Understanding S/PDIF cables: optical vs coaxial guide

Simplified technical guide | IEC 60958 | AES3

This guide explains, without formulas, what the simulator computes. The models and equations are in the technical documentation, and the simulator compares two real cables at the length you choose.

Spdif, S/PDIF, Toslink: what are they?

S/PDIF (often written Spdif) is the consumer digital audio link defined by the IEC 60958 standard. It comes in two forms: a 75 ohm coaxial cable with RCA or BNC plugs, and an optical fibre called Toslink. The "SPDIF output" of a TV, a set-top box or a player sends the audio samples as they are to a DAC or an amplifier. Its professional version is AES/EBU (AES3): 110 ohms, balanced, XLR plug.

Can a digital cable change the sound?

In a normal setup, no. In the simulator, a 10 m Belden 1694A (75 ohm coaxial) adds about 0.05 ns of peak-to-peak jitter, without picked-up noise. The lowest published mean audibility threshold, 10 ns RMS, was measured in the most demanding case: a full-scale 20 kHz sine with sinusoidal jitter at 17 kHz (Benjamin and Gannon, 1998). On chosen music excerpts, thresholds start at a few tens of nanoseconds, and none of the 23 listeners in Ashihara et al. (2005) detected 250 ns of random jitter with music. The exceptions are electrical: a ground loop, a cable that is not 75 ohms over several tens of metres, excessive length. The studies are summarised on the Blind tests page.

Optical or coaxial: the short answer

Both carry exactly the same bits. Choose optical if you hear hum (it breaks the ground loop), coaxial for long runs or 192 kHz.

For coaxial, the added jitter stays hundreds of times below published thresholds. For optical, the modules allow up to ±15 ns of distortion (not modelled), which the receiver PLL partly attenuates.

1. The cable model

What is being simulated?

The coaxial S/PDIF signal is a square wave coded in BMC (Biphase Mark Code), at 0.5 V peak-to-peak on 75 ohms. Every bit starts with a transition; a 1 bit adds one in the middle. The signal therefore switches between two fundamentals: 2.82 MHz for a run of 1s and 1.41 MHz for a run of 0s (at 44.1 kHz). A cell, half a bit, lasts 177.2 ns.

The simulator sends this signal through a model of the cable, in this order: line losses (which depend on frequency), reflections if the cable impedance is wrong, then picked-up noise. Jitter is not added by a formula: it is measured on the signal leaving the cable, as you would with an oscilloscope.

An S/PDIF link transmits continuously. The simulator therefore sends the signal through the cable twice in a row and only analyzes the second pass: the cable is then in steady state, as in a real system, without the cold-start effect that would distort the first edges of a long cable.

The simulator also compares cables that are not S/PDIF cables: a cheap audio RCA lead of unspecified impedance (about 42 ohms estimated) and a generic coaxial. Their parameters are estimated, since no manufacturer datasheet exists.

1.1 Bandwidth limitation

What happens physically: cable losses grow with frequency. The high-frequency components of the square wave are attenuated more than the fundamental, and the edges round off.

The model: a single loss law, α(f) = a_s·√f + a_d·f: one term in the square root of frequency (conductor skin effect) and one term proportional to frequency (insulation losses). Both coefficients are fitted to the losses at 5 and 10 MHz. When the manufacturer's datasheet does not give them, they are derived from its other measurement points, or estimated. For the Canare DA206 and DA202 (AES3), a single point is published, at 3 MHz: the losses at 5 and 10 MHz are extrapolated from it as the square root of frequency, and remain uncertain. For the Belden 1800F, a twisted pair that does not follow this law over the whole range, the fit uses the full manufacturer table (16 points) between 1 and 12.3 MHz: the error stays below 1% at the two BMC fundamentals. The term proportional to frequency then serves as an empirical fit: it does not represent insulation losses.

The -3 dB bandwidth is not a setting: it follows from these losses. With skin effect, it varies as the inverse square of the length. In the model, fitted on the losses at 5 and 10 MHz, a Belden 1694A has a bandwidth of about 1.6 GHz over 10 m, 65 MHz over 50 m and 16 MHz over 100 m; the full table of the Belden datasheet gives about 2 GHz, 85 MHz and 17 to 18 MHz. Treating it as a separate filter would count the same phenomenon twice.

