Monday, March 11, 2019

Quieting an insanely (RFI) noisy LED floodlight

A friend of mine recently installed some inexpensive Chinese-made floodlights to illuminate his backyard, but was dismayed to discover that when they were on, his 80, 40 and 20 (shortwave - 3.5-14.5 MHz) reception "went away", replaced with a very strong noise that was "20 over" - a degradation of apparent sensitivity of much more than 20dB.  As it turned out, almost every frequency below and above this range he checked was also affected to a similar degree.
RF noise from "grow lights" - the same phenomenon

Several years ago, there was some noise (pun intended) in the Amateur press about LED power supplies being sold that caused a tremendous amount of RF interference - and many of these stories also included anecdotes of many of these interference sources having been tracked down and found to have been "grow" operations.  Later, some stories surfaced where law enforcement officers were able to locate some of these "grow ops" simply by finding the source of RF interference.
The LED power supply described on this page is of the same type that was found to cause these very high levels of RF interference.

Even though these lights aren't turned on very often, he decided that their flaws went firmly against his eternal crusade against RFI-generating devices at his house.  After all, when it comes to RF interference, one should remember this cardinal rule:

Most RFI begins at home!

To be sure, there are many cases in which there are noisy power lines or a neighbors plasma TV - just two in a long list of things that can cause interference, but the worst offenders in generating interference are likely in one's own house.   The main reasons for this are simple and (for the most part) obvious:
  • They are nearby.  If a noise generating device is in your house, it's very close-by - and the closer it is, the more your antenna is likely to intercept "grunge" from that device.
  • They are connected to the same wiring as everything else in your house.  There's nothing like a piece of copper to convey RF all over the place with minimal loss, and if a noise generator is powered from the mains, it's likely conducting much of that noise into the same mains connections that power your radios.
  • If you are like most amateurs, you probably have radiating feedlines on your HF antennas.  By their very nature, almost all HF antennas tend to radiate a bit of RF on their feedlines.  For some antennas (e.g. dipoles, yagis, loops) this is incidental - often due to inadequate balun design, but other antennas (offset-fed antennas like Windoms, end-fed antennas) this is often by nature or design.  If the feedline of your HF antenna isn't very well-balanced (often using a "current mode" choke) some of your "noise" from the devices in your house wiring is being conducted from your shack, onto the feedline and then into your antenna.  Fixing this problem certainly warrants a series of articles itself, but suffice it to say, "noisy" devices will seem worse because of this issue than they would normally be.
Figure 1:
The constant-current LED driver with added filtering.  This LED driver
is typical of what is seen in these devices:  A rather generic, potted module
of likely-questionable lineage and quality.
Click on the image for a larger version.
What are these things?

As is typical with these inexpensive LED lamps, the power supply is a constant current module that uses PWM/switching techniques to regulate the current applied to the LED array to some value.  As can be seen Figure 1, this is simply a box with two sets of wires:  The AC (main) input on one side and the DC output to the LEDs on the other.

Because these are constant current supplies, they can be used over a wide range of LED module voltages:  22 to 36 volts, according to its label of that in Figure 1.  Noting the official "50 watt" power rating, we can do the math and see that with a constant 1500mA, the power being delivered to the LED array can vary from 33 to 54 watts, depending on its actual operating voltage.  Depending on the design, these supplies may or may not have their DC outputs isolated from the mains input via an internal transformer, so it is best to assume that they are not isolated and that the DC outputs will be line-referenced and hazardous (even lethal!) to touch.

In this example, the red and black DC leads disappeared into the body of the case where it would connect to an LED module that is (presumably!) insulated from the lamp's case.  Because you can't be sure what to expect, one must always make sure that the safety ground of these lamp housings is actually connected to the case (the ground wires in these devices are often not connected to the case at the factory!) and that it is plugged into a GFCI-protected outlet.

How bad was it?

In the case of these LED floodlights, the only connection that they had to the rest of the universe was via their power connections, so it was clearly via its power leads that they were radiating their "grunge".  To determine in some quantitative way how noisy this device was, a simple test fixture was constructed to measure the energy imparted on the mains power lead, represented schematically in Figure 2, below:
Figure 2:
Test fixture to analyze the amount of RF being conducted from the LED's current supply to its mains leads..
"Ca" and "Cb" are 0.1 to 0.47 "X" class "safety" capacitors used for mains filtering and "La" is a bifilar mains choke of at least 1 milliHenry per winding, these constituting a filter to decouple noise already present on the mains from the test fixture:  One of the filters depicted in Figure 5 could have been used for this purpose.
RF coupling transformer "Ta" consists of a Mix 31 clamp-on ferrite choke with a single wire going to the "Device Under Test" as the primary and 6-8 turns of smaller wire as the secondary to couple RF from it.  The box marked "protection" is simply two back-to-back 5.1 volt Zener diodes in series to protect the analyzer from voltage transients caused by turn on/off transients.
Click on the image for a larger version.
In this circuit we see a common-mode line filter using Ca, Cb and La forming a circuit to attenuate noise that might already be on the mains.  The goal is that when we measure RF noise via coupling transformer Ta, we are (mostly) seeing the noise from the device being tested and not that which may already happen to be on the mains.

