Noise Troubles With BNC Connectors: Investigating Causes & Solutions

[nitin.jain]

Hi!
I guess this thread (could) bear similarities to the discussion in the "Low noise low pass" https://www.physicsforums.com/showthread.php?t=332819"
thread, but I didn't want to hijack that :) so beginning a fresh one.

I find that the electronic noise gets aggravated, when I connect the output of an amplification stage to a BNC connector.
To elaborate, I need to have the (opamp's) output high-pass filtered first (a simple RC filter, built using a SMD capacitor and resistor, with fCutoff being ~ 100 kHz) and then it reaches to the BNC connector with a ~2.5 cm long shielded cable. The ends of the cable are soldered to the PCB and the connector, so that portion (< 0.5 cm on both ends) is unshielded, of course.
I've attached the noise spectrum for these two measurements. The top figure is using a probe at the output pin of the opamp and the bottom being the output from the BNC connector cable (i.e. after the signal has passed thru the HPF and shielded cable) - you can observe the gain peaking in 0-120 MHz frequency band and the radio-frequency peak (around 910 MHz) becoming quite worse in the bottom figure.

The circuit as such is placed inside a steel box, with its chassis sharing the power-supply ground. I've attached a schematic of it, most of the details are in this thread:
https://www.physicsforums.com/showthread.php?t=327510&highlight=cascading"

In essence, it serves to amplify the difference of photocurrents (Idiff) using two successive opamp stages. The idea is to be able to do this fast enough (i.e. have a high bandwidth of operation) so that, if the input to the photodiodes is an optical pulse train (with a repetition rate in MHz, typical to most pulsed lasers these days), then the statistics of Idiff can be collected quite fast & efficiently.

Coming back to the noisy troubles, my question is now, is it really something like bad impedance matching and/or poor shielding, that causes the noise to aggravate so nastily? Or could it be that a normal BNC cable is the culprit, because the probe-cable (that I use for measuring the opamp output) would be pretty well shielded.
Lastly and most relevantly, how should I try to fix this? :)

Thanks aprior!

 

[waht]

Wow that's a cool op-amp OPA 847 with a bandwidth of 3.9 GHz. Didn't know they made these already.

I assume the output impedance of the op-amp will be high, and the impedance of the transmission line (2.5 cm of coax) will be lower (~50 ohms). To match the high impedance to low consider using an L match or a Pi filter instead of the RC high pass after the output stage. These filters can be easily designed to match high-low impedances.

Also, when you took the measurement was the BNC connected to a load? That could make a difference.

Basically, the added connector with shielded cable works as a quarter wave or has enough LC to resonate around 100 MHz in this case. You have to tune it out or replace the cable completely perhaps with a semi-rigid UT-141 or 086.

 

[Bob S]

Hi Nitin-
Do you have more than 1 ground connecting the pc board to the chassis? I would suggest having only one common ground. Right now you have one where the dc comes in and another where the signal leaves. The dc power should be bypassed to the chassis where it enters. Often, a noise signal occurs because of ground currents flowing in coax shields. These can be minimized by putting ferrite beads or chokes around the coax where it goes from the pc board to the chassis. The noise bump at 100 MHz could be related to a reflected mismatch on a 5 ns (~40") long cable.
Is your vertical scale in dBm below 1 milliwatt? What is your analyzer bandwidth? How far above kTB are you?
Bob S

 

[Mike_In_Plano]

Not knowing all the issues, I can see one right away. You need to decouple the output op amp to the BNC, so that the last op amp doesn't unusually low impedance or reactance on the BNC. It's tradition (with good reason) to place a source and termination resistance at either end of the BNC. Most analyzers have 50 ohm terminations, add about 50 ohms between your op amp and BNC and see if that helps.

Also looking, I can see a couple of other things, but it's bedtime here, so if you need or want any more input, perhaps wee can discuss it on the morrow.

Best Wishes,

Mike

 

[Mike_In_Plano]

Hello Nitin,

I apologize for my foggy reply. Last night, it was near bed time...

Anyway, to start, the op amps need to be treated as the RF parts they are. This implies bypass capacitors from each supply to ground plane very close to the package and decoupling of the power supply.

Typically, I'd attempt to scrunch a pair of 1nF 0603 caps quite close to the power leads and bring the grounding of the caps together promptly and then on to a via to the ground plane. Then, on redundant traces connect the supply leads to your typical .1uF ceramic caps which also ground together and go to the plane.

