Ringy Voltage Divider at 10MHz and higher

[Mechatrommer]

Ringy Voltage Divider at 10MHz and higher

Hi. its like i've been hit by a dark force again, my imagination is exhausted as to what cause this effect. I tried to compensate a voltage divider for high speed, sim looks good, do it on "copper board" and voila! it ringy, its passive, and its ringy! i thought DSO input-Z screwing things up on this circuit, i made it buffered with a LM7121 10Mohm input voltage follower, still the same. any idea what cause this? seeking help and advice from high speed gurus and friends here.

attached are the circuit schematics, and the real "copper boarded" picture 1-3. i dont thing this sloppy setup/board/soldering cause the ring, but i'm not 100% sure, who knows, i tried shortest traces possible with available board, this is what i came up with, consider it the best i can do . i used the same board previously without any issue (just as voltage amplifier).

picture 4-9 is the dso capture from 10mhz, 20,30,40,50 and 100mhz.
YELLOW trace = CH1_DSO
BLUE trace = CH2_DSO
PURPLE trace = XY plot of CH1 and CH2

hope to hear some advice and C&C, thanks.
time to take a break :|

OT: since at it, does anybody have microstrip, trace impedance matching and transmission line pcb design technique guide reference? i dont have any in my archive. thanks again.

[jahonen]

Do you have hefty bypass capacitor near the opamp, connected using low-inductance fashion (directly at the VCC pin(s)). Lack of proper decoupling causes all kinds of strange problems. Put some SMD capacitors there if you don't already have those.

Regarding microstrips, just use your favorite calculator found on the net (TXLINE etc), or just use the rule of thumb for 50 ohms: trace width of 2x dielectric thickness results about 50 ohms impedance Then use same techniques than with any transmission line, series or parallel termination etc.

Regards,
Janne

[w2aew]

I agree about the local supply bypass for your buffer amplifier.  Make sure to include a good RF-quality decoupling cap (1-10nF) in parallel with your bulk decoupling.  Also, the buffer amp is mainly loaded by the scopes input capacitance.  Try adding a resistive load to ground on your board (even though the part is specified for stability into unlimited capacitive load, this will likely help too).  BTW - what does the signal look like at the input to your buffer?

[Mechatrommer]

i have decoupling smd caps near Vs+ and -ve (next to green wires there) but i think thats 1uF and 60nF something like that, not really accurate at picking the value, i thought it doesnt matter. and the return gnd path is not in "low inductance" fashion since there's trace going criss cross there, so i have to run the wire on top of the air there (top green wire Vs+ and decoupler). and larger caps near psu output (covered with red and black tape on 2nd picture).

the problem is i tried probing without buffer amplifier earlier, the ringing effect is somewhat more pronounced than whats shown above (with buffer amp) iirc. the ringing problem starts when i add 100pF parallel to 1Kohm to ground. before that everything just fine (except noisy ) hence i believe ringing occurs at the buffer input, ie at the voltage divider circuit, not at the output.

i will try the terminating R to gnd when i got home, thanks for the advice. but will that solve it? since the ringing occurs at the input? i will capture the ringing without buffer amp later.

Make sure to include a good RF-quality decoupling cap

err, whats the good RF quality? smd ceramics not good?

[free_electron]

your 1 megaohm resistor is a spiral cut metal film... that's essentially an inductor ...

time to switch to L or R cut thin film surface mounted resistors ....

[Mechatrommer]

i believe thats carbon. i broke one of those intentionally. it has whitish ceramics like inside, no metal no film no coil, just white. but yeah, it exhibits inductance (or capacitance) at higher frequency, i believe all R are.

[robrenz]

I think that ceramic is what the metal film is on the outside of.  the film is extremely thin. Carbon composition is usualy fairly dark color.  FWIW

[free_electron]

the white ceramic is no tthe resistor. you need to scrape off the paint to see the resistor.
the metal film is a vacuum deposition system . it is only a few microns thick. ater that a spiral is cut around the body to tune it to the correct value.

