Basic concepts of high-speed digital design

Basic concepts of high-speed digital design

Posted on 19:11:47

Issues to consider

From the signal waveform,High-speed digital design needs to consider the vertical and horizontal aspects: the quality of each signal, the mutual timing between signals.

The quality of each signal mainly involves three aspects: signal integrity, power supply integrity, and crosstalk.

Signal integrity problems refer to that PCB cabling cannot be used as a conductor of a simple connection circuit with extremely low DC resistance at high speeds. The capacitance and inductance of other cables (including other signals and locations) must be considered for PCB cabling. Online PCB transmission is no longer a simple step signal, but the superposition of sine waves containing multiple frequency components and back-and-forth reflection attenuation like water ripple.

Power supply integrity refers to the rapid turning of a large number of door circuits, switching large power supply current in a short period of time. This requires considering the reaction rate of the power supply, that is, the size of the AC internal resistance.

Crosstalk exists in low-speed design, but it is usually very small. It becomes very important in high-speed design.

The time available for accelerating clock frequencies in each cycle becomes very limited, causing mutual timing problems between signals.

1.2 Basic Concepts

Some nouns are often used in high-speed digital design. It is helpful to grasp their meanings accurately. Some important concepts are listed below. Read it for the first time and check it later.

The following concepts are common in addition to section 1.2.4. They are applicable not only to the microline and cable structure of PCB, but also to coaxial and twisted pair cables.

1.2.1 Increase time, bandwidth

Rising time: the time when the signal has risen from 10% of the High Level to 90%. It is also stipulated that 20%-80%, 10%-90% is more commonly used.

Bandwidth: For signals, the highest frequency component to be considered minus the lowest frequency component; an instrument is the highest frequency component of an effective output signal when a full-band signal is input minus the lowest frequency component. The valid meaning is that the amplitude is greater than a specified value. Generally, (FH-fl)> FL or FL is close to zero. Therefore, in many documents, the upper limit of frequency (FH-fl) is used to represent the bandwidth, this article will also follow this rule.

1.2.2 Set total system and Distribution System

Total system: the system parameters do not change with the internal positions of the system. The parameters of each point are the same. You only need to consider a specific point.

Distributed System: system parameters vary with the positions in the system. All points must be considered separately.

1.2.3 Controllable transmission lines and Impedance

Transmission Line: a wire used as a distribution system. The capacitance and inductance of the wire are not ignored. It can be divided into ideal transmission line (lossless transmission line) and actual transmission line (lossy transmission line ).

Controllable impedance: an appropriate structure is used to change the impedance of the transmission line unit length as a constant with the position, facilitating the end connection. For example, the PCB cabling is designed to be a micro-strip structure and a two-wire structure, and the interconnection wires are designed to be coaxial cables and twisted pair wires.

1.2.4 Microline, cable, pile line (TLine), reference plane, point-to-point Wiring

Microline: 1, a transmission line with controllable impedance. Only one reference plane is emphasized.

Figure 1 microline

Trunk cable: 2, a transmission line with controllable impedance. There are two reference planes up and down.

Figure 2 geographic line

Pile wire: short branch in PCB cabling, usually used for connector pin.

Reference plane: refers to the power supply layer or formation close to the signal layer, providing a reflux path with low impedance (mainly low sensitivity) for the signal.

Point-to-Point cabling: A drive connects to the cabling structure of a receiver.

1.2.5 Resistance, Impedance

Resistance: the effect of the conductor on current (including conducting current and displacement current) that does not cause phase changes. Simply put, if the voltage at both ends of the loaded conductor is, the current flowing through the conductor is. If a = B, there is a amount of resistance, which is represented by u/I, equal to the DC resistance plus the AC resistance. When the frequency is high, the impact of the Skin Effect on the AC resistance will become greater.

Impedance: the blocking effect of the transformer on the phase changes 90 degrees due to the current (including the conducting current and the displacement current. It is the definition of capacitance and inductance. Negative Numbers indicate 90 degrees ahead of phase, and positive numbers indicate 90 degrees behind phase.

Impedance: A combination of resistance and impedance. The exact definition is the complex impedance z = "R" + x * j. X indicates the impedance, and J indicates the imaginary number unit in mathematics. Sometimes in order to simply represent it with a real number, it is the modulus of the complex impedance. The complex impedance fully reflects the blocking effect of the conductor on the current.

1.2.6 End-to-end and impedance matching

End: The Source end of the transmission line is connected to a component with a specific impedance or a component with a specific impedance connected to the terminal.

