Technical guide | IEC 60958-3 | AES3
The S/PDIF coaxial signal is a square wave encoded in BMC (Biphase Mark Code) at 0.5 V P-P on 75 ohms. The cell frequency at 44.1 kHz is 5.645 MHz (128 cells per stereo frame, per the IEC 60958 structure). The cable behaves as a distributed RLGC network that filters, attenuates, distorts and injects noise into the signal before the receiver decodes it.
The simulator applies five physical phenomena in cascade: low-pass filtering, skin effect, attenuation, reflections, EMI noise -- plus ISI jitter and triple transit.
What happens physically: skin effect and dielectric losses in the cable form a distributed low-pass filter. The longer the cable, the lower the cutoff frequency, and the more the edges of the BMC square wave become rounded.
The model: the effective cutoff frequency follows a square-root law over length. The bandwidth at 1 m (manufacturer specification) decreases according to 1/sqrt(1 + L/8). At 8 m, bandwidth is reduced by sqrt(2), i.e., 3 dB extra. At 24 m, it is halved. This model is calibrated against typical attenuation curves for RG-59/RG-6 coaxial cables.
Concrete impact: a Belden 1694A (BW 250 MHz at 1 m) stays at 233 MHz over 1.5 m -- edges remain nearly square. A generic RCA cable (BW 25 MHz at 1 m) drops to 17 MHz over 10 m -- edges become sinusoidal and the decoder has difficulty locating transitions.
| Quality | Effective BW | Signal effect |
|---|---|---|
| Excellent | > 100 MHz | Sharp edges |
| Good | 30 - 100 MHz | Slight rounding |
| Poor | 10 - 30 MHz | Rounded edges, increased jitter |
| Bad | 5 - 10 MHz | Severely degraded |
What happens physically: at high frequency, current concentrates in a thin surface layer of the conductor (the skin depth). For copper at 5.6 MHz, this depth is approximately 28 µm. At 56 MHz (10th harmonic), it drops to 8.8 µm. The effective conductor resistance therefore increases with the square root of frequency.
Direct consequence: the high-order harmonics of the BMC square wave (3rd, 5th, 7th) are disproportionately attenuated relative to the fundamental. The 3rd harmonic (16.9 MHz) is attenuated 1.73x more than the fundamental. The 7th (39.5 MHz), 2.65x more. These are precisely the harmonics that give the signal its square shape and sharp edges -- rounding them increases ISI.
When to enable it: skin effect is negligible on short cables (< 3 m) with good copper section. It becomes significant on thin cables (AWG 24-26, e.g. Mogami 2964) or long distances.
What happens physically: resistive losses in the conductor (proportional to sqrt(f)) and dielectric losses in the insulator (proportional to f) reduce signal amplitude as a function of length and frequency.
The model: the simulator interpolates attenuation linearly between manufacturer measurements at 5 MHz and 10 MHz, then scales to the actual cable length. Linear interpolation introduces less than 5% error versus the exact sqrt(f) model over this narrow interval.
The critical threshold: the S/PDIF receiver stops decoding reliably below approximately 0.2 V P-P (nominal signal is 0.5 V). That is 8 dB of maximum attenuation before drop-out. A Belden 1694A over 2 m loses less than 0.05 dB. This is not the limiting factor on short runs.
For AES/EBU: signal level is 2 to 7 V P-P versus 0.5 V for S/PDIF. The simulator applies a 20 dB bonus on the attenuation margin (ratio 20 log10(5/0.5)), which explains why AES/EBU cables like the Belden 1800F easily handle tens of meters despite higher per-meter attenuation.
What happens physically: when the cable impedance differs from the load impedance (75 ohms for S/PDIF), a fraction of the signal is reflected at each end. This fraction is quantified by the reflection coefficient Gamma = (Z_cable - Z_ref) / (Z_cable + Z_ref). A cable at 45 ohms gives Gamma = -0.25, meaning 25% of signal amplitude reflected on the first bounce.
What this produces concretely: reflections arrive with a delay equal to the round-trip propagation time (2L / velocity), and superimpose on the useful signal. Each subsequent bounce has amplitude multiplied by Gamma-squared (two reflections per round trip). The simulator models up to 5 bounces. The most visible effect is the appearance of time-delayed "ghosts" on the waveform, and an eye diagram showing multiple traces.
