USB Hub Noise Filter: Reduce 500mV Amplitude & 10MHz Frequency

[chakr]

Hi,
I designed a small wireless sensor that can run from 5V USB power. I use a USB hub to power my device via a USB cable. There is so much noise on the 5V USB line that my device readings turn inaccurate. Using the oscilloscope I can see noise frequencies of 10MHz and down to 100kHz, assuming the scope gives me accurate frequency values. The noise amplitude is almost 500mV.
Can anyone recommend a circuit that can help to get rid of the noise?

Thank you.

 

[berkeman]

Hi,
I designed a small wireless sensor that can run from 5V USB power. I use a USB hub to power my device via a USB cable. There is so much noise on the 5V USB line that my device readings turn inaccurate. Using the oscilloscope I can see noise frequencies of 10MHz and down to 100kHz, assuming the scope gives me accurate frequency values. The noise amplitude is almost 500mV.
Can anyone recommend a circuit that can help to get rid of the noise?

Thank you.

Can your sensor run on 3.3V? If so, use a linear Low-Dropout Regulator (LDO) to make a quiet 3.3V supply for your sensor.

If not, use a boost DC-DC converter to boost the 5V input up to 8V or so, then use a linear regulator to make a quiet 5V rail for your sensor.

What kind of sensor is it? If it is wireless, why not run it off of a battery instead of the USB power?

 

[chakr]

Hi, thank you for your reply. The sensor has an internal LDO which steps the voltage down to 3.3V. I have a 10 and 100 uF decoupling caps at the output of the LDO. The noise still gets through. The unit can run off of a battery, but when USB is plugged in the battery gets disconnected from the main circuit.
If I connect unit ground to AC ground the problem seems to go away, but this is not a viable solution in my case.
Thank you again

 

[berkeman]

Hi, thank you for your reply. The sensor has an internal LDO which steps the voltage down to 3.3V. I have a 10 and 100 uF decoupling caps at the output of the LDO. The noise still gets through. The unit can run off of a battery, but when USB is plugged in the battery gets disconnected from the main circuit.
If I connect unit ground to AC ground the problem seems to go away, but this is not a viable solution in my case.
Thank you again

There are some subtle tricks in laying out an RF circuit/sensor. I've worked on product designs for RF sensors in the past, and we had to get a few things right or our RF section was desensitized badly (many 10s of dB). If done correctly, you should not be getting noise through the LDO regulator -- that is what they are used for. It does sound like the grounding may be the issue. You need to be sure that the RF circuit and any digital circuit on the layout do not share any ground impedance, for example...

 

[chakr]

I see I need to verify the ground inside the unit. The Sensor unit is powered from a USB hub and it has a common ground with the Sensor internal circuitry. There is a long USB cable between the USB hub and the unit. I wonder what it does to the unit ground? The long USB cable, does it make the ground noisier or less effective?
Thank you.

 

[Borek]

I have a 10 and 100 uF decoupling caps at the output

I always thought one uses inductors to filter out high frequency noise, what am I missing?

 

[CWatters]

May not be a cure but perhaps try some 47 or 100nF caps as well as the existing 10/100uF. Ideally should also be caps on both sides of the LDO.

 

[chakr]

Hi, that is what I want to try next. The noise frequency is up to 10MHz with 0.5V amplitude. It is a very noisy USB hub. I am not sure how to select a propper inductor. I know that inductors are not perfect and there is a stray capacitance present in them. I want to try a relatively large inductor, but I am worried that parallel capacitor will become a dominant component.
Thank you.

 

[berkeman]

What is the dropout voltage of the LDO? Does it have any specs for input-output noise rejection over frequency?

Can you share the schematic for this board? Can you share the layout?

 

[gnurf]

If you have a working prototype, I'd try to shunt off the noise with capacitors as close as possible to the connector (on the input side of the LDO). Try to minimize the mounting inductance, but even a less-than-perfect implementation should give you instant hope if there is any. You already have plenty inductance in the cable which will act as the L in your LC filter, so I'd worry about adding bulky inductors later.

 

[chakr]

The circuit is attached. Below is the link to the component datasheet:
http://www.ti.com/lit/ds/symlink/tps737.pdf

 

[chakr]

If you have a working prototype, I'd try to shunt off the noise with capacitors as close as possible to the connector (on the input side of the LDO). Try to minimize the mounting inductance, but even a less-than-perfect implementation should give you instant hope if there is any. You already have plenty inductance in the cable which will act as the L in your LC filter, so I'd worry about adding bulky inductors later.

