April 21, 2020
Charts, Graphs, and Tools
What are those weird charts? How do I read them? What do they mean? Is this really necessary? People have been asking, so here are the answers.
I take audio testing very seriously, and fortunately, PC-based tools can make the testing economical. I've had to build some specialized equipment, however, to analyze the sound coming from the earbuds and to properly test the audio performance of digital audio players under load.
This chart is a spectrogram, and it shows the audio spectrum over time on the horizontal axis against frequency on the vertical axis. The intensity of the sound is plotted as a color, and the legend is on the right, with red being the loudest and blue being the quietest. Here's a larger view of a different player under test:
The larger chart shows extensive harmonic distortion. The slanted bars above the main bar are harmonics of the fundamental frequency. There are at least ten of them, so that's a lot of distortion. It's also pretty strong. By matching up the predominant color in the harmonic bars to the key at the right, you can see that the first harmonic is 10 to 15dB below the test tone, which means that it's plainly audible. Once the harmonics are into the blue range, they're very hard to hear. They still color the music somewhat, and they also rob power from the fundamental frequency.
The blue vertical bars in the lower right corner are faint noise bursts, not harmoncially related to the test tone, but a symptom of some other problem with the player. they sound like a faint, soft-edged popping noise.
This next chart is the frequency response, plotted in 1/3 octave increments:
This plot can either be stereo or mono. Here it's just one channel. The vast majority of players have absolutely identical left/right performance. You can see the roll-off in the low frequencies, which in this case starts at 100Hz. That's in the midbass region, and there's a 10dB drop down to 20Hz. There isn't much music below 40Hz (the low E on an electric bass or upright bass), but organ pedals, 5-string basses, and some bass drums have fundamentals down there.
You often see this kind of plot in audiophile magazines centered, with plus and minus dB regions, greatly magnified. I don't think those little half-dB increments are very important to digital audio player performance.
This plot is made by recording the peaks of bars as the test tone sweeps through the audible range. It cannot show noise or harmonic distortion.
In another spectral view, the peaks are recorded continually, not in 1/3 octave increments. I sometimes do a screen capture while running the sweep tone, as in this case:
The main wave moves across the screen, creating the peak trace at the top. But it's also pushing waves of higher harmonics in front of it. In this particular case, the left and right channels have diverged. (They were convergent at lower and higher frequencies.) The multicolor spectrogram generally does a better job of recording these anomalies, except that it can't display the left/right divergence.
Here's a graph you won't likely see anywhere else. I created this test to explore the low-frequency performance of a number of players, prompted by the essentially perfect bass performance of the iPod shuffle. It's a 40Hz square wave, instead of the usual sine wave, which is much harder for the player to reproduce accurately. Music is full of harmonic content, and a square wave contains an infinite number of harmonics. Adding harmonics that aren't in the source is a form of distortion, and removing harmonics that should be there is another.
In theory, looking at low frequency square wave performance is not much different than the "droop" at the low end of the spectrum that you see in the octave and spectral views. In practice, it tells me about the stability of the player's power supply, its bass performance, and the characteristics of the output section. The player above has no earphone load, but the square wave is still slightly flawed.
This is the same player, with the provided earbuds plugged in. The load of the earbuds prevents the player from sustaining the square wave over time. If the wave sinks all the way down to zero, the bass performance will likely be noticeably lifeless, either lacking in lows or unable to "punch" the bass frequencies convincingly.
May 17, 2010
Bass Performance in Digital Audio Players
This article is a companion to my column, "Shuffle's Got a Secret" in PC Magazine.
It's an attempt to quantify the performance differences among several digital audio players. While all have essentially flat (or flat-enough) frequency response with sine wave sweep tones and less than 0.1% THD at 1KHz, there are audible differences when driving professional headphones or self-powered studio monitors. All can drive a standard set of earbuds (in this case, the standard Apple earbuds) to 100+dB in-canal loudness using a rock test track that fills virtually the entire audible spectrum with a constant din.
Most of the difference is in the bass response, but little differences are discernible from one unit to the next with a 40Hz sine wave (roughly low E on the electric bass). After performing many different tests, I came up with the following test tracks, which I loaded onto each of the players:
This is really a test of the output stage and possibly the power supply, but it's been noted many times that an amplifier is nothing more than a modulated power supply.
All of the digital audio players do a decent job of reproducing the square wave without a load. With the earbuds plugged in, they generally have difficulty sustaining the voltage, and it collapses back towards zero. Some sustain better than others. One in particular sustains remarkably well and subjectively has noticeably better bass response.
