---
Title: Headphone amplifier, step by step – AudioLab
---

# Headphone amplifier, step by step

A headphone amplifier measured from its line input to its headphone jack: what to connect, what to set in AudioLab, in which order, and why. The figures come from a desktop headphone amplifier with a line input and a volume control, measured through a load box. Allow two hours, the warm-up included. Every reading of that measurement, with its chart and CSV: [the full protocol](measured-aune-t1.html).

Measure your interface first ([Measure your interface first](step-interface.html)): its floor is the line every result here is read against.

## What you need

- The interface calibrated, with its loop stored (loop correction) – see the page above.
- A **load box** for headphone amplifiers: a switch with 16, 32, 300 and 600 Ω (or a pair of resistors, 32 Ω rated 2 W and 300 Ω), and a divider that feeds the interface. Ours: [the load box](headphone-amp.html#a-load-box-for-headphone-amplifiers), a 1 kΩ + 1 kΩ + 1 kΩ divider per channel, read balanced across the middle resistor.
- A **true-RMS multimeter** (AC and DC volts, ohms).
- A cable from the headphone jack to two jacks (tip → In 3, ring → In 4) for the readings without the box.
- **No headphones** on the amplifier while measuring: some steps drive it to clipping.

## Levels in this guide

Levels are given in **volts at the amplifier's line input** – what the device sees – with the dBFS they were on our interface (2.45 V at full scale) in brackets. On your interface the dBFS differ; the volts are what count ([Levels](levels.html#volts-for-the-device-dbfs-for-the-interface)).

- **Calibrate the outputs first** ([step 1 of the interface page](step-interface.html#step-1-calibrate)): only then does the line under *Level* say the volts (*≈ 245 mV at the output*), and the keys 100 mV, 316 mV, 0 dBu, 1 V and 2 V play what they say. Without calibration AudioLab knows only dBFS.
- One level: move *Level* until the line under it shows the voltage.
- A sweep (*From* / *To*) is set in dBFS: dBFS = 20 · log₁₀(volts / full scale). With 2.45 V at full scale, −20 dBFS is 245 mV, −6 dBFS 1.23 V, −40 dBFS 25 mV.

## Wiring

![Stereo, references first: each interface output goes through a Y splitter to the device input and to the reference input (OUT L to IN 1, OUT R to IN 2); the device outputs go to IN 3 and IN 4](wiring/stereo_reference_first.jpg)

- OUT 3 → Y → the amplifier's **line input L** and **In 1** (reference L); OUT 4 → Y → **line input R** and **In 2**.
- The **headphone output** → the load box → its taps → **In 3** (L) and **In 4** (R), with TRS cables.
- CONNECTIONS: *Wiring* **Stereo, references first**, *Sample rate* **48 kHz**, *Attenuator before the device* **None**, *On the device output* **Nothing**. Settings → Audio: *Generator outputs* **3–4**, *Generator output limit* **−6 dBFS**. SIGNAL: **997 Hz**. ANALYSIS: *Loop correction* on.

## Step 1 – Warm up, then DC at the output

**What:** the DC voltage on the headphone jack – it would flow through the headphones' voice coils.

**Do:** switch the amplifier on and leave it for 20–30 minutes. Nothing on the headphone jack. Multimeter on **DC volts**, tip against sleeve (L), ring against sleeve (R), at volume 0 and at maximum.

**Read:** a few millivolts is fine (ours: 2–5 mV at both ends of the volume control). **Above about 50 mV, stop**: do not connect headphones or the interface until it is understood.

## Step 2 – The load box's divider

**What:** how much the box's divider takes from the signal, so that AudioLab reads the volts at the headphone jack, in front of the divider.

**Why it is measured, not calculated:** 1 k + 1 k + 1 k should give 9.54 dB, but the interface's input lies across the middle resistor and lowers it: ours measured **10.58 dB (L) and 10.34 dB (R)**.

**Do:**

1. Volume about halfway. CALIBRATION → **Calibrate…** → **Dividers only ▸** → **Measure dividers…**. AudioLab plays 1 kHz at −10 dBFS and lists every calibrated input, in dBFS and volts.
2. The headphone jack **straight** into In 3 / In 4 with the two-jack cable, no box. Wait until the readings stand still → **1 · Read direct**.
3. Now **through the box**, its load switch **off**. Wait again → **2 · Read through divider**. The *DIVIDER [dB]* column fills in for In 3 and In 4; the references keep 0.