1.1b Skin effect

What happens physically: at high frequency, current concentrates in a thin surface layer of the conductor. For copper, this layer is about 39 µm thick at 2.8 MHz and 12 µm at 28 MHz. The conductor resistance grows as the square root of frequency.

Consequence: the harmonics of the square wave lose more than the fundamental. The 3rd harmonic (8.5 MHz) loses 1.7 times as many decibels as the 2.8 MHz fundamental, the 7th (19.8 MHz) 2.6 times as many. Edges spread out, and the shape of each edge depends on the bits before it: this is intersymbol interference (ISI, section 1.5).

The "Skin effect" checkbox: checked by default, it applies this full model. Unchecked, the cable applies the same loss at all frequencies. This simplification is useful for comparison: amplitude drops, but edges hardly spread and jitter largely disappears.

1.2 Attenuation

What happens physically: the signal loses amplitude along the cable, in proportion to length.

The model: the displayed attenuation is the loss at 2.82 MHz (the highest BMC fundamental at 44.1 kHz; it follows the sample rate). A Belden 1694A loses 1.25 dB per 100 m at this frequency, and less than 0.03 dB over 2 m.

The threshold: the standard does not set a minimum amplitude but an eye mask: at the receiver input, the eye must stay open by at least 200 mV over half a cell, around the decision threshold (section 2.11). Starting from 0.5 V, it takes about 8 dB of loss to fall to 200 mV. Over a few meters, attenuation is never the limiting factor.

For AES/EBU: the transmitter is simulated at 5 V peak-to-peak (the standard allows 2 to 7 V), ten times the S/PDIF level. The cable attenuation stays the same, uncorrected; the starting level is what gives 20 dB of extra margin against noise and against the 200 mV mask. At the minimum allowed by the standard (2 V), this margin is only 12 dB.

1.3 Impedance mismatch reflections

What happens physically: the transmitter and the receiver are at 75 ohms (110 ohms for AES/EBU). If the cable has a different impedance, part of the signal is reflected at each end. An edge travels out, is sent back by the receiver, then by the transmitter, and comes back: this is an echo.

What it produces: the echo arrives 2L/v after the main edge (L the length, v the speed in the cable). Its amplitude is the product of the two reflections, i.e. the square of the reflection coefficient: 4% for a 50 ohm line, 8% for a cheap 42 ohm RCA cable. The simulator computes the exact response, all bounces included.

A cable at the right impedance, between a 75 ohm transmitter and receiver (110 ohms for AES/EBU), produces no echo, whatever its length. This is why S/PDIF uses 75 ohms and AES/EBU 110 ohms.

1.4 EMI noise and electromagnetic environment

What happens physically: the cable picks up ambient electromagnetic fields, and its shield attenuates them. The longer it is, the more it picks up.

The three environments:

• Pro studio (0 dB penalty): linear power supplies, no dimmers.
• Home Hi-Fi (10 dB): Wi-Fi, switching chargers, LEDs.
• Industrial / stage (25 dB): drives, motors, parallel mains cables.

The model: the noise level is set relative to the transmitted level. It does not drop when the cable attenuates the signal: on a long cable, the signal-to-noise ratio at the far end therefore degrades twice. This noise model is an order-of-magnitude heuristic, not a measurement: it is meant to compare well and poorly shielded cables, not to predict an exact level.

What it changes: a Belden 1694A (shielding estimated at 90 dB) is not affected by the environment in this model: shielding of 72 dB or more makes picked-up noise ineffective there, up to 200 m (the application's maximum length) and even in an industrial setting. Datasheets do not give a shielding effectiveness in decibels: the cable values are estimates, based on their construction (foil, braid, spiral). A poorly shielded RCA cable picks up clearly more: in a living room, its signal-to-noise ratio goes from 61 dB at 1.5 m to 43 dB at 10 m.

1.5 ISI jitter (Intersymbol Interference)

What happens physically: since edges spread out, the level reached before a transition depends on the previous bits, and so does the time at which the signal crosses the threshold. Transitions no longer fall exactly on the clock grid: this is ISI jitter, the main source of jitter on long cables. Chris Dunn and Malcolm Hawksford (1992) showed, on a band-limited link, that this jitter depends on the data and stays correlated with the audio signal.

