Audio Measurements: The Useful vs. the Bogus in Consumer Audio

Originally Published: April 07, 2011

It's every audiophile's dream to own a loudspeaker that measures ruler flat from 20Hz to 20kHz.  It's not uncommon for the audio magazines to  regurgitate the manufacturers claims either in text or with measurements to emphasize product positives. It's my belief that most of the manufacturers and audio magazines aren't flat out lying or being deceptive as much as they aren't giving you the whole story likely because of lack of proper measurement facilities or techniques at their disposal or a combination of both. 

It's easier to furnish pretty graphs because quite frankly to paraphrase Jack Nicholson "you (consumers) can't handle the truth". In most cases these measurements do have a grain of truth if you place a lot of conditions and caveats on them that are often not disclosed.  In this article and recently related YouTube Video discussion we explore various measurement and graphing techniques. 

Can you handle the truth?  If so, read on...but before you do, we recommend checking out our recently added YouTube video discussion on this topic at hand for further insights

Measurement Manipulation in the Audio Industry YouTube Video Discussion

Frequency Response

Let's start with frequency response.  The intention is that this should tell us something important about how a loudspeaker might sound in a room.  If so, we will need a whole collection of curves, describing the direct (first) sound to arrive at our ears, and then others to describe sounds that are reflected by walls, floor and ceiling, reaching our ears a bit later. While this is demonstrably true, long ago the industry decided that a single curve, the on-axis curve describing the direct sound, was all they would show us. One can understand this because the on-axis curve is typically the best looking of all the curves one can measure, and one always wants to look good, right?

The standard SPL amplitude vs Frequency response graph we are all so used to seeing when looking at measurements of loudspeaker and subwoofers is supposed to show us the linearity of the speaker under its bandwidth of operation. 

Below are measurements of two loudspeakers.  Take a look at the graphs and ask yourself which speaker you would prefer assuming the goal is the most accurate and "linear" frequency response from 20Hz to 20kHz.

Scaling

 
 

SPL vs Frequency Response of SPKA (left pic); SPKB (right pic)

After looking at the graphs, surely most consumers would prefer SPKB because the measurement is much flatter and smoother .  Truth be told you are looking at the exact same speaker and the exact same graph.  The difference is actually in how I chose to display the data collected, but be assured, the data is exactly the same in both curves.  The measurement was taken in-room 8 feet away from my right front reference speaker (RBH Sound T-30LSE).  The measurement was not time gated and thus taken steady state so the room influenced the measurement. The only difference between these two above graphs is the dB scaling, which is the vertical axis.  The SPK A graph has a 24dB scale in the vertical axis while SPK B has a 120dB scale (known to engineers in the business as the marketing curve). The larger the scale (dynamic range) in the vertical axis, the less detail you see about the measurement.  I can't tell you how many so called "professional" reviews I see on other A/V websites and print publications that use 120dB or greater scales to give the illusion of a smooth linear response for the speaker they are reviewing.  We personally like to use 60dB scales for all of our loudspeaker measurements and often use smaller vertical scales if we want to zoom in on a problem area.  (Most engineers use either a 40 or 50dB log scale for displaying data.  This goes back to (possibly precedes) the available parts used in the old analog style graphic level recorders made by the sound measurement company, Bruel & Kjaer).The two most popular potentiometers they sold for audio measurements were 40 and 50 dB. You could even slow the measurement speed down on the old B&K plotters, effectively smoothing  the printed plot to yield a nicer looking graph.

Resolution & Smoothing

SPL vs Frequency Response of No Smoothing (left pic); 1/3 Octave Smoothing (right pic)

Here we have the same speaker, same measurement again with the scale set to 60dB.  The left pic is  not smoothed while the right picture uses 1/3 octave smoothing.  In order to get an accurate measurement, it's important to have a large enough collection of data points in the measurement, else the finer details will be missed and the overall measurement may not reveal the problem areas of the speaker under test.

Why not use 1/3rd Octave Resolution?