![Calibration and dividers: In 1 and In 2 at 0 dB, In 3 at 10.58 dB and In 4 at 10.34 dB; the divider measurement below with the levels now at the inputs](figures/step_hp_divider_wizard.jpg)

**Check with the multimeter:** with the tone still playing, measure AC volts at the headphone jack (box input), tip and ring against sleeve. AudioLab's level meter must show the same: ours read 0.798 / 0.788 V on the meter and 0.797 / 0.789 V in AudioLab.

**Watch out:** a plug not fully home changes the divider by tenths of a dB, and only the multimeter shows it. Push every plug fully home and turn it once, and **measure the divider at the start of every session** – do not rely on the stored value. AudioLab takes a reading only when the level stands still, but a contact can also sit still at the wrong value.

**Afterwards:** the divider stays until you change it. Set it back to 0 when you measure without the box.

## Step 3 – The loads as the amplifier sees them

**What:** the actual resistances, for power and output impedance.

**Do:** amplifier off, the box plugged in and connected to the interface as for measuring, an ohmmeter at the headphone jack, the switch at each position: with the divider in parallel the 300 Ω position read 271.6 / 271.8 Ω, the 32 Ω one 31.9 / 31.4 Ω, and **off** about 2.9 kΩ – the divider alone. Enter them with **Measured…** next to *On the device output*: the rows for 32 and 300 Ω, and the **Off** row for what stays on the output with the box switched off.

**Why the Off row:** Output Z takes the "open" reading as truly open. With 2.9 kΩ still on the output, Z into 300 Ω reads about 10 % low (13.8 Ω for 15.2 Ω), into 32 Ω about 1–2 %.

## Step 4 – Volume for a gain of 1

**What:** a fixed, repeatable position of the volume control for all the steps that do not need power.

**Do:** box switch **off**, *On the device output* **Nothing**. **Level meter**, 997 Hz at **245 mV** (−20 dBFS), the generator on. Turn the volume until **In 3 reads the same voltage as In 1** (about 0.245 V); the *gain* line under each output shows it. Mark the knob.

![Level meter at unity gain: In 1 and In 2 at 247 mV, In 3 at 241 mV with gain −0.21 dB and In 4 at 247 mV with gain +0.01 dB](figures/step_hp_unity_gain.jpg)

**Why:** with the gain at 1 the generator's −6 dBFS limit gives 1.2 V out – enough for distortion and noise without straining the amplifier – and the channel balance at that position is known (ours: within 0.1–0.2 dB).

## Step 5 – THD+N at unity gain

**Set:** **THD+N meter**, 997 Hz, **245 mV** (−20 dBFS), *On the device output* **Nothing**. **RUN** for about ten seconds, **STOP**, **Export**.

![THD+N meter through the load box at unity gain: the tone at 0.25 V; a line at 100 Hz at −78 dBc with weaker ones at 200 to 500 Hz; H2 to H10 all at about −148 dBFS, the noise](figures/step_hp_thdn.png)

**Read:** ours showed THD+N **0.025 %** and THD **−99 dB** – but H2 to H10 all lie at the same height: that is the noise in their bins, not distortion. The largest single part of the THD+N is a line at **100 Hz**, twice the mains frequency.

**Watch out:** through the box the interface's floor is **10.6 dB higher**: its 14.4 µV of input noise are 49 µV at the headphone jack. At this level the amplifier's own distortion lies below what the box lets you see. Steps 7 and 9 work around it.

## Step 6 – Output impedance

**What:** the amplifier's source resistance. Headphones whose impedance changes with frequency then change the response; with Z above about an eighth of the headphone impedance it starts to matter.

**Do, in this order:**

1. Box switch **off**, *On the device output* **Nothing**, **Output Z**, 997 Hz at **245 mV** (−20 dBFS), **RUN**. Wait about five seconds.
2. **In AudioLab first:** *On the device output* → **300 Ω**. The status says *Reading the output with the 300 Ω load*.
3. **Then the box:** switch to **300 Ω**. Wait five seconds, **STOP**, **Export**.
4. The same again for **32 Ω**.