A properly matched 75-ohm cable produces Gamma = 0: no reflections, at any frequency, at any length. This is the entire rationale for the 75-ohm impedance requirement in IEC 60958-3.
| Quality | Gamma | Effect | ||
|---|---|---|---|---|
| Excellent | < 0.01 | Negligible reflections | ||
| Good | 0.01 - 0.05 | Very minor | ||
| Poor | 0.05 - 0.15 | Echoes visible on eye diagram | ||
| Bad | > 0.15 | Severe distortion |
What happens physically: the cable acts as a receiving antenna for ambient electromagnetic fields. The shield attenuates them by its shielding effectiveness (SE). In real environments, the incident EMI field is more intense than in the lab: the effective SNR is therefore SE minus the environment penalty, which grows with cable length (larger antenna area).
The three environments:
Concrete impact: an unshielded RCA cable (25 dB SE) in a living room hits the modelling floor at 20 dB SNR -- noise alone can cause decoding errors. A Belden 1694A (90 dB SE) in the same environment stays at 85 dB SNR: entirely harmless.
What happens physically: a bandwidth-limited cable causes the energy of one symbol to "spill over" into adjacent symbols. This spillover shifts the instants of threshold crossing relative to the ideal time grid. This is the dominant degradation mechanism at long distances, and the direct cause of the cliff effect in digital links.
The model: ISI jitter is proportional to sqrt(L) -- statistical accumulation of perturbations along the length -- and inversely proportional to effective bandwidth. The coefficient K is 20 ns for S/PDIF coaxial and 7 ns for AES/EBU (differential link reduces effective ISI by a factor of approximately 3 through common-mode rejection and greater dV/dt slope at the decision threshold). These coefficients are calibrated against technical literature (Dunn 1992, AES-12id-2020) and manufacturer specifications (Canare DA206 rated to 300 m).
The cliff effect: at short distance, ISI jitter is well below the half-cell period (88 ns at 44.1 kHz): CER = 0%. At the critical distance, ISI jitter becomes comparable to that half-period: transitions spill into the adjacent cell and CER rises sharply. For the Belden 1694A, this transition occurs between 200 and 250 m.
Transmitter intrinsic jitter is not included by default: it is a property of the source device, not of the cable, and it would add identically to both compared cables. To simulate the complete chain, the "Transmitter jitter" and "Edge asymmetry (DCD)" options can be enabled in the simulation options (off by default); the AES-12id standard specifies a typical transmitter jitter of 2 ns RMS.
| Quality | J ISI RMS | Distance (Belden 1694A) |
|---|---|---|
| Negligible | < 1 ns | < 50 m |
| Moderate | 1 - 5 ns | 50 - 150 m |
| Critical | 5 - 30 ns | 150 - 230 m |
| Failure | > 30 ns | > 230 m |
What happens physically: when the cable is mismatched, the signal makes multiple round trips between the ends. The triple transit -- the signal that bounces off the load, reflects back from the source, then arrives again at the load -- creates an echo delayed by 3L/velocity with amplitude |Gamma_load x Gamma_source|.
Why this differs from simple reflections: this echo superimposes on the useful signal at the comparator of the receiver. At each BMC transition, the echo has not yet followed -- it arrives with its delay. The received signal therefore shows a transient "step" before settling: the "horn" visible on the waveform. Its height is proportional to the product of reflection coefficients at both ends; its duration is the triple transit delay.
The resulting deterministic jitter: the horn shifts the threshold-crossing instant. This jitter is deterministic -- it is reproducible and of fixed value for a given cable length. The PLL filters it according to its frequency: a wide-band PLL (CS8412) passes it almost untouched, a narrow PLL (WM8805) or an ASRC rejects it. The simulator combines it in quadrature with the filtered cable jitter to establish the audibility verdict.
Null and critical lengths: when the triple transit delay is an integer multiple of the cell period, the echo arrives exactly in phase with a transition -- zero TT jitter. The shift is maximum a quarter step away from the null lengths (the midpoint between two nulls is itself a null). The step between one null length and the next is typically 10 to 14 m depending on propagation velocity. For domestic cables of 1 to 3 m, no null length exists within this range: TT jitter is always present. The only solution is impedance matching (75 ohms) which zeroes the reflection coefficients.
What happens physically: nearly all S/PDIF outputs go through a transformer or capacitor that blocks DC. This coupling acts as a high-pass filter. The BMC code is balanced, but the frame preambles deliberately violate the coding: their low-frequency content makes the signal baseline drift (baseline wander). Since the receiver compares against a fixed threshold, this drift shifts the detection instants: it is data-correlated deterministic jitter, identified as early as 1992 by Dunn and Hawksford as a structural flaw of the interface.
