---
Title: Measure your interface first – AudioLab
---

# Measure your interface first

Before any device, measure the audio interface itself: its outputs cabled straight back to its inputs, nothing between. It takes about an hour, and every later measurement leans on it. It answers three questions:

- **Is the setup sound?** Calibrated levels, no hum, no ground loop, no dropouts.
- **Where is the floor?** The loop's own noise, distortion, crosstalk and IMD are the lowest figures the setup can read. A device that reads within a few dB of them shows the interface, not itself.
- **Do the reference and the measuring inputs agree?** When they are different kinds of input, their difference is in every response. AudioLab stores it once as a **loop** and takes it out.

The figures on this page come from an M4 measured this way: OUT 3 / OUT 4 as the generator, the front inputs In 1 / In 2 as the references, the rear line inputs In 3 / In 4 as the measuring inputs. On another interface the numbers differ; the steps do not. Every reading of that measurement, with its chart and CSV: [the full protocol](measured-motu-m4.html).

## What you need

- A **true-RMS multimeter** (AC volts) for the calibration.
- Two **Y splitters** and short cables: each output to its reference input and to its measuring input.
- Two **shorting plugs** for the measuring inputs: a mono (TS) jack with the tip joined to the sleeve by a short wire inside. In a balanced (TRS) input the sleeve also touches the ring, so both legs are shorted to ground – as they should be.
- Nothing else connected to the interface; the computer as you will measure with it.

## Set once

![Interface loop: each output of the interface goes through a Y splitter to its reference input (OUT L to IN 1, OUT R to IN 2) and through a short cable to its measuring input (IN 3, IN 4); no device between](wiring/interface_loop.jpg)

| Where | Setting | Why |
| --- | --- | --- |
| CONNECTIONS → *Audio interface* | your interface (ASIO) | |
| CONNECTIONS → *Sample rate* | **48 kHz** | the rate you will measure devices at; the loop and its correction belong to one rate |
| The interface's own driver panel | ASIO buffer **1024** samples | a small buffer drops samples under load, and a dropout reads as noise or a glitch |
| Settings → Audio → *Generator outputs* | **Outputs 3–4** | on an M4 the rear line outputs, at a fixed level: no volume knob can upset the calibration |
| Settings → Audio → *Generator output limit* | **−6 dBFS** | near full scale an interface's own output distorts more (on an M4 its 7th harmonic shows at 0 dBFS); raise it only where a step says so |
| CONNECTIONS → *Wiring* | **Stereo, references first** | references on In 1 / In 2, the measuring inputs on the better In 3 / In 4 – the wiring you will use for devices |
| CONNECTIONS → *Attenuator before the device* | **None** | |
| CONNECTIONS → *On the device output* | **Nothing** | no load in the loop |
| SIGNAL → *Frequency* | **997 Hz** (the *997 Hz* key) | interference lines at whole kilohertz (some USB devices have them) land on the harmonics of a 1 kHz tone; 997 Hz (AES17) keeps them apart |

## Step 1 – Calibrate

**Levels on this page** are in dBFS – the interface's own scale, since the interface is what is measured – with the volts they gave on ours. After this step the line under *Level* says the volts as well; the guides for devices give their levels in volts at the device's input.

**What:** how many volts 0 dBFS is on each output and input, so every later reading is in volts.

**Do:** CALIBRATION → **Calibrate…**. The wizard takes one input at a time: connect the output straight to that input with the multimeter across the same connection, read the multimeter, type the value, **Save & next**. The outputs are calibrated from the same readings. On the last page every *Divider* must be **0** – the loop has no divider.

**Check:** **Level meter**, 997 Hz at −20 dBFS: all four inputs show the same voltage, about 0.245 V on an M4 (2.454 V at 0 dBFS from OUT 3 / OUT 4).

**Watch out:** the calibration holds only while no gain knob on the interface moves. AudioLab compares the reference with what the calibration expects (from 100 Hz to 10 kHz) and warns when it is more than 1 dB off.

## Step 2 – Noise of the measuring inputs, shorted

**What:** the measuring inputs' own noise, with nothing at all connected to them.

**Connect:** pull the cables out of In 3 and In 4 and put the shorting plugs in. The references stay on their Y splitters.

**Set:** **Noise & S/N**, *S/N against a test tone* **off**. **SINGLE**, then **Export**.

**Read:** the noise in µV, unweighted and A-weighted, for In 3 and In 4.

![Noise density with In 3 shorted by a plug, 10 Hz to 21.6 kHz: a flat floor near −150 to −160 dBFS/√Hz from 200 Hz up, rising slowly below; no hum lines on the measuring inputs](figures/step_interface_noise_shorted.png)

On an M4 the line inputs read **14.4 µV, 10.8 µV A-weighted**: 114 dB below their full scale of 7.5 V.

