Electronics Guide

Articles in This Category

The Instrument Is Part of the Measurement

An instrument does not observe a circuit from outside it. The connection perturbs the circuit, and the acquisition path filters and distorts what it captures. Understanding those effects quantitatively is what separates a measurement from a picture.

Bandwidth and Rise Time

An oscilloscope's bandwidth and its rise time are inversely related through a constant set by the shape of the instrument's frequency response. Instruments with a Gaussian roll-off, common below roughly one gigahertz, follow the familiar relationship in which bandwidth multiplied by rise time is approximately 0.35. Higher-bandwidth instruments use digital signal processing to flatten the passband and sharpen the cutoff, and for these the product typically falls between about 0.40 and 0.45. For a Gaussian-response system the displayed rise time is approximately the root-sum-square of the signal's true rise time and the instrument's own, so an oscilloscope whose rise time equals the signal's will report an edge roughly forty percent slower than reality. A common guideline is to select bandwidth at least three to five times the highest significant frequency content of the signal, which keeps the instrument's contribution small enough to ignore.

Probe Loading and Access

A probe becomes part of the circuit it measures. Tip capacitance loads the node, slowing edges and shifting resonances, while the inductance of a long ground lead resonates with that capacitance and adds ringing that does not exist in the unprobed circuit. Conventional passive probes carrying roughly ten picofarads at the tip are unsuitable for fast edges. Low-capacitance active and differential probes, with solder-in or browser tips and the shortest available ground return, are the practical answer. Where a probe cannot reach, observability must be designed in: dedicated test points, characterization coupons, spare connector footprints, and lanes routed to a test connector all cost far less during layout than they do after a failure appears.

Noise Floor, Resolution, and Repeatability

The instrument's own noise adds in root-sum-square fashion to the noise of the device under test, so a measured jitter or amplitude-noise figure always overstates the true one until the instrument's contribution is characterized and removed. Vertical resolution matters for the same reason; the highest-bandwidth real-time oscilloscopes now provide ten-bit converters rather than the traditional eight, which meaningfully lowers the quantization floor when small eye openings are being resolved. Repeatability deserves equal attention. Connector torque, cable flexure, thermal drift, and even the order in which ports are connected all move results, and a measurement that cannot be reproduced twice on the same fixture cannot support a design decision.

Matching the Instrument to the Question

No single instrument answers every signal integrity question. Each class trades bandwidth, dynamic range, acquisition time, and the kind of signal it can accept.

• Real-time oscilloscopes digitize a waveform in a single pass, which makes them the only choice for one-shot events, intermittent faults, and non-repeating data. The sample rate must be roughly two and a half times the specified bandwidth to avoid aliasing, and the fastest commercial instruments now reach about 110 GHz of bandwidth at sample rates near 256 gigasamples per second.
• Equivalent-time sampling oscilloscopes reconstruct a repetitive waveform from one sample per repetition. They cannot capture a single-shot event, but they achieve very high bandwidth with far lower noise and timebase jitter than a real-time instrument of comparable reach, and at lower cost per gigahertz. They remain the standard platform for time-domain reflectometry and for optical eye measurements.
• Vector network analyzers sweep a sinusoidal stimulus and measure magnitude and phase to produce S-parameters. Their wide dynamic range and calibrated accuracy make them the reference tool for characterizing passive interconnects, and their output feeds directly into channel simulation.
• Time-domain reflectometers and transmission measurements launch a fast step and display the returning reflections, yielding an impedance profile plotted against distance. State-of-the-art step generators produce edges of roughly five to fifteen picoseconds. Spatial resolution follows from that edge: two discontinuities separated by less than about half the distance the edge travels during its own rise time merge into one feature, which is why a dense via field appears as a single broad excursion rather than a series of distinct events.
• Bit-error-ratio testers transmit a known pattern and count the errors the receiver actually makes. Sweeping the sampling instant across the unit interval produces a bathtub curve whose extrapolated walls estimate the eye opening at error ratios far below what could be observed directly, and stressed-eye generation exercises a receiver against the impairments the standard says it must tolerate.
• Spectrum and phase-noise analyzers address the frequency-domain character of clocks and of coupled interference, quantifying reference-clock phase noise and locating narrowband aggressors such as a switching regulator's fundamental and its harmonics.

Time-domain reflectometry and vector network analysis measure the same interconnect and can be transformed into each other. The network analyzer generally offers better dynamic range and a more rigorous calibration, while the reflectometer offers an immediate spatial picture that shows an engineer where along the channel a problem lives. Serious characterization work usually employs both and requires them to agree.

Calibration, Fixtures, and De-Embedding

Calibration establishes where the measurement begins. Without it, a result describes the instrument, its cables, and the fixture just as much as it describes the device under test.

Moving the Reference Plane

Vector network analyzer calibration measures a set of known standards and solves for the systematic errors between the receivers and the desired reference plane. Short-open-load-thru calibration, built on a twelve-term error model for a two-port measurement, is the most common approach and depends on how accurately the calibration standards are characterized. Thru-reflect-line calibration instead relies on transmission-line standards and requires only that the launches be identical and that the thru and line segments share impedance, loss, and propagation constant while differing in length. That property makes it attractive on printed circuit boards and on wafer, where precision coaxial standards do not exist. Line-reflect-match variants relax the length constraints that limit the usable frequency span of a single thru-reflect-line standard.

