1
\$\begingroup\$

I have a device I designed which utilizes 32 SK6812 LEDs (RGBW neopixels).

I am experiencing noise in the 1.2khz range (SK6812 LEDs use a 1.2khz PWM signal to control color and brightness) and its associated harmonic frequencies. This noise is most apparent in the 5v line on my device (the SK6812 LEDs are powered on this line).

The device is powered by a modular rack which has +12v, Ground , -12v. The 5v power for the SK6812 is supplied by a switched power supply that drops +12v to 5v5, which goes through an LDO and then to the 5v line where the SK6812s are connected.

The noise is apparent on other modules connected to the same power rails. These are mostly analog audio modules, and when they are completely silent, when the gain is boosted 65db, the noise becomes apparent.

Here is a photo of the spectrum of the noise recorded:

enter image description here

Here is a photo of the spectrum of the rack with my device disconnected completely:enter image description here

I am pretty certain that this noise is coming from the LEDs because (1)the noise is only apparent when the LEDs are on, and changes slightly when different LEDs change color, and (2) because most of the noise heard is at the same frequency as the PWM rate for the LEDs.

One obvious solution is to increase bypass capacitors, and possibly use higher capacitance aluminum electrolytic in addition to my 0.1uF ceramic capacitors

My questions are these:

1) What value (and type) of capacitor do you recommend to lower low frequency noise in the 1khz - 5khz range? With 30 LEDs, how many would you reccomend? 2) are there any clever (and cheap) ways to isolate my device completely where it connects to the rails? Ferrite Beads? Inductors? What type do you reccomend? 3) I plan on moving to a solid ground plane (I am currently using a hatched - which I recently found out doesn't help) as well as adding a ground plane to the top layer (2 layer board). Do you think adding additional copper on ground might help the bypass capacitors work better? 4) Anything else I am forgetting to ask?

Thank you!

Edit: When measuring with a scope I get 530mVpp on the 12v rail with LEDs on, and 480mVpp with LEDs off

on the 5v line, I get 540mVpp with LEDs on and 500mVpp with LEDs off

\$\endgroup\$
4
  • \$\begingroup\$ Why not use a completely separate supply for the LEDs? \$\endgroup\$ May 14, 2017 at 3:05
  • \$\begingroup\$ I tried this, the rails also provide a 5v line (which I do not currently use) and the noise still persisted on the 12v line. I think its because they all share the same ground, and I believe the noise is being fed back through the grounds somehow. \$\endgroup\$
    – cosmikwolf
    May 14, 2017 at 3:10
  • 2
    \$\begingroup\$ If the ground is common it's not separate enough- you can tie it at one point but the LED current should flow only to the LED supply. \$\endgroup\$ May 14, 2017 at 3:12
  • \$\begingroup\$ Why does the new plot have a noisy Red line added above the lower spectrum? PLs identify sources. I asked for DC supply noise spectrum (ac coupled) \$\endgroup\$ May 14, 2017 at 4:42

2 Answers 2

1
\$\begingroup\$

That is the nature of these neopixel devices. They have no output inductance or capacitance on the PWM switching. They generate a lot of EMI.

With a PWM frequency of 1.2Khz, the 1.2khz noise you are experiencing is expected.

It would be best if the 5.5 switcher, 5v LDO, and LEDs had their own ground plane.

Follow the PCB layout and decoupling guidelines to reduce conducted EMI.
This Vishay Engineering Note covers them well.

Insert an input filter on the 12V line in to the switcher.

This is the EMI filter TI recommends on the Vin on their TPS92511 LED driver to suppress conducted EMI. For frequency between 1.2khz and 2.4khz the values would be around 10µH and 68µF. enter image description here

\$\endgroup\$
0
\$\begingroup\$

No that won't help your noise is broadband with harmonics of 20Hz with some phase noise around 1.2kHz . Any clues why?

enter image description here

The problem appears to be load regulation noise , which does not follow the SMPS loop bandwidth or the slope of a capacitor.

What is your peak current and supply current rating?.

An answer is forthcoming but some obvious reduction of conducted noise is needed of "digital" LED noise and Analog supply. This means the source impedance of the supply including wire causes voltage spikes from pulsing the LEDs with ~1.2kHz PWM and some 20Hz causes yet to be defined.

The usual fixes are;

  • isolated wiring directly from DC output to each load, analog and digital, so there is no current sharing and thus no drop on the power and ground for analog along the digital power wires.

  • for audio. use low ESR cap 100uF Tantalum with a small series R with 100mV drop max.

  • Use better quality regulator with low ESR caps in your 5 V supply.

and the most obvious.

Separate regulators for Audio and LED power. (with grounds tied but not sharing current.) (Show block diagram)

enter image description here

I was just watching PGA PLayers Tourney on DVR and decided to show you how I would have demonstrated to a design team, a Root Cause Failure Analysis due to a poor choice of X7R capacitors for filtering pulse noise with low ESR LED's even though fairly low current and good Murata brand parts. These are not a wise choice for low frequency switched loads and low f PWM at 1.1kHz

  • much greater than 20kHz PWM would be better (if possible) to avoid audio crosstalk , even better don't share 12V sources.

schematic Although not self explanatory to most young designers, X7R is the poorest cheapest Ceramic Cap type with high ESR and Dissipation Factor

Better choices are ; Tantalum SMD, ultra low ESR Alum, Plastic Film, NP0/C0G Ceramic for low Dissipation Factor @f Hz or ESR

\$\endgroup\$
11
  • \$\begingroup\$ Ok, this made me think a bit more about how I measured that noise. I measured it using a rack that contained other modules, and noise may be being generated by other modules. I will isolate it to just two modules and then make an update. Are there any other things you would reccomend that I measure to help pinpoint this more accurately? \$\endgroup\$
    – cosmikwolf
    May 14, 2017 at 3:29
  • \$\begingroup\$ Chose your 0V reference wisely. Keep in mind modular triple supplies use single feedback from the primary output +5? and cross load regulation errors can exist from 5 to 12V Then AC couple each DC output and record the audio spectrum with coax pref. Since 5V LDO<5.5<12V the 12V ripple should be terrible. Then share the worst. \$\endgroup\$ May 14, 2017 at 3:37
  • \$\begingroup\$ 5V LED load regulation error on 12V source shared with audio was a bad idea. Give PSU specs pls. \$\endgroup\$ May 14, 2017 at 3:43
  • \$\begingroup\$ Updated spectrum photos where I removed all other modules from the rack. Provided a reference spectrum when my module is disconnected. The 1.2khz and 2.4khz ranges are the most problematic, as they are very audible. Tony - I will measure the Vpp on the 12v line as well as the 5v line and give you a comparison \$\endgroup\$
    – cosmikwolf
    May 14, 2017 at 3:44
  • \$\begingroup\$ 1.1kHz may be most audible due to your speakers it may sound like a 20Hz buzz with dominance around 1.1kHz but the cause of the problem is broadband high DC load modulated noise not just PWM. but the change in LED flashing perhaps at 20kHz rate.err I meant 20Hz \$\endgroup\$ May 14, 2017 at 3:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.