The model: this jitter is not computed by a formula. It appears in the signal filtered by the cable and is measured afterwards. For a Belden 1694A, peak-to-peak jitter is about 0.05 ns at 10 m, 0.76 ns at 100 m and 4.7 ns at 300 m (without noise). Half a cell is about 89 ns: even at 300 m, the eye is far from closing.

Transmitter edges: the transmitted signal has 19 ns edges (10 to 90%) at 44.1 kHz, within the 15 to 20 ns that Julian Dunn measured on a typical AES3 output. They follow the cell duration (8.9 ns at 96 kHz), without going below 5 ns (5.2 ns at 176.4 and 192 kHz). A slower edge makes the threshold crossing time more sensitive to ISI, echo and noise.

The cliff effect: a digital link works without errors as long as the eye stays open, then errors come quickly. The simulator locates this limit with the AES3 eye mask (section 2.11).

Transmitter jitter is not included by default: it is a property of the device, not of the cable, and it would apply equally to both compared cables. The "Transmitter jitter" and "Edge asymmetry (DCD)" options let you add it. The default field value (2 ns) is a working order of magnitude, not a value from the standard.

1.5b Triple Transit

What happens physically: this is the echo of section 1.3, seen from the jitter angle. The direct edge reaches the receiver after L/v; the echo, after an extra round trip, arrives after 3L/v, that is 2L/v after the edge.

Why a short cable has little echo jitter: the echo shifts the time at which the signal crosses the threshold, by an amount that depends on the signal level a little before the edge. If the echo comes back in less than one cell, it falls on the flat level preceding the edge, of opposite sign to the edge. On a lossless line, all edges are then shifted by the same amount: the echo only delays them, and jitter stays negligible (0.002 ns over 10 m, for a 4% echo). On a lossy line, the level before the edge depends on the previous bits, and the echo copies this ISI: 0.016 ns with the losses of a Belden 1694A (same 4% echo, 10 m), about 0.57 ns peak-to-peak at 10 m and 0.12 ns at 5 m for the 42 ohm RCA cable. Later bounces also leave a small residue. This holds up to a length L0 = v × cell / 2: 21.8 m for a cable at 82% of the speed of light, 14.6 m for a cable at 55%, at 44.1 kHz. At 96 and 192 kHz the cell is shorter and so is L0 (10 m and 5 m at 82%).

Beyond L0: the level before the edge depends on the data, and the echo adds data-dependent jitter. It is about twice the echo amplitude divided by the edge slope: 0.74 ns for a 4% echo with this formula, 0.78 ns with the full calculation.

In the application: the triple transit calculator isolates this effect (length L0, echo jitter versus length). It simulates the same line, with its losses, mismatched and then matched, and compares the two edge by edge, all bounces included. It is not added to the verdict, because it is already in the simulated signal and therefore in the measured jitter. The fix is still a cable at the right impedance.

1.5c AC coupling (transformer) and baseline drift

What happens physically: most S/PDIF outputs go through a transformer or a capacitor that blocks DC. This coupling acts as a high-pass filter. Transformers designed for digital audio cut off between about 2 and 20 kHz: 2.4 kHz on 75 ohms for the Pulse PE-65612 that Cirrus Logic recommends, 7 to 20 kHz for the Scientific Conversion range (2015 catalogue). Jon Paul, of that manufacturer, recommends a cutoff between 5 and 20 kHz (AES, 1998). A 100 nF capacitor between 75 and 75 ohms cuts off around 10 kHz.

The BMC code has no DC component, and neither do the frame preambles. But a preamble contains three-cell flat runs, whereas the rest of the code never has more than two. At every subframe, the high-pass therefore makes the signal baseline swing a little (baseline wander), and another part of this offset depends on the data. Since the receiver compares with a fixed threshold, this offset shifts detection times. The cause lies in the interface itself, not in the cable.

