1-1. Introduction
During my university days, there was always noise coming from my FM radio whenever I listened to it beside my PC. The radiation noise from the PC entered and hindered the FM radio. At that time, I didn't know the cause and never even dreamed that I would face this problem after joining the company. When I started actual work on noise suppression, it was so difficult that it took several months in some cases. The major reasons are as follows:
- (1) It is unknown where the noise is generated from (noise source) and how it is conducted
- (2) Appropriate noise suppression methods are difficult to grasp
When the correct noise suppression method was unknown, various things crossed my mind: for example, maybe the part dealt with at that time was wrong and I should deal with another part. It is not rare that a noise suppression method effective for a set, such as filtering, is not effective for another set, which left me scratching my head.
Therefore, I felt the necessity for organized theoretical noise suppression methods, and correct filter selection methods. As part of my research, I investigated the causes and other factors of the difference in the noise suppression effect on the same noise filter depending on the circuit, and summarized them as specific examples. This document describes the contents.
1-2. Cause of and countermeasure against noise from digital signals
In the noise I experienced in my student days, the clock frequency of the PC was about 4MHz. However, for example in Japan, the frequency band used for FM broadcasts is 76MHz and higher. The clock frequency of the noise source does not overlap the broadcast frequency. So why did the failure occur? A digital signal is a square wave composed of a fundamental sine wave and harmonic of its integral multiple as shown in Fig. A. The sharper the rise of the square wave is, the higher the higher-order harmonic component is contained. This means that the actual digital signal had a harmonic component overlapping the band of the FM broadcast, which caused the acoustic noise in the sound of FM radio.
Therefore, the basic noise suppression method is to remove the harmonic of the digital signal. When the harmonic is removed, the rising time and falling time of the waveform are delayed (deformation of the waveform). As such, it is necessary to select a filter which removes the harmonic to an appropriate extent that does not hinder the operation of the circuit.
1-3. Representative noise filters and usage examples
1-3-1. Representative noise filters
You will come up with a capacitor, resistor, and ferrite bead as representative filters used for signal lines. The following shows simple descriptions of their functions:
- - Capacitor
A capacitor reduces the impedance as the frequency becomes higher. Therefore, with a parallel connection (shunt connection) between the signal line and GND, the high-frequency component (noise) of the signal is bypassed to GND.
- - Resistor
A resistance component that absorbs energy.
- - Ferrite bead
A ferrite bead is a type of inductor. Its impedance rises as the frequency becomes higher. Therefore, it is series-connected with the signal line to absorb or reflect the noise. In a general inductor, the reactance component X is dominant, but a ferrite material to increase the resistance component R is selected for ferrite beads in order to increase the loss of energy. Therefore, ferrite beads are superior for noise suppression than general inductors. The reactance component X is a lossless component.
Column: What is filter insertion loss?
To show the effect of a filter, the insertion loss characteristics (IL) are used. This indicates to what extent the installation of a filter attenuates the signal. This method is specified in MIL-STD202, etc. Note that the values were measured with an input/output impedance of 50 ohms. If the impedance of the actual circuit is not 50 ohms, the effect of the filter will be different.
Column: What is the cut-off frequency of a filter?
The cut-off frequency is the frequency at which (1) power that can pass through the filter balances (2) power that cannot pass through the filter.
In other words, it is the frequency at which the output power is 1/2 of the input power. Since W is V2/R, it is 1/√2 in voltage. If I am asked about the appropriate cut-off frequency of a filter, I answer that it is about 3 to 5 times higher than the basic frequency. This is because the harmonics in about the 3rd to 5th order should be retained in order to keep the square wave. However, if the input/output impedance of the circuit is not 50 ohms, the cut-off frequency will be different, so this is just a guideline.
For reference, the 1GHz band in the specifications of the oscilloscope indicates this cut-off frequency. This means that the 1GHz signal is observed with 3dB attenuation. Since the signal is also attenuated in the probe, it is necessary to use measuring equipment with a broader bandwidth than the measurement frequency range.
