USB 2.0 VBUS Filter

Question

I am reading an old (2013) FTDI application note AN_146 about hardware design for USB 2.0. On the subject of VBUS filtering, it says this:

2.4.4 Ferrite bead use and placement – USB peripheral devices

The USB specification prohibits the use of ferrite beads on the USB DP and DM data signals. It does, however, recommend them on the USB power signal (VBUS). It’s common to add bulk and decoupling capacitors as shown in Figure 2.5.

The 10nF capacitor and ferrite bead should be placed as close to the USB connector as possible.

I have a few questions regarding this circuit:

1. Why is C1 connected to the left of the ferrite, while C2 (decoupling) and C3 (bulk) are connected to the right of the ferrite?

2. The combination of C1 and the ferrite look like a low-pass LC filter to me. So, this design would still allow low-frequency A/C noise to pass through, correct?

3. Why does C3, the bulk capacitance, need to be a polarized capacitor? Were 4.7 μF ceramics just not available in 2013? Or is there some deeper reason why the bulk capacitance needs to be a polarized cap (e.g. something to do with ESR)?

4. What does 5V0 mean? I've seen this net name used in other USB circuit diagrams (figure 1 here for example). Is 5V0 a convention used to indicate something different from regular old 5V or +5V?

Answer 1

Letters and Decimal Points

What does 5V0 mean? I've seen this net name used in other USB circuit diagrams (figure 1 here for example). Is 5V0 a convention used to indicate something different from regular old 5V or +5V?

In electrical engineering, there's a tradition to abuse a letter as a decimal point because decimal points often disappear after many generations of photo-copying. Often, the letter from the SI prefix or component designator is abused for this purpose. For example, "6u8" for "6.8 uH inductor" or "6.8 uF capacitor", "1k15" for "1.15 kΩ" resistor, "0R1" for "0.1 Ω resistor", "3V3" for "3.3 V supply", etc.

As Tim Williams has reminded us, its origin is the RKM code, a European standard on resistor and capacitor markings. Since then, this style has gone beyond the original scope of this standard. It's strange at first but newcomers just get used to it after a while, and even start using it in their own schematics.

Pi Filter

Why is C1 connected to the left of the ferrite, while C2 (decoupling) and C3 (bulk) are connected to the right of the ferrite?

This circuit is called a "pi filter", named because it has a similar shape to the Greek letter π - one series inductor in the middle of two shunt capacitors.

It's a common topology in high-frequency EMI filters. It's simply an LC filter with an additional capacitor placed in front of it. The idea is that the first capacitor can already remove some noise before it enters the filter. One can argue that the pi filter really is a 2-stage LC filter, the first "hidden" inductor is the power cable.

Digital Devices need HF, not LF filters

The combination of C1 and the ferrite look like a low-pass LC filter to me. So, this design would still allow low-frequency A/C noise to pass through, correct?

Yes.

In a small pure-digital device, such as a USB dongle, low-frequency noise is usually a non-issue. Low power consumption means that even a small bulk capacitor is sufficient as a reservoir capacitor. Furthermore, the voltage regulator already acts as a low-frequency active filter. Below 1 MHz, any 5 V to 3.3 V linear regulator can provide a Power Supply Rejection Ratio around 20 dB to 60 dB, switched-mode DC/DC buck converters are not as good but they're still regulated in a similar way. Finally, a single cycle of 50/60 Hz noise is an eternity for a 10 MHz or 100 MHz processor.

_{Source: Minimizing Switching Regulator Residue in Linear Regulator Outputs, by Jim Williams}

The real problem is high-frequency noise (i.e. EMI) from 1 MHz to 1 GHz due to rapid switching of digital logic and DC/DC power converters. As wires act as antennas, if conducted interference is not filtered, they may become radiated interference. Thus, it's considered a good habit to always include an LC filter at the entry of the powerline to stop high-frequency interference from leaving (the main concern) or entering our system.

Ferrite Beads vs. Inductors

To remove high-frequency differential-mode noise (common-mode noise is another can of worm, beyond the scope of this article), a ferrite bead is the component of choice in an LC filter. At low frequencies, ferrite beads behave like small inductors - but it's not their true purpose. At high frequencies above 100 MHz, they act like resistors as its magnetic material is deliberately designed to produce core loss.

_{Source: Minimizing Switching Regulator Residue in Linear Regulator Outputs, by Jim Williams}

Alternatively, one can also use an inductor around a few microhenries. But it comes with some possible downsides. Because an inductor has many turns in the winding, they generate more DC power loss than ferrite bead - which only has a few turns. A higher number of turns also means higher parasitic capacitance, making the inductor unable to filter anymore at some point above 100 MHz. Meanwhile, a ferrite bead operates primarily through core loss, theoretically they can be effective even at 1 GHz (but only with perfect layout and shielding, don't expect to see it in a real circuit).

