| Design Articles
Reducing EMI in Digital Systems through Spread Spectrum Clock Generators
High-speed data = high-speed clocks = EMI. It doesn’t have to be that way.
Any device capable of generating signals with frequencies in the RF range is a potential source of Electro-Magnetic Interference (EMI). These signals can cause interference in the normal operation of electronic devices such as radios, televisions, cell phones and other types of equipment. The primary sources of EMI in most systems are the clock generation and distribution circuits.
Interference is caused by electro-magnetic waves that are produced by charged particles moving in an electric field. This condition occurs wherever electric signals exist. There are regulatory agencies that require devices that produce EMI to adhere to a certain set of rules and regulations. Among these rules and regulations is a requirement that the source of radiation not be greater than a pre-determined level at a certain distance from the source within a fixed frequency range. In the United States, the regulatory agency that governs the control of EMI is the Federal Communications Commission (FCC).
Causes of EMI
Clock sources can contribute to EMI in two ways. EMI can be produced through the repetitive nature of a synchronous clock and from an improperly terminated trace. The energy from the clocks radiates into a field through an antenna. An antenna might be in the form of PCB traces, PCB rework wires, components with insufficient shielding, connectors, cables (shielded or unshielded), or improperly grounded equipment.
In high-speed digital devices, fixed frequency clocks are the primary source of EMI because they are always operating at a constant frequency that allows energy to increase to higher levels. Signals that are non-repetitive or asynchronous will not generate as much EMI.
Figure 1: Slower transition rates reduce EMI
As the need for higher throughput has driven faster clock frequencies, signal transition rates have also increased. But with the faster rise and fall times comes an even larger increase in the energy level of the radiated signal. Figure 1 shows two signals that have the same frequency, amplitude, duty cycle, and phase. However, they differ in the signal transition rate. The clock with the faster rise time will have a measurably higher amount of radiated energy than the slower transitioning signal.
The second factor that contributes to EMI is an improperly terminated trace. A trace can exhibit overshoot and undershoot when there is an impedance mismatch. When this condition occurs, radiated energy will increase. Depending on the severity of the overshoot and undershoot levels, this could represent as much as 3 to 4 dB of EMI at a particular signal, or node in EMI terms. If there are ten to twenty nodes with severe overshoot then passing the FCC compliance test is in jeopardy.
There are several methods with which to solve an EMI problem in a digital system. The designer could choose to shield the design, filter a signal, or remove the energy from the offending source. These methods could be used individually or in conjunction with others.
The first method, shielding, is not an electrical solution but a mechanical implementation. Shielding uses metallic packaging to keep the EMI from escaping the unit. This method has been used often in the past but it can sometimes be a costly solution. It also doesn’t lend itself to an easy fix when an EMI problem is found shortly before a product release.
The remaining methods, filtering and energy removal, isolates the trace that is radiating the EMI. To identify which trace (or traces) is causing the problem, a test in the anechoic chamber or an EMI simulation should be performed. From this testing, an emission report will identify which frequencies exceed the specified limits. These particular frequencies are typically called hot spots. By knowing the frequencies (including the harmonics) the clock trace can be identified.
Since poorly terminated signals can cause hot spots, the first solution is to ensure all signals are properly terminated. The signals that are causing the EMI should be simulated and the traces should be analyzed for overshoot and undershoot. If there are exceptional amounts, then adjust the termination values to create a better waveform.
If all the signals are properly terminated and little to no overshoot is present, then the transition rate of the clock needs to be addressed. A substitution of a slower speed buffer may supply the answer. Many clock buffers have an option for high-speed or low-speed outputs. Often, these parts are either pin-for-pin replacements or the device has a programmable slew rate. If the lower drive is acceptable for the system, this may be the best solution. This method directly addresses the clock trace that is causing the problem and typically there is no additional cost to implement.
If a slower device is not available, filtering is a common way to slow the edge rate of a signal. This usually involves adding a capacitor to the signal that will soften the edge rate based on an RC time constant. The values of the capacitors generally range from 5 to 15 pF. Often designers will include these capacitors, which need to be placed near the source, in their schematics but not populate them unless an EMI problem is exposed. If the clock trace uses series termination, the capacitor can be placed on either side of the resistor to reduce EMI. However, for optimal termination and signal integrity, the capacitor needs to be placed between the driver and the resistor.
Although this method reduces EMI, it does however, degrade the signal integrity of the clock. Instead of sharp, clean edges suitable for high-speed clocks, the edges become rounded. Also, capacitors may need to be added for every clock copy in the design.