The result of this measurement can be seen in Figure 3, below, covering the range from nearly DC to 1 GHz, with the cyan trace being with the unit turned off and the yellow trace with it turned on:

Figure 3:
Noise from the power supply as seen from 0 to 1 GHz.  The blue trace is with the LED power supply powered down while the yellow trance shows it powered up.  As can be seen, it is a potent noise generator well into the UHF spectrum - but particularly at and below 100 MHz!
The various signals on the cyan trace are off-air signals, including AM, FM and TV broadcast and 800 MHz - plus some leakage from the noisy mains through the Figure 2 filter:  Ingress of these signals is the inevitable consequence of the rather simple lash-up and not conducting these tests inside an RF-screened room!
Click on the image for a larger version.
While this test fixture isn't perfect (e.g. some leakage from the mains through the filter, some couple of broadcast signals directly into the fixture over the air and the fact that the coupling coefficient is unknown because I didn't bother to determine it!) it did the job of giving a relative indication of how much "grunge" the LED's power supply put into the mains - and this same information would later be useful to get a general idea as to how much our mitigation efforts reduced this noise.  As can be seen, below 100 MHz the added noise (in a 3 MHz detection bandwidth) is nearly 50dB (100000 x) higher than the noise floor of the analyzer and the test fixture.

Refocusing on a smaller frequency range with different analyzer settings, let's take another look at how bad it is over the lower HF range:

Figure 4:
A re-done plot over the range of 0-100 MHz, this time with an 8 MHz resolution bandwidth.  The higher resolution bandwidth results in a higher reading from the QRM generator as its output is broadband noise.
Over much of this range, the base noise level (in cyan) is below the measurement sensitivity of the analyzer.
Click on the image for a larger version.
From the plot in Figure 4 we can start to get a picture of how bad the situation really is.  As can be seen at Marker #1, we measured a power level of about -3dBm - or 0.5 milliwatts within an 8 MHz bandwidth, but if we were to integrate this energy over the entire 0-100 MHz range we can see that there may be, perhaps a couple of 10s of milliwatts of noise being coupled into the mains:  We can only guess at the true amount of conducted RF owing to the comparative crudity of our test fixture and its unknown coupling coefficient across the RF spectrum, but we can be reasonably sure that what we see on this trace is but a fraction of the total energy present.

Figure 5:
Some board-mountable Shaffner mains filters from the Electronic
Goldmine, item G21844 (no longer available - sorry...)
Click on the image for a larger version.
As noted earlier, the entire purpose of these measurements was not to determine an absolute level of RF energy, but rather to have a means of repeatably measuring how bad things are - and also to be able to determine if our mitigation methods are having the desired effect.

"Fixing" the problem:

One solution to this problem (aside from not getting cheap, uncertified devices in the first place - but even then, one is never sure what one is really buying!) is to add known-to-be-effective filtering to the mains leads.

At about the time my friend brought these lamps to me, I noticed that the Electronic Goldmine had, on sale, some small, board-mount mains filters, so I suggested that he buy at least two for each of his three lights (for a total of six) - so he bought 10 of them.  These particular devices were attractive because they were relatively inexpensive, potted (helpful, because this will be mounted outdoors where moisture ingress could be a problem) and small enough to fit in the limited-space enclosure in the back of the floodlight.  Being that the lamps were only "50 watt", the 1.6 amp rating of these filters would be more than adequate.

Figure 6:
Another view of added filtering and their integration into the enclosure.
The Shaffner filters were mounted "dead bug" (leads up) and held in place
using both the ground wires and silicone (RTV) sealing compound.  The
lug at the lower-right was added to help make sure that this plate was
electrically bonded to the main body of the LED floodlight.
Click on the image for a larger version.
As can be seen in Figure 1 and Figure 6, two of these filters were installed "back to back" in the back of the lamp housing, using direct-soldered connections between the ground terminals of the filters and the metal plate itself with short pieces of heavy (8 AWG) copper wire to keep the impedance of these leads as low as possible:  Even a few inches/centimeters extra was found to significantly reduce the efficacy of these filters at VHF and higher frequencies.

You may notice something else about the layout:  The wires going in and out of the LED driver are bundled together with plastic wire ties and routed to the "far" side of the power supply, as distant from the mains filters and wires as possible - this to minimize the amount of RF energy that might be coupled from these "noisy" wires into the power cord - something that would surely "un-do" some of our hard-won efforts in minimize the amount of conducted RF noise.

The result:

The results of this effort can be seen in Figure 7, below:
Figure 7:
"Before" and "After" traces over the 0-10 MHz range.  The cyan trance is with the LED unit powered down, the yellow trance is without filtering and the magenta trace is including filtering.  Note the lower resolution bandwidth (91kHz) as compared to the other figures which will tend to reduce the apparent level of broadband noise from LED driver and accentuate those of "coherent" signals such as broadcast stations.
The strong signals at about 1.0-1.4 MHz are due to ingress of local AM broadcast stations into the lashed-up test figure, the level exceeding that of the leakage through the filter.
Click on the image for a larger version.
In Figure 7, above, we can see multiple traces - with the explanation below:
  • The Cyan (blue-ish) trace is our baseline measurement with the LED driver module powered down.  The signals below about 1.5 MHz are ingress from strong, nearby AM broadcast stations, some of which are nearly as strong as the noise at specific frequencies.
  • The Yellow trace is with no filtering of the LED driver module, showing the relative energy from the LED driver module over the 0-10 MHz range.
  • The Magenta (purple-ish) trace is with the LED driver module powered up with the added filtering.
"Could you have just snapped ferrites on the power cable?"

In reading this article, one might wonder if we could have solved the problem simply by putting snap-on ferrites on the power cord.

I doubt it.

Snap-on ferrite devices are very good about reducing the amount of RF conducted on wire, but with the extreme nature of the interference of these devices, it would never have been enough at HF.  The reason for this is that in order to adequately quash the QRM to the "point of undetectability" it would take at least several k-ohms of impedance on the power cable to solve the problem.
While it is possible that one can do this, it would take several large-ish cores (probably mix 31) with a dozen or more turns on each just to add that much reactance - but that material and winding topology would only work to the high end of the HF spectrum, so you'd need another core or two with windings on different materials - say 43 and 61 mix.

To make matters worse, you'd have to keep these chokes well-separated physically or else RF energy would be conducted around them - or even radiate directly from this rather large structure:  You certainly wouldn't have been able to easily fit it in the back side of the lamp's enclosure.

Self-contained filter modules like the ones used are specifically designed to quash RF over a very wide frequency range:  Not only are bifilar inductors used, but capacitors are also used to force the interfering energy to common mode so that the inductors can best do their job, plus there are other capacitors that do an excellent job of shunting RF to the case to "completely" contain that energy.

In other words:  In such an extreme case, you'd be far better off using an L/C filter like that depicted in Figure 5 without even bothering with ferrite chokes.
Interpreting these results we can see that over much of this range that the filtering reduced the amount of conducted noise to just above that of the cyan line, knocking the noise down by roughly 20dB over the range.  These filters start to lose their effectiveness below 1 MHz which is why, at very low frequencies (below 500 kHz) one starts to see more conducted energy - but these frequencies don't radiate very well, anyway so they are of generally less concern in most amateur stations.

When this plot was taken, the circuit depicted in Figure 2 was very close to the filter networks and it was believed that some energy was directly coupling into it from the LED driver module.   After the lamp was assembled (the cover put on and the power cord fitted) another test was done and no difference at all could be seen in the "on" and "off" traces - except at frequencies below about 1.5 MHz:  I somehow managed to omit capturing this trace.

Did it help?


My friend reinstalled these lights and was happy to report that upon listening on various HF bands from 160 through 10 meters, he was unable to detect when the lights were on or off, indicating that the modification was successful.  It is possible that within a few feet/meters of these lights that some low-level direct radiation of noise could have occurred on VHF/UHF frequencies, but this energy was demonstrably not being conducted via the power cord, and emissions would not likely be detectable more than a few feet/meters away, anyway.

Would just a single  filter have done the job?

Probably - but since the lights were a bit of a pain to take down and put up again it was decided to use two of these filters just to avoid the possible hassle of having to take them down (and apart) again if just one filter hadn't been enough!

* * * * * * * * * * *

Links to other articles about power supply noise reduction found at

This page stolen from