In the simulator: the "AC coupling (transformer)" box enables a first-order high-pass with a cutoff you choose: 1, 2.5, 5, 10 (default) or 20 kHz. On a 2 m Belden 1694A, a 10 kHz cutoff adds about 50 ps RMS (250 ps peak-to-peak) of jitter, and a 20 kHz cutoff roughly twice that. This is the same order of magnitude as the ISI of a 100 m 1694A (0.2 ns RMS), and very far below audibility thresholds. At 1 kHz, the effect does not exceed a few picoseconds. It does not depend on the cable: it applies to both branches of the comparison.

1.6 Custom cable

The interface lets you define a cable by its parameters: attenuation at 5 and 10 MHz (in dB/100 m or dB/100 ft), impedance, propagation velocity and shielding. Bandwidth is not a parameter, since it follows from the losses. Useful to test a cable whose datasheet you have, or to isolate the effect of a single parameter.

1.7 Optical links

An optical link (TOSLINK, glass fiber) has no impedance and no picked-up noise, and the receiver module regenerates a clean signal. The simulator only models the spreading caused by the fiber (modal dispersion). A TOSLINK plastic fiber offers at least 40 MHz over 100 m under restricted launch (IEC 60793-2-40, category A4a.2), far more than needed over a few metres. This is therefore not the limiting factor: the usable length is set by the optical budget of the modules (enough light must reach the receiver without saturating it). The reference audio pair, Toshiba TOTX147/TORX147, is rated for 0.2 to 5 m; beyond that, the simulator flags the result as outside the specified range. For glass fiber, the limit also comes from the transceivers, which are not modeled. These modules add their own jitter: up to ±15 ns of pulse width distortion for the Toshiba pair. This jitter is not modeled; the DCD option can represent it. The same pair accepts 15 Mb/s NRZ, i.e. 7.5 Mb/s in biphase: a 96 kHz stream (6.1 Mb/s) passes, a 192 kHz stream (12.3 Mb/s) does not.

2. Analysis metrics

2.1 Cell Error Rate (CER)

What it is: the fraction of BMC cells read differently from what was sent. Neither S/PDIF nor AES/EBU corrects errors: each subframe carries a parity bit that detects some of them. The professional (AES/EBU) channel status adds a CRC check; the consumer channel status has none. A detected error is usually concealed by the receiver, which repeats the previous sample, sets it to zero or lets the converter interpolate it. An undetected error on a most significant bit produces a click.

How it is computed: the analyzer aligns the received signal on the reference by correlation, then compares cell by cell, in blocks of one subframe. If the decoder gains or loses a cell, the alignment is readjusted and this slip is counted separately.

What it means: CER = 0% means all bits arrived intact. At 44.1 kHz, 5.6 million cells go through every second: a CER of 0.01% already means 565 errors per second. Two cables at CER = 0% deliver exactly the same bits. Without an ASRC or an external clock, however, the receiver derives its clock from the received signal: a jitter difference can remain, very small with matched cables (section 2.2).

The application reads the CER on this scale:

2.2 Jitter RMS and Peak-to-Peak

What it is: jitter measures the deviation between the actual time of each transition and an ideal clock. The analyzer finds each threshold crossing, fits the best regular clock over all transitions, then measures the deviation of each transition from that clock (TIE, *time interval error*). This is the quantity a receiver PLL filters.

RMS and peak-to-peak: RMS summarizes the whole distribution; peak-to-peak gives the gap between the two extreme transitions, the one that matters for the eye opening.

The role of the PLL: this is interface jitter, not converter jitter. The receiver PLL tracks slow jitter and rejects fast jitter. The simulator filters cable jitter according to its actual frequency content, then adds the intrinsic jitter of the receiver clock (its floor).

An ASRC attenuates interface jitter very strongly but does not remove it: the residue becomes a small amplitude error. With a Word Clock, the converter clock comes from elsewhere and cable jitter does not reach it.

Example: a 100 m Belden 1694A produces about 0.2 ns RMS of interface jitter. After a CS8412, about 205 ps remain, most of it from the receiver floor (200 ps); with the other receivers, only their floor remains.