_{Source: Minimizing Switching Regulator Residue in Linear Regulator Outputs, by Jim Williams}

On the other hand, inductors do have a real advantage over ferrite beads: they provide filtering from 1 MHz to 50 MHz, this range is the blind spot of ferrite beads, as they only start becoming lossy around 100 MHz. In a personal design of mine, I found a ferrite bead provides 20 dB attenuation above 50 MHz, but was completely useless at stopping the 1 MHz switching spike and its harmonics below 50 MHz, while an inductor offered over 40 dB attenuation from 1 MHz to 300 MHz. Inductors may not work at 1 GHz like ferrite beads, but they can definitely work with careful component selection (select a good low capacitance, high SRF part) and layout. The price to pay is its larger size: a ferrite bead can be as small as a 0603 resistor, but an inductor is a coil.

Filter Design Tips

1. If you have a choice, place the input filter as close to the connector as possible, do not place the filter near the load. Otherwise, noise may jump across the filter via parasitic coupling.

2. Use ferrite beads and inductors near the power source. In particular, do not connect them in series with the (digital) VCC/VDD pins of a digital chip like a processor, unless you really know what you're doing (analog pins are okay). It stops the chip from receiving the high-frequency current it needs - in some old designs, a slow chip may still work. For modern chips, it may cause the power supply impedance to go out of spec.

3. Do not use a ferrite bead at its rated DC current. Always select a ferrite with a large margin - DC current degrade the impedance of the ferrite because as the core is going closer to saturation.

4. Ferrite bead is usually rated using its impedance in ohms at 100 MHz. You can find ferrites with huge impedance, over 1000 Ω. But there ferrites have a very low current rating, and they're not useful for powerline filtering. A 50 Ω ferrite is a more realistic choice. Like everything else, it's a tradeoff.

5. Other parameters of the ferrites are not directly listed, such as its low-frequency inductance. Instead, they can be found indirectly from the resistance vs frequency and reactance vs. frequency charts. Alternatively, just download its SPICE model or its S-parameters file and import it in your circuit simulator.

Overall, if the filter performance is not critical, don't spend your time overanalyzing it (like what I'm doing right now...), just add a ferrite bead as a good habit, at least, include the filter (and its RC damper, to be shown next) in the schematics and PCB layout. It can be further tuned or even deleted later: during production, if a device is found to pass radiation test without the ferrite bead, it's often removed and replaced by 0 Ω resistors to cut the costs down to the last cents... Not a practice I'd recommend, but it's the reality of low-cost consumer electronics.

If it's critical, analyze it on a case-by-case basis and test your filters and circuit boards with an impedance analyzer.

Damping an LC filter

Why does C3, the bulk capacitance, need to be a polarized capacitor? Were 4.7 uF ceramics just not available in 2013? Or is there some deeper reason why the bulk capacitance needs to be a polarized cap? (e.g., something to do with ESR)

First, "use electrolytic capacitors as bulk capacitors, use ceramic capacitors as high-frequency capacitor" is just an outdated habit for many people. I'm not a fan of it. For such a small value, on the surface, there's no real advantage other than saving perhaps a few cents.

But there's indeed a deeper reason. The high ESR of the electrolytic capacitor forms an RC damper to suppress possible resonance of the LC filter (the same can also be implemented by a 6.8 uF ceramic capacitor in series with a 0.25 Ω resistor, which is the solution I prefer).

The usual practice is using a generic (not low-ESR) wet-electrolyte aluminum electrolytic capacitor. As a rule of thumb, use a large bulk capacitor 10x the size of all the other capacitances combined. So this value is around 100 to 1000 uF.

However, in this particular example, since the maximum capacitance of a USB device is restricted by the specification to avoid excessive voltage droop at the USB port, only a 10 uF can be connected at a time. Due to high ESR of small aluminum electrolytic capacitors (it needs to be high, but not too high, definitely not much greater than 1 Ω), this trick only has very limited effect and can even backfire. Solid-electrolyte aluminum capacitors or tantalum electrolytic capacitor may have just the right ESR for this purpose, but I haven't investigated it so I'm not sure.

The ceramic capacitor in series with an SMD resistor is a better solution. For such a small capacitor value, there's no need to use electrolytic capacitors.