In the simulator: the "AC coupling (transformer)" checkbox enables a high-pass filter whose corner you choose. A properly sized transformer (300 Hz or 1 kHz corner) only creates a few picoseconds; an undersized capacitive coupling (10 kHz) adds about 200 ps of jitter even on a perfect 2 m cable. The effect does not depend on the cable: it applies to both branches of the comparison.
The interface allows defining a cable by its six physical parameters directly: impedance (ohms), attenuation at 5 MHz and 10 MHz (dB/100 m), bandwidth at 1 m (MHz), propagation velocity (% of c), and shielding effectiveness (dB). Useful for testing a cable from a datasheet, or for isolating the influence of a single parameter.
What it is: the CER is the proportion of BMC time cells corrupted during transmission. It is the fundamental metric: S/PDIF has no error correction (unlike professional AES/EBU which has parity detection). A single error on a most-significant bit of a 16-bit sample produces an audible click immediately.
How it is calculated: the analyser aligns the degraded signal to the reference via cross-correlation (256-cell window over 768 reference cells), then compares cell by cell. The CER is the number of mismatched cells divided by the total cells compared.
What it means: CER = 0% means bit-perfect transmission. With 44,100 Hz x 128 = 5.6 million cells per second, a CER of 0.01% already represents 565 errors per second -- frequent audible clicks and unusable audio. Between two cables that both yield CER = 0%, there is no difference in digital transmission. The bits are identical.
What it is: jitter measures the deviation between actual signal transition instants and the ideal time grid (integer multiples of the cell period). The analyser detects each threshold crossing by linear interpolation between two consecutive samples, computes the interval between successive transitions, and derives the deviation relative to the nearest cell-period multiple.
RMS vs. Peak-to-Peak: jitter RMS (root of the mean square of deviations) is the reference statistical metric -- it captures the weight of the full distribution. Jitter P-P (total range) captures extreme events that may trigger isolated errors even when RMS is low.
The role of the PLL in the chain: the jitter measured here is interface jitter (TIE, Time Interval Error). It is not the jitter the DAC experiences. The PLL of the receiver filters jitter as a low-pass filter: only a fraction of cable jitter reaches the converter. The simulator models this filtering through the Butterworth equivalent noise bandwidth.
| Receiver | PLL cutoff | Fraction of cable jitter reaching the DAC |
|---|---|---|
| CS8412 (1990) | 25 kHz | ~7% |
| VCXO (2000) | 200 Hz | ~0.6% |
| WM8805 (2010) | 90 Hz | ~0.4% |
| ASRC (modern) | 1 Hz | < 0.1% -- floor 20 ps |
| Word Clock | 3 Hz | < 0.1% -- floor 5 ps |
Audibility thresholds: the two reference studies yield very different results because their protocols differ fundamentally. Benjamin & Gannon (AES 4826, 1998) tested sinusoidal jitter (artificial, periodic): minimum detected threshold approximately 10 ns RMS with 17 kHz tone, 20 ns RMS on music. Ashihara et al. (2005) tested random jitter (closer to real cable conditions) with 23 professional listeners in double-blind controlled conditions on music: at 250 ns, none detected anything; at 500 ns, 6 out of 23 perceived something. These thresholds are far above the few nanoseconds produced by a well-matched short coaxial cable.
SNR degradation from jitter: the Dunn (1997) formula gives SNR = -20 log10(2 pi x f_audio x J_RMS). At 10 ns RMS at the DAC and 20 kHz, this yields 64 dB -- only perceptible via a CS8412. With an ASRC, 10 ns at input becomes < 0.1 ns at the DAC: 104 dB of jitter SNR. The interface noise entirely disappears into every other source of error in the signal chain.
What it is: the simulator measures separately the average voltage of samples at the high level and at the low level, plus the peak-to-peak amplitude. It separates the two populations around the decision threshold (midpoint of the excursion).
Why it matters: attenuation reduces V_PP by compressing both levels toward the threshold. Below 0.2 V P-P (the IEC 60958-3 200 mV P-P threshold), the receiver may not decode reliably. Asymmetry between high and low voltages indicates non-linear distortion or a DC offset.
What it is: the standard deviation of samples around their mean level, computed separately for the high and low populations. High noise reduces the margin between the two logic levels and increases the probability of a decision error.
Link to SNR: SNR_signal ~ 20 log10(V_PP / (2 x max_noise_RMS)). Strong asymmetry between high and low noise may indicate that reflections preferentially affect one type of transition.