![Output Z into 32 Ω: the level falls 3.39 dB (L) and 3.37 dB (R) with the load; Z 15.2 and 14.9 Ω](figures/step_hp_output_z.png)

**Why the order:** while *Nothing* is chosen every reading goes in as the open one. We switched the box first, the open reading was overwritten by the loaded level and Z read *below 0.16 Ω* for a 15 Ω output. AudioLab now keeps the open reading when its level falls and says so in red – but the right order avoids it.

**Read:** ours, with the Off row in the formula: **15.2 Ω into 300 Ω and 15.1–15.5 Ω into 32 Ω**. The same into both loads: the output behaves as a plain 15 Ω resistance. Into 32 Ω headphones it takes 3.4 dB of the level (damping factor 2.1); it suits headphones from about 120 Ω.

## Step 7 – THD vs level and power

**What:** how clean the amplifier is as the level rises, and how much power it gives before it clips.

**Into 300 Ω, at the mark:** *On the device output* **300 Ω**, the box at 300 Ω. **THD vs level**, 997 Hz, −40 to −6 dBFS (25 mV to 1.23 V), 2 dB steps, **RUN**, **Export**. Ours: THD down to **−105 dB (0.0006 %)** at 0.6–0.8 V, no rise up to 1.2 V (5.4 mW). At unity gain the generator cannot drive the amplifier any further.

**Into 32 Ω, at full volume:**

1. *On the device output* **32 Ω**, then the box at 32 Ω. **Volume to maximum** – no headphones.
2. **THD vs level**, −40 to −6 dBFS (25 mV to 1.23 V), **RUN**.
3. If it does not clip by −6 dBFS: Settings → Audio → *Generator output limit* **0 dBFS**, and run −12 to 0 dBFS (0.62 to 2.45 V). Once you see where it clips, run again across it in **0.5 dB** steps (ours: −5 to 0 dBFS).
4. **Set the limit back to −6 dBFS** and the volume back to the mark.

![THD+N and THD into 32 Ω at full volume, 0.15 to 373 mW: THD+N falling 1 dB per dB; THD at −102 to −103 dB around 9 mW, rising from 25 mW to about −90 dB and staying there](figures/step_hp_thd_power.png)

![Around clipping in 0.5 dB steps: clean at −88 to −92 dB up to 590 / 574 mW, then a jump to 1.1–1.2 % at 636–653 mW and on to 17 % at 0 dBFS](figures/step_hp_clipping.png)

**Read:** the bar gives the lowest THD+N and the power at the distortion limit (*Distortion limit [%]*). Ours: clean, about **−90 dB (0.003 %)**, up to **590 / 574 mW**, then a hard clip – **about 640 / 630 mW at 1 %**. The output power is V² / R across the measured load, so steps 2 and 3 matter here.

**Watch out:** at full volume the amplifier's own noise is higher than at the mark – THD+N there is still the noise.

## Step 8 – Frequency response

**Set:** volume at the mark, the box at **300 Ω**, *Loop correction* on. **Stepped sine**, **245 mV** (−20 dBFS), 20 Hz – 20 kHz, 31 points, **RUN**, **Export**. **Gain & phase**, **SINGLE**, as a check.

![Stepped sine into 300 Ω from 20 Hz to 20 kHz: both channels flat, the phase turning by −2.9° at 20 kHz](figures/step_hp_response.png)

**Read:** the curve's level is the ratio of the two inputs' digital levels – it includes the ranges of the inputs and the box's divider – so only its shape counts. Ours: **flat within 0.01 dB** from 20 Hz to 20 kHz, −2.9° at 20 kHz. That is below the ±0.1 dB the loop correction itself vouches for: flat within the setup.

**Watch out:** earlier versions said *Reference … reads −1.6 dB against its calibration* at 20–25 Hz – the front inputs' own bass roll-off, not a fault. AudioLab now checks the calibration only from 100 Hz to 10 kHz.

## Step 9 – Noise: through the box, then straight

**Through the box:** the box at 300 Ω (or 32 Ω), volume at the mark. **Noise & S/N**, *S/N against a test tone* off, **SINGLE**, **Export**; again at minimum volume.

![Noise through the load box at the mark: a flat floor with lines at 100 and 200 Hz](figures/step_hp_noise_box.png)

Ours: 55 µV at the mark, 49 µV at minimum – and the interface's floor through the box is 49 µV. **At minimum it is the floor itself**; the box hides the amplifier's noise.