Thank you, I agree about the cable. I do have a 10uF cap 15-20 mil from the part, almost on the part. Should I make it larger?

 

[chakr]

What is the dropout voltage of the LDO? Does it have any specs for input-output noise rejection over frequency?

Can you share the schematic for this board? Can you share the layout?

 

[gnurf]

Thank you, I agree about the cable. I do have a 10uF cap 15-20 mil from the part, almost on the part. Should I make it larger?

No, you should make it smaller. As in physically smaller which means low inductance which is what you need to have a low impedance return path for the high-frequency harmonics (noise) you're most likely seeing. Actually, I'm not talking about the bulk 10uF capacitor on your LDO input (the location of which is probably irrelevant here and in general), I'm talking about adding a filter to your connector in an attempt to attack the problem as close as possible to the source.

I'd try this first and go from there.

 

[chakr]

No, you should make it smaller. As in physically smaller which means low inductance which is what you need to have a low impedance return path for the high-frequency harmonics (noise) you're most likely seeing. Actually, I'm not talking about the bulk 10uF capacitor on your LDO input (the location of which is probably irrelevant here and in general), I'm talking about adding a filter to your connector in an attempt to attack the problem as close as possible to the source.

I'd try this first and go from there.

I can add a separate filter circuit between the USB cable and the Sensor node. Do you think it will help?

 

[jim hardy]

Switching power supplies are electrically noisy.

Make sure you're not seeing stray pickup on your scope
by connecting both the 'scope probe and its little ground clip to same signal common point on your gizmo

If it's quiet, ie no noise, then you have capability of decent 'scope measurements.
If it's quiet then move scope probe to +5 line but leave the ground clip on common.

C33 is probably electrolytic and will need to be bypassed by a small high frequency capacitor.
That's because electrolytics are made by rolling a strip of aluminum into a coil and inserting that into a tube.
That coil has a tiny amount of inductance and at high frequency the inductance becomes significant impedance.
Seems crazy to bypass a 10 with a 0.1 but that's how it is done.

Used to be we wouldn't use a switcher with analog measuring equipment.
Of course i might be all wet. But i have chased a lot of noise over the years.

old jim

 

[chakr]

May not be a cure but perhaps try some 47 or 100nF caps as well as the existing 10/100uF. Ideally should also be caps on both sides of the LDO.

Thank you, I can try that.

 

[chakr]

Switching power supplies are electrically noisy.

Make sure you're not seeing stray pickup on your scope
by connecting both the 'scope probe and its little ground clip to same signal common point on your gizmo
View attachment 229990

If it's quiet, ie no noise, then you have capability of decent 'scope measurements.
If it's quiet then move scope probe to +5 line but leave the ground clip on common.

C33 is probably electrolytic and will need to be bypassed by a small high frequency capacitor.
That's because electrolytics are made by rolling a strip of aluminum into a coil and inserting that into a tube.
That coil has a tiny amount of inductance and at high frequency the inductance becomes significant impedance.
Seems crazy to bypass a 10 with a 0.1 but that's how it is done.

Used to be we wouldn't use a switcher with analog measuring equipment.
Of course i might be all wet. But i have chased a lot of noise over the years.

old jim

Thank you, Jim
C33 is a ceramic cap (GRM188R60J106ME47D) . I will try to place my scope the way you recommend. C33 is very close to the LDO. C33 goes straight into the ground via a VIA connection.

 

[chakr]

If you have a working prototype, I'd try to shunt off the noise with capacitors as close as possible to the connector (on the input side of the LDO). Try to minimize the mounting inductance, but even a less-than-perfect implementation should give you instant hope if there is any. You already have plenty inductance in the cable which will act as the L in your LC filter, so I'd worry about adding bulky inductors later.

 

[gnurf]

[...] C33 is a ceramic cap (GRM188R60J106ME47D) [...]

This makes me less optimistic about my initial thoughts about adding low-inductance capacitance right at the input of the connector, as the sole solution. I'd still attack the noise on VBUS close to the connector, but perhaps a (bulky) series inductor will ultimately be necessary in addition to the cap. The cap-only solution is very easy/cheap to try, which is why I suggested that first. Does it have any effect at all? Also note that the effective capacitance of GRM188R60J106ME47D [1] is 3uF with a 5V DC bias, and not 10uF as you maybe think it is. The little snippet of your PCB layout does not reveal enough to be helpful. In fact, there's a lot to be said about layout here, for all we know you've split your ground plane in some ludicrous way etc etc--small things that alone could kill your design.

