cables/ethernet-cable-audio · v1.0

Ethernet Cables & Audio Quality

An Ethernet cable always delivers data 100% correctly — but the RF energy carrying that data can leak off the cable and travel into the DAC. Here is the path, step by step.

PICHITCHAI OPADWORARAT · MUSIC ENTHUSIASTSIGNAL PATH: LAN → STREAMER → INTERFACE → DAC1000BASE-T · PAM-5 · 125 MBaud

Physical structure and components of a LAN cable

A LAN cable is not ordinary wire — it is a transmission line built to carry RF in the 100–800 MHz band, and every layer of its construction directly affects its electromagnetic behaviour. It holds eight copper conductors in four twisted pairs. Permanent runs use solid conductors (lower DC resistance, better skin effect at HF); patch cords use stranded for flexibility but with more total surface.

Characteristic impedance of a transmission line
Z₀ = √( L' / C' )
L′ = inductance/length, C′ = capacitance/length · Cat 5e: Z₀ = 100 Ω ±15% (IEC 61156) · Cat 7/8 with tight geometry: ±5 Ω — a uniform impedance along the run reduces reflection noise
GradeZ₀ toleranceConductorReflection
Cat 5e100 Ω ±15%solid/strandedlooser geometry
Cat 6A100 Ω ±~8%solidtighter pitch control
Cat 7/8100 Ω ±5%solid, shieldedmost uniform

Skin effect and conductor material

At high frequency the current crowds into the surface of the conductor — the surface matters more than the bulk.

Skin depth (δ)
δ = √( ρ / (π · f · μ₀ · μᵣ) )
ρ(copper)=1.68×10⁻⁸ Ω·m · at 1 kHz: δ≈2.1 mm · 1 MHz: 66 μm · 125 MHz (1000BASE-T): 5.9 μm · 500 MHz: 3.0 μm
AC resistance from skin effect
R_ac ≈ ρ / (π · d · δ) → R_ac/R_dc ≈ d / (4δ)
d = conductor Ø (AWG24 = 0.51 mm) · at 125 MHz: R_ac/R_dc ≈ 22× — ETP copper RAC is ~5–15% higher than OFC due to a rougher oxide surface
R_ac / R_dc (skin effect)

Fig 1. AC-to-DC resistance ratio of an AWG24 conductor — barely above DC below ~50 kHz, but at 125 MHz it climbs to ~22× because the current is squeezed into a 5.9 μm surface layer

OFC copper (ASTM B170, under 10 ppm oxygen) has a smoother surface than ETP (200–400 ppm), so lower surface resistance. At RF the point isn’t bit errors (the margin is huge) but lower RF emission, because a more uniform current distribution radiates less EMI that couples into the DAC chassis.

Twist pitch, symmetry and crosstalk

Twisting makes induced noise equal on both conductors (common-mode) so it cancels at the receiver — but only as far as the symmetry of the twist holds. Asymmetry converts common-mode back into differential.

Mode conversion from twist asymmetry
Vdm = Vcm · (ΔC / 2C) · jωRs
ΔC = difference in capacitance-to-ground between the two wires of a pair · the larger ΔC and the higher f, the more common-mode leaks into differential · a good cable keeps ΔC low through uniform geometry
Near-end crosstalk (NEXT)
NEXT (dB) = 20 · log₁₀( Vdisturbing / Vinduced )
higher is better · TIA-568: Cat 6A NEXT ≥ 54 dB @100 MHz, Cat 5e ≥ 35.3 dB · a good audio-grade cable should beat spec by ≥ 6–10 dB

Each pair uses a different twist pitch to avoid resonance between pairs at the same frequency. Cat 6A controls pitch more tightly, so NEXT is lower.

PAM-5 line coding and the digital domain

Gigabit Ethernet doesn’t send binary; it uses PAM-5 (five levels: +2,+1,0,−1,−2 V) across all four pairs full-duplex, so the symbol rate drops to 125 MBaud/pair instead of running at 1000 MHz.

1000BASE-T throughput
1 Gbps = 4 pairs × 125 MBaud × 2 bit/symbol
PAM-5 + Trellis coding gives net 2 bit/symbol · fundamental = 125 MHz, but PAM-5 harmonics reach 500+ MHz (IEEE 802.3 §40)

Before transmission the data passes a scrambler that whitens the spectrum, so the on-cable spectrum looks like broadband noise spanning 0–500+ MHz continuously, not a single spike.