Impedance Matching: the process of making the impedance continuous is called impedance matching, and the means is the end connection. The proper impedance of the client is matched, and the unsuitable impedance of the client is not matched.

1.2.7 Mutual interference, mutual interference, remote crosstalk, and near-end Crosstalk

Crosstalk: crosstalk caused by changes in the source voltage. The amplitude of both the remote and near-end crosstalk is always the same as that of the interference source, which can be represented by the model in figure 3. Just like a lot of small capacitors are evenly connected to each other.

Figure 3 intercommunication crosstalk Model

Figure 4 compatibility crosstalk Waveform

Mutual crosstalk: crosstalk caused by changes in interference source current. The remote crosstalk is opposite to the interference source. The amplitude of the near-end crosstalk is the same as that of the interference source, which can be represented by the model in Figure 4. Just like the coupling between the primary and secondary Coils of the transformer.

Figure 5 mutual interference model

Figure 6 mutual inductance and Crosstalk Waveform

Crosstalk occurs at each point of the signal. The superposition of crosstalk at all points is the result of measurement. The effect of crosstalk is a small spike that spreads forward. The crosstalk at each point is transmitted to both the source and terminal directions.

Remote crosstalk: the crosstalk transmitted to the terminal is the remote crosstalk relative to the interference source.

Near-end crosstalk: Compared with the interference source, the crosstalk transmitted to the source end is near-end crosstalk.

Figure 7 remote and near-end Crosstalk

1.2.8 Ground bombs, orbital collapse, decoupling

Ground play: voltage fluctuation on the inductance in the signal return path. The inductance refers to the inductance that goes through the line, not the coil. Generally, it is often used to analyze the ground bullet of the chip, because the inductance of the ground pin of the chip is greatly affected. Power supply pins also have the same phenomenon.

Rail collapse: refers to the fluctuation of power supply voltage and ground caused by ground bombs. Do not be misled by the word "collapse", including fluctuations in the high and low directions.

Decoupling: coupling is commonly used to indicate the mutual influence between two signals. The literal meaning of decoupling is to reduce coupling. This mainly refers to reducing the mutual influence caused by the collapse of power supply and ground. The common means is to parallel capacitance between power supply and ground.

1.2.9 RLCModel, resonance frequency

RLC model: replace authentic components with a combination of ideal resistors, capacitors, and inductors. It is a common model for analyzing the frequency response of passive devices.

Common RLC models include the RLC series model and the RLC parallel model.

Figure 5 shows the capacitor RLC series model.

Figure 5 RLC Series Model

Resonance frequency: the response of the RLC model has an extreme value with the change of frequency. The frequency corresponding to the extreme value is the resonance frequency. At this frequency, the impedance of the RLC model is the smallest and the impedance is zero.

For the RLC series model, the approximation is as follows:

The larger the resistance, the smaller the Q value. The larger the inductance, the larger the Q value. The larger the capacitor, the smaller the Q value.

1.2.10 RLCModelQValue

The Q value of the RLC model:, which reflects the speed of energy attenuation in a cycle.

1.2.11 Increase time and bandwidth

The two most important parameters for processing digital signals are the increase time and bandwidth. Many system design rules are based on these two parameters. The shorter the increase time, the more difficult it is to process. How can we determine the system's rising time? This depends on the key chips used in the system. These parameters are provided in their data manuals.

Find the minimum increase time of the system. This is what we need to do first.

Sometimes the concept of bandwidth is also mentioned, which is often used in signal measurement. Consider the typical situation: clock signal.

If the signal bandwidth is mentioned, it is to perform Fourier analysis on the signal. I need to consider the largest and smallest frequency components. We only need to consider the components within this range, and ignore other components. This accuracy is sufficient for us. Generally, low frequency processing is very easy, so you don't need to consider it. When it comes to bandwidth, it is often replaced by the maximum frequency. Select a clock frequency of 5 times for the bandwidth here (expanded with a triangular series to take the first five harmonic waves ). If it is not a clock signal, the signal waveform is divided into a small segment to find the fastest changing frequency.

If an instrument is mentioned, it must be determined whether the input bandwidth is greater than the signal bandwidth. This is an accurate measurement condition. You can compare the bandwidth of the instrument directly. If the increase time of the instrument input is given, you can convert the time according to the following formula before comparison.

It can also compare the rising time, requiring that the rising time of the instrument be less than the rising time of the signal.