Audibility thresholds: Benjamin and Gannon (Dolby, AES, 1998) tested sinusoidal jitter. On a full-scale 20 kHz sine with jitter at 17 kHz, the mean threshold is 10 ns RMS: this is the most demanding case. On a 4 kHz sine, it rises to 100 ns. On music, the vast majority of programs do not make jitter audible; on excerpts chosen to be revealing, individual thresholds range from about 20 to 370 ns. Ashihara and colleagues (2005) tested random jitter on music: nobody detects it at 250 ns, 6 listeners out of 23 at 500 ns. A matched 10 m coaxial cable (0.05 ns peak-to-peak) stays hundreds of times below the lowest published mean threshold, and thousands of times below the thresholds measured with random jitter on music.

The application places the jitter at the converter clock, after the PLL, relative to these thresholds: below 10 ns, "below published thresholds"; from 10 to 500 ns, "within the range of published thresholds" (depending on the signal and the listener); above 500 ns, "beyond published thresholds".

These thresholds come from listening tests. Theoretical criteria, which compute the worst possible case, go much lower: Dunn (1992) asks for less than 20 ps peak-to-peak for jitter at 20 kHz, and Benchmark wants sidebands 23 dB below the converter noise, that is a few picoseconds. These are design targets, not thresholds measured on listeners; opinions differ, and the application does not settle the matter.

Jitter and signal-to-noise ratio: a jitter J on the converter clock limits the signal-to-noise ratio of a signal of frequency f to -20 log10(2 pi f J). At 20 kHz: 78 dB for 1 ns, 58 dB for 10 ns, 98 dB for 100 ps. This is a worst case. Physically, it concerns the jitter that reaches the converter clock (after the PLL); the application also shows it on the raw interface jitter, as a pessimistic bound.

2.3 Voltages (high, low, P-P)

What it is: the mean voltage of the high level, that of the low level, and the peak-to-peak amplitude.

Why it is useful: attenuation brings both levels closer to the threshold. The standards (IEC 60958 and AES3) do not set a minimum amplitude but an eye mask: at least 200 mV of opening over half a cell, around the decision threshold (section 2.11). Since the eye is always smaller than the peak-to-peak amplitude, an amplitude approaching 200 mV points to decoding that is already compromised. An asymmetry between the high and low levels points to a DC offset, for instance from AC coupling.

2.4 Noise RMS

What it is: the noise measured on the high and low flat levels, from the differences between successive samples, leaving out samples close to the edges so that transitions are not counted as noise. A median makes the measurement insensitive to the few remaining edge points: it is zero on the ideal reference and accurate to within 1% for white noise. High noise reduces the margin between the two logic levels.

Link with SNR: SNR ≈ 20 log10(V_PP / maximum noise).

2.5 Parity errors

What it is: each 32-bit subframe ends with a parity bit, chosen so that bits 4 to 31 hold an even number of 1s. The analyzer counts subframes that break this rule. S/PDIF and AES/EBU both have this bit.

The limit: parity only detects errors affecting an odd number of bits. Parity errors indicate a signal that is already clearly degraded.

2.6 Eye diagram

What it is: the overlay of signal pieces two cells long. The pieces are aligned on the measured clock, so that transitions fall at 0, 1 and 2 cells. They are drawn as thin overlaid traces, like the persistence of an oscilloscope. The AES3 mask is drawn in each opening: a rectangle 200 mV high over half a cell, centered on the decision threshold (the level midpoint). The 200 mV is a minimum eye height, not an absolute level.

What it reveals: the eye height at the center gives the voltage margin (reduced by attenuation, ISI and noise); its width gives the timing margin (reduced by jitter).

2.7 Waveform overlay

What it is: three synchronized panels: global view, automatic zoom on the region where the two cables deviate most from the reference, and difference signal (cable minus reference).

2.8 Automatic interpretations

Under each chart, a text summarizes what the measurements show. For the eye, the opening is estimated as the gap between levels minus 6 times the noise (3 sigma on each side); causes (reflections, noise, attenuation) are identified by thresholds. For jitter, the text places the result relative to published detection thresholds (scale in section 2.2), and notes for an optical link that module jitter is not modeled.

2.9 S/PDIF and AES/EBU Connectors

The simulator does not model connectors: at S/PDIF frequencies they are discontinuities too short to matter.