**Straight:** set both dividers to **0** (CALIBRATION → Calibrate… → Dividers only ▸, type 0, Tab, Finish). The headphone jack straight into In 3 / In 4 with the two-jack cable, *On the device output* **Nothing**. **Noise & S/N** at the mark and at minimum.

![Noise with the headphone output straight into the interface at the mark: the floor near −150 to −160 dBFS/√Hz, the 100 Hz line at −122.5 dBFS/√Hz](figures/step_hp_noise_direct.png)

**Read:** ours 18.8 µV, 12.5 µV A-weighted at the mark; with the interface's 14.4 / 10.8 µV taken out as powers the amplifier's own noise is **about 12 µV, 6 µV A-weighted** – 98 dB(A) below 1 V. The unweighted part is mostly the 100 Hz line, which was stronger at minimum volume and through the box with its load.

## Step 10 – Crosstalk, into two loads

**Set:** volume at the mark. **Crosstalk**, **245 mV** (−20 dBFS), 20 Hz – 20 kHz, 31 points, **RUN** – once with the box at **32 Ω**, once at **300 Ω**.

![Crosstalk into 32 Ω: both directions flat at −28.3 dB from 20 Hz to 20 kHz](figures/step_hp_crosstalk.png)

**Read:** ours **−28.3 dB into 32 Ω and −44.3 dB into 300 Ω**, both flat across the band and the same at any volume. Flat means no capacitive leak (that would rise with frequency); depending on the load means a **resistance shared by both channels' return** – the sleeve of the jack and the ground behind it. The driven channel's current makes a voltage across it, and that reaches the other channel through its output impedance: crosstalk = R / (R_load + Z_out). Both loads give the same **1.8 Ω**.

**Why two loads:** with one, a shared ground and a leak inside the amplifier look alike. Where the 1.8 Ω lie – amplifier, jack contact, box – needs a further reading; headphones share the same path up to their plug.

## Step 11 – IMD

**Set:** the box at 300 Ω, volume at the mark. **IMD vs level**, SMPTE, −40 to −6 dBFS (peak 25 mV to 1.23 V), **RUN**, **Export**; then CCIF.

**Read against the loop:** ours SMPTE 0.0017–0.0021 % (the loop alone 0.0011–0.0016 %), CCIF 0.0006–0.0009 % (the loop 0.0005–0.0011 %). CCIF lies within the interface's own range: nothing of the amplifier's shows. SMPTE is 1–4 dB above it: the amplifier's share is about 0.001 %, an estimate so near the floor.

## Step 12 – Channel balance over the volume control

**Set:** the box **off**, **Level meter**, 997 Hz at **0.78 V** (−10 dBFS). Turn the volume to five or six positions from near the minimum to the maximum; at each, **Copy** the readings (the *gain* of In 3 and In 4).

**Read:** the difference of the two gains. Ours stayed within **0.72 dB** over 46 dB of the control, the left channel the louder at every position, most in the lower half. The maximum gain from the line input: +12.3 dB.

## What this amplifier showed

| What | Result |
| --- | --- |
| DC at the output | 2–5 mV |
| Output impedance | 15 Ω |
| Power into 32 Ω at 1 % | about 640 / 630 mW, clean (−90 dB) up to 590 mW, then a hard clip |
| Response, 20 Hz – 20 kHz | within 0.01 dB |
| Noise | about 12 µV, 6 µV A-weighted, at unity gain |
| Crosstalk | −28 dB into 32 Ω, −44 dB into 300 Ω, from a 1.8 Ω shared return |
| IMD SMPTE / CCIF | at or just above the interface's floor |
| Channel balance | within 0.72 dB |

Clean and flat; what sets it apart are the 15 Ω output, which suits higher-impedance headphones, and the crosstalk into low-impedance ones.

## The order, in short

1. Interface loop and calibration first.
2. Warm-up, DC.
3. Divider – measured, then checked with a multimeter.
4. Loads at the jack, the Off row included.
5. Volume for unity gain, marked.
6. THD+N, Output Z (setting before switch), THD vs level and power (no headphones, limit back afterwards), response, noise (through the box and straight), crosstalk into two loads, IMD, balance.
7. Dividers back to 0, the output limit to −6 dBFS.
