Buck converter that draws a constant current

[sveinb]

Hey!

I have this challenge: My circuit has a USB interface and some MEMS microphones on the same circuit board. They are powered by separate switched voltage regulators, both of which are powered by the 5 V USB supply voltage. The USB interface pulls more current when transmitting than not, and it transmits about 30 % of the time, in an 8 kHz cycle. This causes a low but audible 8 kHz noise to be picked up by the microphones. I've identified two routes that this noise takes: 1. The 5V supply voltage gets an 8 kHz ripple which leaks to the microphone supply voltage and 2. The current variation on the USB voltage rail causes mechanical vibration of the circuit board which is in turn picked up by the microphones. I cannot move the components to different circuit boards.

A good deal of ripple is acceptable on the USB voltage rail, but not on the MEMS voltage rail. What I want is a voltage supply for the USB interface which draws a constant current from the 5V source. This would eliminate both paths of the noise. But how can I make such a circuit? It should be a switched regulator (for efficiency), and the USB supply voltage should remain between 1.6 V < VDDUSB < 2.0 V.

[tom66]

This sounds like a classic XY problem.  You are trying to stop the current on the USB interface varying to reduce the noise, but why not address why the varying current on the USB rail causes the noise in the first place?  Most likely you have a common mode grounding issue, which is causing a radiated emission interfering with the sensitive microphone supply.    The microphonic effect is more difficult but might be addressed with mechanical changes.

Suggestions would include:

• Ensuring the USB current is balanced and common mode paths are eliminated where possible.  This means the shield on the USB should be connected to the PCB ground (directly, not with any "filter" components) and the PCB ground also connected to the USB cable/connector ground.  If your product uses a pigtail connector, the shield pigtail needs to be as short as possible and kept wrapped on the cable length as much as possible.
• Keeping sensitive circuits away from any sources of common mode noise. 
• Adding a shield over the noise generating part of the circuit, which is grounded all around the edges, to provide a path for common mode current.  You could also consider shielding the microphone/regulator circuit though you will need a port for the audio. 
• Adding additional low frequency filtering to the sensitive supplies.  Using ideally non-microphonic components for these parts.   
• Choosing an LDO with better low-frequency rejection to reduce the noise on the microphone supply. 
• Select a MEMS microphone with better PSRR, if possible. 
• For the mechanical problem, adding a way to mechanically isolate the microphone, such as putting it on a cut-out section of the PCB, or adding epoxy in the area of the board with components that appear to be generating the noise (if it's coil whine for instance).   Or perhaps you can build a baffle into the product case which provides support, or mount the microphone on a different PCB (you said you can't change the position of the microphone, but can you have another PCB with the microphones on it, within the same volume of the existing board, connected by a wire?)

There are plenty of digital devices which manage to keep audio noise low, think about a modern smart phone for instance - you will not hear the CPU switching on and off during a phone call but it will be doing that a great deal as audio compression doesn't use much CPU - and a modern phone has plenty of MEMS microphones in it.  They do not use any smart tricks like constant current regulators, they just segregate the noisy parts and minimise paths for common mode noise by shielding sections as needed.

[tszaboo]

I don't think the idea you are trying to do would work.
You can replace your original regulators with an LDO, it would be my first choice trying to eliminate the noise.
Other than that, some regulators go into low power modes, like pulse skipping more, or decrease the frequency when light load. Don't use those.
I don't think efficiency is paramount for an USB powered gadget.

[sveinb]

I should maybe add that the electrical noise path is mostly a problem when the device is connected through long USB cables with active repeaters. In that case we are talking about 20-30 meters of cable, with the inductance and resistance that comes with that. Suggestions that people should use shorter USB cables are to be expected, but not useful :-)

[tom66]

I should maybe add that the electrical noise path is mostly a problem when the device is connected through long USB cables with active repeaters. In that case we are talking about 20-30 meters of cable, with the inductance and resistance that comes with that. Suggestions that people should use shorter USB cables are to be expected, but not useful :-)

This suggests the problem is due to common mode noise from the cable system returning through the product, OR, your cable is so resistive that the voltage drop is significant when the USB transmit is active.  If it is the former, proper shielding discipline (including using shielded cables) should help.  If it is the latter, you could solve the problem by adding an LC filter that rolls off well before the 8kHz operating frequency, though the components may be reasonably large.