RCA: impedance not standardized, around 30 to 60 ohms depending on the geometry. But the discontinuity is short: 1 to 2 cm for the plug, 2 to 4 cm with the socket. Seen by an edge of about twenty nanoseconds, its effective reflection is 0.2 to 0.6% (up to about 2% with 5 ns edges), whatever the cable length. It also delays the transmitted wave by 7 to 46 ps, but by the same amount for every edge: it adds no jitter. The "Belden 1694A + 50 ohm RCA plugs" preset therefore keeps a 75 ohm cable.

BNC: the IEC 61169-8 standard describes the 50 ohm BNC; the 75 ohm version only appears in annex A, for its dimensions. The quality of a 75 ohm BNC therefore depends on its maker: broadcast models state a negligible reflection (for instance Kings 2065 series: return loss better than 36 dB up to 1 GHz).

XLR (AES/EBU): uncontrolled impedance, but about 3 cm long, i.e. 0.10 to 0.15 ns of travel: even with a reflection coefficient of 0.3, the effective reflection stays below 0.5%. A quality plug (Neutrik NC3MXX and NC3FXX, for instance) is rated for more than a thousand insertions and at most 3 mΩ of contact resistance.

Toslink: no electrical reflection. The jitter of an optical link comes from the transmitter and receiver modules, not from the connector (section 1.7).

In practice: a clean connector, whatever its type, changes nothing. An oxidized contact can cause errors at any length.

2.10 Cable-Metric Sensitivity Matrix

The matrix varies one parameter at a time around the chosen configuration and measures the effect on each metric (CER, jitter, amplitude, SNR, eye opening).

Length: losses, hence ISI and jitter; picked-up noise.

Attenuation: eye height, then CER when the eye approaches the 200 mV mask.

Impedance: echo; data-dependent jitter beyond the length L0, a smaller residue below it.

Shielding: noise and SNR.

Propagation velocity: no effect on a matched cable, which is expected: it only sets the echo delay when there is an echo.

Bandwidth is not in the matrix: it follows from the losses, it is not an independent parameter.

2.11 AES3 eye mask and practical limit

The standards set the minimum eye at a receiver input: 200 mV high over half a cell wide, centered on the decision threshold. This is the AES3 mask (EBU Tech 3250, §6.3.3), also found for the consumer interface (IEC 60958-3:2003, then IEC 60958-1), which also requires pulses of at least 0.8 UI: the simulator applies this rule to S/PDIF. It measures the eye without noise, then removes the margin needed for noise to cause only one error in a million million bits. The practical limit shown is the largest length, up to 200 m, that meets this mask. The search assumes that the signal degrades as length increases, which holds since losses and picked-up noise grow with it; "> 200 m" means "beyond 200 m", the search limit. The cause shown is noise only if, without noise, the mask would still be met at a 10% longer length; otherwise, it is loss.

Examples (living room, 44.1 kHz): Belden 1694A beyond 200 m; Van Damme about 190 m, Mogami 2964 about 135 m and generic coaxial about 120 m, limits set by losses (noise only removes 8 to 9 mV from the eye, so they depend little on the noise model); cheap RCA cable about 20 m, a limit set by noise (35 m without noise), hence by the noise model, which is approximate.

In AES3, the Belden 1800F, Canare DA206 and DA202 exceed 200 m: this is the effect of the simulated 5 V level. At the minimum of the standard (2 V), the 1800F holds about 179 m at 48 kHz and the DA202 about 195 m, which matches the manufacturers' distances: Canare states 180 m for the DA202 and 360 m for the DA206 at 48 kHz, Belden 203 m for the 1800F at 6 MHz. The model remains optimistic: it targets the minimum mask, without margin, with an ideal receiver.

The AES3 standard provides for 100 m without equalization, for sample rates up to 50 kHz; beyond 100 m, the EBU allows optional equalization at the receiver. For an optical link, the modules set the limit (section 1.7).

The cable ranking uses this mask: a cable that fails it is ranked last; the others are separated by a score combining jitter, SNR and eye height.

3. Global verdict

The verdict combines the CER and the RMS interface jitter:

The verdict gives a quick reading. For borderline cases, look at all the measurements and at the eye diagram.