**Why:** this is the floor under every noise reading of a device. A device with 20 µV of noise reads 24.6 µV here, one with 10 µV reads 17.5 µV: the two add as powers. Short the input once at the end of your cable and once straight in the socket – if both read the same, the cable adds nothing.

## Step 3 – Noise of the loop

**What:** the noise with the output connected but silent – what the output adds.

**Connect:** the short cables back in In 3 / In 4.

**Set:** **Noise & S/N** as in step 2, **SINGLE**. You do not switch the generator off: the task plays silence itself.

![Noise density of the loop: the same flat floor as with the inputs shorted, no mains lines on the measuring inputs](figures/step_interface_noise_loop.png)

On an M4: **14.7 µV, 11.1 µV A-weighted**, 0.2 dB above the shorted inputs – the silent output adds about 3 µV. The loop's noise is the input's.

**Watch out:** lines at 50, 100 or 150 Hz on the measuring inputs mean a ground loop – usually the computer and another mains device on different outlets, or a USB device sharing the computer's ground. Find it now; with a device connected it would read as the device's hum. The export marks this reading *limited by interface*: in the loop that flag says nothing, the loop *is* the interface.

## Step 4 – THD+N at three levels

**What:** the loop's distortion and its noise at a low, a middle and a high level.

**Set:** **THD+N meter**, 997 Hz; ANALYSIS → *Subtract reference THD* **off**. Run it at **−20 dBFS** (0.245 V), **−6 dBFS** (1.23 V) and – with the output limit raised to 0 dBFS for this step – **−1 dBFS** (2.18 V). Let each run for about ten seconds, **STOP**, **Export**.

![THD+N meter, loop at −20 dBFS: the tone at 0.245 V on all four inputs; on the measuring inputs H2 and H3 near −142 dBFS at the noise; on the reference inputs H2 at −120 dBFS](figures/step_interface_thdn.png)

| Generator | THD on In 3 / In 4 | THD+N | H2 on the reference inputs |
| --- | --- | --- | --- |
| −20 dBFS (0.245 V) | −107 dB (0.0005 %) | −84 dB (the noise) | −94 dBc |
| −6 dBFS (1.23 V) | −106 to −108 dB | −97.5 dB | −80 dBc |
| −1 dBFS (2.18 V) | −109 to −111 dB | −101.5 dB (0.00084 %) | −74 dBc |

**Why:** THD here is the lowest distortion the setup can read. THD+N is mostly the noise from step 3, so it improves with the level.

**Watch out:** the front inputs that carry the references add their own second harmonic, rising with the level (−94 dBc at 0.25 V, −74 dBc at 2.2 V on an M4). It does not enter a device's reading, which is taken on In 3 / In 4 – but *Subtract reference THD* would subtract that distortion from a device that never had it. Leave it off unless the reference inputs are as clean as the measuring ones. Set the output limit back to −6 dBFS.

## Step 5 – THD vs level

**What:** the floor of every THD vs level sweep, level by level.

**Set:** **THD vs level**, 997 Hz, *From* −60, *To* −1 dBFS (2.45 mV to 2.18 V; output limit 0 dBFS for this sweep), *Step* 2 dB. **RUN**; the sweep ends by itself. **Export**, and set the limit back to −6 dBFS.

![THD and THD+N of the loop against the generator level: THD+N falling 2 dB for every 2 dB of level; THD falling to about −107 dB at −22 dBFS and staying between −105 and −111 dB above; the reference inputs' THD+N turning up above −16 dBFS](figures/step_interface_thd_level.png)

**Read:** THD+N falls a straight 1 dB per dB – it is the noise, the same at every level. THD reaches the loop's own distortion, **−105 to −111 dB**, from about −22 dBFS (0.2 V) up; below that the harmonics sink into the noise (−89 dB at 24.5 mV, −69 dB at 2.45 mV).

**Why:** a device's THD at a given level can only be read where it lies above this curve. Keep the export: it is the line to put under every sweep you make later.

## Step 6 – Response, and the loop correction

**What:** how the reference input and the measuring input differ in level and phase over the band – and storing that difference so it is taken out of every response.

**Set:** ANALYSIS → *Loop correction* **off** first. **Stepped sine**, −20 dBFS (0.245 V), 20 Hz to 20 kHz, 31 points, **RUN**.

