Monday, August 26, 2019

Cleaning a Power Bank

One common option to supply the small TTGO LoRa boards is to use a small power bank. For supply small devices like these boards I use one really cheap power bank, so cheap it can be obtained for 0.80 euros from China (with no 18650 li-ion cell included, of course).

The 0.80 euros power bank

It seemed to work just fine but powering a 433 MHz TTGO board I noticed the power bank itself radiates a lot of noise in this frequencies.  UHF noise from a switching mode power supply is a new record for me!

But... Can it be cleaned?

The power bank uses a HT4928S integrated circuit from HOTCHIP in a small PCB

Power bank's PCB

The circuit is the one in the HT4928S's datasheet. Verbatim:

HT4928S application circuit

The circuit is so simple that there is only one suspect point: pin 6, the output from the switching mosfet to the coil. I connected the oscilloscope there and I got this beautiful ringing:

Ringing on pin 6. Scale is 1 V/div and 20ns/div

This is the source of the noise at UHF and other bands as I checked later. Can this unwanted oscillation be suppressed? At first try I installed different capacitors at the footprint marked C1 with the hope to filter out this ringing but none of them worked. The noise was still there.

This is a common EMI problem, and it is usually solved installing a snubber. A snubber is just a resistor and capacitor in series designed to absorb (or damp) these unwanted oscillations. There is a lot of information in the net, but for its simplicity I followed this document from Nexperia:

NXP Semiconductors. AN11160. Designing RC snubbers

The application note shows a method to calculate the parasitic L and C values of the resonant LC producing this ringing, and then another method to calculate the RC values for the snubber. The PDF is so clear and straightforward I will not reproduce it here. Just read it, and then follow this post.

The first step is to measure the approximate frequency of the ringing. Using the oscilloscope this was easy: Fring0 = 94.3  MHz.

To calculate Fring1 I used a 4.7 nF capacitor placed at C1 footprint. This reduced the ringing frequency (just as expected) to 25.5 MHz (so Fring1 = 25.5 MHz).

These values led to X = 3.69, and therefore, Clk = 372 pF and Llk = 7.6 nH.

At this point we know the parasitic inductance is 7.6 nH and the parasitic capacitance is 372 pF. These two combined created a parasitic LC circuit with a resonating frequency of 94.3 MHz. This parasitic LC pumped by the switching mosfet is the one that produce the broadband noise.

Continuing with the PDF the next step is to calculate the damping resistor, the one that will be in the snubber. Using a damping factor of 1 (to dissipate the stored energy in the LC as fast as possible) results in a Rs value of 2.2 ohms.

Taking this value, and the ringing frequency of 94.3 MHz lead to a 760 pF (820 pF) capacitor for the snubber.

So the snubber for this circuit is a 820 pF in series with a 2.2 ohms resistor placed as close as possible to the switching device, in this case, as close as possible to pins 6 and 7 of the IC. Once I installed the two small SMD components I got this:

Ringing with the first snubber: 2.2 ohms + 820 pF

Pretty clean, but not enough. UHF noise disappeared completely, but there was still a very noticeable noise in the low VHF band. Then I noticed this paragraph:

The cut-off frequency of the snubber must be low enough to effectively short-circuit the undamped oscillation frequency, but not so low as to present a significant conduction path at the operating frequency of the circuit (for example 100 kHz or whatever). A good starting point has been found to be Fc = Fring0

In my case, Fc = Fring0 is not enough. So I increased the capacitor from 820 pF to 3300 pF. With the 2.2 ohms resistor this produces a cut-off frequency of 22 MHz. Low enough to pass the 94 MHz signal with almost no attenuation but at the same time high enough to block the main switching frequency of 1 MHz. Then, I got this with the oscilloscope:

Ringing with the second snubber: 2.2 ohms + 3300 pF

With these values noise totally disappeared in UHF and VHF at the expense of a diminution of 2% of the converter efficiency. Not bad for a two components solution.

Miguel A. Vallejo, EA4EOZ


  1. Enhorabuena y gracias por la información. Recientemente estuve enredando con un banco similar pero con otro circuito integrado y efectivamente tenía una red como esa que tú has añadido.
    Creo que la numeración de los condensadores en el esquema no coincide con la de la PCB. Creo que C1 en el esquema corresponde a C2 en la PCB, C2 en el esquema corresponde a C3 en el PCB y el condensador en paralelo con el transistor de conmutación (donde tú has puesto la red RC) viene numerado C1 en la PCB y no figura en el esquema.
    EA1AWY. Javier Muriedas

    1. Sí, una cosa es el PCB y la otra el datasheet. La numeración de los componentes no coincide pero el esquema si. Saludos.

  2. Hey, do you know something about the status LED´s and its meanings?

    1. My unit does:

      Charging: red/blue blinking

      Charged: blue

      Discharging: blue

  3. Look at emissions radiating from the boost inductor. Cheap ferrite inductors are open around the outside edge. This design has a high saturation current, but it leaks RF energy. Consider a shielded type inductor with air gap on the inside of the core.

    1. The snubber works fine in this case, probably in higher power converters more attention should be paid to the coil.

      Thank you for the tip

  4. thank a lot for the information
    wahhab yi3whr