The five players I tested, in order the quality of their perceived bass performance in listening tests were:
All but one of these players use single-ended, capacitively coupled output stages. It's an inexpensive and effective way to deliver acceptable, if not superb performance, but the size of the coupling capacitor and the impedance of the headphones have a significant effect on the player's ability to sustain a complex bass tone.
The screen shots below are just a small portion of the testing I did, but should give you an idea of the differences among players. The spectrum analyzer and oscilloscope are part of the Acoustic Analyzing System 5E, from www.ymec.com.
I used an M-Audio Transit USB, an external USB sound card, as the audio interface. It has a high-impedance input that doesn't load the circuit under test, and it has no external adjustments for gain--no knob settings to screw up; identical results from test run to test run. It also has better than 100dB signal-to-noise ratio and can sample at 96KHz, sufficient headroom for tests like these.
This is the Dell DJ20, identical in performance to the current DJ30. Apart from some overshoot, it does a pretty good job of generating an unloaded square wave. It has good subjective bass performance. But unloaded square wave performance doesn't really tell you all that much--unless the player can't form a decent square wave, as one below couldn't.
This is the Dell DJ20 with the headphones plugged in. You can see how the voltage can't be sustained with the load, but players with curves like this still manage to sound pretty good. If the voltage doesn't sink all the way to zero, the player sounds noticeably better in bass response.
This is the Dell DJ20 with pink noise. The top line is the response without the headphone load; the bottom line is the response with the headphones plugged in. The load of the headphones has little effect on the upper frequency response, but the drop is noticeable at lower frequencies.
Above, the 15GB iPod does a near-textbook job of generating the square wave...
... and falls down badly when it comes to sustaining it under load. In my testing, players that were dragged down to zero by the load of the earbuds didn't sound as full as the ones that had a little left in in the tank at the end of the cycle.
My son took the 15GB iPod back to college with him before I could run the pink noise tests. I did, however, record a 1/3 octave sweep test. It doesn't show no/load load. While the iPod is generally well-regarded, bass performance is not stellar. This sweep was run with the headphone load, and you can see the fall-off in the lower octave. The rest of the frequency response curve is flat, however.
Above is the original iPod Mini (not the new 6GB Mini). It has trouble making a clean square wave or even sustaining it when the player is unloaded.
Once I plugged in the earbuds, the waveform of the iPod Mini really deteriorated.
Audiophiles and high-end headphone manufacturers agree that Apple undersized the output capacitors on the Mini, whether for space, cost, or power reasons.
With the pink noise test, this Player 3 shows a larger variance between the load and no-load lines than any other player in the test. The divergence, as expected, is greatest at low frequencies.
The Zen Micro does a pretty good job of generating the unloaded square wave, above.
The Zen Micro's signal sinks when it hits the earphone load, but it doesn't sink to zero. Subjective bass response on this player is pretty good, better than most.
The Zen Micro shows little divergence under load from its no-load performance. This player did very well in listening tests and is well regarded.
Above, the iPod Shuffle. The square wave is just about perfect--the overshoot might be heard as a small amount of harmonic distortion, but is immaterial.
With earbuds, the iPod Shuffle's signal looks darn near identical to the no-load signal. I checked and rechecked this result because I couldn't believe my eyes. The iPod Shuffle sounds great, with a solid low end, and no need for bass boost.
The reason for this sterling performance is that the left and right channels each have two transistors, one pushing, one pulling, and no capacitor that gets discharged over time.
The pink noise performance of the iPod Shuffle is, as you might expect, exemplary. The load/no load performance is very close, even at the deepest bass frequencies.
You might be wondering why these pink noise charts seem to sink down as frequency goes up. The reason is that you're looking at a logarithmic chart, and the intensity of pink nose tapers off as frequency goes up. This is to avoid the problem of white noise, which has the same amount of power at every frequency. That sounds like a good thing, but in fact would mean that fully half of the sonic power would be above 10KHz, and sound just isn't like that in the real world. Pink noise is more representative.
Since I wrote the column, Apple introduced the iPod Mini 6GB. I subjected it to some of the same tests as the other players, and as you can see from the screen capture above, Apple has obviously improved the audio amplifier. The 6GB Mini sounds noticeably better than the original Mini.
The curve gets pulled down to zero with standard Apple 32 ohm earbuds. We could have hoped for more capacity, but it's still better than the original iPod Mini.
I didn't get a chance to run the load/no load sweep test on the 6GB Mini, but I did run a 1/3 octave sweep with the earbud load. You can seen from the graph that the lowest frequencies are still somewhat deficient, but there's definitely more bass from 30Hz on up.
You can see how much the earphone load affects audio players other than the iPod shuffle. I still haven't reverse-engineered the shuffle design to tell you what Apple did right, but the differences are obvious, both in casual listening and under the testing I've done here.