Skip to content 
What Is Nyquist Frequency in High-Speed Cable Links？
- andy
A cable can pass continuity, look mechanically perfect, and still fail the moment the SerDes starts running at full speed. That is because a high-speed cable assembly is not just a collection of conductors; it is part of an analog transmission channel carrying a digital waveform. As data rates rise, engineers need a common reference that connects symbols per second with insertion loss, reflections, equalization, and eye opening. Nyquist frequency is one of the most useful reference points for making that connection, but it is often misunderstood when bit rate, baud rate, bandwidth, NRZ, and PAM4 are treated as if they were interchangeable.
In high-speed serial-link engineering, Nyquist frequency is commonly defined as one-half of the symbol rate. For NRZ, one symbol carries one bit, so the Nyquist frequency is usually half the physical bit rate. For PAM4, one symbol carries two bits, so the same raw bit rate uses about half the symbol rate and therefore a lower Nyquist frequency. The actual interface specification still determines the exact electrical lane rate.
The formula is simple; the engineering judgment is not. A 28 GHz Nyquist point does not mean frequencies above 28 GHz are irrelevant, and two cables with the same insertion loss at 28 GHz can behave very differently if one has a smooth response while the other has reflection-driven ripple. That is where real cable development becomes interesting. The useful question is not merely, “What is the Nyquist frequency?” but, “What does the complete cable channel look like around that frequency, and can the transmitter and receiver still recover the intended data with enough margin?”
What Is Nyquist Frequency?
In a high-speed serial cable link, Nyquist frequency is one-half of the symbol rate and corresponds to the fastest repeating alternating-symbol pattern. Engineers use it as a practical reference for channel loss and signal integrity because attenuation near this frequency strongly influences edge shape, intersymbol interference, equalization demand, and available receiver margin. It is a link parameter, not a permanent frequency rating of the cable itself.
Nyquist in a Serial Link
Consider a 10 GBd NRZ stream that alternates continuously between 1 and 0. Each symbol occupies 100 ps, but one full electrical cycle needs two symbols: one high state and one low state. The repeating 1010 pattern therefore has a period of 200 ps, corresponding to 5 GHz. That 5 GHz point is what signal-integrity engineers normally call the Nyquist frequency of a 10 GBd serial link. The same logic applies to higher symbol rates even when the waveform is no longer visually close to an ideal square wave after traveling through a practical connector-and-cable channel.
This definition is useful because it joins the digital and analog descriptions of the same link. The digital side talks about symbols, unit intervals, coding, lane rates, and error probability. The physical channel is measured in impedance, GHz, dB, delay, reflection, crosstalk, and S-parameters. Nyquist frequency gives both teams a common coordinate on the frequency axis. It does not replace full channel analysis, but it helps engineers quickly relate a data rate to the region where cable attenuation and discontinuities begin consuming meaningful signal margin.
Bit Rate, Symbol Rate, and Bandwidth
Bit rate is the number of bits transported per second, while symbol rate is the number of signaling symbols transmitted per second. The two are equal for ideal binary NRZ because each symbol represents one bit. PAM4 changes the relationship because four amplitude levels can encode two bits per symbol. This distinction is the reason “data rate divided by two” is a safe shortcut only when the actual line signaling uses one bit per symbol and the number being quoted is the electrical lane rate rather than an aggregate system throughput number.
Bandwidth is different again. It describes a usable frequency range according to a defined response or performance criterion, and it should not be confused with Nyquist frequency. A cable evaluated for a link with a 14 GHz Nyquist point may still need controlled behavior well above 14 GHz because real transmitters have finite but fast rise times and therefore generate useful spectral content beyond the fundamental alternating-symbol frequency. Saying “the Nyquist frequency is 14 GHz” is therefore not the same as saying “nothing above 14 GHz matters.”