Overview

An LC circuit is good at removing noise, but it also has the side-effect of creating resonance at its natural frequency. Worse, modern ceramic capacitors are so great that they're very close to ideal capacitors in terms of ESR, so this resonance would have a high Q - great when you actually want a resonator, but just like many well-known examples from mechanical engineering, an unintentional resonance can also be detrimental in electronics.

_{Source: Input Filter Interactions with Switching Regulators, by Christophe Basso}

There's a chance that the external noise or the operating frequency of your circuit happens to match the filter's resonance frequency, in this case, instead of removing the noise, the filter ironically causes a significant increase of noise voltage. If your circuit contains a switched-mode power supply and you're especially unlucky, the worst-case scenario is the complete destabilization of the power supply - a voltage-regulated power supply has negative resistance, turning it into an oscillator.

The series LC filter is not the only unwanted resonator lurking in a circuit, there are more! Another common source of unwanted resonance is when a large bulk capacitor is used in combination with a small bypass capacitor. Due to parasitic inductance, at high frequency, the big capacitor acts like an inductor. Thus, they form an parallel LC resonator. At the resonant frequency, the power supply impedance goes to the moon, from 0.01 Ω to 1 Ω or more. The circuit behaves as if it has no capacitor installed at this exact frequency.

_{Source: Electromagnetic Compatibility Engineering by Henry Ott, fair use.}

In many designs, both kinds of resonance problems are often overlooked or ignored entirely: LC filters or ferrite beads are often added with a "just in case" mindset, and similarly, the suggestion is often given to use "multiple values of capacitors, 10x apart, such as 0.1 uF and 1 uF" , without additional thought. These circuits usually work without any apparent problems - resonance only occasionally causes problems. However, a good circuit design should be robust, thus it's a good design practice to suppress resonance in an LC filter by installing a damper.

This is usually achieved by connecting a small resistor in series with a large capacitor. The small resistor creates energy loss in the "LC resonator", killing the oscillation. The large capacitor ensures the resistor only dissipates at AC, not at DC, otherwise it would be a short-circuit.

Simulation

Notes on modeling:

1. At low frequency, ferrite beads behave just like small inductors, at high frequency, they behave like resistors. The inductance value is often not obvious in the datasheet, but it can be calculated from the impedance chart, alternatively, I just downloaded the SPICE model (S-parameters also work) of the ferrite bead for a quick simulation.

2. The ESR of ceramic capacitors is extremely low in general, thus I used 0.05 Ω, in series of 5 nH parasitic capacitance, which is typical for SMD parts placed as close as possible. The 4.7 uF bulk capacitor has 50 nH of parasitic capacitance, an optimistic value of parasitic inductance from the layout. For an electrolytic capacitor, an additional 50 nH is added - electrolytic capacitors have high ESR and ESL, and useless for high-frequency filtering - they're only used as a bulk capacitor.

3. These simulations are only for illustrative purposes. Your mileage may vary, especially your circuit parasitics. If filter or power delivery performance is really important, sweep your board with an impedance analyzer.

As you can see, the ferrite bead is highly effective, creating an attenuation of 15 dB at 1 MHz and 20 dB at 100 MHz. However, at 130 kHz, it instead creates a 8 dB of voltage gain due to resonance from the LC filter.

Since a ferrite bead is used, the result is actually better than I expected. For inductors, this gain can easily reach 20 dB. If any operating frequency or noise locates near this frequency, the result would be disastrous. Are you feeling lucky? It's also interesting to note that due to different capacitor values in parallel, another resonance occurs at 2.3 MHz with 0 dB attenuation, rendering the filter useless. But a ferrite bead is meant to filter high-frequency noise around 100 MHz or so, not low-frequency noise, so I would give it a pass...

Then, watch what happens when a 0.25 Ω resistor is inserted in series with the 4.7 μF capacitor. The 130 kHz resonance is suppressed, with only 1.3 dB of gain. The 2.3 MHz resonance also disappears.

However, the attenuation at 350 kHz reduced from 20 dB to 10 dB, but such is the price of removing resonance, and it's usually worthwhile. Combining bypass capacitors of different values creates deeper attenuation at many frequencies, but at the expense of high peaks without attenuation.

Now, try a 4.7 uF electrolytic capacitor instead. For a small capacitor, the ESR can be as high as 50 Ω at mains frequency. However, it's difficult to find clear data on high-frequency ESR. A Vishay datasheet seems to suggest 3 Ω for a good-quality aluminum capacitor with wet electrolyte (but you probably can't rely on this value as it can vary greatly depending on the model and vendor).