To ensure that one stage doesn't talk to another via the supply lines, put in a decoupling network. I usually make these using a ferrite bead, resistor, and capacitor on each power line. I choose a ferrite bead that's at least 50 ohms in the range of interest, place a resistor that's about 1/2 - 3/4 this value to keep it from ringing, and place a .1uF ceramic cap to ground on the power supply side. Be sure to route to the cap and then from the cap to power and keep the ground trace of the cap short.

Also, the power supply lines to the PIN diodes can get you in trouble. I see you've placed a series resistor on the top PIN, but you need to do the same with the bottom. You cannot count on capacitors to get the noise off of the power supply lines by themselves.

Back to the basic function, transconductance amps can be stubborn designs. For example, I see that yours has a noise peak down around 100MHz. I've found that dealing with capacitance on the inverting input of that first stage can be troublesome, particularly if your demanding a lot of gain off that stage.

One fix that I've had success with is to get some of your gain from T divider. That is, divide the output of the first op amp using fairly low value resistors (i.e. 4k and 1K) by about five and then use that as your feedback to the input. This allows you to decrease the value of your feedback resistors and increase the value of the feedback cap.

This method will give you a more reproducible design, but there is a cost. The noise increases. So, it's best to stick with about 4X or 5X gain using this technique.

Getting back to the original response, op amps don't appreciate reactive loads, particularly capacitive. A few tens of ohms at the output of each well help with capacitance downstream and the BNC load.

 

[nitin.jain]

Hi everyone,
Thanks very much for your prompt replies. Firstly, some apologies for a few missing details as well as small mistakes in the schematic I sent you, I've updated that now and have also added a pic of the front side of the actual circuit (on PCB), showing various portions that have been discussed in here. The pic is not very clear, unfortunately, but I still hope it shall be possible to relate it with the schematic.
In the next few replies, I'll try to answer the questions or add to whatever I can, and I hope you can tell me if it sounds correct or not?

 

[nitin.jain]

I can't remove the earlier schematic for some reason. Putting in the new one!

 

[Mike_In_Plano]

Nitin,

Looking at your PCB, it looks as though you have a single sided board with large, long traces. This sort of construction will have a great deal of parasitic inductance and capacitance. I think it would be challenging to get a functional circuit with the bandwidths you're working with.

You could probably help the design a bit by placing the power supply bypass capacitors on top of the ICs. Aside from that, I don't know. Without ground planes and controlled traces it's hard to work in the UHF and beyond.

- Mike

 

[nitin.jain]

Do you have more than 1 ground connecting the pc board to the chassis? I would suggest having only one common ground.

I have GND planes on both the front & backside of the PCB. On the backside, the GND plane covers around 80% of the board and is continuous too. It's removed out near the opamps (their feedback paths, to be precise) and the resistor bridge between them.
I've interconnected the various small and disconnected planes on the front side (hopefully visible in the pic) with the back GND plane, by soldering-thru a few holes in the PCB. So in a sense, it should be one common GND plane only, right!?
This plane is then finally connected to the chassis of the box using a normal (unshielded) cable, as shown in the pic I've attached.

The dc power should be bypassed to the chassis where it enters. Often, a noise signal occurs because of ground currents flowing in coax shields. These can be minimized by putting ferrite beads or chokes around the coax where it goes from the pc board to the chassis.

Yes, I am using ferrite beads on the +/- 15V supplies, before they enter the bus on the board.

Is your vertical scale in dBm below 1 milliwatt? What is your analyzer bandwidth? How far above kTB are you?

Yes. The values in the spectra are typically between -60 to -90 dBm, so that would mean something like a 1-1000 pW.

For the measurements I had shown in the first post, the analyzer's video bandwidth = resolution bandwidth = 5 MHz.

Umh, is kTB somehow related to thermal noise? Sorry, I don't get the context completely.

 

[nitin.jain]

Hi Mike, you were too fast! :)
The board is double-sided, I've explained a few things in my latest post. I've also attached a pic of how the backlayer looks like.

On the front side, and between the signal or supply traces, I intentionally left copper (with sufficient clearance) to form disjointed GND planes, that I later connected to the back GND plane (sort of the "mother" GND plane).