What the Nyquist Point Tells You
The Nyquist point is especially useful when examining insertion loss. If the cable assembly already attenuates the signal heavily at Nyquist, the receiver has less high-frequency content available to reconstruct the transmitted symbols. Transitions become slower, pulse energy spreads into adjacent unit intervals, and the equalizer must work harder. The acceptable amount of loss is not universal, however. It depends on the transmitter swing and pre-emphasis, receiver sensitivity and equalization, modulation, reflections, crosstalk, noise, and the channel budget defined by the applicable interface.
A single Nyquist-loss number can also hide important problems. Imagine two assemblies that both measure 12 dB of insertion loss at 28 GHz. The first has a smooth, monotonic loss curve from low frequency through 28 GHz. The second reaches the same value but contains a deep notch around 22 GHz caused by a connector resonance. Those channels may not behave similarly in the time domain, because sharp frequency-selective features create different pulse responses and can be harder for equalization to compensate. Nyquist frequency is therefore a reference point, not a complete quality grade.
How Do You Calculate Nyquist Frequency?
Calculate the Nyquist frequency from the actual symbol rate: fNyquist = Symbol Rate / 2. For NRZ, one symbol normally carries one bit, so symbol rate is approximately the physical lane bit rate. For PAM4, each symbol carries two bits, so the symbol rate is approximately half the raw bit rate. Always verify protocol encoding, FEC, lane count, and the specified electrical rate before finalizing the calculation.
The Basic Formula
The core relationship is straightforward: Nyquist frequency equals one-half of the symbol rate. If the symbol rate is 20 GBd, the Nyquist frequency is 10 GHz. If the symbol rate is 56 GBd, the Nyquist frequency is 28 GHz. The same result can be calculated from the unit interval, because the unit interval is the time allocated to one symbol. A 56 GBd signal has a unit interval of about 17.86 ps, so two unit intervals occupy about 35.71 ps, corresponding to approximately 28 GHz.
The calculation becomes less straightforward when the starting number is a product-level throughput label rather than the actual lane rate. A system described as 100G, 200G, or 400G may divide traffic across multiple electrical lanes, and coding or forward error correction can make the electrical symbol rate different from a simple division of the payload throughput. For cable development, the engineering input should therefore identify the per-lane signaling rate and modulation, not just the aggregate data number used in marketing or system architecture documents.
NRZ and PAM4 Calculations
For NRZ, the simplified relationship is convenient because one symbol represents one bit. A 25 Gb/s NRZ lane is therefore approximately 25 GBd and has a Nyquist frequency of 12.5 GHz. A 56 Gb/s NRZ lane is approximately 56 GBd and has a 28 GHz Nyquist frequency. This is the origin of the familiar rule that Nyquist frequency equals one-half of the bit rate, but that rule should only be used when the quoted bit rate truly represents a one-bit-per-symbol electrical lane.
PAM4 carries two bits per symbol, so the same raw bit rate can be transmitted at roughly half the symbol rate. In a simplified 112 Gb/s PAM4 example, the signal is approximately 56 GBd, producing a 28 GHz Nyquist point. A simplified 224 Gb/s PAM4 example is approximately 112 GBd, producing a 56 GHz Nyquist point. Actual standards may use rates such as 53.125 GBd or other non-round values, so the standard-defined line rate should take precedence over a rounded “112G” or “224G” label.
Idealized Raw Lane Rate | Modulation | Bits per Symbol | Symbol Rate | Unit Interval | Nyquist Frequency |
10 Gb/s | NRZ | 1 | 10 GBd | 100 ps | 5 GHz |
25 Gb/s | NRZ | 1 | 25 GBd | 40 ps | 12.5 GHz |
56 Gb/s | NRZ | 1 | 56 GBd | 17.86 ps | 28 GHz |
56 Gb/s | PAM4 | 2 | 28 GBd | 35.71 ps | 14 GHz |
112 Gb/s | PAM4 | 2 | 56 GBd | 17.86 ps | 28 GHz |
224 Gb/s | PAM4 | 2 | 112 GBd | 8.93 ps | 56 GHz |
Common Calculation Mistakes
The most common mistake is taking a system data rate and immediately dividing by two without checking what that number actually represents. A 400 Gb/s link may be composed of several lanes, so 200 GHz is obviously not the right Nyquist frequency for one physical pair. Another mistake is treating PAM4 as though it were NRZ and using raw bit rate as symbol rate. A third is ignoring protocol overhead and assuming every advertised throughput number maps exactly to the electrical line rate used by the SerDes.