Column: What is characteristic impedance?
I think you hear very often that the characteristic impedance is 75 ohms for TV antenna cables and 50 ohms for measuring equipment. Simply put, the characteristic impedance indicates the impedance per unit length. If the loss caused by the resistance component can be ignored, it is calculated from (1) the capacitance per unit length (F/m) and (2) the inductance per unit length (H/m). Both the numerator and denominator have (/m) and are canceled, which makes it difficult to understand.
1-3-2. Insertion loss of a capacitor, resistor, and ferrite bead
Fig. A shows the insertion loss characteristics of a capacitor, inductance, and resistor. These were measured with a 50 ohm system. The shapes resemble the impedance curves.
Fig. B shows an example of an equivalent circuit of a digital signal circuit. In this example, the output resistor is 20 ohms, output capacitance is 10pF, and load capacitance is 5pF. This means that it is not a 50 ohm system.
1-4. Usage examples of representative filters
This section introduces the waveforms with each product used in an actual digital circuit which is not a 50 ohm system, and examples of actual measurements of the radiation noise suppression effects. The radiation noise was measured with an antenna installed 3m away from the set. The bold black lines drawn on the graphs of the radiation noise indicate the limit values of the international standards for noise, CISPR 32.
1-4-1. Example of noise suppression with a capacitor
First, I will introduce the noise suppression result with a capacitor.
Since the load capacity is 5pF, it was necessary to make the impedance of the capacitor to be installed lower than the load capacity, so I connected 100pF to the shunt. Fig. B shows the result. Thanks to the installation of the capacitor, the radiation noise is attenuated by about 4dB.
Fig. C shows the result of the measurement of whether the installation of a capacitor reduced the current flowing to the signal line. It shows that, after the installation, the current was reduced in the signal line between the capacitor and the load.
The concern here is that the current between the transmission IC and the capacitor is higher than the initial state. This is because the addition of a capacitor increased the load for IC. Therefore, if a capacitor is used for a signal line of a clock, etc., it may increase the power consumption. In addition, the current flowing to GND will also increase, so it will raise the noise level of GND. Therefore, it is rare to line up capacitors on a signal line in a PCB such as a bus line.
However, the problem with an increase in the power consumption is not significant in a low-speed line such as an external interface. Moreover, the addition of a capacitor improves the resistance of a circuit against static electricity. Therefore, it is effective to add a capacitor to a low-speed external interface connection part.
1-4-2. Examples of noise suppression with resistors
A resistor is a product which has often been used for the noise suppression of a signal line.
Fig. B shows the waveforms and radiation noise with a resistor.
The installation of a resistor reduces ringing of the waveform (overshoot/undershoot). Ringing is caused by multiple reflections of the signal between the output and the load. A resistor suppresses ringing by decreasing the multiple reflections.
A resistor also reduces the radiation noise by suppressing the current. It is easy to use because it does not increase the output current of IC like a capacitor.
However, if the resistance value is increased in order to improve the noise suppression effect, the deformation of the waveform may become larger (the rise and fall times of the signal may become longer), or the crest value may drop. In addition, it reduces the supply voltage in a power line.
1-4-3. Examples of noise suppression with ferrite beads
The ferrite bead is developed in order to gain a better noise suppression effect with less deformation of the signal waveform than a resistor. Its impedance is lower at the signal band and higher at the noise band. Though it is a type of inductor, a material which is mainly composed of a resistance component (material which increases loss) is selected in order to consume the noise component.
Generally the specification of an inductor is represented in inductance (H: henry) and that of a ferrite bead in the resistance value (ohm). Its purpose is to make it easy to imagine the influence on the signal and noise when replacing a damping resistor. The reason why the impedance at 100MHz is shown is that the signal frequencies of digital equipment ranged about up to 10MHz when the ferrite bead was developed and the concerned noise frequency range was about 200MHz to 300MHz, so 100MHz, the frequency in between, was selected as the guideline frequency. Recently, the noise in the high-frequency range has increasingly caused a problem, so ferrite beads are also available for which impedance at 1GHz is specified in addition to impedance at 100MHz.
