Engineers have it rough: The standard for our performance is perfection. A hockey goalie who stops 9 out of 10 shots is considered a demi-god. A quarterback who completes 60% of his passes is destined for the hall of fame, and if a major league batter gets a hit 40% of the time, he is named the MVP. An engineer who designs 100 bridges, only one of which falls down, is considered a disgrace to our profession. The designers of the space shuttle Challenger are not remembered for any of their successful launches, and who spends time praising Microsoft for the bug-free features of Windows? Because of this, engineers tend to be risk adverse criticism and shy away from projects that look like they would be difficult. Mixed-signal design is one of those tasks. A right-minded engineer looks at the requirements and determines that the benefit of a successful design (a photo-copied recognition certificate in a handsome plastic frame), is overcome by the risk of failure (public humiliation and possible unemployment.) The engineer then pronounces that the project is either not possible, or explains that the budget and schedule for such a task are outside what would be acceptable to management.

Unfortunately for us some of our managers will see through this ploy, and will force us into a mixed-signal design. With this paper in hand, you will be able to reduce the risks associated with the project, and improve your chances of being able to decorate your cube with another certificate. If it doesnýt work, check out: /centers/carcenter.html

Mixed-Signal Design

A digital system can tolerate a certain amount of noise, with no degradation in its performance, due to noise immunity. If a digital signal is sent with a 16-bit word, each bit having nominal levels of 0 V and 4 V and there is 200 mV of noise on each bit, there will be no errors on the receive end. Upon receiving a signal with a level of 0.2 V the receive end ýknowsý that this is supposed to be a 0 and, passing the signal along, it removes the noise present. If there is an analog signal with a range of 0 V to 4 V and 200 mV of noise is added to the signal, then the loss of information would be equivalent to losing the 12 LSBs of the 16-bit word. As a result of the high level of noise immunity, several practices that are detrimental to analog system design have become common place in digital design. If you have some of these bad habits, it would be a good idea to rid yourself of them before you get much further.

Please click here for full picture.

 In the oscilloscope photo (above) on the left the upper trace is the input to a digital gate, and the lower trace is the output. You can see how the noise that is present on the signal is reduced by the gate. In the photo on the right the noise has been faithfully passed from the input to the output by an analog amplifier.

Even if the system has been rid of the common faults of digital design, and is made clean, there are several interactions between the digital circuitry and the analog circuitry that must be considered. Digital circuits and signals are very different from analog in many fundamental ways. Care must be taken to make certain that these signals are each treated appropriately. To see some of the differences let'us compare signals in the analog and digital domains and see how they differ in various attributes.

The biggest difference between analog and digital representations of similar signals is how the energy that is used to represent these signals is distributed in the frequency domain. Consider a video signal, which has an analog representation of a 1-V p-p signal, and a digital signal, which is sampled at 4x the subcarrier frequency (about 15 MHz for NTSC.) You can examine the spectra of these two signals on a spectrum analyzer and you will see that all of the energy in the analog signal lies at frequencies below 5 MHz. The digital signal consists of harmonics of 15 MHz and therefore has the bulk of its energy at frequencies well above that of the analog signal. This is important when you consider that the ability of a signal to capacitively couple itself into another path increases as the frequency increases. The result is that digital signals find it much easier to find their way into parts of a circuit where they do not belong than do analog signals. Furthermore, as we discussed above, analog signals are much less tolerant of interference than are digital signals. Therefore the problem of digital signals coupling into analog signals is a major one in mixed-signal systems.

Another big difference between analog and digital signals is how they dissipate power. An analog signal dissipates power at an approximately constant rate with respect to time. Consider the same video signal examined above in the digital domain, and assume that each output bit transitions on average every other clock cycle (bits look like a 3.75MHz square wave.) We will also consider that this signal is being driven onto a bus with an impedance of 10 kW and 30 pF. The power dissipated during the high portion of the output bit is V²/R, or 2.5 mW, during the rising edge of the signal; if we assume a 1 ns rise time and 4 V high signal then dV/dt = 4E9 V/s, driving the 30 pF of capacitance, then there is a surge current of 120 mA, a surge power of 600 mW. This means that there is a factor of almost 200 between the average and peak powers dissipated in the digital signal.

We can see from the difference in the way that analog and digital have their energy distributed, and from the difference in how they consume power, that they are not alike. We have also seen that analog signals are much more vulnerable to interference. To reduce the amount of corruption of an analog signal by digital signals, we need to understand the mechanism of this corruption.

A digital circuit has pins some of which are identified as inputs and, in general, it is very difficult to influence the output of the circuit by injecting foreign signals into other pins (power pins, output pins, no-connects etc.) An analog IC, however, has as many inputs as it has pins. It is true that it will generally be more sensitive to unwanted signals if they are on its true inputs, but the output can be easily influenced by signals injected through any pin on the chip. The two networks with which we need to concern ourselves the most are the input signal, and the power supply network.

The primary culprit causing corruption of an analog signal is capacitive-coupling from another source. These sources can be digital signals, power supply lines or ground planes. Capacitive-coupling is worst when the corrupting signal has higher frequency components and for larger signal swings. Accordingly, the most common problem encountered is that of a digital signal (or clock signal) coupling into your analog signal.