The limit of the digital domain

TCP/IP has CRC-32 + retransmit, so the audio data reaching the DAC is 100% correct — but that only guarantees data, not that the RF energy carrying it stays on the cable. Every component in the 100–500 MHz band can couple into a circuit sensitive to that frequency.

Impedance mismatch and signal reflection

When impedance changes abruptly (a kink, a poor RJ45 termination, non-uniform geometry), the signal reflects, creating a brief high voltage on the cable → more radiated EMI.

Reflection coefficient (Γ)
Γ = (ZL − Z₀) / (ZL + Z₀)
Z₀ = 100 Ω · ZL = 115 Ω (poor termination): Γ = 0.07 → 7% of the voltage reflects → more EMI + waveform distortion
Return loss (RL)
RL (dB) = −20 · log₁₀|Γ|
TIA-568: RL ≥ 20 dB @100 MHz (Cat 5e) = |Γ| ≤ 0.1 · a high-quality cable should hold RL ≥ 26 dB along the whole length

Common-mode noise and mode conversion

This is the main mechanism by which digital-domain noise leaks into the DAC’s analog domain. Ethernet is differential — noise equal on both wires is subtracted out (CMRR) — but asymmetry converts between differential and common-mode.

Common-mode rejection ratio (CMRR)
CMRR (dB) = 20 · log₁₀( Adiff / Acm )
CMRR = 60 dB → common-mode cut 1,000× · but CMRR falls with frequency — at 100 MHz it may be only 20–30 dB
Longitudinal conversion loss (LCL) — IEC 62153-4-3
LCL (dB) = 20 · log₁₀( Vdifferential / Vcommon-mode )
measures how much differential leaks into common-mode; higher is better · TIA-568: Cat 6A LCL ≥ 40 dB @500 MHz · low-quality cable = low LCL → more common-mode → easier entry into the DAC ground path
Differential
clean signal
Asymmetry
ΔC / twist error
Common-mode
leaks out
DAC ground
noise enters

Ground loops and shield topology

Two connected devices usually sit at slightly different ground potentials (different SMPS, different outlet circuits). Even a 10–100 mV difference drives current through the conductor that joins them.

Ground loop current
Iloop = ΔVgnd / Zloop
ΔV = 50 mV, Zloop = 2 Ω → Iloop = 25 mA — this current enters the PCB ground plane and modulates the DAC chip’s supply rail
TopologyGround loopNote
UTP (unshielded)no conductive loopbut radiates RF fully
S/FTP, shield both endsloop forms automaticallyshield current may exceed UTP
S/FTP, one end (DAC side)breaks the looprecommended

RF radiation and capacitive coupling into the chassis

Even without direct contact, the electric field from the RF cable couples into the DAC chassis through stray capacitance.

Quarter-wave resonance
L = c / (4 · f · √εr)
f = 125 MHz, εr ≈ 2.3 (PE) → L ≈ 0.50 m — a 0.5 m patch cord is exactly a λ/4 dipole at 125 MHz; unshielded UTP radiates RF most freely
Capacitive coupling current into the chassis
Icouple = Cstray · dV/dt = Cstray · V · 2πf
Cstray = 10 pF, V = 1 V, f = 100 MHz → Icouple ≈ 6.3 mA — enough to significantly modulate the DAC’s PCB ground plane

PHY chip, Ethernet transformer and interwinding capacitance

The PHY chip (Realtek/Intel/Marvell) is mixed-signal; the current it draws from the supply changes with the data pattern → data-dependent switching current creates supply noise synchronised to the data.

IEEE 802.3 mandates an isolation transformer on every port. The transformer blocks DC and audio frequencies completely, but at RF its weak point is interwinding capacitance.

Leakage current through interwinding capacitance
Ileak = Cstray · ( dVcm / dt )
Cstray (Ethernet transformer) = 5–15 pF · at 100 MHz, ΔV = 1 V: Ileak ≈ 6.3 mA · a high-quality PCB transformer with Cstray ≤ 3 pF cuts Ileak 3–5×
PHY
data-dependent I
Supply noise
sync'd to data
Transformer
Cstray 5–15 pF
DAC ground
common-mode through

In other words: above 100 MHz the transformer behaves like a 5–15 pF capacitor that lets common-mode current pass straight into the DAC’s ground plane.

RF rectification in the op-amp and DAC analog stage

Once RF reaches the ground plane it enters the analog stage two ways: through the op-amp input pin and through the supply rail. Every input pin has ESD diodes with a nonlinear I–V curve → they rectify RF into a DC offset.