Removing the Fixture

Calibration ends at a coaxial connector, but the device of interest usually sits behind a launch, a length of board trace, and possibly a connector. De-embedding is a second-tier calibration that removes that fixture mathematically. The most widely used method characterizes a two-times-thru coupon, a structure exactly twice the length of the fixture half to be removed, and derives the fixture model from it. The method is fast and requires no calibration kit, but it carries a firm requirement: the fixture in the coupon and the fixture in the measurement must have the same impedance and construction. When they differ, the extracted model is wrong in a characteristic way, producing non-causal S-parameters that show a response before the signal could physically have arrived. Automatic fixture removal techniques relax this constraint by extracting the fixture from the measurement itself.

IEEE Std 370-2020, IEEE Standard for Electrical Characterization of Printed Circuit Board and Related Interconnects at Frequencies up to 50 GHz, codifies this territory. It specifies how to design a test fixture, how to remove fixture and instrumentation effects, and which quality metrics must be computed and reported alongside the S-parameter data. Its practical contribution is a common vocabulary: a supplier and a customer exchanging a Touchstone file can now agree on what makes that file trustworthy instead of arguing about it.

Judging Whether the Data Can Be Trusted

A measurement file that opens without error is not the same as a measurement that is correct. Passive interconnects obey physical constraints, and checking measured data against those constraints catches a large fraction of fixture and de-embedding mistakes before they propagate into a simulation.

• Passivity requires that a passive structure never deliver more power than it receives. Measured data that shows gain at some frequency indicates a calibration or de-embedding error, and it will destabilize any time-domain simulation that uses it.
• Causality requires that no response precede its stimulus. Impedance mismatches between a de-embedding standard and the actual fixture are a frequent cause of violations, and the symptom appears as energy before time zero in the transformed impulse response.
• Reciprocity requires that transmission be identical in both directions through a passive network. An asymmetry between the forward and reverse paths points to an instrument or connection problem rather than to a property of the device.
• Consistency across domains and instruments is the strongest available check. When a network analyzer sweep and a reflectometry trace of the same coupon disagree, or when a measurement and a correlated model diverge, the discrepancy is itself a finding. It commonly exposes a wrong dielectric constant, an omitted via stub, an inaccurate connector model, or a fixture artifact that had been quietly corrupting the analysis.

Uncertainty deserves an explicit number rather than an implicit assumption. Residual calibration error, connector repeatability, noise floor, and drift each contribute, and their combined magnitude sets the smallest difference a measurement can honestly resolve. Reporting a one-decibel improvement in insertion loss from a setup whose repeatability is half a decibel is not a result; it is noise given a name.

From Waveforms to a Verdict

Characterization eventually has to produce a decision. Compliance testing converts acquired data into a pass or fail against the limits a standard defines, and the methods used have shifted substantially as data rates have risen.

The eye diagram remains the most legible summary of link health, overlaying many unit intervals to reveal the opening a receiver has to work with. Its limitation is that it displays only the errors that occurred during the acquisition. A bit-error-ratio tester measures the link as the receiver experiences it and, through bathtub curves in both the time and voltage dimensions, distinguishes a link that passes with comfortable margin from one that passes only because the test was short. Statistical confidence follows a simple rule of thumb: observing roughly three divided by the target error ratio in error-free bits establishes about ninety-five percent confidence that the target is met, so a target of one error in a trillion requires on the order of three trillion error-free bits.

Two developments complicate direct observation. First, at the highest rates the receiver's decision point sits behind an equalizer inside the package, where no probe can reach; transceivers answer this with on-die instrumentation such as internal eye monitors, error counters, and the receiver lane margining that PCI Express defines, which shifts the sampling point in time and voltage on a live link. Second, standards increasingly specify channels through computed figures of merit rather than through simple masks. Channel operating margin, introduced by IEEE 802.3bj and specified in Annex 93A, combines insertion loss, return loss, crosstalk, transmitter and receiver equalization, and package models into a single statistical margin figure computed from measured S-parameters. Effective return loss, adopted in later Ethernet projects, similarly replaces classical return-loss masks with a metric that weights reflections according to the harm they actually do to an equalized receiver. Under these methods the measurement is no longer the verdict; it is the input to the calculation that renders one.

About This Category

Together these topics span the complementary perspectives an engineer needs to characterize a high-speed channel end to end. Probing technique governs how faithfully a signal can be acquired at all. Time-domain and frequency-domain measurements describe the same interconnect from different vantage points, one showing where a problem lives along the channel and the other showing how it varies with frequency. Compliance testing translates the resulting data into a verdict against the governing standard. The common thread is skepticism applied systematically: know what the instrument contributes, calibrate it away, check the result against physics, and treat a number as trustworthy only after it has survived that scrutiny.

Related Topics