There is also a more subtle mistake: calculating the correct Nyquist frequency and then assuming that this one number completely defines the cable requirement. It does not. The cable engineer still needs the required frequency range, channel loss limits, impedance target, connector system, allowable length, return-loss or crosstalk limits, and any eye or BER criteria. A good RFQ for a high-speed cable assembly therefore includes the protocol, actual lane rate, modulation, connector part numbers, target length, and available channel specification whenever those details are known.
How Do NRZ and PAM4 Differ?
NRZ uses two voltage levels and carries one bit per symbol, while PAM4 uses four levels and carries two bits per symbol. PAM4 can therefore double raw bit throughput at the same symbol rate and Nyquist frequency. The tradeoff is smaller vertical decision margin, tighter linearity requirements, and greater sensitivity to noise, reflections, crosstalk, loss, and other channel imperfections that disturb the spacing between its four amplitude levels.
Bits per Symbol and Frequency Demand
The difference starts with information density. NRZ has two possible states, so each symbol represents one bit. PAM4 has four possible amplitude states, which allow each symbol to represent two bits. If both links operate at 56 GBd, both have a 28 GHz Nyquist frequency, but the idealized raw throughput differs: approximately 56 Gb/s for NRZ and approximately 112 Gb/s for PAM4. From the cable’s point of view, the symbol timing is the same, even though the amount of information carried by each symbol is different.
This is valuable because copper-channel attenuation generally becomes more severe as frequency rises. If a design team attempted to carry a simplified 112 Gb/s stream using one-bit-per-symbol NRZ, the symbol rate would be about 112 GBd and the Nyquist frequency about 56 GHz. Using PAM4 allows the same idealized raw bit rate at about 56 GBd and 28 GHz Nyquist. That lower frequency burden can make the channel more practical, especially when connector transitions, dielectric loss, conductor loss, and available equalization are already limiting the reach.
The Cost of Four Levels
PAM4 does not make the channel “easier”; it moves part of the difficulty from the frequency axis to the voltage axis. NRZ primarily asks the receiver to distinguish between two levels separated by one decision threshold. PAM4 has four amplitude levels and three decision thresholds. If the overall signal swing is constrained, the voltage separation between neighboring levels becomes much smaller, leaving less vertical margin for random noise, crosstalk, reflection-induced distortion, transmitter nonlinearity, and receiver imperfections.
That smaller spacing makes seemingly modest interconnect defects more consequential. An impedance discontinuity that produces a manageable reflection in an NRZ system may disturb one or more PAM4 levels enough to reduce eye height materially. Pair imbalance can convert useful differential energy into common-mode energy. Crosstalk from an adjacent lane can shift decision levels. A poorly controlled connector transition can introduce frequency-selective distortion that complicates equalization. The result is that PAM4 may use a lower Nyquist frequency for a given raw bit rate while still demanding a better-controlled channel.
Choosing the Right Engineering View
It is tempting to compare NRZ and PAM4 by asking which one is “better,” but the useful engineering question is what tradeoff the system is making. NRZ consumes more bandwidth for a given bit rate but offers larger level spacing and simpler detection. PAM4 improves spectral efficiency but requires more sophisticated transmitter and receiver behavior and tighter channel control. The cable assembly sits between those decisions and must preserve enough amplitude, timing, symmetry, and return-path quality for the chosen SerDes architecture to work as intended.