Did you explore the noise rejection ratio of your LDO as berkeman suggested?

[1] https://psearch.en.murata.com/capacitor/product/GRM188R60J106ME47#.html

 

[Rive]

I use a USB hub to power my device

How is that hub powered? Can you attach your sensor to a PC for a short test? Are you sure you don't have a ground loop somewhere (maybe through your oscilloscope)?

Ps.: How does the power looks like on the input side of the hub? How does it looks like on the output side without your device?

 

[jim hardy]

Thank you, Jim
C33 is a ceramic cap (GRM188R60J106ME47D) .

Well i didnt expect that ! Bravo ! good choice.

 

[jim hardy]

just fishing...

from TI datasheet

to your question about an inductor
even a ferrite bead would help.
Dont obsess about its stray capacitance, your 10μf should swamp it. I'd worry more about Z of that single via..

Scope check might reveal something.

old jim

 

[gnurf]

[...]
to your question about an inductor
even a ferrite bead would help.
Dont obsess about its stray capacitance, your 10μf should swamp it. I'd worry more about Z of that single via..

Jim,
I think you'll be hard pressed to find a practical ferrite that will be effective across the bandwidth in question. I agree that stray capacitance of a series L is not a concern, but not for the reason you state; the unwanted parasitic capacitance will be in parallel with the inductive element, providing a low-impedance path past it, so the downstream capacitance (10uF) will not swamp it. The reason I agree to not worry is that I don't think it will be a problem at these low frequencies (unless you completely botch the layout of course, but that's another story). What do you think about that?

 

[jim hardy]

I agree that stray capacitance of a series L is not a concern, but not for the reason you state; the unwanted parasitic capacitance will be in parallel with the inductive element, providing a low-impedance path past it, so the downstream capacitance (10uF) will not swamp it.

That would make a capacitive voltage divider . Picofarads of stray capacitance and microfarads of intentional capacitance in series - which one incurs the majority of voltage drop ?

The reason I agree to not worry is that I don't think it will be a problem at these low frequencies (unless you completely botch the layout of course, but that's another story). What do you think about that?

i think we're dancing around the same concept. If your C33 doesn't have another impedance in series from stray inductance then its impedance will be way less than that of the stray capacitance around your inductor...

I think you'll be hard pressed to find a practical ferrite that will be effective across the bandwidth in question.

For your 100khz switching noise a bead certainly is underkill.
My thought was to pirate one out of some scrap electronics and see if it reduces the high frequency hash your 'scope is reporting...

First thing though is to see if 'grounding' scope probe at the circuit common cleans up the trace.
If it does you've probably been picking up radiated noise through the air and might call that 'measurement error' ..

http://www.analog.com/en/analog-dialogue/articles/ferrite-beads-demystified.html

old jim

 

[essenmein]

I can add a separate filter circuit between the USB cable and the Sensor node. Do you think it will help?

Adding a filter is what I'd do, it should actually be your default reaction to any off board signals.

Do you know if you are dealing with common mode or differential mode noise? I'd bet its common mode, which is a pain.

Anyway, Id put a small cap directly on the 5v to gnd where it comes on your board. Then add series BLM in series with both the 5V and GND, then keep the 10u ceramic and add a few smaller caps in parallel to give you better performance at higher freq. The 10u has a self resonance at ~2Mhz, 0.1u 0603 sits around 20Mhz, 10n ~90Mhz.

Place the filter as close to where the offending signal are coming on to the board and do not put any internal traces near these unfiltered dc inputs to avoid coupling the noise back in via the PCB. You'll want the small caps close to the filters rather than the LDO, they just need some large capacitance to for stability and aren't overly fussy when it comes to high speed decoupling, esp if your load is slow or sufficiently decoupled.

Then if you're not drawing much current from the 5V rail, even series resistance of a few ohm to 10's of ohm in series with each BLM will really help the caps filter noise from your supply, you can put down a couple of 0603 resistors and put zero ohm links in if the BLM's are enough.