Placing the power supply bypass capacitors on top of the ICs might be something I can try... Since the mother GND plane is absent "below" the opamps as such, I can maybe connect them to the front planes, using tiny wires... what do you say?

 

[Bob S]

For the measurements I had shown in the first post, the analyzer's video bandwidth = resolution bandwidth = 5 MHz.

Umh, is kTB somehow related to thermal noise? Sorry, I don't get the context completely.

Hi Nitin-
For a 5 MHz bandwidth, the thermal noise at the source (if it were resistive) would be -107 dBm. If you had a 3 dB noise figure for your first amplifier input, the noise would be -104 dBm (for 5 MHz BW). If you had 30 dB gain, then the noise power at output would be -74 dBm per 5 MHz bandwidth. See
http://en.wikipedia.org/wiki/Johnson–Nyquist_noise

Bob S

 

[nitin.jain]

It's tradition (with good reason) to place a source and termination resistance at either end of the BNC. Most analyzers have 50 ohm terminations, add about 50 ohms between your op amp and BNC and see if that helps.

Yes, sorry I forgot to mention that I was using a 50 ohm connection between the BNC and the output of the high pass filter stage (after the opamp) as well. I've updated this in the schematic now.

To ensure that one stage doesn't talk to another via the supply lines, put in a decoupling network. I usually make these using a ferrite bead, resistor, and capacitor on each power line. I choose a ferrite bead that's at least 50 ohms in the range of interest, place a resistor that's about 1/2 - 3/4 this value to keep it from ringing, and place a .1uF ceramic cap to ground on the power supply side. Be sure to route to the cap and then from the cap to power and keep the ground trace of the cap short.

I have been using the ferrite beads and the bypass capacitors too, though the values were chosen rather without much logic :) I shall try what you suggest in above. Could you maybe elaborate on the very last sentence in above, though?

Also, the power supply lines to the PIN diodes can get you in trouble. I see you've placed a series resistor on the top PIN, but you need to do the same with the bottom. You cannot count on capacitors to get the noise off of the power supply lines by themselves.

Yes, sorry again for being lousy while having posted the schematic. I indeed use identical series resistors and capacitors on both the photodiodes. In the current board pic, 1-4 are the bypass capacitors on the two opamps, 5 and 6 are on the diode-biasing supplies (+/- 15 V).

One fix that I've had success with is to get some of your gain from T divider. That is, divide the output of the first op amp using fairly low value resistors (i.e. 4k and 1K) by about five and then use that as your feedback to the input. This allows you to decrease the value of your feedback resistors and increase the value of the feedback cap.

I've also put the value of the feedback resistances I am using in the last schematic... based the on the datasheet of the opamp (which mentions a detailed application for making a trans-impedance amplifier), I believe I am using rather low values, but I'll be grateful if you could confirm that? If you think it still could be a problem, then I shall try the T-network solution.

Getting back to the original response, op amps don't appreciate reactive loads, particularly capacitive. A few tens of ohms at the output of each well help with capacitance downstream and the BNC load.

Okie! I guess I should try increasing R19 then, right?

 

[Mike_In_Plano]

I found something interesting about the op amp you chose. It has a high closed loop gain requirement (12?).

This type of amp gives wide bandwidth at higher gains (and advertises a crazy high gain bandwidth), but is incapable of operating at lower gains. Conversely, a more common op amp works at unity gain, but has less gain bandwidth.

Ti probably sells a couple of different op amps that are identical, except for a compensation capacitor.

I'd like to plug on this in Spice, to see what makes it stable. Can you give me some capacitance versus voltage data for the PIN diodes?

 

[nitin.jain]

I found something interesting about the op amp you chose. It has a high closed loop gain requirement (12?).

This type of amp gives wide bandwidth at higher gains (and advertises a crazy high gain bandwidth), but is incapable of operating at lower gains. Conversely, a more common op amp works at unity gain, but has less gain bandwidth.

Ti probably sells a couple of different op amps that are identical, except for a compensation capacitor.

I'd like to plug on this in Spice, to see what makes it stable. Can you give me some capacitance versus voltage data for the PIN diodes?

Thanks Mike, the photodiode I'm using is a custom-made product, so I have capacitance values at only 4 discrete voltages, I hope that shall do? Here they are:
0 V - 10.5 pF (at 100 kHz)
5 V - 5 pF
10 V - 2.5 pF
20 V - 2.2 pF

I am using a reverse bias of ~15 V, as such.