For that reason, a cable supplier should not qualify a high-speed assembly only by saying it has the “right impedance.” Differential impedance matters, but so do insertion loss, return loss, crosstalk, mode conversion, skew, connector transitions, and repeatability of pair geometry through the termination area. SINO-CONN treats the interface, lane rate, cable construction, connector system, shielding, and validation requirements as connected inputs during high-speed development, because no single specification can represent the whole transmission path.
How Does Nyquist Loss Affect Signal Integrity?
Insertion loss at Nyquist shows how strongly the channel attenuates a strategically important part of the data spectrum. As loss increases, high-frequency content is reduced relative to low-frequency content, edges become slower, and pulse energy spreads into neighboring unit intervals. The result is more intersymbol interference and a greater equalization burden, leaving less margin for reflection, crosstalk, jitter, noise, temperature variation, and production tolerance.
Insertion Loss in Practical Terms
Insertion loss is normally expressed in decibels and varies with frequency. For a matched channel, a 6 dB loss corresponds to an output voltage amplitude of roughly half the input amplitude, while 20 dB corresponds to roughly one-tenth. These simple ratios help explain why an extra few decibels can matter in a high-speed link. The receiver does not see “a little less signal” in a linear sense; it sees a logarithmically smaller amplitude at the affected frequency, on top of whatever reflections, noise, and crosstalk are already present.
The number is meaningful only when the measurement conditions are defined. Cable length, conductor size, dielectric system, connector construction, termination geometry, temperature, bend state, fixtures, and de-embedding can all change the result. A statement such as “8 dB insertion loss” is therefore incomplete without a frequency and a defined assembly. A more useful statement is “approximately 8 dB at 14 GHz for this specified cable length and connector configuration under the stated test setup,” followed by the complete loss curve rather than a single point.
Insertion Loss | Approx. Voltage-Amplitude Ratio | Approx. Power Ratio | Practical Interpretation |
3 dB | 70.8% | 50% | Noticeable attenuation, often manageable in many systems depending on budget |
6 dB | 50.1% | 25% | Half the voltage amplitude remains at that frequency |
10 dB | 31.6% | 10% | Equalization and remaining channel budget become increasingly important |
20 dB | 10.0% | 1% | Strong attenuation; feasibility depends heavily on the complete SerDes/channel specification |
Loss, ISI, and the Eye Diagram
A digital pulse is made from many frequency components, and a copper channel does not attenuate them equally. High-frequency components are usually attenuated more strongly than low-frequency components, so the pulse arriving at the far end is broader and slower than the one that left the transmitter. Part of its energy extends into the next unit interval, while previous symbols can continue influencing the voltage at the current sampling point. This pattern-dependent overlap is intersymbol interference, or ISI.
The timing scale becomes unforgiving at high symbol rates. A 10 GBd signal has a 100 ps unit interval, while 56 GBd provides only about 17.86 ps and 112 GBd only about 8.93 ps. A discontinuity or delay spread that looks small on a mechanical drawing can consume a large fraction of the available eye width at those rates. The eye diagram makes this visible by overlaying many symbol transitions, revealing how channel loss, reflections, jitter, noise, and ISI reduce the vertical and horizontal margin available to the receiver.
Equalization Helps, but It Has Limits
Modern SerDes links are designed with equalization because a passive copper channel often behaves like a low-pass filter. Transmitter FFE can shape the outgoing waveform with weighted taps so that the received pulse is less affected by predictable pre-cursor and post-cursor ISI. Receiver CTLE relatively boosts higher-frequency content to compensate for channel roll-off, although that boost can also raise high-frequency noise. DFE uses previously detected symbols to estimate and cancel post-cursor ISI without simply amplifying the entire high-frequency spectrum.
Equalization works best when the channel response is reasonably smooth and predictable. It is much less effective against deep notches, strong resonances, excessive crosstalk, severe mode conversion, or poor signal-to-noise ratio. That is why “maximum acceptable insertion loss at Nyquist” cannot be answered with one universal dB value. The real limit belongs to the complete transmitter-channel-receiver system and its compliance budget. The cable’s job is to remain inside that budget with enough margin for manufacturing variation and operating conditions.