![Gain of In 3 against In 1 without loop correction: flat at −3.18 dB from 200 Hz to 5 kHz, rising 2.2 dB towards 20 Hz with −43° of phase, and 0.2 dB towards 20 kHz with +18°](figures/step_interface_loop_raw.png)

**Read:** the level, −3.2 dB, is only the ratio of the two inputs' ranges (7.52 V against 5.21 V on an M4). The shape is what counts: on an M4 the front inputs lose bass the line inputs keep – **+2.2 dB and −43° at 20 Hz** – and lag a little at 20 kHz (+0.2 dB, +18°). Without a correction every device measured this way would carry that curve.

**Store it:** **Gain & phase**, −20 dBFS (0.245 V peak), **SINGLE**, then **Save as loop**. Under the option each channel now lists its loop (*In 1 → In 3 …*). Tick *Loop correction* and run **Stepped sine** again:

![The same with the loop correction: flat within about 0.1 dB from 20 Hz to 20 kHz, the phase near 0°](figures/step_interface_loop_corrected.png)

The two inputs now agree within about **0.1 dB and 0.4°** over the band, by Gain & phase and Stepped sine alike.

**Why and when:** the loop correction changes only **Gain & phase and Stepped sine**; THD, noise, crosstalk, IMD and levels are the same with it or without. It is needed when the reference and the device output are on different kinds of input (*Stereo, references first*); not with *Stereo* (both on the same kind) or *External DAC / ADC* (no reference). Details: [Loop correction](response.html#loop-correction).

**Watch out:** a loop belongs to one pair of inputs and one sample rate. Store it again after changing cables, the wiring or the rate, and before a precise response measurement: on an M4 a loop stored 18 days earlier had drifted by 0.1 dB at 20 kHz.

## Step 7 – Crosstalk

**What:** how much of one channel the loop itself leaks into the other.

**Set:** **Crosstalk**, −20 dBFS (0.245 V), 20 Hz to 20 kHz, 31 points, **RUN**. It drives one channel at a time.

![Crosstalk of the loop: below 700 Hz only noise-floor bounds of −107 to −117 dB; above, a leak rising 6 dB per octave to −94.5 dB from 8 kHz up](figures/step_interface_crosstalk.png)

On an M4: below about 700 Hz nothing leaks above the noise (better than −107 dB); above it a leak rises 6 dB per octave to **−94.5 dB** from 8 kHz. Points marked ≤ lie at the noise and say only "at most".

**Why:** a device's crosstalk can be read down to this curve. Most devices lie far above it.

## Step 8 – IMD

**What:** the floor for two-tone intermodulation: **SMPTE** (60 Hz + 7 kHz at 4 : 1) and **CCIF** (19 + 20 kHz at 1 : 1).

**Set:** **IMD vs level**, *IMD method* SMPTE, *From* −40, *To* −6 dBFS (peak 25 mV to 1.23 V), *Step* 2 dB, **RUN**, **Export**; then the same with CCIF. The task drives the generator itself.

![IMD SMPTE of the loop against the level: falling from 0.014 % at −40 dBFS to 0.0012 % at −18 dBFS and staying between 0.0011 and 0.0016 % above; the reference inputs rising from −28 dBFS](figures/step_interface_imd_smpte.png)

On an M4: SMPTE **0.0011–0.0016 %** from −20 dBFS (0.25 V) up, CCIF **0.0005–0.0011 %** from −24 dBFS (0.15 V) up; below that the noise sets the reading.

**Watch out:** IMD has no reference subtraction; the reference inputs' IMD is drawn alongside and rises with their own second harmonic. It is not the device's.

## What the loop gives you

| Floor | An M4 through In 3 / In 4 | A device reading near it… |
| --- | --- | --- |
| Input noise | 14.4 µV, 10.8 µV A-weighted | adds as power: 20 µV reads 24.6 µV |
| THD | −105 to −111 dB from 0.2 V up | is at most this clean; more cannot be read |
| THD+N | the noise: −84 dB at 0.25 V, −101.5 dB at 2.2 V | is limited by the noise, not by distortion |
| Response, with the loop | ±0.1 dB, ±0.4° | is flat within the setup |
| Crosstalk | below −107 dB to 700 Hz, −94.5 dB above 8 kHz | |
| IMD SMPTE / CCIF | 0.0011–0.0016 % / 0.0005–0.0011 % | |

Keep the exports of this hour. Each later protocol can then say which of its figures belong to the device and which to the setup – for a load box with a divider, or any attenuator, the noise floor rises by its loss, and the same floor applies at the point in front of it.

## When to measure it again

- **Steps 2–3** (noise) after changing the interface, its cables, the computer, a USB hub or the power arrangement.
- **Step 6** (the loop) after changing cables, the wiring or the sample rate, and before a precise response measurement.
- **All of it** on another interface, or after a driver or firmware update.
