High frequency noise for 10 Gbps circuit

[raebrm]

Hi,

I am designing a 10 Gbps laser driver using commercially available ICs. At the moment, I noticed that there is significant noise in the resulting eye diagram. This noise is reduced when I touch the circuit (to be precise, the interface line between the IC and the laser) with my bare hands - pictures attached

hence, can someone help me with these questions?

What's the effect of touching a circuit using our bare hands?
How does this reduce high frequency noise?
Can we reduce this noise using passive components? (i dont want to have to touch it all the time haha)

Anyway, thanks in advance if u can help
raebrm

 

[karesz]

Hi, you should read pls. from Linear Technology, Jim Williams Apllication Notes -breadboard or RF testboards...
In such from hes scripts (I remember 30-40 pp) is a lot to find over "hand as best RF-detector" or similares...
Sorry, I snow it over years.
My opinion: you maust not touch a RF line, and so less an line which bandwith is at 10GHz!
Alone if you are moving in hes direction_you will make some coupling/for example humm!_maybe you see it as "normal noise" in your signal, pls. make an FFT of your signal_you will know lot more over your "effect"...
What means pls. "I touch the IF lines between IC & Laser"? Are their signals or "only" the control lines?
K.

 

[biff44]

Sounds like you have an oscillation. Do you have the recommended bypass caps installed?

 

[karesz]

Or reflections in "undamped" lines?...

 

[rfmw]

How did you measure these two eye diagrams, using a probe or direct connection from the laser driver to the scope input?

Your (electrical) eye amplitude is very low, I wonder what extinction ratio you get in the optical eye?

Post your schematic, so we can help better...

Btw, I've done 2.5 and 5Gb/s laser drivers with far better eyes than yours.

 

[raebrm]

Hi thanks for your replies

I measure the eye diagram from the electrical output of a high speed receiver photodiode. I attached a laser (VCSEL) to the laser driver circuit and coupled the emitted light to this high speed receiver. Not sure what the extinction ratio is but that is the optimum eye i got.

I attach the full circuit here and highlighted the area where i touched the circuit to give the improved eye. (i did not change the extinction ratio to get the better eye, just touched the circuit)

Any idea what happened or do u notice something wrong with the circuit?

Thanks for the inputs!

 

[biff44]

are c5, c6, c8, and c9 very close to the driver chip, and do they have good via holes to a real ground plane on the board? Try finding other manufacturers chips to replace those capacitors...they might have poor high frequency performance. Did you solder the big thermal pad on the bottom of the driver to your ground plane thru many via holes? Did you try putting a pair of series matching resistors between the diode and the driver? I would try a pair of 10 ohm resistors, close to the diode. Do you have the microprocessor registers correct to ignore the PD input that you have open circuited?

Try temporarily putting a 22 uF cap across the voltage regulator output.

Operating on the assumption that something is unstable in the driver, don't just touch the whole area with your finger. Get something metal, like some tweazers, and touch one part at a time and see while parts are sensitive to touch.

Rich

 

[raebrm]

hi and thanks for the reply

c5,c6 are within 15 mm to the driver chip
c8,c9 are within 2 mm to the driver chip

all have direct via connection to the ground plane

big pad of the laser driver is soldered to a ground pad with 4 direct via connections to the ground plane

register is set so that the driver operates in open loop

i tried 20 ohm series matching resistors and it improved performance a bit. but it's still better if i touch the area.

will try with 10 ohms later. any reason why you suggest a value of 10 ohm?
will also look into the 22 uF across voltage regulator and metal tweezer test.

thanks again for the help everyone, appreciate it lots

 

[biff44]

c5 and c6 are too far away. Put them right on the chip's lead! We are talking about a 10 Ghz signal here! 1.5 cm might as well be a mile away! Myabe there is a way to add a test capacitors right at the driver chip to see if the ringing goes away? Try a 10,000 pF ceramic chip. You need a short connection to the ground plane, though.

I would be worried about c7 also, if it is too far away from the driver's pin. Also, you might want to try a 1 uF cap right on the VCC side of L7.

I would also be tempted to try a 10K to ground at the PD pin....just in case.

If you tried the 20 ohm resistors, that is good enough. I was just wondering if there was some driver output instability, and some series loss would have probably stopped it if it was there.

 

[karesz]

Hi, I would in all cases pack at 2mm to both pins (Nr16&20) a plus 1nF good HF -qualityporcelan (ATC) or min. NPO capacitors, than L4 on Vcc; is also to good bypass(1nF II 100nF) + a small tantal of couple of uFs!!
At 10GHz are the 100nFs for me illusions!:-(, but more usable good 1nF, ev. 10nF II on it & possible you need on all bypassed pins a 100pF more, than typical capacitors arnt enoug good in the full bandwith up to min. 20GHz! So can a 100pF + 1nF parallel better be as 100nF with 10nF in x mm away from pins....
For me are the seriell coppel capacitors a problem too; than they must be acceptable quality parts at full frequency band too, but a 100nF is not (extrem)HF part!! Pls solder on all minimum a parallel (good HF) 1nF too...
Than; as Biff asked you some lines over their: are you sure, that the thermal pad(IC backsides big solder pad)is true soldered to your PCB? Try it to reflow-resolder pls...
In my opinion are he output stage of driver IC & the VCSEL biasing too_ clearly not good bypassed .
K.

 

[biff44]

Technically, it is 10 gbps, not 10 Ghz. He probably has important harmonics at 30 and 50 Ghz! It is a wonder these laser drivers work at all!

 

[karesz]

Exactly, that is the reason why I wrote more times over needed bandwidth for the bypassing over/minimum 20GHz
You are writing to first over 10 GHz by distances of capacitors from driver IC.
K

 

[rfmw]

If you touch the L1 or L4 alone (not any other component) does the eye change? Seems you have a hidden resonance somewhere...

 

[karesz]

Hi raebrm,
What is progress pls?
K.

 

[raebrm]

Hi,

Thanks for the replies. I've been away for a while so I could not work on this.

To summarize, these are the things that I should try right?
1. Put 22 uF capacitor after regulator - biff
2. Put the bypass caps closer - biff
3. 10 kOhm resistor to GND on PD - biff
4. Use combination of 1pF + 1nF, 100nF, tantal of capacitors a couple of uFs for bypass - karesz
5 Touch L1 and L4 alone to see if the noise reduces - rfmw

To answer karesz:
1. yes, big pad is soldered to ground using reflow
2. L4 is required since that's the bias line. no high freq current must flow there. but yes, bypassing Vcc there will help.

The one i tried:
1. Try a 10 Ohm resistor in the driver line to damp oscillation - helps, Q-factor now 8.5. Eye diagram attached.

Will try to implement the rest slowly.
Thanks guys!

 

[karesz]

...
4. Use combination of 1pF + 1nF, 100nF, tantal of capacitors a couple of uFs for bypass - karesz...

Hi,
tnx for your note!
No, Im writing over 100pF-not 1pF...
Good luck!
K.

Added after 9 minutes:

Some days ago loaded somebody up on another topic 100nF MW-Extrem Broadband capacitor TypeNr; I found datasheets from ATC &AVX in these subject; the solution is me new, but maybe helps for you?
Both products are (in my opinion the same, but ; its life