What blind listening tests say

We found no peer-reviewed study that compares, blind, two digital cables carrying the same bits. The public Ars Technica test (2015: 6 people out of 7 did not recognise the expensive cable) involved an Ethernet cable: the data goes through a buffer, with no clock transmitted, which says nothing about an S/PDIF link. Its protocol was rough, as its author admits: one trial per person, the answer "no difference" counted as a miss, an audience of skeptics told that such cables were pseudoscience, and only one of the two network legs tested. The available studies mostly measure the level at which jitter becomes audible:

• random jitter on music (Ashihara et al., 2005, 23 professional or semi-professional listeners): nobody detects it at 250 ns, 6 listeners at 500 ns;
• sinusoidal jitter (Benjamin and Gannon, 1998): 10 ns on average on a full-scale 20 kHz sine, the most sensitive case; 20 to 370 ns depending on the listener on chosen music excerpts.

In the simulator, a 10 m Belden 1694A adds about 0.05 ns peak-to-peak (without noise): hundreds of times less than the lowest published mean threshold, thousands of times less than the thresholds measured with random jitter. In a double-blind test, inserting a 16-bit / 44.1 kHz conversion into high-resolution playback was not detected (Meyer and Moran, 2007: 276 correct answers out of 554, chance). A negative test does not prove that no difference exists for anyone, and ground loops carried by an electrical cable remain a separate effect, which optics removes.

Bibliography

Standards

Papers and books

• Dunn, C. & Hawksford, M.O.J. (1992). *Is the AES/EBU/SPDIF Digital Audio Interface Flawed?* AES preprint 3360, 93rd convention. Band-limited link, data-dependent jitter.
• Dunn, J. (1992). *Jitter: Specification and Assessment in Digital Audio Equipment.* AES preprint 3361. aes.org, author's PDF
• Dunn, J., McKibben, B., Taylor, R. and Travis, C. (1993). *Towards Common Specifications for Digital Audio Interface Jitter.* AES preprint 3705 (origin of the AES3 jitter specifications). author's PDF
• Dunn, J. *Measurement Techniques for Digital Audio.* Audio Precision, Application Note 5, 2001-2004. Measured AES3 edges (15 to 20 ns), eye mask, interface jitter.
• Paul, J.D. (1998). *The Effect of Transformers on Transmission of Digital Audio Signals.* AES preprint 4840, 105th convention. Low cutoff of interface transformers.
• Benjamin, E. & Gannon, B. (1998). *Theoretical and Audible Effects of Jitter on Digital Audio Quality.* AES preprint 4826. aes.org
• Ashihara, K. et al. (2005). *Detection threshold for distortions due to jitter on digital audio.* Acoustical Science and Technology 26(1). doi.org
• Meyer, E.B. & Moran, D.R. (2007). *Audibility of a CD-Standard A/D/A Loop Inserted into High-Resolution Audio Playback.* J. Audio Eng. Soc. 55(9), 775-779. aes.org
• Hutchinson, L. (2015). *The audiophile's dilemma: strangers can't identify $340 cables.* Ars Technica, 30 July 2015. Ethernet cable. arstechnica.com
• Adams, R.W. (1994). *Clock Jitter, D/A Converters, and Sample Rate Conversion.* The Audio Critic, No. 21. biline.ca
• Macadie, D. (2005). *Jitter performance of S/PDIF digital interface transceivers: is meeting standards enough?* Wolfson Microelectronics, white paper, rev. 1.1. WM8805 PLL bandwidth.
• Crystal Semiconductor, Cirrus Logic (1998). *CS8411, CS8412 Digital Audio Interface Receiver*, datasheet DS61F1. Jitter attenuation curve (fig. 3).
• Pozar, D.M. *Microwave Engineering*, 4th ed., Wiley, 2011. Transmission lines: reflections (§2.3, §2.5, §2.6), lossy lines (§2.7), transients (§2.8); skin depth (§1.4).
• Ott, H.W. *Electromagnetic Compatibility Engineering*, Wiley, 2009. Cable shielding, braids and spirals: chapter 2, §2.13-2.14 (the shielding values of the simulator's cables are estimates, not values from this book).