Speaker Measurement System

Introduction:

I put together this speaker measurement setup in early 2013 in order to compare 2"-4" diameter cone loudspeaker drivers as objective comparisons of drivers this size were few and far between.
Manufacturers provide datasheets for their drivers however these usually only include Thiele-Small parameters (useful only for designing enclosures) and frequency response plots produced by an undocumented test setup. Harmonic distortion plots are rarely provided and without information about the test conditions cannot be directly compared with other HD plots.
This leaves it up to 3rd parties to measure and provide comparable performance data for drivers from different manufacturers.

Probably the largest database of 3rd party speaker driver measurements is provided by John Krutke (http://www.zaphaudio.com/). My test setup and data format are inspired by his and i have tested a few of the same drivers as he has so loose comparisons can be made between his results and mine.

Hardware

Speaker Baffle:

Without an anechoic chamber available, the next best thing is a large baffle in a large, quiet room. The baffle should preferably be as large as possible, flat and rigid.

Some practical compromises had to be made as it would have to fit through a standard size doorway and be light enough to be moved by one person. I ended up constructing a baffle out of 1200x900x16mm MDF, braced on the rear side to prevent panel resonances. There is a 20x20cm MDF insert for the speaker under test to be mounted as some speakers require different sized cut-outs. The rear off the cut-out is bevelled as to not restrict airflow from the rear of the speaker.

Making a baffle smaller has the effect of worsening baffle edge diffraction. Placing the microphone closer to the speaker driver and windowing the measurement can help eliminate the effects of baffle edge diffraction however there are practical limits as to how close you can place the microphone due to path length differences from the microphone to the radiating parts of the speaker, and also subjecting the microphone to an excessive sound pressure level.
The larger the cone of the speaker under test, the further the microphone has to be located to prevent destructive interference at higher frequencies and therefore the larger the baffle has to be to reduce the edge diffraction effects.

Using 'The Edge' to simulate baffle edge diffraction, mounting the speaker at (50mm,75mm) from center provides a slight reduction in the severity of baffle effects due to destructive interference between different edges of the baffle.

The baffle is propped up in the middle of a lounge room which has high ceilings and provides a few meters of clear space around the baffle. By windowing the measurement, room reflections can be eliminated down to about 3Khz. Wall resonances and reflections only become a major issue under about 500Hz however this is below the main frequencies of interest for the small midrange drivers that I have tested. Sources of noise within the house (e.g. Fridge compressor) are avoided where possible while making measurements.

From DAC to ADC:

A Creative Labs E-MU 0204 USB is used as both the digital-to-analog and analog-to-digital converter. This is an affordable USB sound interface which has very low distortion and low noise. The only downside to using this unit is that it doesn't provide phantom power on it's XLR input which is needed to power the microphone.

The output of the E-MU 0204 is run straight into a Dick Smith Electronics A2760 Speaker Amplifier. I'm using this amplifier for no other reason than that I had one lying around! The amp has unacceptably high distortion from the factory however with a few minor modifications it can be turned into quite a low distortion amp, at least at the low power outputs needed for testing. The speaker amp drives the speaker under test directly. An 8ohm resistive dummy load and a multimeter are used to calibrate the output voltage of the amplifier.

The microphone used is a Dayton Audio EMM-6. After the baffle, this is probably the second biggest weak point in the setup. Noise floor is not the best and it suffers from second order harmonic distortion when subjected to over 100dB SPL. I'm lead to believe this is due to the internal FET in the mic capsule being overdriven. Otherwise it is a reasonable mic for the price and is even provided with an individualised frequency response calibration.

Since the EMM-6 needs phantom power, I built a 48V phantom power supply based on Rod Elliott's design. The powersupply is complete overkill for this application (it could easily power all the mics for a rock concert) but it is a simple linear supply that can be built from jellybean parts and the output ripple is very low when driving a single mic.

Software

I use HOLMImpulse to generate and record sweeps. HOLMImpulse has all the basic features needed to plot frequency response and harmonic distortion however it has some limitations. While it is very flexible in windowing/gating frequency response, the windowing is fixed for harmonic distortion plots and resolution is limited. Mixing near and far field measurements is also a clunky process. It also lacks the ability to plot more than one harmonic at once, so multiple plots have to be overlaid with Photoshop or similar.

HOLMImpulse has the option of saving the untouched recording to a WAV file, so I wrote a MATLAB script to process the WAV file with better suited windowing and produce the plots needed.

The script works by applying frequency-dependant windowing around a number of linearly distributed points from the start of the sweep to the end. The window width for a given frequency was found by experimentation; A window too narrow is susceptible to transient background noise and a window too wide results in reflections influencing the measurement.

Since the sweep is logarithmic and the length, starting and ending points are known from HOLMImpulse, the fundamental frequency is known for each point. An FFT of each windowed clip is taken and the magnitude of the fundamental and harmonics can be extracted. A microphone frequency response calibration is applied and the data is plotted.

MATLAB Scripts inc. example recordings (.zip 4.44MB)

Synthetic Test, Harmonic Distortion 2nd to 5th Order Harmonics (-40, -60, -80, -100dBFS):

Real World Test:

Measurement Results: