•  
• EDN Europe MAGAZINE
•  
• Free SUBSCRIPTION
• Subscription Renewal
• View Media Kit

• EDN
• EDN Australia
• EDN Asia
• EDN China
• EDN Japan

Cutting the buzz in LED Drivers, part one

Audible noise created by PWM dimming at frequencies within the audio band is a serious problem with high power LED drivers. Since it is a mechanical issue, it can be very difficult for electronics engineers to deal with.

Various techniques exist for reducing audible noise, and the most commonly-employed is to move the dimming frequency beyond the range of human hearing. For the sake of our feline, canine, and other friends in the animal kingdom, frequencies in the range of 25 to 30 kHz or more are needed to be truly ultrasonic. Yet dimming at such frequencies is not always possible. The principal drawback to PWM dimming at high frequency is the loss of contrast ratio (sometimes referred to as dimming ratio) which is inversely proportional to frequency. Applications that require high contrast ratio, such as lighting for aircraft cockpits, will not achieve the necessary performance when the dimming period is reduced to 33.3 μsec (as it would be with a PWM dimming frequency of 30 kHz.) Figure 1 shows a graphical explanation of contrast or dimming ratio. Linear dimming, where the average current through the LEDs is modulated, is achieving increasing acceptance, but does not deliver the precision or contrast ratio possible with PWM.

Sources of audible noise

LED driver circuits for currents of 350 mA and above predominantly use switch-mode power supplies (SMPS). There are two main classes of component found in every SMPS that are most likely to generate audible noise. The first is magnetic components, specifically the inductors and transformers. The cores of inductors and transformers exhibit a mechanical deformation called magnetostriction when they are magnetised. This causes the core to swell and shrink each time the magnetic field changes. Since the magnetic field is proportional to the current flowing, each time current changes the core vibrates, mechanically contacting the windings, the shielding or bobbin, and the PCB on which the device is mounted.

The second component that is likely to convert electrical energy into acoustic energy is the multi-layer ceramic capacitor (MLCC), which exhibits a piezoacoustic effect when AC current flows through it. MLCCs have become more and more popular in SMPSs due to their long life, high reliability, low loss, and availability of greater and greater capacitance at higher and higher working voltages. Unfortunately, every switching converter forces heavy AC current through at least one capacitor; compounding the problem, the higher the AC current the more appealing MLCCs become for the designer. In the case of the buck converter, it is the input capacitor(s) that carry the heaviest AC currents.

Both types of components can be likened (in the context of audible noise generation) to the electromagnet in a loudspeaker, and the PCB is analogous to the cone or body of the loudspeaker. If the PCB is bolted or otherwise attached to heatsinks, frames, enclosures, or any other solid object, then these too become part of the “cone”. In general, the larger the mass connected to the sources of noise, the louder the noise will be. Higher power also leads to greater AC current and hence more noise. Generation of audible noise is not easy to simulate, which means many solutions are based upon trial-and-error. What seems like an insignificant buzz in a small PCB in a laboratory can become very irritating in a final product where the PCB is attached to the metal enclosure or frame of the power supply or the LED lamp itself. It is also important to note that linear regulators, often thought to be noise-free in both the acoustic and electronic domains, will also generate noise if dimmed by PWM within the audio range. All it takes is one magnetic component or one MLCC with AC current running through it.

This report describes the results of testing a step-down (buck converter) constant current source for driving LEDs based on the LM3404, with various combinations of inductors, PWM dimming frequencies and methods of applying the PWM dimming signal. The goal was to find a combination that eliminated or at least minimised the audible noise, as well as to identify trends that will make it easier to avoid excessive audible noise in future circuits.

TEST CONDITIONS

Four identical development kit PCBs for the LM3404 (LM3404FSTDIMEV, for which the bill-of-materials is listed in the web version of this article) were modified in turn, in this case by replacing the inductor L1 with the alternative components listed in Table 1, respectively;

ID	Part Number	Type	Size	Parameters	Vendor
A	7447789122	Shielded Ferrite Core	7.3 x 7.3 x 3.2 mm	22μH, 1.4A, 0.19Ω	Wurth
B	IHSM4825ER220L	Epoxy Encap-sulated Ferrite Core	13.7 x 6.4 x 5.7 mm	22μH, 1.7A, 0.15Ω	Vishay
C	IHLP2020CZER220M11	Powd- ered Iron Core	5.2 x 5.5 x 3.0 mm	22μH, 1.7A, 0.26Ω	Vishay
D	030KM2502SM	Kool-Mu Core, Toroidal	11.2 x 11.0 x 9.1 mm	25μH, 2.2A, 0.07Ω	Gowanda

Table 1 The inductor L1 on the standard development board was substituted with a selection of alternative components.

A. L1 on the first 1st board was replaced with a Wurth 7447789122 22. The PCB is supplied with a 33-μH ferrite device, but 33-μH inductors could not be found at the required power level for the other types of inductors, hence the change to a shielded, 22-μH part.

B. L1 of the second board was replaced with Vishay IHSM4825ER220L

C. L1 of the third board was replaced with Vishay IHLP2020CZER220M11

D. L1 of the fourth board was replaced with Gowanda 030KM2502SM.

Test conditions for each PCB were a 24.0-V input, powering a string of five white LEDs in series at an average current of 600 mA. PWM dimming frequencies of 200 Hz, 1 kHz and 10 kHz were applied with dimming duty cycles of 50% and signal levels of 0 to 2.5V. From experience and testing, 50% duty cycles were found to generate the most audible noise.

For the first set of tests the PWM dimming signal was applied to the DIM pin of the LM3404 as shown in Figure 2. This method is referred to as “Enable Pin Dimming” since the DIM pin of the LM3404 is a shutdown/enable pin which turns off the switching of the internal power MOSFET.

For the second set of tests the PWM dimming signal was applied to a gate driver which powered a discrete N-MOSFET placed in parallel to the chain of LEDs as shown in Figure 3. This method is referred to as “Parallel FET Dimming.” Parallel FET dimming differs from enable pin dimming in that the current through the inductor flows continuously when the LEDs are off, even though the output voltage drops to near-zero. This method provides much faster slew rates than enable pin dimming, provided that the capacitance in parallel to the chain of LEDs is minimised.

The PCBs under test were placed in a vice and the audio environment recorded for eight seconds each with an industry-standard dynamic microphone. The tip of the microphone was placed 20 mm above the surface of the PCBs, with measures taken to maintain the microphone in the same position over each PCB. In expectation that the inductor and input capacitor would each produce audible noise, the central axis of the microphone was placed halfway between these two components. By default the LM3404FSTDIMEV does not have any output capacitor, however if one was used, it would likely be a third source of noise, despite the much lower AC current that flows through output capacitors in buck regulators.

Each recording is an uncompressed WAV file recorded at 44.1 kHz, 16-bits. The WAV files were then played and the audio signal captured in a series of oscilloscope screen captures. Then, the WAV files and screen captures were combined into mp4 files for easy access via the internet. Finally, although admittedly subjective, the author´s notes are included on what could be heard (or could not be heard) with the unaided ear during the testing. Every effort was made to keep the lab as quiet as possible, however the signal generator used for the PWM dimming signal used a cooling fan that could not be turned off, and some external, environmental noise inevitably leaks into any audio recording. For this reason reference recordings of each combination of input capacitor and inductor were also taken, with accompanying oscillographs. A complete list of equipment used can be found in the appendix in the web version of this article.

Table 2:

Enable Pin Dimming
Inductor Type	Reference (No dimming)	200 Hz	1 kHz	10 kHz
Shielded Ferrite Setup Photo Oscope Capture	Link to mp4	Inaudible Link to mp4	Barely Audible Link to mp4	Faint Link to mp4
Encapsulated Ferrite Setup Photo Oscope Capture	Link to mp4	Inaudible Link to mp4	Inaudible Link to mp4	Audible Link to mp4
Kool Mu Toroid Setup Photo Oscope Capture	Link to mp4	Barely audible Link to mp4	Faint Link to mp4	Audible Link to mp4
Powdered Iron Setup Photo Oscope Capture	Link to mp4	Inaudible Link to mp4	Inaudible Link to mp4	Faint Link to mp4

Table 3:

Parallel FET Dimming
Inductor Type	Reference (No dimming)	200 Hz	1 kHz	10 kHz
Shielded Ferrite Oscope Capture	Link to mp4	Inaudible Link to mp4	Barely Audible Link to mp4	Audible Link to mp4
Encapsulated Ferrite Oscope Capture	Link to mp4	Inaudible Link to mp4	Inaudible Link to mp4	Faint Link to mp4
Kool Mu Toroid Oscope Capture	Link to mp4	Inaudible Link to mp4	Faint Link to mp4	Audible Link to mp4
Powdered Iron Oscope Capture	Link to mp4	Inaudible Link to mp4	Barely Audible Link to mp4	Audible Link to mp4

Tables 2 and 3 present, at a first level, a purely subjective assessment of the noise level resulting from each circuit configuration and drive set-up. In the corresponding Tables in the web version of this article at www.edn-europe.com/ article.asp?articleid=5328, you will find live links to mp4 audio files, and to photographs of test set-ups and oscilloscope screen captures for each configuration.

For the unaided ear testing, and to minimise variability in the subjective assessment, the author put his left ear as close to the PCB as possible without disturbing the test setup, and then toggled the PWM dimming signal generator on and off several times. These results were cataloged, in order of softest to loudest, as: inaudible, barely audible, faint, audible.

OBSERVATIONS

The following observations were based mostly on the audio recordings, with support from the oscilloscope captures:

1. Regardless of dimming frequency or method, the shielded ferrite and encapsulated ferrite inductors were clearly the noisiest, and were about equally noisy

2. The Kool Mu toroid was noticeably quieter than either ferrite, and the powdered iron inductor was by far the quietest of all. The powdered iron inductor was also by far the smallest device, while the Kool Mu toroid was the only one fitted into a plastic bobbin, possibly providing some mechanical isolation from the PCB

3. Two distinct tones can be heard in almost every recording. One is more harmonic and correlated to the PWM dimming frequency. The second is a more mechanical tone. One likely comes from the magnetic component, the other from the input capacitor. By comparing the enable pin dimming recordings for all four inductors, it seems that the mechanical sound comes from the capacitor, as it appears to stay relatively constant, whereas the harmonic tone changes noticeably in volume between the ferrite and non-ferrite inductors

4. Parallel FET Dimming was noticeably quieter than Enable Pin Dimming for the ferrite inductors at 1 kHz and 10 kHz, and only slightly quieter at 200 Hz: however;

5. Enable Pin Dimming was noticeably quieter than Parallel FET Dimming for the Kool Mu toroid and powdered iron inductor, at all three frequencies

6. Dimming at the lowest frequency, 200 Hz, produced noise that was more difficult to hear with the unaided ear. While the microphone was able to pick up a distinct harmonic tone, the sound to the unaided ear was only the mechanical, buzzing tone. This noise seems more easily masked by ambient noise.

CONCLUSIONS

Prior to beginning the research for this article series, the following general trial-and-error observations and solutions had been made and applied with varying degrees of success:

1. Non-shielded drum-core inductors are quieter than shielded types. The suspected reason behind this is that the shielding is also resonating with the magnetostriction of the core.

2. Toroids are quieter than drum core inductors

3. Certain inductor manufacturers’ inductors are less noisy than others. These tend to be the well-known brands, and the difference may well be due to consistency in manufacturing, and factors such as more liberal usage of glue or potting to hold the windings to the core.

4. Adding glue to the windings of the inductor helps muffle noise.

5. Potting the inductor and any MLCC capacitors helps muffle noise, and potting the entire PCB helps even more.

6. The smaller the MLCC, the quieter. Two smaller capacitors make less noise than one larger one.

7. Placing two MLCCs on the top and bottom of the PCB, one directly above the other, can generate some phase cancellation of noise.

With the data already taken it is clear that enable pin dimming with ferrite core inductors generates a large amount of audible noise. Distributed air-gap cores such as Kool Mu and powdered iron are significantly quieter. This fits with the expectation that powder cores exhibit less magnetostriction than ferrite cores.

One conclusion that did not fit with expectation was the higher noise with Kool Mu and powdered iron when using parallel FET dimming. Where the parallel FET dimming was slightly quieter than enable pin dimming for the two ferrite core devices, the complete opposite was true for the two powder cores. Since average inductor current stayed nearly identical, and since the LM3404FSTDIMEV uses a circuit to maintain a constant peak-to-peak AC ripple current in the inductor, according to initial expectations the magnetostrictive effect should have been much less, and hence the result quieter. Nonetheless, listening to the audio files and observing the scope captures tells a different story. It appears that the fast edges generated by parallel FET dimming translate to sharp edges in the audible noise waveform. Comparing the plots and videos of MLCC Powder EN 1kHz and MLCC Kool Mu EN 1kHz with MLCC Powder Parallel 1kHz and MLCC Kool Mu Parallel 1kHz show this effect particularly well.

Part 2 of this article series will reexamine the same set of four magnetic components, but with the MLCC input capacitor replaced by a film (polypropylene) capacitor.

Author Information

Christopher Richardson is a Field Applications Engineer for the South Europe region with National Semiconductor, now part of Texas Instruments.