Is Nyquist Frequency the Only Frequency That Matters?
No. Nyquist frequency is a valuable reference point, but real digital signals contain useful spectral energy above it, and the complete channel response beyond Nyquist can still influence rise time, eye opening, jitter, and BER. Engineers therefore examine insertion loss, return loss, crosstalk, mode conversion, and resonances across the frequency range required by the interface rather than treating Nyquist as a hard upper-frequency boundary.
Energy Above the Nyquist Point
The alternating-symbol pattern establishes the Nyquist reference, but a digital waveform is not a single sine wave. Sharper transitions require additional high-frequency components. If those components are removed, the receiver may still see the basic high-low pattern, but the edges become rounded and timing uncertainty increases. This is why a channel can have an acceptable insertion-loss value exactly at Nyquist and still produce disappointing eye performance if the response immediately above that point falls rapidly or contains strong resonances.
The upper frequency that should be measured or modeled depends on the interface specification and the purpose of the analysis. Some compliance methods define explicit frequency ranges beyond Nyquist because higher-frequency behavior affects channel modeling, return loss, crosstalk, or transmitter edge fidelity. The correct practice is not to invent a generic “2x Nyquist” or “3x Nyquist” rule and apply it everywhere. The applicable protocol, channel specification, and measurement method should determine how far the frequency sweep needs to extend.
Rise Time and Electrical Length
Rise time offers another intuitive way to understand the importance of higher-frequency content. A commonly used first-order relationship for a single-pole response is bandwidth approximately equal to 0.35 divided by 10-to-90-percent rise time. Under that rough model, a 100 ps rise time corresponds to about 3.5 GHz, 50 ps to about 7 GHz, 20 ps to about 17.5 GHz, and 10 ps to about 35 GHz. These values are illustrative rather than protocol limits, but they show how faster edges demand broader analog behavior.
Fast edges also make short physical structures electrically significant. A connector launch, untwisted differential pair section, solder transition, shield breakout, or small geometry change may occupy only a few millimeters, yet the signal can “see” it as a discontinuity when the transition time becomes short enough. The important design question is therefore not simply whether the cable has the correct nominal impedance, but whether that impedance and return path remain controlled through the whole termination, connector, and routing geometry that the fast edge encounters.
Look at the Shape, Not Just One Point
Two channels with the same loss at Nyquist can produce different pulse responses because the shape of the frequency response matters. A smooth attenuation curve is generally easier for a linear equalizer to compensate than a response containing sharp dips and peaks. Return loss adds another dimension because energy reflected by an impedance discontinuity can arrive back at the receiver with pattern-dependent timing. Crosstalk can inject unrelated energy from neighboring lanes, while differential-to-common-mode conversion can reduce useful signal and increase EMI sensitivity.
This is also why bending, connector changes, or production modifications should not be judged solely by continuity. A tight bend can change local pair geometry or coax geometry and alter impedance and insertion loss. A connector alternative with the same pin count and mechanical fit can still have different high-frequency performance. A longer strip length may make assembly easier while increasing the uncontrolled transition region. At multi-gigabit rates, those mechanical and manufacturing details become part of the electrical channel, so the complete response deserves attention.
How Do Engineers Validate High-Speed Cable Links?
Engineers validate a high-speed cable link by first defining the interface, per-lane symbol rate, modulation, impedance, cable length, connector system, routing conditions, and applicable channel limits. They then combine appropriate frequency-domain and time-domain measurements, such as VNA S-parameters and TDR, with eye, BER, compliance, or system tests where required. The validated construction must finally be translated into controlled production dimensions, materials, processes, and inspection criteria.
Define the Link Before Testing It
Testing starts with requirements, not instruments. A request that says only “we need a 112G cable” leaves too many unanswered questions. The engineering team still needs to know whether 112G refers to one lane or aggregate throughput, which modulation is used, the actual electrical symbol rate, the intended cable length, the connector part numbers, target differential impedance, bend and routing conditions, and whether the customer has insertion-loss, return-loss, skew, crosstalk, eye, or BER limits. Without those inputs, a test result can be precise yet irrelevant to the real system.
A practical high-speed requirement package usually includes the interface or protocol, lane count, data and symbol rate, modulation, impedance target, minimum and maximum length, connector identities, PCB mating information, shield and grounding strategy, environmental conditions, and expected production configuration. If the link has a formal compliance specification, its limit masks and fixture requirements should be identified early. If it is proprietary, the transmitter and receiver budget or simulation targets become even more important because there is no generic “high-speed cable” specification that can replace the system design.
Use VNA and TDR for Different Questions
A vector network analyzer characterizes how energy is transmitted and reflected across frequency. Depending on the port configuration and project, engineers can use it to examine differential insertion loss, return loss, crosstalk, and mixed-mode behavior. The output is especially useful because the S-parameter data can be used in channel simulation and compared with frequency-domain masks. However, a VNA result is only as good as its calibration, reference-plane definition, fixtures, adapters, and de-embedding, so test setup information should travel with the measurement rather than being treated as an invisible detail.
TDR answers a different but complementary question: where is the discontinuity? By launching a fast edge and observing reflections versus time, TDR can reveal impedance changes near a connector launch, solder or crimp transition, splice, shield breakout, pair untwist, or other local structure. That physical localization is extremely useful during prototype debugging. A VNA might show poor return loss in a particular frequency band, while TDR helps the development team identify whether the dominant issue is in the cable body, termination, connector transition, or fixture interface.
Match Tests to the Failure Risk
A high-speed validation plan normally has several layers. Basic electrical inspection confirms pinout, continuity, open/short conditions, connector orientation, and dimensional requirements. Signal-integrity validation then checks the properties that determine channel behavior, which may include impedance, insertion loss, return loss, skew, crosstalk, and mode conversion. At the next level, system testing can include eye diagrams, receiver compliance, BER testing, or operation in the customer’s real hardware. Environmental or mechanical stress may be added when the installed cable will operate bent, flexed, vibrated, or across a wide temperature range.
Validation Item | Typical Output | What It Helps Confirm | Typical Use in a High-Speed Project |
Differential impedance / TDR | Ohms vs. time or distance | Geometry consistency and discontinuity location | Prototype, FAI, failure analysis |
Insertion loss | dB vs. frequency | Forward attenuation and channel roll-off | Prototype, channel qualification, sampling plan |
Return loss | dB vs. frequency | Reflection from impedance mismatch | Prototype and connector-transition validation |
Crosstalk | dB vs. frequency | Coupling between adjacent channels | Multi-pair or dense cable assemblies |
Skew | ps | Timing mismatch within or between pairs | Parallel differential channels and clock-sensitive links |
Eye diagram | mV, ps, UI, mask result | Combined voltage and timing margin | System/compliance validation where applicable |
BER / functional link test | Error rate or pass/fail | End-to-end data recovery | System validation and customer hardware correlation |
Control the Validated Geometry in Production
The final engineering challenge is repeatability. A carefully built prototype can pass because a skilled technician preserves pair twist, shield overlap, strip length, and conductor geometry with unusual care. Production introduces more operators, larger material lots, tooling wear, process variation, and possible substitutions. If the dimensions that control signal integrity are not transferred into drawings, work instructions, fixtures, and inspection criteria, the production cable can drift electrically even while every unit still passes continuity and looks acceptable from the outside.
Important controls may include strip length, exposed pair length, untwist length, shield termination position, drain-wire path, conductor spacing, crimp or solder geometry, overmold transition, connector orientation, cable length tolerance, and no-bend zones. Material substitutions also deserve review because two cables with the same nominal impedance can have different conductor, dielectric, shield, and attenuation characteristics. For custom high-speed programs, SINO-CONN can support the progression from requirement review and prototype validation into controlled pilot and repeat production, with the test depth matched to the actual interface and project risk rather than applying every laboratory test indiscriminately.