What it is: each IEC 60958 sub-frame of 32 cells contains a parity bit (bit 31). The standard defines that the parity of bits 4 to 31 must be even. The analyser checks each sub-frame and counts violations.
The limitation: parity can only detect errors affecting an odd number of bits per sub-frame. Errors on an even number of bits pass undetected. In practice, if parity errors are detected, the signal is already significantly degraded.
What it is: the eye diagram superimposes all signal segments lasting 2 cell periods (2 UI). The analyser splits the signal into windows of 2 x SPC samples (SPC = samples per cell) and overlays them as a 2D histogram (200 x 120 bins) displayed as a heatmap.
What it reveals: the vertical opening at the center indicates the voltage margin (distance between high and low levels, reduced by attenuation and noise). The horizontal opening indicates the timing margin (width of the stable zone, reduced by jitter). Multiple traces signal reflections. Trace thickening indicates EMI noise.
| Appearance | Cause |
|---|---|
| Wide open | Clean signal |
| Narrowed vertically | Attenuation or high noise |
| Narrowed horizontally | Jitter or insufficient bandwidth |
| Closed | Cumulative degradation, decoding compromised |
| Multiple traces | Reflections (impedance mismatch) |
| Thick traces | EMI noise |
What it is: three synchronized panels -- global view of the full signal, automatic zoom on the region of greatest divergence between the two cables, and the difference signal (cable - reference). The zoom is centered on the maximum of the smoothed absolute divergence (moving average), systematically pointing at the most analytically relevant zone.
The analyser generates a text interpretation below each graph. For the eye diagram, the estimated vertical opening is V_PP - 6 x max_noise_RMS (3 sigma on each side). An opening > 0.35 V with noise < 5 mV corresponds to a wide-open eye. Degradation causes (reflections, noise, attenuation) are identified automatically by thresholds on Gamma, sigma, and V_PP.
Connectors affect the signal differently depending on their type. The simulator does not model them, but manufacturer specifications (IEC 61169-8, IEC 61076-2-103) allow quick impact assessment.
RCA (phono) : not RF standardized, impedance varies 40-70 ohms. Each pair creates a localized discontinuity. On short cable (< 1 m), echo damps out ; beyond 3-5 m, echoes accumulate and become visible on the eye diagram.
BNC 75 ohm : standardized (IEC 61169-8), impedance 75 ±5%, VSWR < 1.05 at 10 MHz. Negligible reflection (Gamma < 0.025). S/PDIF impact imperceptible at all practical lengths.
XLR (AES/EBU) : balanced connector, impedance non-critical. Length < 30 mm, negligible at 5.6 MHz. Minimal impact.
Toslink (optical) : no electrical reflection, optical jitter negligible. Only optical attenuation (< 2 dB plastic, < 0.5 dB glass).
Operational summary : BNC and Toslink best choices. RCA short (< 2 m) acceptable. RCA long (> 10 m) risks parasitic traces if connector oxidizes.
Each cable parameter affects each metric differently. High sensitivity = small parameter change strongly impacts the metric.
Length : CER and SNR dominant (√L growth of ISI, larger antenna).
Attenuation : CER dominant (signal approaches 200 mV decoding threshold).
Impedance (Gamma) : impact localized to connectors, low on well-matched 75 ohm cable.
Bandwidth : ISI jitter critical (inversely proportional to BW).
Shielding : SNR and noise RMS (direct relation to shielding effectiveness in dB).
Velocity : no direct impact on signal integrity ; affects only propagation delay and reflection echoes.
This matrix helps identify which parameter dominates which metric — essential for choosing between, e.g., a shorter cable vs. a better-shielded cable.
The synthetic verdict combines CER and jitter RMS in a fixed hierarchy:
| Verdict | Condition | Meaning |
|---|---|---|
| Signal intact | CER = 0 AND J_RMS < 0.5 ns | Perfect transmission |
| Light degradation | CER < 0.001% AND J_RMS < 2 ns | Inaudible, no impact |
| Degraded signal | CER < 1% | Clicks possible |
| Corrupted signal | CER >= 1% | Transmission unusable |
The verdict is a quick summary. For complete evaluation, examine all individual metrics and the eye diagram, especially in borderline cases where CER is zero but jitter or reflections degrade signal integrity.
| Standard | Access |
|---|---|
| IEC 60958-1: frame structure, BMC | EBU Tech 3250-E |
| IEC 60958-3: S/PDIF coaxial | EBU Tech 3250-E |
| AES3-2009: professional interface | aes.org/publications/standards-store/?id=13 (paid — free equivalent: EBU Tech 3250-E) |