bipolar level shifted circuit?

Sorry for being unclear. The circuit is based on the „DC to 480MHz Voltage-controlled amplifier using active level shifting”  circuit found on page 6-45 in the AN-212 App Note.

I wonder if there could be a problem with the substrate you are using.  Things can get weird when maximal bandwidth flatness is a requirement.  Humidity might be used as a test to see if the substrate is absorbing water which is a common problem.

This could be a clue, as all my ICs are from old stock (early 90’s), although the chip itself is running rather hot, so it should evaporate the humidity pretty fast

What was their Ft?  There is a deplorable availability of fast PNP transistors now since NXP discontinued theirs.

At my first trial I used BFT95 transistors (ft: 5GHz) for the common base stage, but what I’ve got was a nice untamable oscillation. Then I stepped one back and used 2SA1765 transistors (ft: ~1GHz) while these were stable they were not giving any performance lead compared to 2N3906, so eventually in the end I used the 2N3906 as originally suggested.

-I have varied R31-32 resistors

Which are those?  The two schematics are not marked.

Designators referenced to the VNB_DB_AD834_NO_GAMMA.pdf

It seems like some type of equalization adjustment would be called for but the effect you are observing is pretty subtle.

Subtle, but not unreasonable I think. Also in figure 15 (in AN212) their version does look better although with 10dB/div it is a little bit hard to tell... Just for sure I also have a VNA shot with 10dB/div

You might consider using a more complex termination for the AD834 (or bipolar level shifter) involving a t-coil although construction can be difficult.

-I have tried several opamps on the output (HFA1100, CLC449, THS3201, LMH6703) along with different feedback resistors, with these I could control the amplitude of the last peak before the final roll off, but overall had little effect on the flatness.

Honestly I would be tempted to drop using the operational amplifier as a differential to single ended converter and replace the output stage with some sort of transconductance (series feedback) based bipolar transistor based version.

Can you elaborate here a bit more what do you think about? (eg. the link is broken)

[duak]

For what it's worth, here's a couple of thoughts:

-  R36 & R45 seem high for collector load resistors for the desired bandwidth.  The MMBT3906 has a Cob(max) of 4.5 pF which gives a corner frequency of 75 MHz.  This roughly agrees with the observed results.  Does reducing the values of R36 & R45 make a difference in flatness?  ie., does the frequency of the first drop increase?

- capacitance on the inverting input node of U6 can cause gain peaking.   Since there are two 470R in parallel with similar capacitance the corner frequency of the feedback network would be 150 MHz.  This agrees roughly with the observed results.  A small capacitor across R49 might reduce the level of the bump past the dip.

Here is a more precise analysis of the peaking phenomonon: http://www.ti.com/lit/an/sloa013a/sloa013a.pdf page 7.

[dzseki]

For what it's worth, here's a couple of thoughts:

-  R36 & R45 seem high for collector load resistors for the desired bandwidth.  The MMBT3906 has a Cob(max) of 4.5 pF which gives a corner frequency of 75 MHz.  This roughly agrees with the observed results.  Does reducing the values of R36 & R45 make a difference in flatness?  ie., does the frequency of the first drop increase?

- capacitance on the inverting input node of U6 can cause gain peaking.   Since there are two 470R in parallel with similar capacitance the corner frequency of the feedback network would be 150 MHz.  This agrees roughly with the observed results.  A small capacitor across R49 might reduce the level of the bump past the dip.

Here is a more precise analysis of the peaking phenomonon: http://www.ti.com/lit/an/sloa013a/sloa013a.pdf page 7.

Thanks, for the good idea. It is easy to try different R36 & R45 at least. if your assumption is right then the real deal would be to tune the input peaking of the opamp and the collector resonance to the same frequency so they would cancel out each other.
With the exact results however I have to wait next week, since I don't have a VNA at home (unfortunately )

[David Hess]

I wonder if there could be a problem with the substrate you are using.  Things can get weird when maximal bandwidth flatness is a requirement.  Humidity might be used as a test to see if the substrate is absorbing water which is a common problem.

This could be a clue, as all my ICs are from old stock (early 90’s), although the chip itself is running rather hot, so it should evaporate the humidity pretty fast

I have not heard of it being a problem with IC packages except when surface mount soldering.  Printed circuit board substrates can be a major problem if they are sensitive to humidity changes.  Water is a very lossy dielectric.

What was their Ft?  There is a deplorable availability of fast PNP transistors now since NXP discontinued theirs.

At my first trial I used BFT95 transistors (ft: 5GHz) for the common base stage, but what I’ve got was a nice untamable oscillation. Then I stepped one back and used 2SA1765 transistors (ft: ~1GHz) while these were stable they were not giving any performance lead compared to 2N3906, so eventually in the end I used the 2N3906 as originally suggested.

Well, you will not find anything better than the BFT95s.  Might have to add a lossy element into the base or collector circuit to quash oscillation with such a fast part.

If the 2SA1765s and 2N3906s performed the same then I would assume the problem lies elsewhere.  Some 2N3906s can be pretty fast.

You might consider using a more complex termination for the AD834 (or bipolar level shifter) involving a t-coil although construction can be difficult.