There are also more accurate measurement methods. In this case, the signal rising time, the rising time of the oscilloscope, and the measured rising time meet the following requirements:

Therefore, you do not need to determine whether the rising time is satisfied. After the measurement is completed, the rising time of the signal can be calculated directly. In addition, this is more accurate.

1.2.12 Set total system and Distribution System

From the perspective of time, it takes some time for the digital signal to complete the level change. It also takes some time for the source level changes to be reflected on the terminal. in a period of time, signals are transmitted and reflected forward in a form similar to water waves. There is a parameter similar to the wavelength, called the length of the rising edge.

If the length of the strip in the system is <the length of the rising edge/N, it can be processed as a collection system. This usually means that the PCB design is greatly simplified, and even high-speed signals can have no end connection, and do not need to adopt the impedance controllable microline or strip structure.

The signal quality of the centralized system depends on the Q value of the equivalent RLC circuit. When the Q value is lower than 0.5, the signal quality is good. When the Q value is greater than 0.5, corresponding measures should be taken to reduce the Q value. You can increase the series resistance to reduce the Q value.Whether there is a signal quality problem depends on whether the working frequency is close to the resonant frequency. If the Q value is greater than 0.5, it only indicates that there is a hidden danger..

If the length of the cabling in the system is> the length of the rising edge/N, it should be processed as a distribution system. This usually means that the PCB design is complicated, and high-speed signals must be connected to each other. To facilitate the connection, the impedance-controlled strip and strip structures must be adopted.

The value of N is not fixed. 6 ~ for PCB cabling ~ 4. 6 for strict signal quality requirements and 4 for non-strict requirements; 6 for Interconnected Cables ~ 3.

Whether a distributed system or a centralized system is the basis for determining whether a high-speed digital system requires special processing. It is more meaningful than the absolute frequency. The most typical example is the telephone line and PCI plug-in card. We cannot tell exactly what the telephone line signal frequency is, but at least it will not be higher than the 33mhz clock of the PCI bus. However, the telephone line needs to be 600ohm, while our common PCI interface Nic is a simple two-layer PCB Board, and even the controllable impedance of the microline and cable are not used.

1.2.13 Controllable impedance and Client Connection

If the system must be designed as a distributed system, it is necessary to consider the high-speed signal. We need to know the Characteristic Impedance of PCB cabling to achieve the end connection. Therefore, we need to adopt the controllable impedance microline and strip line structure. This usually means that at least four layers of PCB structure are required, because each signal layer requires at least one reference plane. The feature impedance is determined by the ratio of the strip width to the reference plane distance. This is the main design basis of multilayer PCB structure.

Of course, there are also special cases. For example, a high-speed signal test system can be designed with a low-cost two-layer PCB. Designed as a microline, one layer is the signal layer, and the other layer is the reference plane. The thickness of the board is the distance to the reference plane. The thickness can be 1.0mm (about 39mil. If you want to design a feature impedance of 50 ohm, it will take about 2mm-width strip. In practical applications, 2mm-width transmission lines are unacceptable.

The end connection aims to eliminate reflection, ensure signal quality, and achieve signal integrity. Sometimes the end-to-end resistance also has the effect of Attenuation Reflection (not to eliminate reflection) and improving the signal amplitude.

1.3 Noise

Noise is another important factor affecting the design rules of High-speed signals. The main sources of noise are intercommunication and mutual interference. Necessary measures include isolation, meeting the 3 W principle, and differential cabling. This is the main reason why the high-speed digital system strictly controls the cabling distance and carefully designs the reference plane.

1.4 Power supply integrity

The third important factor affecting the design rules of High-speed signals is power supply integrity. Power integrity mainly refers to rail collapse and high-frequency noise decoupling of power supply. High-frequency noise decoupling of power supply is a frequent problem for high-speed IC such as DSP and FPGA.

1.5 Problem1Answer

Problem (1). If you need to design a PCB, it is enough to consider two layers of PCB from the aspect of wiring resources. So we adopt two or more layers. How can we make judgments? Why?

Answer: First, find the fastest rising time of the system signal. Find the unit length of PCB cabling delay data and estimate the shortest length of the rising edge. The formula is as follows:

Length of the rising edge = rising time/unit length cabling Delay

Different from multi-layer panels, two-layer panels are calculated once. The longest line of the estimated signal. a pcb pre-layout can be performed. If the maximum length of the strip is <the shortest length of the rising edge/N, you can select two layers of PCB without considering the signal. If the maximum cabling length is greater than the shortest rising edge length/N, select at least 4 layers of PCB, and ensure signal integrity through signal connection.