A ferrite bead can suppress a decrease in the power supply voltage by reducing the DC resistance when used for a power line.
1-5. Difference in the noise suppression effect depending on the transmission line length
1-5-1. Difference in the noise suppression effect of a ferrite bead depending on the transmission line length
The examples of noise suppression with ferrite beads shown in the preceding section show the data with a transmission line length of 5cm.
To investigate the reason why the noise suppression effect varies among sets, I also prepared substrates with 10cm- and 20cm-long transmission lines. Their circuit compositions and substrate outer shapes are the same. Only their transmission line lengths are different. Fig. B shows the measurement results of the radiation noises.
Their radiation noises before installation of a filter were a little different, but their peak frequencies and levels were similar. However, the noise suppression effect after the installation of a ferrite bead was much different depending on the transmission line length. At the peak of the radiation noise, 375MHz, the reduction was 13dB in the case of a transmission line length of 5cm, but it was only 2dB in the case of 20cm.
Like this, the noise suppression effect of the filter changes when only the transmission line length, which is a circuit condition, is different. This is why a noise suppression method effective for one set is not so effective for another set.
1-5-2. Analysis of the cause of differences in the noise suppression effect
We measured the voltage and current distributions on the transmission line to investigate the cause of the change in the noise suppression effect of the ferrite bead depending on the length of the transmission line. The voltage was measured using a voltage probe and spectrum analyzer. The current was measured using a magnetic field probe and spectrum analyzer. The conversion from the measured value of the magnetic field probe to the actual current value was derived using a calibration substrate.
Fig. B shows the voltage distribution and current distribution measurement results with a transmission line length of 20cm. Since impedance matching is not performed, the signal is reflected between the input and output, and the voltage and current vary depending on the position on the transmission line. There was a tendency that the higher the frequency was, the bigger the difference became depending on the position on the line.
We measured the voltage and current distributions on the transmission line to investigate the cause of the change in the noise suppression effect of the ferrite bead depending on the length of the transmission line. The voltage was measured using a voltage probe and spectrum analyzer. The current was measured using a magnetic field probe and spectrum analyzer. The conversion from the measured value of the magnetic field probe to the actual current value was derived using a calibration substrate.
Fig. B shows the voltage distribution and current distribution measurement results with a transmission line length of 20cm. Since impedance matching is not performed, the signal is reflected between the input and output, and the voltage and current vary depending on the position on the transmission line. There was a tendency that the higher the frequency was, the bigger the difference became depending on the position on the line.
1-5-3. Change in the current distribution caused by the installation of a ferrite bead
I compared the current distributions before and after the installation of a ferrite bead because the radiation noise in this substrate was current-based. In the prototyped evaluation substrate, 375MHz showed a particularly big difference in the noise suppression effect of a ferrite bead depending on the position on the transmission line, so I focused on 375MHz. Fig. A shows the measurement result with each transmission line length. As with the radiation noise, the shorter the transmission line is, the smaller the current distribution at 375MHz becomes. With a transmission line length of 5cm, the overall current decreased, and the peak current decreased by 13dB as with the radiation noise. With a transmission line length of 20cm, the current did not decrease so much, and the peak current decreased by only 2dB as with the radiation noise.
1-5-4. Analysis of the cause
It turned out that the change in the current distribution is associated with the change in the radiation noise, so I compared the current distributions between different transmission line lengths before the installation of a ferrite bead.
Focusing on the current distributions at the ferrite bead installation position, I found that the current was large with transmission line lengths of 5cm and 10cm, which showed great noise suppression effect. On the other hand, with a transmission line length of 20cm, which showed little noise suppression effect, the current at the filter installation position was extremely small, and the peak of the current was away from the filter installation position, in other words a little closer to the load side.