[GridWork]

First pass, sounds like an EMI problem. Filtering, lots of filtering. A common mode choke, capacitor, differential mode choke, capacitor is a classic method for basic filtering. You could build the emi filter up in a dusb dongle style for testing. Depending on current draw and voltage requirements, a previous suggestion of an LDO on the MEMS rail could be acceptable, especially one with a high power supply rejection ratio.

[David Hess]

I've identified two routes that this noise takes: 1. The 5V supply voltage gets an 8 kHz ripple which leaks to the microphone supply voltage and

So apparently the line regulation of the regulator for the microphone supply is insufficient.  I would double check that this is not caused by a ground loop.  Make sure that the board uses a single point ground for the separate supplies.

Do you really need the higher efficiency of switching converters?  What are your supply voltages?

and the current variation on the USB voltage rail causes mechanical vibration of the circuit board which is in turn picked up by the microphones. I cannot move the components to different circuit boards.

It seems unlikely that the board itself is the problem, however ceramic capacitors (piezoelectric) and sometimes inductors (magnetostriction) can produce mechanical vibration from their changing voltage and current.  This can be made worse with certain switching regulator modes like variable frequency, burst mode, or if subharmonic oscillation is present in a current mode converter.

Besides choosing more suitable switching regulators, better ceramic capacitors could be used, or ceramic capacitors could be replaced with some other capacitor type.  I would be very tempted to use polymer tantalum or solid tantalum capacitors.

[magic]

I've identified two routes that this noise takes: 1. The 5V supply voltage gets an 8 kHz ripple which leaks to the microphone supply voltage and 2. The current variation on the USB voltage rail causes mechanical vibration of the circuit board which is in turn picked up by the microphones. I cannot move the components to different circuit boards.

How exactly do you know it, particularly the latter, or is it only guesswork?

Is this a digital mic or analog, feeding an ADC in the USB chip? If the latter, I suppose you may need to consider the ADC's PSRR too.

As a quick and dirty experiment, I would try powering the mic with external batteries and see if it helps.

[Smokey]

Doing USB "right" becomes an legit high speed design issue. 
Are the USB traces impedance controlled to the correct value?  Are they super long?  Are they routed over splits in the ground plane?

https://www.silabs.com/documents/public/application-notes/an0046-efm32-usb-hardware-design-guidelines.pdf

Anything sensitive and analog probably deserves it's own local voltage regulation.  Pick an LDO with a good PSRR and put it right next to the microphones. 

Since it sounds like this is a mixed analog/digital design, have you been exceedingly careful to segregate the analog and the digital stuff so they don't run through the same sections of the solid ground plane? 

[tom66]

I've identified two routes that this noise takes: 1. The 5V supply voltage gets an 8 kHz ripple which leaks to the microphone supply voltage and

So apparently the line regulation of the regulator for the microphone supply is insufficient.  I would double check that this is not caused by a ground loop.  Make sure that the board uses a single point ground for the separate supplies.

If you're talking about splitting the ground plane - apologies if I misread you - generally this is a bad idea.  Everything should return to a single common ground, with as low of an impedance as possible (which means no splits and ideally a dedicated plane layer.)

There are cases where split ground planes are necessary such as with precision analog.  But audio usually doesn't require it when designed right.  Most high speed ADC datasheets will have analog and digital grounds going to the same plane. 

[NiHaoMike]

Try adding a large (1mF or more) capacitor at the USB power input? Note that if that quick test worked, to stay within inrush current limits, you'll need to add a precharge resistor and bypass MOSFET in the final design.

[sveinb]

Quote from: sveinb on January 25, 2024, 11:32:34 am

I've identified two routes that this noise takes: 1. The 5V supply voltage gets an 8 kHz ripple which leaks to the microphone supply voltage and 2. The current variation on the USB voltage rail causes mechanical vibration of the circuit board which is in turn picked up by the microphones. I cannot move the components to different circuit boards.

