Quasi Anechoic Speaker Testing with dScope Series III
RELEVANT PRODUCTS |
Webinars |
Principles of Operation
The basic principle is very similar to MLS analysis, where an impulse is derived by cross-correlation with a known stimulus. The stimulus is passed through the speaker under test, via a microphone/pre-amplifier and into the Analyzer. The first stage of analysis involves correlating the signal recovered with a reference. This produces an 'impulse response' of the EUT - i.e. its theoretical output if stimulated by a perfect impulse. In dScope, the correlation is performed using an FFT-based process. It is generally more practical to derive the EUT's impulse response by this method than by stimulating it with an actual impulse, which contains very little energy and so would produce a result with poor signal-to-noise ratio. Using a stimulus with its energy spread out over time and deriving the impulse response mathematically gives much better noise immunity and reduces the risk of damaging drive units. Using a pink weighted (equal energy per octave) stimulus is ideal for getting the best signal to noise ratio from real world audio transducers as it has similar weighting to music (and typical background noise). Somewhat paradoxically (and very conveniently), the frequency spectrum derived from the impulse response is effectively white (equal energy per Hz) which gives a flat measured frequency response when using FFT analysis. For more detail about this, see the impulse response application note which covers this in more detail.
Getting a frequency response
In assessing acoustic spaces, the impulse response can be interesting in itself, whereas for measuring transducers it is mainly a means of deriving the device's frequency response independent of the environment in which it is measured. The frequency response of the EUT can be calculated as the FFT of its impulse response; a 'perfect' impulse containing all frequencies in equal measure.
A common requirement in transducer measurement is to be able to measure devices in anechoic conditions, and free from background noise (which might compromise the reliability of the results). To do this in real anechoic chambers is expensive and inconvenient in a production environment, particularly for low-cost devices. Impulse response windowing can help to solve this problem. This involves applying a "window" to the impulse response prior to performing the final FFT. Basically, the windowing operation allows only the desired period of the impulse response to be included, i.e. later parts which contain reflections from nearby surfaces can be excluded, as well as the intervals before and after the impulse response which may contain background noise. By carefully time-positioning a suitable Window function over the impulse response prior to calculating the frequency response, it is possible to simulate anechoic measurement conditions very successfully.
Window functions
dScope allows any of the normal FFT Window functions to be selected for impulse windowing, in either a full or 'half-window' configuration. A half-window has only the 'right-hand' side of the specified function; i.e. it has unity gain at its start, and reducing gain thereafter. The selected window or half-window can be precisely positioned at the right time, and scaled to the desired duration, to provide optimum selectivity. The resulting frequency response is essentially the frequency response of the device in the absence of reflections, and with a significant portion of the noise windowed out.
Setting up dScope for loudspeaker testing
dScope's implementation of quasi anechoic testing is a little different from some of the traditional systems such as MLSSA, but is immensely powerful when combined with the ability to set up user defined FFT Detectors and scripted analysis. dScope's impulse response testing is essentially a different mode of viewing time domain data. Normally dScope displays the time domain audio sample data in a Scope Trace exactly as it is received by the Digital Inputs or sampled by the Analogue Inputs. In impulse response mode, the displayed time domain data is the impulse response derived as described previously. Whichever mode the dScope is in, the FFT data displayed is based on the time domain data that is showing at the time. In the steps that follow, we will describe how to set up an impulse response measurement with the dScope connected back to back, and then with a loudspeaker.
- Configure the hardware: The first step is to load the default Configuration and then connect the Analogue Outputs of the dScope to its Analogue Inputs, either using cables or by selecting 'Generator' as the 'Source' setting in the Analogue Inputs dialogue box.
- Set up the Generator for Swept sine: In the Signal Generator dialogue box, set the 'Function' to 'Swept sine', with the default parameters (20Hz to 20kHz, log, 4k points with 250 sample space, 100 sample ramp up and down, playing continuously).
- Set up the FFT parameters: In the FFT Parameters dialogue box, make sure that the 'Number of points' is set to 4k, set the 'Window function' to 'None (Rectangular)' and 'Trigger', 'Mode' to 'Gen wavetable'.
- Check the Trace Window: Look at the Trace window; make sure that the Scope and FFT Traces for channel A are displayed, and that the Scope Trace is fully zoomed out on the X axis - if not, repeatedly click the Trace Toolbar icon with the Scope Trace selected. The Trace window should now look something like this:
The Scope trace shows the Swept sine waveform, triggering reliably at the same point; the FFT Trace shows the FFT of the stimulus, which steadily falls in-band owing to its logarithmic progression, and then rolls off out-of-band.
- Set the Impulse response parameters: Now open the Impulse Response Parameters dialogue box from the Analyzer menu and check 'Create impulse response in sample buffer'. The Scope Trace should be replaced by an impulse at the left-hand side of the Trace window. To make this a bit clearer, go back into the FFT Parameters dialogue box, click in the 'Trigger at' box, enter '100' and hit [Enter]. The impulse should move to the right a little (to sample 100, in fact).