[Mechatrommer]

the white ceramic is no tthe resistor. you need to scrape off the paint to see the resistor.
the metal film is a vacuum deposition system . it is only a few microns thick. ater that a spiral is cut around the body to tune it to the correct value.

i didnt scrape off, i broke it to pieces with hammer. i didnt aware any spiral, but maybe its there i just didnt see it too small and fragile.

anyway. made a capture without buffer amp, ie direct connection from voltage amplifierdivider to dso... i've added 1MHz, yeah i think thats noise from surrounding or psu. and on higher freqs, the effect is much exageratted. is it possible this noise is resonating and the 100pF form RLC bandselect filter?

[Mechatrommer]

and this is 100pF removed, direct to dso... everythings clean.

[Mechatrommer]

as per amspire advice on another thread, i made the 100pF in series with some resitance. some attenuation in oscillation noticed, but still there. picture showing 47ohm in series 100pF. ok, i should revise my (poor) other active/driving circuit (bypass caps and gnd feedthrough), dont make any assumption yet .

[Mechatrommer]

and the frequency respond in sim of this 47ohm in series is quite nasty (nastier for bigger resistance). compensating single cap for flatter respond is almost impossible. pls also note that 3 elements (RLC) is actually my model for a single resistor that i tried to get from real situation i got earlier (3 resistor = 9 elements). and err... assuming pure/ideal 100pF capacitance for capacitor and resistance of dso input (c1 and c2) .

[amspire]

My suspicion is still the amplifier is oscillating. I am not completely sure how you are connecting the scope.

High speed amplifiers are usually designed to get as much gain as possible while remaining stable. It doesn't take much to make them unstable. Believe it or not, it is much closer to being unstable in a voltage follower mode then a x10 gain mode.

The 75 ohms on the output is good, but I would like to see you adding a 50 ohm termination on the scope input. This will attenuate both voltages, but you can calculate the effect, and the waveforms will be much more accurate. It stops the scope looking capacitive.

Also have you added a resistor between the capacitor and the amplifier input?  How can an input oscillate - why would a signal be coming out the input pin anyway? The miller capacitance of the input transistors mean you can get AC current coming out from the input pins. A capacitor across the input can add a 90 degrees phase shift. A series resistor reduces this shift to less then 90 degrees.

Also, I couldn't make sense of your simulation. A 1 meg resistor will have less then 1uH inductance - not 1 milliHenry. The 1K resistor will be much less then 1uH.

I think once you get the amplifier stable, everything else will start working. Running the amp at a gain of 2 or more will make it less likely to oscillate. Look closely at the datasheets. Using a gain of 2 means you can add a few pFs compensation which also may help.

Richard.

[Mechatrommer]

ok i improved the board. 10nF bypasses as per datasheet, put more vias (diy) to the board. the ringing respond is... improved (without series 47ohm). i may have pointed my finger at wrong direction here. my pcb/setup is not robust enough. the grounding system is not good enough. now what? barricading the ground? 100MHz? still ringing (freq doubler) i didnt know how serious this is, the board is just 3cm x 3cm. maybe i'm playing with the dark lord matter here the psu also may not good enough, now i think i know why pc motherboards have built in regulator instead of relying to external crappy atx psu.

below is comparison before (left) and after (right). instead of ringing at rising edge, now the ring is attenuated and at the falling edge of signal.

[slateraptor]

How is your LM7121 powered, viz. voltage level for V+ and V-?

[Mechatrommer]

yes richard. maybe its my failure to fortified the amp supply and io and pcb. janne and w2 already made the advice i just didnt see what they are talking about. i'm newbie i thought 1/1000X divider using cheapo parts is unfeasible. i maybe (and surely) wrong. thanks guys for the advice, sure i will digest it in my brain whether i understanbd it or not.

How is your LM7121 powered?

dont ask please! you made me found more guilty. i have 1 psu shared among fg, my diy eval board and the lm7121. i probed the v+/v-, yeah its not that great, how "not great"? well, lets just say its one of the culprit. i will learn to make onboard regulator later.