Then, the impedance was calculated by dividing the voltage value by the current value. The impedance at the filter installation position was a little less than 100 ohms with transmission line lengths of 5cm and 10cm, but it was extremely high, about 1 kiloohm, with a transmission line length of 20cm. Since a ferrite bead is an impedance component, low impedance at the installation position leads to a great noise suppression effect, but if the impedance at the installation position is high, it is difficult to achieve a sufficient noise suppression effect. It seems that a sufficient noise suppression effect was not realized because while the impedance of the ferrite bead was 166 ohms, the impedance at the filter installation position was high, 1 kiloohm, with a transmission line length of 20cm.
1-5-5. Difference in the noise suppression effect depending on the transmission line length
The above-mentioned investigations show that the current and voltage of the transmission line vary according to the position on the line and their distributions also vary according to the frequency. It also turned out that the differences in the distributions influence the noise suppression effect of a ferrite bead.
In these investigations, I had been focusing on the frequency of 375MHz. As the next step, I investigated how the noise suppression effect varies depending on the frequency. The peak current insertion loss was used for the investigation. This is the peak current of the transmission line without a filter minus that with a filter at each frequency as shown in Fig. B. It can be used as a guideline for the radiation noise suppression effect of the filter. Fig. C shows the measurement results of the peak current insertion loss with transmission line lengths of 5cm, 10cm, and 20cm. It turned out that the frequency at which the noise suppression effect of a ferrite bead becomes smaller varies depending on the transmission line length.
Fig. C. shows that in the case of a general C-MOS digital circuit, it is likely to get sufficient noise suppression effect with a ferrite bead at a frequency of 1GHz or less with a transmission line length of 5cm. On the contrary, when the transmission line length is longer, the frequencies at which a sufficient noise suppression effect of a ferrite bead is difficult to acquire are likely to become obvious.
1-5-6. Effect of the combination of a ferrite bead and capacitor
At the frequency of 375MHz in question, a sufficient noise suppression effect of a ferrite bead could not be obtained because the impedance at the filter installation position was high. In this case, a capacitor works effectively since it bypasses the noise current from the transmission line to GND by reducing the impedance between the transmission line and GND works effectively. Therefore, I considered using a capacitor.
First, I removed the ferrite bead and installed a capacitor with a relatively small capacity, 10pF, between the transmission line and GND. As a result, the noise was suppressed at some frequencies, but not at other frequencies.
When the impedance at the installation position is low, a ferrite bead can realize a great noise suppression effect. On the other hand, when the impedance at the installation position is high, a capacitor can realize a great noise suppression effect. Therefore, I installed an inductor and a capacitor. In this case, a great noise suppression effect was realized at a wide range of frequencies. At 375MHz, the noise was 18dB smaller than without a filter. Like this, if a ferrite bead alone cannot realize a sufficient noise suppression effect, a combination with a capacitor can lead to a great noise suppression effect.
I also confirmed the waveforms with a ferrite bead only and in combination with a capacitor.
Since the added capacitor had a relatively small capacity, 10pF, there was little influence of the deformation of the waveform even after the addition of the capacitor25MHz.
1-6. Policy on filter selection for signal lines
So far, I have explained the details of the investigation on how to select a filter for a signal line for efficient noise suppression.
If the transmission line is short, an impedance component such as a resistor or ferrite bead is likely to suppress the noise. However, if the transmission line is long, an increase in the impedance may not lead to a sufficient noise suppression effect. In this case, consider adding a capacitor. This countermeasure is also expected to reinforce the resistance to electrostatic discharge (ESD) in the cable connection.
Summary
- - Noise suppression in digital circuits using a product is realized by removing the harmonic of a digital signal.
- - Major products used for noise suppression are capacitors, resistors, and ferrite beads.
- - Even with the same countermeasure, the effect may vary depending on the pattern length of the noise path, installation position of the product used for the countermeasure, etc.
- - When the pattern length is long, a combination of a ferrite bead and capacitor is likely to be effective.
Next: Chapter 2 How to Select Ferrite Beads Considering the Characteristics of Digital Circuits