-I have tried several opamps on the output (HFA1100, CLC449, THS3201, LMH6703) along with different feedback resistors, with these I could control the amplitude of the last peak before the final roll off, but overall had little effect on the flatness.

Honestly I would be tempted to drop using the operational amplifier as a differential to single ended converter and replace the output stage with some sort of transconductance (series feedback) based bipolar transistor based version.

Can you elaborate here a bit more what do you think about? (eg. the link is broken)

Whoops, I am not sure what happened.  Here is the link again:

https://www.electronicdesign.com/analog/what-s-all-t-coil-stuff-anyhow

[David Hess]

I like Duak's suggestions.

Here is a link I received today to a comprehensive book about wideband amplifier design which may be helpful.  It even has an in-depth discussion about t-coils but beware, it also has lots of math.

[duak]

I was going to suggest that book too.  At one time I would have understood more of the math, but not anymore.  Now I just read these books for the stories.

[David Hess]

Now I just read these books for the stories.

That is what guys said about Playboy magazine also.  But we all know the truth.

[dzseki]

I have lowered the collector resistance down to 240R, with a quick test I could not see any improvement unfortunately.
Paddling on the board had some (good) effect around the output opamp, so perhaps adding some "input peaking" to the opamp (along with a much larger feedback resistor to counteract higher frequency peaking) could be a quick and dirty method for the goal.

[dzseki]

The ratio of R31/R27 and R32/R28 trims the gain of this circuit (references to my schematic). R31 and R32 also there to decouple somewhat the open collectors of the AD834 preventing unwanted oscillations, so they are better to be not very low values. I had the idea to poke around there...
I increased R31-R32 to 100R, and put an RC network accross to each resistor, I choose 39pF and 33R after some calculations and from the available parts around. The intention was to tune the circuit to around 120MHz. Obviously refinement of the RC values are needed because now I have +1.2dB at 120MHz, but the things have changed in the right direction at least, perhaps adding a coil (making the compensation to RLC) would be benefical to tame high freq peaks (or increasing the opamps feedback even further).

[duak]

Interesting - I believe increasing the values of R31 & R32 will reduce the voltage gain of Q4 & Q5, and probably increase bandwidth and stability.  The application note discusses this in the first two paragraphs of p.6-46 but does not go into the model of the circuit showing the parasitic inductances and how to mathematically determine the values.

This circuit is also known as a Folded Cascode and is often used in fast opamps.

I would try using the faster transistors now and adjust R31 & R32 to stabilize the circuit.  I believe the AD834 uses a 1 GHz or faster process so it doesn't make sense to use 200-300 MHz parts for Q4 & Q5 if faster ones are available.  I remember trying to build a low power FM transmitter with 2N2907 or 2N3906 transistors at 100 MHz.  No way - not enough gain.  When I used a VHF transistor from an old radio, it worked perfectly.

[dzseki]

Here is a VNA screen capture with the revised tuning network. Now I've used 49R9+15pF across 100R at the emitter of the transistors.
I am somewhat concerned about the sharp peaks in the response above 100MHz, but that might be a VNA  thing (an Agilent E8358A) because it changes suddenly above that point.
Also I may try to further increase the feedback resistor of the LMH6703 to dampen the peak around 300MHz a little, but overall this is already much better than it was.

[dzseki]

I would try using the faster transistors now and adjust R31 & R32 to stabilize the circuit.  I believe the AD834 uses a 1 GHz or faster process so it doesn't make sense to use 200-300 MHz parts for Q4 & Q5 if faster ones are available.  I remember trying to build a low power FM transmitter with 2N2907 or 2N3906 transistors at 100 MHz.  No way - not enough gain.  When I used a VHF transistor from an old radio, it worked perfectly.

I was very thorough checking 2N3906 and MMBT3906 datasheets, and I only found two which contained an ft curve at all, those curves were satisfactory however. One Taiwanian manufactuer (Comchip) suggested that typical ft should be around 500MHz @ Ic=10mA, other datasheet from ROHM suggests also >400MHz around 10mA, sot this should be OK I think.

[David Hess]

I would try using the faster transistors now and adjust R31 & R32 to stabilize the circuit.  I believe the AD834 uses a 1 GHz or faster process so it doesn't make sense to use 200-300 MHz parts for Q4 & Q5 if faster ones are available.  I remember trying to build a low power FM transmitter with 2N2907 or 2N3906 transistors at 100 MHz.  No way - not enough gain.  When I used a VHF transistor from an old radio, it worked perfectly.

I was very thorough checking 2N3906 and MMBT3906 datasheets, and I only found two which contained an ft curve at all, those curves were satisfactory however. One Taiwanian manufactuer (Comchip) suggested that typical ft should be around 500MHz @ Ic=10mA, other datasheet from ROHM suggests also >400MHz around 10mA, sot this should be OK I think.

2N3906s can be selected for low collector-base time constant yielding high Ft (1) however in general, a faster part like an MPSH82 would be used for 100 MHz operation.

(1) Tektronix did this to find 500 MHz parts for 100 MHz designs.