Step 1 - Make Your Digital Signals As Benign As Possible

One of the first things that you can do to reduce the amount of coupling from digital signals is to reduce the peak energy dissipated in the digital logic. As pointed out above most of the power in a digital circuit is dissipated when the outputs are making transitions. The two main factors determining the edge energy are the rise/fall times of the logic, and the size of the swings. To reduce the energy in the edges, and hence the capacitive-coupling, you want slow rise and fall times, and small logic swings. If an 'LS' gate with 5 ns rise times will work, use that rather than an 'F' series gate with 1 ns rise times. If you can migrate your digital design from 5 V to 3 V, you will get a 40% reduction in the edge energies automatically. ECL is better than CMOS or TTL, for noise purposes, and in cases where exceptionally low-noise digital circuitry is required differential ECL is a great alternative.

In systems where there is an ADC sampling the analog signal, another game that you can play is to adjust the phase of the sampling clock, relative to the rest of the system. When the digital clock transitions the whole circuit has a hiccup, with a surge of current flowing through the ground planes, bypass capacitors discharging and starting to charge up again and generally much more excitement than is good for an analog signal. This disturbance has probably all settled down by the time the next clock edge comes along. If you can manage to sample your analog signal during one of these periods of relative calm you will end up with a cleaner signal.

If you have a digital signal travelling any appreciable distance (my rule of thumb: if the signal trace is longer than your thumb, the distance is appreciable) then you need to think about possible transmission line effects. For a primer on transmission lines, read Transmission Line Basics. The things that you want to avoid in a mixed-signal design are poorly terminated transmission lines which result in ringing. When there is ringing there is additional energy available, at high frequencies, that can couple itself into places where it is not supposed to be.

Often times, the digital signals will find their way into the analog signal paths through the power supply network. To reduce this, bypass, bypass and bypass some more. By adding bypass capacitors to the analog circuitry, you will be reducing the amount of interference that can get from the power supply into your circuits, and by adding bypass capacitors to your digital circuitry, you will be reducing the amount of disturbance that gets into the power supply network in the first place.

Step 2 - Make Your Grounding Work For You

Contrary to what my son believes, grounding is a good thing, but like any good thing, you can have too much of it. In a mixed-signal design, you will probably have both a digital and an analog ground. The key to success is to have a very-low impedance ground plane for low frequencies, and to have the impedance at high frequencies between these two ground planes as high as possible. Remember how digital logic dissipates lots of power during transitions? This means that the current flow through the digital ground plane will consist of a series of current spikes. These current spikes translate to voltage spikes on the ground plane, which if they get into the analog circuitry, spell bad news. By minimizing the impedance from the digital ground to the power supply ground, you will minimize the size of these spikes; by increasing the high-frequency impedance between the ground planes, you will keep the spikes from getting onto the analog ground plane. One way to accomplish this is by connecting the analog ground plane to the power supply through an inductor. This will maintain a low impedance path for the return current, but will make it harder for the high-frequency digital interference to make its way onto the analog ground plane.

Tree structured ground networks will also help to keep signals from one part of your circuit from interfering with other parts of the circuit. Try to imagine the path that the ground current will take to get from your circuit back to the power supply. Try and lay out your ground plane in such a way that the current is forced to go directly back to the power supply without getting into any mischief first.

Step 3 - Use Board Layout To Keep Analog And Digital Signals Separate

Board layout can make or break a mixed-signal design. Many of the bad things that happen to a mixed-signal design happen as a result of capacitive-coupling of one signal into another. To minimize this remember that by physically separating two traces or circuits, you decrease the capacitance between them and reduce the coupling. Therefore try to keep all of the analog circuits together on one part of the board, and all the digital parts as far away as is practicable.

In addition to capacitive-coupling there can be inductive-coupling between two traces. This is worst when the traces run parallel to one another. One trick that you can use is to make the routing channels of the analog circuitry run at a 45ý angle to the digital routing channels. This way it will be less likely to have a long digital trace running parallel to an analog trace.

If possible, ring the analog circuitry on the board with a shield, connected to analog ground. This will reduce any capacitive-coupling between things inside and outside the shield.

In many cases, you will be able to have one or more PCB layers dedicated to power and ground. If this is the case remember that if there is an entire layer devoted to analog ground, and another layer devoted to digital ground, that these look like a large parallel plate capacitor, and it is likely that the digital ground noise will be coupled directly into the analog ground plane. It is a good idea therefore, to not have digital ground-plane or power-plane under analog circuitry and vice-versa.

When fabricating a PCB the outer layers are often tin plated to aid in solderability. This means that the conductive layers on the outside are thicker, and lower impedance than those on the internal layers. If you lay out your pad stacks properly you will be able to make prototype boards with power and ground on internal layers (it is much easier to debug with traces on the outside layers), then when the board goes to production swap the layers so that the power and ground planes are on the outside. This will give you lower impedance power and ground planes, as well as shield your traces from the outside world.

Conclusion

The key to successful mixed-signal designs is to keep the digital signals from interfering with the analog signals. This is done my making the digital signals as benign as possible, while at the same time erecting as many barriers to keep them at bay as possible. By doing these things, chances are, you will end up with a working design. If not, try professional baseball, it is a much more forgiving profession!