The FFT won't yet be showing the frequency response of our wire (hopefully flat) because the impulse response is not yet being windowed properly. Click the icon on the Trace Toolbar and an Impulse response Window function Trace should appear, along with the Edit Impulse Window dialogue box. You can drag the Window Trace with the round "handles" on its left, and adjust its duration with the handle on its right. When you position it to include the impulse on the Scope Trace, the FFT Trace should begin to show a flat frequency response, like this:
In the screen shot above, the furthest left portion of the FFT trace is shown as a dotted line. This is because the windowed portion of the impulse response is too short to accurately derive low frequency information so the data in this region is doubtful. Note also that at this stage in the process, the FFT can be running continuously and dragging the window updates the FFT trace in real time, allowing you to see instantly what effect the type and placement of the Window function is having on the results.
- Measurements with reflections: If we now introduce a digital reverb into the signal chain, the individual "reflections" become apparent as in the screen shot below. (This step is mainly for illustration - you can skip to the next one if it doesn't interest you)
Here there is a positive reflection at about 15ms after the main impulse (corresponding to a reflection from a hard surface approximately 2.5m away) and further reflections later. By applying the window such that the first reflection is excluded from the FFT calculations, an anechoic measurement can be made (anechoic meaning "without echoes"). This method of removing reflections is typically more effective than an anechoic chamber. In this case it is possible to see that the frequency response of the direct signal of the reverb is flat up to 20kHz and then drops suddenly. This is because the reverb is sampling at 44.1kHz.
- Measurements with a speaker: If we now set up the dScope with a speaker and microphone as shown in the diagram below:
We can proceed to measure a real world loudspeaker. The illustration shows the direct acoustic path between the speaker and the first reflected path which will typically be from the floor. The reflected path needs to be as long as possible for best low frequency measurements. In order to make the best use of the space available, this will usually mean placing the speaker half way between the floor and ceiling, and at least as far from the nearest wall or large object as from the floor. In the screen shot below, a small computer loudspeaker is being measured with the microphone very close to the drive unit. The impulse response trace (green) is at a high magnification so that the first reflection can be clearly seen. The frequency response of the drive unit is shown in red. This measurement took less than a second to acquire.
Notes
Time Aliasing: When measuring with a repeated stimulus, if the reverberation time of the room in which you are measuring is of comparable length to the length of the stimulus, the sound from the previous stimulus will not have died away sufficiently before the current one starts. This is referred to as time aliasing and manifests itself as pre-echoes (reflections that appear before the main impulse) and noise in the impulse response. To avoid this, use a stimulus with sufficient length that the sound from signal has died below the noise floor before the next stimulus begins. Ideally the "time of flight" section before the direct impulse should be completely silent.
Low frequency measurements: measuring low frequencies is notoriously difficult. Anechoic chambers are not usually very effective at low frequencies because of the wavelengths involved, and quasi anechoic measurement techniques also have their problems. In practice it comes down to the proximity of the nearest surface as this determines how long we have before the first reflection comes back. The lowest frequency we can measure is where the wavelength is equal to the path length difference (the difference in distance between the direct path from the speaker to the microphone and the speaker to the microphone via the nearest surface). From the impulse response trace, this is equal to the inverse of the delay between the direct impulse and first reflection - for example, given a delay of 15ms (0.015s), the lowest frequency that can be measured is 1/0.015 = 66Hz. It all then comes down to increasing the time before the first reflection. One technique that has been used is to measure outdoors and raise the speaker and microphone off the ground on cranes - not always very practical. Another technique is to simply place the speaker on the ground in a large open space (eg, car park) and measure in "half space". This works well with subwoofers where the wavelengths are long, but gives interference effects at higher frequencies. For absolute level measurements, it should be stated that this is how the measurement was done as it will give a 6dB increase in level.
Time invariance: The technique used for deriving the impulse requires that the two signals being compared be synchronous and time invariant. Synchronous means that the two signals (either the generator and analyzer, or two analyzer channels when doing inter-channel measurements) must be derived from the same clock signal so they run exactly in time. Time invariant means that the system must not change over the time the measurement is being taken. Failure of either of these conditions causes distortion of the impulse response. The main practical implication of this is that the measurement set-up must remain completely static while the measurement is taken - this includes moving objects that reflect sound near the measurement set-up. Also included is the air in between the speaker and microphone - measurements of speaker systems outdoors can be severely compromised by wind, particularly when the distances involved are large. Care should also be taken when using the dScope cross-domain (ie, generating digital and analyzing analogue or vice versa) - the two domains are only synchronous when they have the same sample rate.
Microphone and pre-amplifier considerations: The dScope can accept microphone calibration data (sensitivity and frequency response) in order to allow calibrated measurements and to compensate for microphone frequency response variations. The dScope analyzer inputs do not have phantom power, so any measurement microphone must either be self-powered or must be used with a pre-amplifier/phantom power source. Although the dScope outputs are capable of driving headphones directly, they are not capable of driving loudspeakers, and in both cases, the use of an appropriate amplifier is recommended. The gain of this amplifier and of the microphone pre-amplifier can be entered into the dScope so that the measurements can be made effectively after the amplifier on the generator side and before the pre-amplifier on the analyzer side. See the dScope Operation Manual for more information.
Further Information
This application note has only touched the surface of the possibilities that are opened up by dScope's impulse response and quasi anechoic testing features. There are many more options available than there is scope to discuss here. For further details, consult the dScope online help or download the dScope operation manual. For any more information or assistance please contact us using the form below:
App ID: 0014, Resource ID: 0