[slateraptor]

dont ask please! you made me found more guilty. i have 1 psu shared among fg, my diy eval board and the lm7121. i probed the v+/v-, yeah its not that great, how "not great"? well, lets just say its one of the culprit. i will learn to make onboard regulator later.

Don't worry about that. What's the voltage level for V+ and V-? Single-supply 5V? +/- 5V?

[Mechatrommer]

quick update. i changed the psu to lead acid batteries arranged for dual rail supply. just a tiny improvement over the latest pictures. so maybe the major thing is the ground or i didnt put 68uF at psu input as recommended. or ringing at the dso input due to lack of termination. LA battery noise = 40mVpp, my psu 100mVpp++ noise, approx, FWIW.

[slateraptor]

i changed the psu to lead acid batteries arranged for dual rail supply.

So you're using a 12VDC lead acid battery with virtual ground for an effective +/- 6V? Or a single 12VDC battery for V+ and V- each?

[Mechatrommer]

±15V supply, test signal = ±3V (6Vpp), expected output signal (1/1000x) = ±3mV (6mVpp) if the respond is flat. so we are looking at 2mV/div blue trace, 1V/div yellow trace. relieved, i thought you asked something else. 2x12V LA battery in series ±12V. sorry was not clear.
edit: 200MHz oscillation amspire found out is probably the DDS clock from the FG. i can see quantization somewhere in the output signal. that as well maybe the reason. pretty much the whole system is at fault

[slateraptor]

Did you catch the little note on page 15 of the LM7121 datasheet about unity gain stability when powering the LM7121 at or above a rail-to-rail voltage of 6V or greater? It doesn't look like it from your schematic.

[Mechatrommer]

yeah thanks for the highlight i missed that. but the latest ringing pictures i removed the 7121 buffer and scope direct to the divider (so 50ohm termination is also not an option since it will alltogether screwing the hi impedance 1Kohm divider. so no more 7121, infact when using it, the ringing is attenuated.

[Mechatrommer]

let me deal with them first then i'll come back later. i believe the ground is ringing at 200MHz. when i placed the 100pF, the ring is coupled to the divider's output. until then. thanks all.

[Mechatrommer]

ok. the verdict is concluded (without much intefering the board of council ). i've observed and improvement attempted on all involved system. the verdict... there's 200MHz digital/clock/fpga contaminant noise in ground plane of my Hantek 3x25 AWG while generating a signal. whether the "signal" is amplified or attenuated, its there, around 5-30mVpp hard to tell, since its resemble random noise, but closely to 200MHz. its leaked to my attenuator test circuit pictured in the 1st post probably from psu and/or bnc ground connection (what else?) so the small amplitude signal will suffer from this contaminant floor, esp when coupled with capacitor, the noise will leaked to the "signal plane", since i can see my differential amp is unable to reject this "common mode ground" noise (if ever there's such a term ).

solution? none that i'm satified with or can find from. digging mixed signal thread here last night indicates, even the gurus are unable to solve this dark matter unless go to 4th dimension of pcb layer or careful "twinkle little star" gnd method. so, but my mixed signal Hantek 3x25 is 2 layer only, so i guess i cant get rid of it. another solution? buy the more "lab quality" fg. yeah right! thanks all guys for helping me brainstorming. and i'm sorry i've wasted your precious time.

edit: sorry i forgot to include the sample picture. below is 10mV 10MHz sine i tried to generate from 3x25 directly bnc to bnc probed by dso without the mentioned 1/1000X divider, its noise and its there even from 3x25 original output (removing my modded amplifier). and even the fg is disconnected from usb, there's around 20mV max ringing in my scope screen very close to this 200MHz frequency, i believe its also from my ext psu. so, thats about it, the Hantek 3x25 and my poorman psu noise limitation.

edit: striked above, not from the psu, i used the 2x12V LA battery, there's also this 200MHz ring, so not sure what, maybe "natural" harmonics from the system

[amspire]

You may not have this problem of you use proper 50 ohm terminations on your cables. You don't have to keep them short either.

Things don't work well at 200 MHz when you have badly terminated coax.

Richard.