Figure 2: Spread-spectrum clocking (SSC) reduces EMI peaks by
spreading the energy from a clock across a wider frequency band Another method to address signals that emit excessive radiation is clock modulation. Clock modulation, also known as Spread Spectrum, has been used effectively in virtually all PCs for many years, and is ideally suited to many other applications as well. Spread spectrum clocking (SSC) is an efficient, effective and less costly alternative method for controlling EMI. SSC reduces EMI peaks by spreading the energy from a clock across a wider frequency band. Figure 2 shows that as the frequency band is made wider, the energy peak is reduced.
With this technique, peak reductions of 5dB to 18dB are possible. It can be implemented with few or no additional components, and has the advantage that energy spreading is maintained as clock signals are fanned out and propagated to their destinations.
Plotting the clock frequency versus time, we see the shape of the modulation signal. The maximum frequency excursion, Δf, is the difference between the upper and lower limits. This is usually specified as a percentage of the main frequency, and is normally referred to as the spread amount. Typical spread values are between 0.25% and 4%. This is the variable that is fine-tuned to bring EMI within spec.
The shape of the modulation profile is important in producing the maximum amount of db reduction. Although a triangle wave profile is simple to implement and gives good results, the Lexmark profile (see Figure 3) is optimized to produce a flatter frequency profile, as seen in Figure 2, and greater peak reduction. Most SSC devices use either one profile or the other, though some devices may have a selectable or programmable profile.
Figure 3: The Lexmark profile produces a flatter frequency profile
and greater peak reduction than a simple triangle wave The last variable of the modulation signal is the frequency. A standard modulation frequency is approximately 30kHz. In general, a low modulation frequency is desirable because it minimizes many possible negative effects that a spread spectrum clock may have on down-stream devices. If the modulation frequency is too low, however, audio interference becomes a risk. This is why 30kHz is chosen – it is comfortably clear of the audio band. Occasionally modulation frequencies up to 100kHz or more are seen, but these are usually specific applications that are known to be compatible with these modulation frequencies. Most SSC devices operate at a fixed modulation frequency.
The amount of EMI reduction depends not only on the amount of spread, but also on the clock frequency. For example, a 1% spread on a 100 MHz clock means that the peak-to-peak frequency excursion is 1 MHz. In comparison, 1% spread on a 20 MHz clock is only 200kHz. So while the spread percentage is the same for these two cases, the actual frequency spread is not, and therefore the EMI peak reduction is also not the same. A lower frequency clock requires a greater spread percentage to achieve the same EMI reduction as a higher frequency clock.
This relationship to frequency applies to the harmonics as well as to the main frequency. Just as a harmonic is an integer multiple of the main frequency, the width of frequency spreading is similarly multiplied. Going back to our 100 MHz clock with 1% spread, the 3rd harmonic is at 300 MHz and has 3 MHz of spread – three times the spread on the 100 MHz clock. Because of this increased frequency spread for the harmonic, the EMI reduction is correspondingly greater compared to the main frequency.
One more interesting observation is that the relationship between peak EMI reduction and spread amount is non-linear. The most optimum reduction is achieved with a small spread percentage, since increasing the percentage yields diminishing returns. This is one reason why few applications use more than 5% of spread.
Cost Savings with EMI Suppression
Use of an EMI suppression-enabled clock IC can result in a reduction of system radiated EMI of 10 dB of more. This can result in dramatic cost savings for the system, of anywhere from less than $1, to $5–10 or more. Conventional techniques for reducing EMI include shielding ground planes, filtering components and shielding. Going from a two-layer board to a four-layer board to insert additional ground planes could easily cost $5–6. Filtering EMI typically uses ~$.25 worth of resistors, inductors, and capacitors, and often $.70 worth of common mode chokes and toroids. In many cases filtering will not be enough to allow a system to pass EMI tests, in which case costly shielding may be required. Shielding can easily add several dollars to the cost of a system.
Integrating Provisions for EMI
When it comes time to integrate SSC into a design, engineers will find there are a variety of SSC devices. Clock generators produce a specified frequency from a crystal. Some of them can generate a combination of spread spectrum clocks and non-spread clocks, at multiple unrelated frequencies. Single frequency, fully integrated spread-spectrum oscillators (SSXO) are also available. Other low pin-count devices can be placed in the path of an ordinary clock signal to add spread spectrum.
Many SSC devices allow spread to be turned on or off, and some allow the amount of spread to be adjusted by pin selection. Programmable spread spectrum clock generators provide considerable flexibility, allowing the user to specify both the output frequency and the exact amount of spread. Output drive strength is also configurable on most programmables. If EMI testing at the end of the design cycle determines that a change to the spread parameters is needed, a field programmer can be used to quickly program a device with the new settings. In this way, by provisioning for SSC during the design of a system, an engineer can rest assured that full provisions are in place to control clock-generated EMI, and that only a small amount of time needs to be budgeted for EMI fine tuning at the end of the design cycle.
This article originally appeared in the July, 2008 issue of Portable Design. Reprinted with permission.
San Jose, CA