RF-induced DC offset (Shockley diode model)
IDC ≈ Is · ( VRF² / 2VT² )
VT = kT/q ≈ 26 mV @25°C · the DC offset is amplified by the op-amp gain → output error · ref: Analog Devices MT-096, Microchip AN1767
EMI rejection ratio (EMIRR) — Microchip AN1767
EMIRR (dB) = 20 · log₁₀( VRF,input / Voffset,induced )
higher is better · typical audio op-amp EMIRR ≈ 50–80 dB @100 MHz · EMI-hardened parts (OPA2134/OPA1612): EMIRR > 85 dB

DC offset → shifted bias point → more even-order harmonic distortion (2nd, 4th), which the ear reads as “warm/smooth” — but it is measurable distortion, not the recording.

Supply-rail modulation and the noise floor

Besides rectifying at the input pin, a dirty ground plane modulates the DAC supply directly (because the decoupling cap ties supply to ground tightly).

Power supply rejection ratio (PSRR)
PSRR (dB) = 20 · log₁₀( ΔVsupply / ΔVoutput )
a typical DAC has PSRR ≈ 80–100 dB @1 kHz · but it falls with frequency: ~40 dB @1 MHz, 10–20 dB @100 MHz · 1 mV noise @100 MHz, PSRR 15 dB → 180 μV at the output — can exceed a −120 dBFS noise floor
PSRR (DAC supply)CMRR (cable/interface)

Fig 2. Both protective mechanisms — PSRR and CMRR — collapse at high frequency. At 100 MHz only ~15–28 dB remains. This is the quantitative reason RF noise in this band slips into the analog domain even though audio frequencies are buried

The net effect: RF from the cable → ground plane → (1) op-amp input → rectify → harmonic distortion, and (2) supply path → modulate the rail → raise the noise floor across the spectrum → the sound loses “air” and detail sinks.

Mitigation: fixing each noise path

Fix it path by path — not “expensive cable = good.”

Common-mode choke impedance
Zcm = jωLcm (high at HF) Zdm ≈ 0 (differential passes)
Lcm = 100 μH – 10 mH · at 100 MHz, Lcm = 1 mH: Zcm ≈ 628 kΩ → blocks common-mode current effectively · ref: US Patent 10,026,544 “Common mode noise restrainer applicable to Ethernet”
Noise pathFixEffect
common-mode on cablecommon-mode choke (CMC)most cost-effective
conducted + ground loopoptical fiber + media convertercuts all network noise
switching noise at sourcelinear regulated PSU vs SMPSfixes the source
multiple return pathsstar ground + single-end shieldreduces ground loop
Order of value

Work from source to sink: swap SMPS → linear (fix the source), add a CMC at the interface (block common-mode), break to fiber if possible (cut all conducted noise), star-ground (prevent loops). Every step has a clear, measurable mechanism — not belief.

References

  • aesHockanson, D. et al. “Measurements and simulations for ground-to-ground plane noise…,” IEEE Trans. EMC.
  • aesArchambeault, B. et al. “Impact of analog/digital ground design on circuit functionality and radiated EMI,” IEEE Trans. EMC, 2005.
  • stdIEEE Std 802.3-2022, Ethernet — §40 (1000BASE-T PHY).
  • appAnalog Devices “MT-096: RFI Rectification Concepts,” Rev.0.
  • appMicrochip “AN1767: EMI Rejection Ratio (EMIRR) of Op Amps.”
  • appAbracon LLC “Common Mode Chokes: Basics and Applications,” white paper.
  • stdANSI/TIA-568.2-D, Balanced Twisted-Pair Cabling and Components, 2018.
  • stdIEC 62153-4-3, Longitudinal conversion loss (LCL); IEC 61156-5; IEC 60958-1.
  • stdASTM B170, Oxygen-Free Electrolytic Copper.
  • bookOtt, H. W. Electromagnetic Compatibility Engineering, Wiley 2009 (ch.4–5).
  • bookPaul, C. R. Introduction to Electromagnetic Compatibility, 2nd ed., Wiley 2006.
  • bookWilliams, T. EMC for Product Designers, 5th ed., Newnes 2017.
  • patUS 10,026,544 B2 “Common mode noise restrainer applicable to Ethernet,” USPTO 2018.
Edited by Pichitchai Opadworarat Head of R&D — Pyramid Lifestyle Technology Ltd. Part. 2 years in audio engineering (since the company was founded)

Revision history

v1.02026-06-11Migration + skin/rejection charts