It’s a common misconception that 1/3rd octave spectral analysis is a good enough approximation of the perceptual critical bands over the mid and high frequency range.  Some have argued that our hearing discernability is limited to this resolution.  In reality, these bands define the bandwidth over which spectral information is summed for estimates of loudness and in the simultaneous masking of tonal signals by broadband noise (see: Toole, CH19, pg450).  Within these bands the beats from multiple tones influence the perceptual quality of "roughness".  Differences in roughness contribute to the distinctiveness of sounds and timbre.  1/3rd octave smoothed measurements mask associated performance issues in speakers which is why according to Dr. Toole higher resolution is needed (1/20th octave in the frequency domain) to really get a better picture of the loudspeakers true response.  Since the majority of our loudspeaker measurements are done in-room, we typically use 1/12th octave smoothing which isn't perfect but does give you a much better idea of speaker performance than the typical 1/3rd octave resolution graphs found in most print and online A/V magazines. Loudspeaker designers typically use higher resolution measurements in their design process for troubleshooting purposes.   

Smoothing is typically not needed for nearfield measurements, or those done in anechoic chambers since the room is taken out of the picture and  loudspeaker performance can be assessed without the normally significant influence the room plays on the sound which reaches the listener.  Smoothing is a useful technique  for in-room measurements which if not abused can still produce a relatively accurate performance graph of the speaker while removing many of the room reflections and their substantial influence on what we hear in the real world. This is known as smoothing by spectral averaging, which can be interpreted as we know exactly where we measured it but we are not sure what we measured.  The alternative technique is to use spatial averaging which is very useful when measuring in the near field (1-2 meters typical) of a complex source, aka a loudspeaker system, where we know that there will be acoustical interference errors. So, what we do is make a few high resolution measurements at slightly different angles and average them. This can be interpreted as we have a very good idea of what we measured, but we are uncertain as to exactly where in space it applies. In loudspeaker measurements the latter is actually more useful and something we plan on doing more of for our in-room loudspeaker measurements.  We (reviewers/manufacturers)  have long relied more on the “listening window” data for direct sound estimates than on the solitary “on axis” response.

If it looks to good to be true, it probably is!

Some A/V publications go so far as to draw a ruler flat line graph in Excel to represent frequency response while some manufacturers publish simulated frequency responses that they modeled for their product.  If a graph looks too smooth or flat it's more than likely not a very useful or accurate representation of actual product performance but instead a great sales tool for the marketing department.

Velodyne SMS-1 Frequency Response Graph

Be careful with measurement devices such as the Velodyne SMS-1 whose output is limited to 1/3 octave resolution.  This resolution is too crude for performing any type of high Q room correction as the measurement system will likely miss those room modes (in-room peak) and nodes (in-room dip) in the frequency response sweeps.  Such low resolution measurement systems do a poor job of seeing the higher Q problems which typically plague rooms, and need to be removed with a scalpel, not a saw.  It's important to have enough resolution to accurately dial in the response.  It is recommended to instead use an external measurement device with at least 1/12th octave resolution for fine tuning of bass response or limit your corrections to very low Q adjustments. The frequency response scale of your RTA should be set to plot between 10Hz to 200Hz with a maximum of 60dB vertical dynamic range to get a clear picture of what is going on with your systems bass response.

Editorial Note about Calculating Q
For more advanced users utilizing subwoofers with built in equalization and or adjustable subsonic filters and crossover slopes, you may consider manipulating one sub at a time to better tune in your response but there is a lot of guess work by doing so don't waste a lot of time tweaking if you aren't seeing measurably better results.

For example, if you have a measurable bass peak centered at 40Hz that is 10Hz wide, you can setup an EQ band of 40Hz with a Q of 4 to reduce this problem. 

Where Q = fo / BW ; fo = center frequency  and BW = bandwidth

To determine BW, mark -3dB high (f2) and -3dB low point (f1) of the problematic mode and subtract F1 from F2.  Once you have that number, use that for the denominator and the center frequency (fo) of the peak or dip for the numerator and you will find the Q.  (Note:  This mathematical method works better for high Q and high amplitude resonances than it does for low Q low amplitude peaks and dips.)

Editorial Note about Resolution by. Dr. Floyd Toole
1/3-octave resolution is a crude approximation of "critical bandwidths" which represent the bandwidths over which loudness summation occurs.  However, within a critical band multiple tones or overtones interact with each other to produce beats and something called "roughness", both of which are important contributors to timbre/sound quality. “This is why we must measure loudspeakers with more detail than is revealed by 1/3-octave resolution, ideally about 1/20-octave.”  The engineering reality is that bass room modes behave as minimum phase systems and it is necessary to be able to match the shape of a parametric EQ to a peak so that the resonance is damped, not just turned down. This requires higher resolution data than simple 1/3rd octave.