The 130 kHz resonance is completely suppressed without a trace. There's a small resonance at 1 MHz with 3.3 dB gain due to paralleling capacitors, but it's low enough to be safe. The disadvantage is that there's no filtering at all at low frequencies, meanwhile its high-frequency performance is unaffected with 20 dB attenuation at 100 MHz. As a ferrite bead is only meant to filter high-frequency noise around 100 MHz or so, not low-frequency noise, it's an acceptable solution. In my opinion, the bigger problem is the ESR of small aluminum electrolytic capacitors cannot be relied upon.

Thus, overall, in this particular case, I prefer the second solution: 4.7 uF ceramic capacitor with 0.25 Ω resistor. Because a typical 16 V, 0603 ceramic capacitor loses 30% of its capacitance under a 5 V bias, select 6.8 uF instead.

Aluminum Capacitors as RC Damper

In power supply designs, a large, generic and cheap capacitor is often used. As a rule of thumb, use a large bulk capacitor 10x the size of all the other capacitances combined. So this value is around 100 to 1000 uF, with an ESR around 0.5 Ω (if size or cost is a problem, further optimizations are possible, see Reference 4 for design equations).

This kills three birds with one stone, first, it's cheap, next, the high ESR of a cheap electrolytic capacitor serves as an RC damper - it's not a bug, it's a feature. Worse is better... Finally, they act as reservoir capacitors for low-frequency bypassing (but it's the least important aspect).

However, in this particular case, the maximum capacitance of a USB device is restricted by the specification to avoid voltage droop at the USB port, only a 10 uF can be connected at a time (you may switch the capacitors on sequentially, separated by a 100-ms interval). Since small electrolytic capacitors have ESR greater than 1 Ω, with its high-frequency ESR poorly controlled or documented, it limits the usefulness of this old trick and can even backfire. For example, this is what happens when ESR reaches 10 Ω, a resonance at 1 MHz is worse than the resonance at 130 kHz.

Small solid-electrolyte aluminum capacitors or tantalum electrolytic capacitor may have just the right ESR for this purpose, but I didn't investigate it while writing this answer, so I'm not sure, so... exercise for the reader.

Further readings

1. Minimizing Switching Regulator Residue in Linear Regulator Outputs - Banishing Those Accursed Spikes, by Jim Williams, Linear Technology.

   A tutorial on the use of ferrite beads to suppress high-frequency interference. The example given is a switched-mode power supply but it's also applicable when filtering a digital board.

2. Ferrite Bead Demystified, by Jefferson Eco and Aldrick Limjoco, Analog Devices

   A more advanced tutorial on the use of ferrite beads to suppress high-frequency interference, with damping examples.

3. Optimizing Power Distribution Networks for Flat Impedance, Heidi Barnes, Signal Integrity Journal.

   An introduction to the parallel resonance problem and its implications on IC power decoupling.

4. Input Filter Interactions with Switching Regulators, by Christophe Basso, ONSemi.

   It contains an introduction to an advanced circuit analysis technique called Fast Analytical Techniques in Electrical Circuits (FACTS), not for beginners. I don't understand it. But at the end of the presentation, the author derived the formula for designing an optimally damped LC filter, with many practical suggestions.

5. Electromagnetic Compatibility Engineering, Henry W. Ott, Chapter 11: Digital Circuit Power Distribution

   Henry Ott should be required reading for all circuit board designers.

Answer 2

1. The cable has inductance so it is best to have a small capacitor right at the connector for some filtering. It lowers the impedance of cable side. It prevents high frequency AC noise from entering or exiting the cable. Basically for EMI reasons. An L-C-L filer is called a pi filter due to the shape of it. It is better than just LC filter.

2. Yes C1 with ferrite is a low pass filter. But just enough to pass EMI compliance testing. Lower frequencies are not a problem, if you blink a LED then your device anyway has non-DC consumption which varies.

3. Your guess is as good as mine. Sometimes higher ESR bulk caps are used so they don't cause a huge inrush current, which coule cause high voltage spikes due to cable inductance. It damps the ringing and oscillations somewhat. Also it might be that tantalum or electrolytic were just the de facto caps cheap and good enough back then.

4. It just means 5.0 volts, just like 3V3 means 3.3 volts.

Answer 3

3. Why does C3, the bulk capacitance, need to be a polarized capacitor?

Polarization is not the issue. Most ceramic capacitors are very voltage dependent, even at low voltages. The higher values are difficult to obtain in non voltage dependent ceramics like C0G/NP0 types.

Electrolytic capacitors are not voltage dependent and higher values are easily obtained. But they do have higher ESL and so are inductive a higher frequencies, so where important, the electrolytic is often paralleled with a ceramic as is shown in your circuit diagram.