 

[chakr]

This makes me less optimistic about my initial thoughts about adding low-inductance capacitance right at the input of the connector, as the sole solution. I'd still attack the noise on VBUS close to the connector, but perhaps a (bulky) series inductor will ultimately be necessary in addition to the cap. The cap-only solution is very easy/cheap to try, which is why I suggested that first. Does it have any effect at all? Also note that the effective capacitance of GRM188R60J106ME47D [1] is 3uF with a 5V DC bias, and not 10uF as you maybe think it is. The little snippet of your PCB layout does not reveal enough to be helpful. In fact, there's a lot to be said about layout here, for all we know you've split your ground plane in some ludicrous way etc etc--small things that alone could kill your design.

Did you explore the noise rejection ratio of your LDO as berkeman suggested?

[1] https://psearch.en.murata.com/capacitor/product/GRM188R60J106ME47#.html

Hi, I tried to add a capacitor to the USB connector. Unfortunately, it didn't help. The ground areas on the top layer are broken because I have components. Internal layers have continuous ground planes, but I do not have a dedicated ground layer. Maybe that is the problem? My choice of the input decoupling cap seems to be wrong.
Thank you for your advice. Noise is fun to deal with.

 

[chakr]

How is that hub powered? Can you attach your sensor to a PC for a short test? Are you sure you don't have a ground loop somewhere (maybe through your oscilloscope)?

Ps.: How does the power looks like on the input side of the hub? How does it looks like on the output side without your device?

Hi Rive, the hub is powered from straight AC. I have attached an image from my scope that shows the noise. The scope shows noise from the hub even when it is not connected to my device.

 

[essenmein]

Not sure what the policy on links is, but here is your problem, with a solution.

http://andybrown.me.uk/2015/07/24/usb-filtering/

 

[Rive]

Hi Rive, the hub is powered from straight AC. I have attached an image from my scope that shows the noise. The scope shows noise from the hub even when it is not connected to my device.

Ugh. I think what you are trying to repair here is not your circuit but your equipment. Please try to power it from a different source (for example, from the computer you are using to access this forum).

 

[chakr]

Not sure what the policy on links is, but here is your problem, with a solution.

http://andybrown.me.uk/2015/07/24/usb-filtering/

Thank you, this one looks promising.

 

[chakr]

Ugh. I think what you are trying to repair here is not your circuit but your equipment. Please try to power it from a different source (for example, from the computer you are using to access this forum).

Hi Rive,
I have a USB off-line AC power supply I have selected for this product. The power supply works fine. When I power the sensor from the hub temperature reading on my device changes significantly. The noise generated inside the hub affects circuits inside.
Thank you.

 

[chakr]

Jim,
I think you'll be hard pressed to find a practical ferrite that will be effective across the bandwidth in question. I agree that stray capacitance of a series L is not a concern, but not for the reason you state; the unwanted parasitic capacitance will be in parallel with the inductive element, providing a low-impedance path past it, so the downstream capacitance (10uF) will not swamp it. The reason I agree to not worry is that I don't think it will be a problem at these low frequencies (unless you completely botch the layout of course, but that's another story). What do you think about that?

Hi Jim, hopefully, the layout is not too messed up. I have ground plane polygons on the internal layers. They make about 70-80% of the surface area of layers 2 and 3. I think a large inductor or maybe 2 inductors on 5V and GND lines could help to reduce the noise. I am not sure about the values of inductors to use. The load current is about 350mA.
Thank you

 

[chakr]

just fishing...

from TI datasheetView attachment 230012

View attachment 230013to your question about an inductor
even a ferrite bead would help.
Dont obsess about its stray capacitance, your 10μf should swamp it. I'd worry more about Z of that single via..

Scope check might reveal something.

old jim

Hi Jim, I am going to order a demo board for the TPS73701DCQR, TPS73701DRVEVM-529, to verify my layout. If the output from the demo board is not noisy, then I have a layout issue.

 

[jim hardy]

Noise is fun to deal with.

I don't think you can get a good measurement anyplace near that USB hub.
You did the experiment i was hoping you would

that thing is radiating so intensely it induces voltage in your 'scope probe. A few feet of separation might help.

I'd say try powering your gizmo from a battery if you can find one that's not over the 6 volt absolute maximum rating and preferably not over the 5.5 recommended maximum operating.
Maybe series up three AA's ?
If that helps get a 4.5 or so volt "Wall Wart" that's linear (line frequency transformer and rectifier inside) . You can tell them from switchers they're about twice as heavy. Thrift shops are full of them .

First you might try placing a copper clad iron skillet over the USB hub for an EMI shield. That ought to be nice and lossy at 100 khz

Keep on having fun ! It feels so good when you conquer a pesky problem like this.

old jim