How exactly do you know it, particularly the latter, or is it only guesswork?

Is this a digital mic or analog, feeding an ADC in the USB chip? If the latter, I suppose you may need to consider the ADC's PSRR too.

As a quick and dirty experiment, I would try powering the mic with external batteries and see if it helps.

There are several indications that there is an acoustic component. The strongest indication is that the phase of the noise is different in the different microphones on the board. If it had been purely an electrical noise path, I would expect the noise to be in phase.

The mics are digital, and I have tested using an external supply. That didn't solve the problem (this was before I had concluded that there must be an acoustic path also)

[sveinb]

Try adding a large (1mF or more) capacitor at the USB power input? Note that if that quick test worked, to stay within inrush current limits, you'll need to add a precharge resistor and bypass MOSFET in the final design.

This was probably the first thing I tested, and it didn't eliminate the problem. However, this was before I had concluded that there was also an acoustic noise path, so this might solve the electric part of the problem.

[sveinb]

Someone outside of this chat pointed out another thing I hadn't considered: As mentioned, the noise problem gets more pronounced when using long usb cables with active repeaters. The longer cable means that the 5V supply gets softer and more susceptible to ripple caused by the USB interface's varying current draw. However, the active repeaters in the cable itself are also powered off the same 5V supply, so that supply is likely to have an 8 kHz ripple even if my circuit doesn't create it.

[tom66]

Fundamentally you will usually have no control over the USB source.  You can probably assume that the computer could put 100mV of ripple and noise on the USB supply at 8kHz and still be compliant with the USB specification.  As far as I can tell, USB makes NO requirements about the noise on VBUS, it only specifies a minimum voltage under load and things like that.  So you will need to build your product to withstand supply noise and not try to stop it happening in the first place because someone will just invent a noisier source even if you do manage to keep your current nice and close to DC.

[magic]

active repeaters in the cable itself are also powered off the same 5V supply, so that supply is likely to have an 8 kHz ripple even if my circuit doesn't create it.

Right, 8kHz is the microframe interval so anything doing periodic hi-speed transfers is likely to contribute.

There are several indications that there is an acoustic component. The strongest indication is that the phase of the noise is different in the different microphones on the board. If it had been purely an electrical noise path, I would expect the noise to be in phase.

Makes sense, although I wonder if other factors could explain such differences, perhaps physical orientation if it's a matter of RFI pickup?
Mind that speed of sound in solids tends to be a 4 digit number (m/s), corresponding to tens of cm wavelength at 8kHz and not much phase difference to be seen on small distances.

MLCCs would be candidate for the source of vibration.

The mics are digital, and I have tested using an external supply. That didn't solve the problem (this was before I had concluded that there must be an acoustic path also)

In such case is there even any electric component at all, if external supply made no difference?

Try adding a large (1mF or more) capacitor at the USB power input? Note that if that quick test worked, to stay within inrush current limits, you'll need to add a precharge resistor and bypass MOSFET in the final design.

This was probably the first thing I tested, and it didn't eliminate the problem. However, this was before I had concluded that there was also an acoustic noise path, so this might solve the electric part of the problem.

Did it significantly reduce supply ripple without reducing recorded noise?

[sveinb]

Update: I've done more experiments and measurements and found the source of the problem. There are five MLCCs on the 5V rail (4.7 uF + 4 x 100 nF). Replacing them with a single 100 uF surface mounted tantalum cap reduced the noise by 30 dB. Replacing the tantalum cap with a leaded 4.7 uF electrolytic cap reduced the noise by a further 10+ dB (it dropped below the noise floor at the settings I was using). The noise was simply proportional to the ripple on the 5V rail. Using a long USB cable increased the ripple (to 100 mV pp) and thereby increased the noise. Many thanks to everyone who took the time to think about my problem and suggest solutions. And full score to David Hess and magic who suggested that the MLCCs could be to blame!

However, I don't have room for a leaded component. There's plenty of horizontal space, but only 1.5 mm free height for components. So now I'm looking for advice on quiet capacitors that would fit. Since that doesn't match the title of this thread, I'll start a new one: https://www.eevblog.com/forum/chat/quiet-capacitors/.