Nyquist frequency is useful because it turns a vague statement such as “this is a very fast link” into a concrete electrical reference. Once the symbol rate and Nyquist point are known, engineers can ask better questions about insertion loss, return loss, edge behavior, equalization, and measurement bandwidth. The strongest cable designs are not created by chasing one impressive dB number; they come from controlling the complete transmission path from conductor geometry through connector transition, validating it under the right conditions, and then preserving the approved construction through production. For teams developing a custom high-speed interconnect, that disciplined path is usually more valuable than any generic claim that a cable “supports” a particular headline speed.
Frequently Asked Questions
Is Nyquist frequency half the bit rate?
It is half the bit rate only when the physical signaling carries one bit per symbol and the quoted bit rate is the actual electrical lane rate. That is usually a reasonable simplified relationship for NRZ. PAM4 carries two bits per symbol, so its symbol rate is approximately half the raw bit rate and its Nyquist frequency is approximately one-quarter of that raw rate. Protocol coding, FEC, and lane architecture can change the exact numerical relationship, so the standard-defined symbol rate is the safest starting point.
What is the Nyquist frequency of a 112 Gb/s PAM4 link?
Using the simplified raw-rate relationship, 112 Gb/s PAM4 carries two bits per symbol, so the symbol rate is approximately 56 GBd and the Nyquist frequency approximately 28 GHz. Many real 112G-class interfaces use specific lane rates that are not exactly 56 GBd because of protocol architecture and FEC, so the final calculation should use the electrical symbol rate stated by the applicable specification rather than assuming the rounded marketing data rate is exact.
Does a cable need bandwidth only up to the Nyquist frequency?
No. Nyquist frequency is a useful reference, not a hard bandwidth cutoff. Fast digital edges contain useful energy above the Nyquist point, and channel behavior at higher frequencies can influence rise time, pulse shape, reflection, crosstalk, and equalization. The required measurement or model bandwidth should come from the relevant interface specification or channel methodology. Judging a high-speed cable only by one insertion-loss value at Nyquist can miss resonances, notches, or return-loss problems elsewhere in the important frequency range.
How much insertion loss at Nyquist is acceptable?
There is no universal dB limit that applies to every cable and every SerDes. Acceptable Nyquist loss depends on the transmitter output, pre-emphasis or FFE, receiver equalization, modulation, channel length, connector losses, reflection, crosstalk, noise, and the compliance budget of the interface. A smoothly lossy channel may be easier to equalize than a lower-loss channel containing a severe notch. The correct limit should therefore come from the system or protocol channel specification rather than a generic cable rule.
What tests are most useful for a high-speed cable assembly?
The most useful tests depend on the interface and failure risk. TDR is valuable for locating impedance discontinuities, while a VNA can characterize insertion loss, return loss, crosstalk, and other S-parameters across frequency. Eye diagrams and BER or functional tests add system-level evidence when required. Basic continuity remains necessary but cannot establish high-speed performance by itself. A sound validation plan starts from the actual lane rate, impedance, connector system, length, routing, and acceptance criteria, then selects the measurements that prove those requirements.
Related Keywords :Nyquist frequency, high-speed cable links, high-speed cable assembly, signal integrity, baud rate, symbol rate, bit rate, NRZ, PAM4, insertion loss, return loss, S-parameters, eye diagram, intersymbol interference, equalization, CTLE, FFE, DFE, TDR testing, VNA testing
With over 18 years of OEM/ODM cable assemblies industry experience, I would be happy to share with you the valuable knowledge related to cable assemblies products from the perspective of a leading supplier in China.
manufacturer catalogue
Get A Sample Now From Factory→
Get a quote quickly
Here, developing your OEM/ODM custom cable assemblies collection is no longer a challenge—it’s an excellent opportunity to bring your creative vision to life.
